12175 Case Studies onV Women's Employment and Pay in Latin America EDITED BY GEORGE PSACHAROPOULOS AND ZAFIRIS FILE COPY TZANNATOS Report No.:12175 Type: (PUB) Title: CASE STUDIES ON WOMEN'S EMPLOY Author: PSACHAROPOULOS, GEO Ext.: 0 Room: Dept.: BOOKSTORE NOVEMBER 1992 Case Smtdi On Women's Employment ad Pay o La 0m Case Sudiks on WVom©n' Employment and Pay i Latin Amerka GEORGE PSACHMOULOS ZFARRS T7AMMATOS The World Bank Wi*iq,pV% D.C. © 1992 Th= International Bank for Reconstrucrion and Dcvlopment / The World Bank 1818 H Stret, N.W., Washington, D.C. 20433 All rights reecrved Manu&ztured ia the United States of America First printing Novcmber 1992 This holume is a companion to the World Bank Regional and Scctoral Study Woman's Empkymrn and Pay in Lain Aoerica ODvierw and Methodology. The World Bank ReSional and Secoral Studies series provides an outlet fr work that is relatively limited in its subject matter or geographical covcrage but that contributes to the intdlectual foundations of devdopment operations and policy formulation. These studies have not necessariy been edited wtith tLc same rigor as Bank publications that carny the imprint of a university pre& The findinp, intcrprctations, and condusions expressed in this publication are those ofthc authors and should not be attributed ih. any maannr to the World Bank, to its affiliated o.anizations, or to the members of its Board of Ezcuie Directors or the countries rb-y reprcsnt. Thc material in this publication is copyrighrt L Requstsu for pemission to reproduce pordons of it should be sent to the Officc of the Publisher at the address shown in the copyright notice abc.-- The World Bank encourages diseminadon of its work and vsill normally givc permission promptty and, whcn the reproducon is for noncommercial purposcs, without asking a fee. Pcrmission to copy portions for clasroom use is granted through the Copyright Clearance Center, 27 Congress Stet, Salcm, Massachusetts 01970, USA. The cumpicte backlist of publicadons from the Wordd Bank is shown in the annual iduex jePMicauort, whtich contins an alphabcdcal tide list and indcxcs of sbjects, authors, and countries and rcgions. The lat edition is availabk free of charge from Distribution Unit, Office of the Pubishcr, Thc World Bank, 1818 H Str=, N.W, Washington, D.C. 20433, U.SA., or from Publiations, The World Bank, 66, avenue d"IUna, 75116 Paris, Frnce. Gcorge Psacharopolos is the senior hunun resources adviser to the Worid Bank's Latin America and Caribbean Technical Dcpartment He prviously taught at the London School ofEconomic. Zafiris Tiannatos is a hbor coaomist with the Population and Human Resources Dcpartmzca at the World Bank Hc is an honorry research fcllow at the Un;,crsitds of Nottngham and St. AndrewB in the United inmios. Cover d=sz by Sam FrmO library if Conj CataliV4-Psbl"tcai Dam Case snxr= on women's employment and pay in tin America / edited by Gcorge Psachropoulos and Zfiris Tzannatos. p. crn Indude bibliogrphical rferences. ISBN 0-8213-2308-3 1. Wome-Employment-Latin Amcrica-Cas stuldies 2. Wageo- Latin America-Case studies 3. Disaimination in employmcnt- lAtin Amezica-Case tudies. 4. Sex discrimination against women- Latin AmrnicaCse studes I. Psacharopoulos, George. IL Tz oros, Zafris, 1953. HD6100O5.C37 1992 331.4'098-dc2O 92-40880 CIP Contents Acknowledgments vii Foreword ix I Female Labor Force Participation and Gender Earnings Differentials in Argentina I by Y. C. Ng 2 Women in the Labor Force In Bolivia: Participation and Earnings 21 by K. Scott 3 Labor Force Behavior and Earnings of Brazilian Women and Men, 1980 39 by M. Stelcner, J. B. Smith, J. A. Breslaw and G. Monette 4 Female Labor Force Participation and Wage Detemination in Brazil, 1989 89 by J. liefenthaler 5 Is There Sex Discrimination in Caile? Evidence from the CASEN Survey 119 by 1. Gill 6 Labor Markets, the Wage Gap and Gender Discrimination: The Case of Colombia 149 by J. Tenjo 7 Female Labor Market Participation and Wages in Colombia 169 by T. Magnac 8 Women's Labor Force Participation and Earmings in Colombia 197 by E. Velez and C. Winter vW Wm" 'z EpIymeNt ad Pay in Lain Amerwa 9 Female Labor Force Participation and Earnings Differentials in Costa Rica 209 by H. Yang 10 Why Women Earm Less nhan NM I 'osta Rica by T. H. Gindling 223 11 Ihe Effect of Education on F=iale Labor Force Participation and Earnings in Ecuador 255 by G. Jakubson and G. Psacharopoulos 12 Female Labor Forcef Participation and Earnings in Guatemala 273 by M. Arends 13 Women's Labor Force Participation and Earnings in Honduras 299 by C. Winter and T. H. Gidling 14 Female Labor Force Participation and Earnings: The Case of Jamaica 323 by K. Scott 15 Women's Participation Decisions and Earnings in Mexico 339 by D. Steele 16 Female Labor Force Participation and Wages: A Case Study of Panama 349 by M. Arends 17 Women's Labor Market Participation and MaleFem-le Wage Differences in Peru 373 by S. Khandker 18 Is There Sex D scrminatin in Peru? Evidence from the 1990 Lima Living Standards Survey 397 by I. Gill 19 Women's Labor Force Participation and Earnings: The Case of Uruguay 431 by M. Arends 20 Female Participation and Earnings, Venezuela 1987 451 by D. Cox and G. Psacharopoulos 21 Female Earnings, Labor Force Participation and Discrimination in Venezuela, 1989 463 by C. Winter Append A: Contents of Companion Volume 477 Appendix B: Authors of Country Case Studies 479 Acknowledgments We have benefited rom comments and encouragement from many people who read earlier versions of this study and participated in seminars given at the World Bank, the University of SL Andre", and conferences organLed by the Comparative and International Education Society, the International Union for the Scientific Study of Population, and the Eumpean Society fbr Population Economics. In particular we would like to thank Ana-Mara Arriagada, Alessandro Cigno, Deborah DeGraff, Barbara Herz, David Huggart, Emmanuel Jimenez, Philip Musgrove, and Michelle Riboud for their helpful comments, Professor William Greene for providing purpose built routines for LIMEP which facilitated the estimation procedures used in the country studies; Diane Steele and Carolyn Whtea for their reviews of the book; Hongyu Yang for preparing the graphics; and Donna Hannah for typing and preparing the earlier versions of this book and Marta Ospina for tking these tasks over and eventually putting the book into its present form. The completion of the present study would have not been possible without the genes support of the Norwegian Trust Fund. Forewolr Women's role in economic development can be examined from many different porspectives, including the feminist, anthropological, sociological, economic and legislative. This study employs an economic perspective and focuses on how women behave and are treated in the workt force in a number of Latin American economies. It specifically considers the determinants of women's labor force paricipation and male-to-femnale earnings differentials. Understanding the reasons for 'low' labor market participation rates among women, or 'high' wage discriminafion against women, can lead to policies that will improve the efficiency and equity with wbi-ch human resources are utilized in a particular country. The study is in two volumes. The companion volume presents aggregate data on the evolution of .female labor force participation in Latin America over time, showing that in some countries twice as many women (of comparable age groups) work in the muket relative to twenty years ago. This volume uses household survey data to analyze labor force participation rate and wages earned by men and women in similar positions, paying special attention to the role of education as a factor influencing women's decision to wor. The results show that, overall, the more years of schooling a woman has, the more likely she is to participate in the labor force. In addition, more educated women earn significantly more than less educated women. The book also attempts analyses of the common factors which determine salaries paid to men and women in an effort to identify what part of the male/female earnings differential can be attributed to different human capital endowments between the sexes, and what part is due to unexplained factors such as discrimination. Differences in human capital endowments explain only a small proportion of the wage differental in most of the country studies. The remaining proportion thus represents the upper bound to discrimination. !t is our hope that this work will be followed up by a more careful look at labor legislation and the role it plays in preverting women from reaching their full productive potential. S. Shahid Husain Vice President Latin America and the Caribbean Region Tha World Bank Female Labor Force PaSridspation and Gender Earnings Differenflals 'm Argentina rmg Chu Ng 1. Introduction In this study we estimate earnings functions for Argentinian males and females using the 1985 Buenos Aires Household Survey data. Our purpose is two-fold. FRst, we seek to investigate the income differentials between male and female workers. Second, we examine the existence of earnings discrimination by gender. In the following section we provide a brief overview of the Argentine economy and the operation of the labor market. In the third section we discuss the data base used in the analysis and present the main characteristics of male and female labor force participants. Female labor force participation and the factors influencing women's decision to participate are discussed in Section 4. In Section 5 we present the results from earnings functions estimates for male and female workers, and Srction 6 provides an analysis and discussion of the extent to which male/female earnings differentials can be attrbuted to differences in hurrun capital endowments and to discriminatory practices by employers in the labor market. The paper concludes with a discussion of these findings in Section 7. 2. The Argentine Ewonomy and Labor Market Since the 1940s political events in Argentina have had a considerable impact on tha functioning of the economy, the structure of the labor market, and the carnings structure of workers. Government policies favoring import-substitution and the introduction of a wage setding mechan.;.n meant tl'x the growth of relative wages from the 1940s to the 1980s was highest in non tradable activnies. This, plus the fact that wage determination was increasirgly influenced by collective bargaining, has led to a concentration of resources, including human resources, in urban areas. More than 30 percent Cim 1987) of the total population was concentrated in the capital, Buenos Aires. In general, labor force participation rates in the urban markets in Argentina are above 40 percent of the resident population (Sanchez, 1987). However, the Argentine labor market is characterized by cyclical periods in which labor is either scarce or relatively abundant. Two factors explain this. First, there are substantial fluctuations in terns of domestic and foreign migai on. Sc.ond, the fact that unemployment and underemployment rates remain relatively low regardless of whether there is an e s or scarcity of labor suggests that there may t'e a stron.g 'added worker' effect operating. Riveros and Sanchez (1990) provide evidence that this is the case. They report that th-. 'added worker' effect resulted in substantial increases in female labor force I 2 Womm'i F pLcymM and Pay in Lai Ameica partici,;'tion rates, particularly among wonen aged 35 to 49 years, during the economic crisis of the early 1980s. In Argentina there is not only a higher labor force participation rate relative to other Latin American countries, but workers also tend to work longer hours. Dieguez (n.d.) reported that am>- 3.9 million employed individuals in May 1985, 37.6 percent worked between 35 and 45 hours weekly, and 31.4 percent worked over 45 hours per week, while only 17.5 percent worked less than 35 hours per week. Female tvoilenw. Among female workers de Lattes (1983) found thst in both 1960 and 1970 there was a higher participation rate among single and divorced women aged between 25 and 59 years than among married women of the same age. Wainerman (1979) and de Lattes (1983) obund similar results: The probability of single female participation in the work force was at least three times that of married f.males. Education is an impot2;nt variable determining female participation rates. Wairierman (1979) found that more educated women were more likely to participate in the labor fiorce. Data collected from the Instituto Nacional de Estadiq :a y Censos (n.d.) showed that in Buenos Aires in 1970 the proportion of working females with less than primary education was substantially lower than women with higher educational attainment Women with secondary or university education made up the largest portion of the female work force (de Lites, 1983). Female migrants constitute an important proportion of the female labor force in urban areas (Marshall, 1977; de Lates, 1983) and especially in Buenos Aires. Female workers are concentrated in non-agricultural acti'.'ties. More than 65 percent of workers in the noniradable sector were females in the 1960s, and this figure increased to 79 percent of workers in 1980. 3. Date Charaictstles The data used in this study are drawn from the 1925 Buenos Aires Housebold Survey which was undertaken by the National Institute of Statistics (INDEC) and surveyed 15,580 individuals. 'Tough the survey covers only Buenos Aires, it represents more dt 30 percent of the total population of the country. In the present stwiy, we extract females (working and non-working) and working males aged 15 to 65, resuldng in a sample of 7,097 individuals. The descriptive statisdcs and the definitions of the main variables are presented in Table 1.1. The female participation rate is 36 percent. The average education level of the sample is nearly 9 years of schooling for both sexes. Working females average over 9 years of schooling but have less work experience than males. The overall sample characterisdcs are very similar for both ma0es and females. A very large portion of the working population is employed in the dependent employment sector - 80 percent of females and 78 percent of mwes. Females work fewer hours than males on average and the number of part-time female workers is about twice that of part- time males. TBe average earnings of females is about 64.5 percent that of males, and other income (defined as the difference between famiy income and the respondent's labor income) is 1.75 times as much for females as it is for males. Il order to have a closer look at the earning differentials, information on earnings among different employment sectors and employment types by educational level is presented in Tables 1.2A and 1.2B. Regardless of the differences in sec=r and employment type, the higher the Famaks Labor Force Pardcat and GCde Fanbg DWrrenida in Argamriw 3 Table 1.1 Buan Airs, Argenfna Means (and Standard Deviations) of Sa=ple Vriableo Variable Woring Mala WordnB Femles All Femeals Age 37.99 35.71 37.95 (12.39) (12.42 (14.68) Years of Schooling 8.80 9.41 8.23 f3.87 (4.17) (3.73) Leas than Primary 0.01 0.02 0.03 (0.11) (0.13) (0.16) Primary 0.51 0.44 0.53 (0.50) (0.50) (0.50) Secondary 0.34 0.35 0.34 (0.47) (0.48) (0.47) University 0.14 0.19 0.10 (0.35) (0.39) (0.30) Experienc 24.19 21.30 (13.46) (14.06) Experin Squaed 766.36 651.14 3711.62) (712.0) Monthly Incoma 984V3.48 63558.97 (737t.i.89) (52861.05) Weekly Hours Worked 46.31 37.43 (13.62) (15.04) Employee 0.78 0.80 (0.42) (0.40) Public 0.17 0.28 (0.38) (0.45) Part-time 0.12 0.35 (0.88) (0.48) Ovetime 0.41 0.22 (0.49) (0.41) Maried 0.74 0.55 0.65 (0.44) (0.50) (0.48) Size of the Family 7.50 7.28 7.35 (4.28) (4.36) (4.23) Number of Childre 0.84 0.70 0.77 (1.13) (1.04) (1.09) HauselId Ownersip 0.79 0.76 0.80 (0.41) (0.43) (0.40) Head of Household 0.73 0.13 0.09 (0.44) (0.34) (0.29) Number of Tacom E uAners 4.43 4.14 4.27 (2.02) (2.07) (1.99) Ottle Income 61348.40 104887.52 106560.65 (79903.70) (102931.38) (10Q214) Foreip Bom 0.1' 0.08 0.11 (0.31) (0.28) (0.32) N 2,397 1,338 4,700 Now Feale Labor Force Particip6on Rat = 36% Sourcc: Bu Air Houschold Sutrvq, 1985 4 Women 's E5kymem and Pe, in Zaba Arnra Tab!e 1.2A Me= Earnings by EduatioA Level and Type of Employment (Peso pe mUDDh) Femals Males Employee Self-employed Employe Self-temlloyed Less than imy 40,506 25,233 S1,409 80,000 Primary 4Sm026 40,268 74,379 82,629 Seconday 72,721 56,969 100,Sl' 113,,68 UniverEi.; 97.229 106,813 174,427 159.484 E=]Qm=Eg Private 59,957 52,231 94,916 102,787 Public 79,539 50,654 105,894 113,962 overall 66,399 52,060 97,100 103,333 N (1,073) (265) (1,865) (532) Table L2B Meim Eaminp by Eductio and Sector of Einployn (Peso per month) Fessle Males Privat Public Priwt PuN Les than 36,109 50,000 62,269 ' 8,000 Primlay 41,679 57,34S 75,461 80,018 Soomday 70,150 71,459 103,282 106,174 University 108,056 92,611 180,918 152,179 OvcoII 58,027 77,558 96,908 106,422 N (9S9) (379) (2.000) (397) Fenwk bL7br Force Parhion and Gender Eariags Diftrewiab in Argenrina 5 education level the more the individual earns. Males earn about 50 percent more than females w:ch the same education, except where employees have less than primary education or where they are self-employed with less than primary education. Within employment type, different pattens are seen for males and females. Self-e-wqloyed males earn more than wage employees at all levels except males with university educatioi. lThe exact opposite pattern is seen for females: Employees earn more than the self-employed at all levels except females with university education. There is no clear pattern for male earnings by public and private sector. However, female workers in the public sector earn more than those in the private sector for all education levels except for university education. From the sample statistics, there is obviously a wage gap betweon males and females. In economic theory, wage differentials come from two broad sources: (1) differences in 'skill' (attributes) and (2) differences in 'treatment' (wage structure), i.e., from 4iscrimination by nployers. 'Me upper bound of this discrimination can be compi ad by using the Oaxaca (1973) decompos,eion technique. This requires esimating earnings functions for females and males. d. DaterfWnants o? Female Labor For Partdpation In this section, we discuss the determinants of female labor force pardcipation. Important factors deermining women's propensity to participate in the labor force are marital status and presence of young children. It is common for women to vwithdraw firon the labor force during child- bearing and when their children are young. Obviously, the presence of young children increases the value of non-market activities, particularly in developing contries where childcare services are very limited. In certain economic groups it is also common for women vo ceas'- working when they marry. Hence, single females have a relatively high probability of particip&ng in the work force. Age is also a kay variable in explaining the probability of female participation. Greenhalgh (1980) ad Mohan (1985) use the quadratic form to demonstrate that the participation rate of women increases at a decreas~Ag rat a they age. On the other hand, some researchers argue that tie Labor force participaian of women is expected to follow a U d profile with age, indicadtng changes over their iife cycle (Sheehan, 1978; Layard, Barton and Zabalza, 1980; King, 1990). Hence entry wages and poteil market wages of w ,men are associatd with age, which in turn exerts effects on female labor force participation. Aside from demograplhic cra scs, eonomic fadors are found to be highly correlated with the labor force participation of women. Standing (1978) suggests that in the participation functon tbe 'need' for income is the dominant force in explaiing the participation decision of women, other things being equal. He argues that the 'need' for income can be m;'asured by several variables, namely husband's income, family income excluding female's eamirnn, assets or wealth possessions, the m ber of income earnes in the family and the houshold status of women. All of the above, exceping wome's household status, are expected to have a negadve impact on female labor force pard6cip E%idence from empirical stdies in the Unite1 Stes supports the fact tat female pardcip-ion is negaively related to the husband's wage, otht family income and other family income per equivalent adult (Sweet, 1973). As a determina of labor sply, investment in human capital cannot be ignored. Individuals invest in human capital either through schoolng or taining to obtain higher ;> ture earnings. Thus, the higher an individual's educational It vel, the higher the opportmity crst of being out 6 Woen's ExVikyment and Pay in Lain America of the labor force. The expet:.4 positive association between education and labor force participation of women is usually found. Studies of female labor force participation find that the decision to participate can a'so be influenced by area of residence and by migration. In Latin America, it is common for E-urban residents to migrate to urban cities to look for better opportunities. Standing (1982) points out that 'more women than men [traditionally] have gone to the towns and cities and these women have been predominantly young and single ... getting into labor force activities or finding better, higher-paying work, or access to training for employment.' Tlere is also, however, evidence of non-significant effects of migration on the labor force participation decision in Standing (1978) and Behrman and Wolfe (1984). In summary, the probability of participation is affeed by personal characteristics, family composition, educational attainment, and economic factors related .w income 'need.' An abundant literature has been written on the issue of seJectivity bias when estimating wages using only working females (Gronau, 1974; Heckman, 1979). It has been argued that an estimation based only on working females gives rise to biased estimates. The bias is mairly due to the fact at the sample of workers in the labor market is self-seleced, having lower reservation prices than otherwise simiar non-workers. For those non-working females, their wages are unobserved. To correct for such a censoring problem, Heckman (1979) proposed a two-step method. First, a probit equation is used to estimate the probability of a woman being in the work force. The inverse Mill's ratio is computed (denoted here by Lambda) and is added to the earning function as an additional regressor in the second step. Tle empirical work in this study follows the Heckman procedure and the definitions of the variables used are discussed in the following paragaphs. The dummy variable that defines the labor force participation decision of females is set to 1 if ,Aie female is economically active, looking for a job, or temporarily un mployed due to sicknest and job search, and 0 otherwise. Personal characteristcs such as age, ma itl status and education are important explanatory variables. To capture the non-linear relationship balween age and the probability of labor force participation, age splines of 5-year intervals are used and the omitted category is the 60 to 65 age group. A dichotomous variable for marital status is used. Simiarly, to examine how differences in educational levels affect the female participation decision, dummy variables are created for less than primary education (the reference group), primary education, commercial secondary education, technical secondary education, other secondary education, and higher education. Severzl other variables that measure the wealth, income, and household production demands are also included in the analysis. A dummy variable for the proxy of wealth, is assigned the value 1 if the respondent owns the house andlor the land and 0 otherwise. Likewise, information about fai-, acome other than the female respondent's is another independent variatle. Total mumber of income earners within the same household is added to capture the housebold's division of labor between the home and the market. A du:mmy variable for a female headed household is used to proxy the financial responsibility of the female and other socio-economic differences in family types. The size of the family and mLmber of young children undez 6 years in the family are used to account for the effect of household production on the labor fiorce participation decision. Finally, the p equation also includes a variable to see whether being foreign-born affects the participation decision positively since a large portion of the population is non-Argentinean. Owing to data limitadons, migration status and regional residence are not considered. F=ak Labor Force P rf . and Gndgr Ewmgjs D57a0rerals bi a,gemua 7 lTe results of the probit fncton are shown in Table 1.3. As expected, age is found to be an important draerminant in explaining female labor force participation. The probability of participation appears to have a concave profile and peaks between ages 25 to 29 years. In addition, married females are less likely to be in the labor force; the probability of labor force participation is reduced by 31 percent as compared to non-married females. Moreover, the estimates for family sre and the DnUber of children aged under 6 are consistent with the literature. Regarding the 'need' for income, two out of the four variables are statistically significant at conventional levels. Having owned a house and/or land reduces the probability of being in the labor force of women by 10 percem. Possession of other income (fronm husband and/or other family members) leads to a 0.1 percent decline in the labor force participation probability. The insignificance of the two other variables may be due to the high correlation between them and the proxy for wealth. Being forei,gn-born, however, is insignificant in explaining labor force participation among females. Table 13 Probit Regesm Rtalis for Female Labor For:e Participation Variable Coeffiti-t t-raio Partial Dervative Constant -O7: (-4.62) Age 19 or less 0.145 (1.24) 0.053 Age 20 to 24 1.137 (9.98) 0.418 Age 25 to 29 1.292 (11.65) 0.47S Age 30 to 34 1.247 (11.62) 0.459 Age 35 to 39 1.225 (11.38) 0.451 A4e 40 to 44 1.198 (10.96) 0.441 Age 4S to 49 1.182 (10.87) 0.435 Age S0 to 54 0.909 (8.38) 0.334 Age 55 to 59 0.629 (5.75) 0.231 Maried -0.840 (-13.22) -0.309 Own house/ld -0.271 (-5.43) -0.099 Other ILcome -0.004 (-1.87) -0.001 Had of Household 0.002 (0.02) 0.000 Nunber of Earners 40.018 (-1.43) -0.006 Family Size 0.011 (1.88) 0.004 Children adler 6 yeses -0.095 (-4.11) -0.035 Piny EducaioDs 0.299 (2.29) 0.110 RegulAr Soooeday 0.333 (2.40) 0.122 Technicl Secoldaxy 0.700 (3.22) 0.257 Commercl Scmday 0.378 (2.75) 0.139 University 0.976 (6.74) 0.359 Foreip born -0.063 (4.96) -0.023 Notes: Sample sin - 4,700. Log-Likelihood - -2666.9 Similar to findings in other Latin American counries, primary education exerts a significant effect on ftmale labor force participation when compared to less than primary eduction (the omitted cate.-ory). Note that the probability. vparticipation increases with increasing educaional 8 Women ' EpZyzw aad Pay in LZ, Ameria attainment. The highest protability of participation is found for those with completed higher education (36 percent). Te participation probability varies by type of secondary education. According to our estim, of the three types of secondary education, having technical secondary education gives the highest probability of participating (26 percent) while commercial secondary education only inareases the participation p%.i bility by 14 percent. Using the above probit results, we eaamine the effect of cLanges in certain characteristics on female labor force participation by simulation. Given that other sampie characteristics remain unchanged, we predict the probability of labor force participation for different educational levels, marital status, the mLmber of young cildren, the size of the family, the ownership of house and/or land, and age groups. The predicted probabilities are found in Table 1.4. As one might expect, increases in educational levels lead to an increase in participation probability. It is also interesting to find dthg the predicted probabilities for primary education and academic (regular) secondary education differ by only 1 percent. Tlis interesting finding may reflect the impact of compulsory primary education which reduces the reward differences between primary and secondary education (as shown in Table 1.1 over 50 percent of females in the sample have rrimary education). Commercial secondary education, on the other hand, does not increase the participation probability as much as technical secondary education does with respect to academic secondary educatin The probability of participation for non-married women is twice that for married women. The number of young children is another constaint on the participation decision. In Table 1.4, predicted probabilit drops from 37 percent to 23 percent when the number of young children increases from nom to 4, other thinp being equal. Consequently, household responsibilities of women are impor factors affecting the paricipation decision. In contrast, the size of the family increases the probability of participation with a flat rate of 0.4 percentage point. This interesting result may demonsate the fa that as the size of the family increases, the tme needed for childcare m home production from the women is overcome by the 'need' for 'rcome to support the fmily. Owning a house and/or the land causes the probability of participation to decline from 42 percen to 32 percat. Finally, p;edicted probabilities from the age splines demonstrate an invered U-shaped profile with the highest probability for women aged 25 to 29 years. S. Ermin 1 7 Mimcer's basic modd f esdmain earnings functions regresses hourly earnings (wage rates) on schooling, and experience squared. The stndard way of incorpurting educatin is to use a contnwt variable to measre the yearr of schooling so estimates of private rates of return to additional year of schooling can be obtained. According to economic theory, the earnings profile appears to be increasing at a decreasing ratm with years of market attacbmen Labor marltet and its squae are used to test for this. In most cases, significant quadratic effects with eventual dimiishing marginal returns to experience have been found (Shields, 19W, Behman and Wolfe, 1984). In the absence of actual experience information, potential experience was calculated as age minas years of schooling minus six. We estimate a Miner-pe earing function in which years of schooling, experience, experience squared and the nawral logarithm of hours worked are included as regressors. A separate earnings fiunction is esimatd for females and males to account for any wage diffcrences due to sectoral and industrial differences. Femak Labor Force Pwwi4adoni and Gender Eansbgp Differendals bin Argetnma 9 Table L.4 Predicted Participation Probabilities by OChameistic Cbmwasc PA2Wi Pbability Lew than prmazy 21.7 Primsy 31.4 Ragular seondsy 32.6 Commercial codary 34.3 Technical M=oday 46.7 Universty 57.6 Non-married 55.9 Married 24.5 N vmber of Yom' Oiildr NonW 37.2 One 33.6 Two 30.2 Thre 26.9 Four 23.8 Size of ffi Ps,tmv one 31.9 TWo 32.3 1hree 32.7 Four 33.1 Five 33.5 Six 33.9 Ovshpof HQso Nd/r No 42.7 Yes 32.4 Ani 15-19 17 20-24 44.0 25-29 50.2 30-34 48.4 35-39 47?5 4044 46.4 45-49 45.8 50-54 3S.2 55-9 25.5 60-6S 1.0 Actuamn puticpation 36.0 .Note: Bs c the ratlt reported in Tabl 1.3. 10 Wom' Dr SEi.,Iym andPay i LaiAu A -ica For both males and females, the standard Mincer-type regression function includes independent variables measuring the years of schooling (S), potential labor market experience (defined as AGE-S-6)1 and its square and the natural logarithm of weekly hours work. The dependent variable of the earnings functions is the natural logarithm of monthly earnings. In order to account for any structural difference in the labor market that affects mne 'treatmento component of wage differentials, sepac:ate earnings functions for males and females are estimated. With the eueption of a change in the schooling variable (which is replaced by categorical educational levels as described in the probit equation), several independent variables that measure sectoral employment and types of employment are included. In addition to potential experience, several dichotomous variables indicating whether individuals have received on-the-job training are added to capture the impact of specific training.2 Similarly, dummy variables for being employed in the public sector or not, working in the dependent employment sector or not, and whether weekly hours of work is less than 35 or over 45' are included. Earnings differences due to any intra-industrial differentials are taken into account by various industrial sector dummy variables. Finally, being married and/or being a foreign-born individual (FOREIGN) may affect earnings as well. In the case of females, earning functions are estimated by ordinary least squares (OLS) and are presented for the purpose of comparison with respect to the equation with selectivity adjustment. Table 1.5 shows the results for earnings functions of males and females that are consistent with the theory. The inverse Mill's ratio (Lambda) in the female. earnings function in colutmn 3 of Table 1.5 is marginally significant and negative. Consequently, it is noa surprising to find that results of the Heckman procedure are little different from those in the regular OLS. For both types of estimaon, an additional year of schooling increases earnings by about 11 percent. Likewise, potential labor market experience and its square reveal a non-linear earnings profile for females, increasing at an decreasing rate. For males, the earnings elasticity with respect to hours of work is 0.3905 and for females is 0.6589 (uncorrected for selectvity) or 0.6607 (corrected for selectvity). Moreover, the returns to education for males is lower than for females, 9 percent compared to 11 percent. The reward of potential market experience among males is higher than that of females. A different specification of the earnings function, which includes information assrociated with labor market structure and training produces slightly different results for each earnings function. Table 1.6 presest earnings functions for both males and females, with female earning functions estimated using regular OLS and OLS with selectivity adjustment As shown in the table, the I Though the definition has been criticized by various sclolarsr a more ccurRw measmr is unavailable fiom this dala eaL As a result, the re ter should be cautious in the intepreation of the 2 SeW Kugl Md PacharpouloS (1989). Tle purpose of the hors of work variable is to account for any effect of worling overtima nd moonhghtn sine there is a large proportion of respondents working more than 45 hours per weed Female Labor Force Panicapion and Gnder Eavnings Differuetials in Argentina 11 Table 1. Eanungs Functons Men Women Wome (uncorrected (coffected for for Variable selectivity) selectivity) Consant 8.3429 7.0701 6.9662 (63.245) (48.300) (52.276) Schooling (year) 0.08 0.1067 0.1092 (30.006) (22ns75) (24.747) Experience 0.0491 0.0384 0.0394 (15.505) (9.185) (9.S40) Experience squared -0.0007 -0.0005 -0.0006 (-11.482) (-6.472) (-7.050) In (hows) 0.3905 0.6589 0.6607 (11.647) (21.055) (21.066) Lambda -0.0837 (-1.695) R-Sq,ared 0.3463 0.4527 0.4515 N 2,397 1,338 1,338 Notes: Figures in paZretheseS are t-raimo. The mean of the depndent varns. , log-monthly eamings, for me gad workdng women are 11.29 and 10.79, respectively. effects of potential experience, its squared term and hours of work for both sexes are similar but with slightly different magnitudes thin those in Table 1.5. Tho Lambda (inverse Mill's ratio), is again negative but highly significant. The correction fov censoring leads to two main differences in the esimation results. First, the dummy variable for w2rital status is highly significant with a greater effect (17 percent versus 7 percent). Second, the return to higher education is substanially lower in the case of the selectivity result These changes reflect (1) biased estimates obtained firom regressing working females alone without controlling for selectivity and (2) the importance of higher education and marital status in affecting the self-selection process (participation decision). Since both etima in the females earnings functions are quite similar (with the exceptions ,-nentioned above), the following discussion is based on the results of the selecivity equation (Column I of Table 1.6) unless otherwise specified. Notice that the returns to education increase with educational level (average returns of 5 percent to secondary and 9 percent to higher 12 W07en 's Ekpkymeid aRd Pay in Lazb Amca*a education) e pt for primary education2! The insignificant effect of primary education on female's earnings might indicate the 'deflation value' effect which results from compulsory primary education. The experience profile appears to be concave, peaking at 34.33 years calculated at the mean value of the working female sample. A one perceat increase in weekly hours worked leads to a 0.73 percent increase in earnings. The consequence of having any work-relaed training has an interesting impact on earnings. In the case of females, having one training course enhances female earnings by 15 percent while having two or three taining course makes no difference to earnings compared to those having no training at all. Earnings increase by 56 percent if a female has four training courses, while having five or more training courses has a negative impact on earnings, compared to the reference group (no training). This result is puzling but also likely to be imprecise - less than one percent of the females in the sample fell into the latter two groups. For females, earnings do not differ between the public and private sectors, other things equal. On the other hand, if they are an employee (dependent employment) they will earn 10 percent more than the self-employed, holding other favors constant. This is not surprising given the wage detemination system in Argentina. Earnings of part-timers are not statistically differet from dthse of full-time workers. Those who work more than 45 hours weekly, however, earn about 22 percent less than full-time workers, holding hours of work constant. In the female labor market, ethnicity does not affect earings. A working married female earns 17 percent more than a non-married woman, though the probability of participation is negatively associated with marital status. Those who work after marriage tend to have a higher educational attainment or more investment in human capital. across differen types of industy, only finance and services industries exert effects on female earings. Working in the finance industry allows females to earn 22 percent more while serv. e female workers are paid 37 percent less than other types of industry (the omitted group). The latter result is as expected since the relative overaowding and the low skill requirement in ta sector lowers average earnings. For males, the altenate specificadon shows a different picture. The rate of reurn to different levels of education increases with each successive level of education, with the exception of secondary education (an average of 6.6 percent reurn to priay education, 5.3 percent to secondary education and 11.6 percent to university level). The experience profile is concave in shape and has a lower decreasing rate than that of women. Males' earnings are less sensitive with respect to the mber of hours worked per week (elasticity of 0.332). 4 Th avI'am vge retu to each edut nal ll ciaculatd by dividing the change in coefficients of the pd edcaonal level by the difference in year of schooling beween tXc compued Felral Ltabor Force Parwraton and Gend4r Eamndgs Dfferencals m Argentsa 13 Table 1.4 Eamgs Functons with Altereative SIwification (t-toa in parenthess) and Sample Means of Selected Vuriables (standard deviatk9 in pareses) Womn Males Corrcted Unoorrected for for Variable Selectivity Selectivity Mean OLS Mean Constant 7.6189 7.3687 8.8320 (26.779) (27.532) (3S.086) Primay 0.1252 0.1463 0.3946 (0.997) - (1.22:) (4.070) Secondary 0.43U4 0.4796 0.7126 (3.320) (3.817) (7.208) University 0.7899 0.9115 1.1753 (5.435) (6.093) (11.,67) Experience 0.0302 0.0376 0.0396 (5.571) (8.278) (10.94K) Experience squared -0.0004 -0.0006 -0.0006 (4.143) (-6.913) (-9.284) in (hors) 0.7289 0.7372 0.332i (13.398) (13.398) (5.5Th) Tm= One course 0.1522 0.1531 0.22 0.1006 0.16 (3.899) (3.880) (0.42) (3.311) (0.36) Two courses 0.902S 0.00002 0.07 0.1917 0.04 (0.040) (0.000) (0.25) (3.421) (0.20) Three courses 0.0867 0.0826 0.04 0.3109 0.01 (t.039) (0.981) (0.19) (2.624) (0.09) Four courses 0.5565 0.5495 0.004 0.3463 0.003 (2.431) (2.372) (0.07) (1.744) (0.05) Five or nmom couses -0.7796 -0.7865 0.002 -0.1193 0.002 (-1.961) (-1.972) (0.04) (-0.504) (0.05) Public Sector -0.0188 -0.0160 -0.0042 Employee (40.203) (-0.172) (-0.105) Employee 0.1042 0.1075 -0.0509 (2481) (2.562) (-0.177) - contumd 14 Wmlen's Eipioymm and Pay in Lad Amrica Table 1.6 (cnntm Eamin Futons with Alternatve Spefication (t-rao in .sis) and Snmple Means of Seacted Vaibks (sdard devia in pa3thens) women Males Coffected Unconrected for fcr Variable Sdectivity Selectivity Memn OLS Mma Manufactring -0.1010 -0.1083 0.21 -0.0447 0.31 (-1.07) (-1.068) (0.41) (-0.851) (0.46) Construction 0.2287 0.2112 0.034 -0.1739 0.11 (0.929) (0.849) (0.07) (-2.894) (0.31) Commerce -0.08'4 .0.0850 0.14 -0.0331 0.16 (-0.777) (-0.881) (0.35) (0.592) (0.37) Transpont 0.1461 0.1400 0.01 0.0796 0.10 (0.875) (0.830) (0.12) (1.335) (0.30) Finance 0.2231 0.2119 0.09 0.1226 0.09 (2.060) (1.936) (0.28) (2.014) (0.28) Public Sector 0.0248 0.0158 0.27 -0.0321 0.10 (0.188) (0.118) (0.45) (-0.463) (0.29) Recreation -0.2971 -0.3010 0.01 0.1437 0.02 (-1.592) (-1.595) (0.09) (1.482) (0.13) Service -0.3674 -0.3745 0.24 -0.2438 0.08 (-33583) (-3.607) (0.43) (-3.891) (0.27) Married 0.1734 0.0786 f5.1867 (3.392) (2.285) (6.262) Foreign -0.0795 -0.0897 -0.0645 (-1.354) (-1.52) (-1.832) P a-tilme worksr -0.0334 -0.02S4 -0.1248 (.0.642) (-0.483) (-2.742) Works over 45 r Win -0.2151 -0.2206 -0.0292 (-4.756) (4.824) (.0.939) Lambda -0.1945 (-2.555) R-squatd 0.4867 0.4840 0.3572 N 1,338 1,338 2,397 Note: The mean of tbe depmdeut vardblc, Io g.oindhy eaninp, *r am and wordkhig vA= anr 1129 and 10.79, rwp evdy. Female Lk-bor Forte Parrication and Geider Earnings Differernials in Argentina 15 Unlike females, the returns to training increase as the number of training courses received increases, except where the individual has five or more training courses. Sectoral differences have no impact on earnings for males. According to the result, other things equal, part-timne ..ale workers earn 12 percent less than full-timers. Like females, married men earn about 19 percent more than non-married males. On the other hand, males who were not born in Argentina earn less, by 6 percent. In addition to finance and set vices sectors, the construction sector affects males' earnings rzlative to the omitted category (other industries). Earnings in the service and construntion industries are lower by 24 percent and i 7 percent, respectively. On the other hand, males working in the finance sector have an cdvantage and earn '2 percent more. Both sexes gain more b' working in the finance sector and this reflects the profitability of the sector. 6. Discrimination The actual average earnings differential bet m.m working females and working males is 35 percent. Tle Oaxaca (1973) 'upper bound' decomposition can be obtained by the fIbllowing methods: ln(Earnings) - ln(Earningsf) = Xi(b,-bd + b.(X,-X) (I) = Xj(b.-b) + b(X,-X), (2) where X,s denute variables of the earnings functions, bis are the corresponding parameter est aates, and i=f (female) or m (male). The first term of the right hand side of equations 1 and 2 is the difference in earnings due to differences in the wage structure while the second tern refers to the difference due to differences in endowments. Note tba: the upper bound decomposition can be done both ways. This gives rise to the so-called index-number problemL Since economic theory provides little guidance on this, Table 1.7 summarizes both methods' results, denoted I and 2 accordinglv, for regular OIS and OLS with selectivity adjustment. Using the uncorrected OLS regression estimates (columns 1 and 2 of Table 1.5), if females have the saame wage structure as males, differences in endowmentc betv een males and females explain 22 percent of the total earnings differential and 78 perce.a of the total earnings differential is due to differences in the wage structure. Using equation 2, difference in endowments and differences in the wage structure a.ccount for 32 percent and 68 percent of differences, respective!y. Taking selectivity bias Mto account, the rirst decomposition method shows 26 percent and 74 percent of differences being due to endowments and the wage structtrre, respectively. When females are assumed to have the same characteristics as males, differences in the wage structare account for 62 percent of the total earnings differential. Differences in er.owments, on the other hand, explain 38 percent of the total earnirgs differential when the second decomposition method is used. The upper bound of decomposition attributes all o' the unexplained earnings gap to discrimination. Since the regressors do not capture all attr..uts that affect earnings, any left-o variables lead to an upward bias in measuring discrimination.' On the other hand, if any 5 T'he decomposition is calulated for the altrna specification (Table 7.5). The result show tha the unexplaini porion of the eauings differendals diffrs by 10 percenL 16 Wmne a Enwp&yis and PaY n Ltin Ame Tao!2 1.7 Decomposition of th Sex Eamuingo DswreaiA Perun:tar of the diffeti due to difleme3 in Specification Endw=MZA Wag a Stru Not Correced for Sylecthivit Equation1 22 78 Equaion 2 32 68 Corrctd fo Selctivity Equaionl I 26 74 EqueSon 2 38 62 (Wage1,dWager 154%) Nou= Tle decGmposition is based on the rea4ts of Table 1.5, aboe Wage.] age1 is the mtio of the m y e explanatory variables are affected by discrimination such as educaion, sor of employment etc., the measure of discrimination obtained could be biased downwwds instead. Regardless of wvhich upper bound decomposition method we use, par of the gender earing differectials coine from unexplained sources other than indidual's initial eo . In other words, dicruimination appears to exist in the Argenine labor matn 7. Discussion The analysis of female labor force pardcipation indicates a fumre change in the soci>-economic structure. The highest probability of female labor force participaton is found in the prime age group. Morsover, married stau htd number of young cHdren are the key social determinants in the participation decision, which fiurher supports the fact that there is a tendency among women to postpone marriage and/or childbearing. As these cohos age we can expect much higher levels of female labor force pardtipation in the fiture. Educational is found to be an im;po tant fctor affecdng the participation decision. Better opportnities in education, especially technical secondary education and Ligher educadon, stimulate females to enter the labor market. Combining the efes of higher educational attainment with lover ferility, we expect a marked increase in female labor force participation in the near future. In order to facilitate married women in joining the labor market, a greater dcsmnd for extensive childcare services provided by the private and/or the public sector is likely to occur. With respect to wrnigs differentials between males and females, the upper bound decomposition technique provides an insitutional idea on the subject The following argume am based on the second decomposition metod described in the previous secdon. For workig females alone (the uncorrected OLS es), increases in endowments will allow females to earn about 76 percut F=a;o Labo FO7" PFW*OadM anad Oddw Zwxa D&TMdaIr ho ArgFAw 17 of male eamingp. On the other Md, if females are rd as males, the emaings diffmeial drops to 10 percent Hence, to brig abwt gra oquaWity betwoen working males and working fenales, more emphasis should be placed on te treatment of females La the labcr maruk, occupatons and job moblity. References Behrman, !'.R and B.L. Wolfe. Tabor Force Participation anl Eaning Determinants (or Women m the Special Conditions of IXveloping Ccuntries.' Jounal of DewClopmew Economics, Vol 15 (1984). pp. 259-288. de Lates, Z.R. Dynamics of the Funaec Labor Force In Argentaia. Paris: The United Nations Educational, Scientific and Cultural Organization. 1933. Dieguez, H.L. 'Social Consequencs of the Economic Crisis: ArgenSha ' Mimeograph. Washington, D.C.: World Bank, aot dated. Greenhalgh, C. 'Participatien and H urs of Work for Mlarried Women in Great Britain.- 0ord Economic Pape, Vol. 32, no. 2 (1980). pp. 296-318. Grona., R. 'The Effect of Children on the Housewife's Value of Time' in T.W. Schultz (ed.). Economics of the Family. Chicago: University of Chicago Press, 1974. Keckman, J. 'Sample Selection as a Specification Error.' Econowmeirca, Vol. 47, ro. 1 (1979). pp. 153-161. King, E.M. 'Does Education Pay in the Labor Market? The Labor Force Participan, Occupation and Eanings of Peruvian Women.' Living Standards Measurement StD'y Working Paper No. 67. Washington, D.C.: World Bank, '990. Kugler, B. and G. Psacharopoulos. 'Earnings and Education in Argentina An Analysis of the 1985 Buenos Aires Household Survey.' Economics Gf Education Review, Vol. 8, no. 4 (1989). pp. 353--65. Layard, R., M. Barton, and A. Zabalza 'Married Women's Particiption and Hour . Economkca, Vol.47 (1980). pp. 51-72. Marshall, A. 'Inmigraci6n, demanda de fuerza de trabajo y eatructura ocupacional en el drea mer"politana argentina.' Desarroilo Econ6mlco, Vol. 17, no. 65 (1977). Mincer, J. Schooltig, EWrerence and Earnings. New York. Colurnb.a University Press, 1974. Mohan, R. 'Labor Force Farticipation in a Developing Metropolis: Does Sex Matter?' World Bank Staff W.'orking Paper No. 749. Washington, D.C.: World Bank, 1985. 18 Fmalk Zabor Force Parzctoa and Gndgr Earnings Dfferenziab u Argunia 19 Oaxaca, R.L. uMale-female Wage Differentials in Urban Labor Markets.' lntematonal Economic Review, Vol. 14, no. 1 (1973). pp. 693-709. Riveros, L.A. and C.E. Sanchez. 0Argeatina's Labor Markets in an Era of Adjustment.' Working Paper No. 386. Washington, D.C.: World Bank, 1990. Sanchez, C.E. 'Characteristics and Operation of Labor Markets in Argantin;." Development Reurch Department Discussion Paper Report No. DRD272. Washington, D.C.: World Bank, 1987. Sheehan, G. 'Labor F orce Participation in Papua, New Guinea' in G. Standing and G. Sheehan (eds.). Labor Force Parfctparlon in Low-income Countries. Geneva: Intemational Labor Organizaton, 1978. Shields, N. 'Women in the Urban Labor Market of Africa: The Ca5e of Tanzania.0 World Bank Staff Working Paper No. 380. Washington, D.C.: World Bank, 1980. Standing, G. 'Femnale Labor Supply in an Urbanising Economy' in G. Standing and G. Sheehan (eds.). Labor Plorce Parnlcipaton In L come Coumres. Genea Intnatioa1 Labor Organization, 1978. . Lor Force Panlcoadon !. d Deweopment. Geneva: lntenazonal Labor Organizaion, 1982. Sweet, I.A. Women In he Labor Force. New York: Seminar Press, 1973. Wainernamn, C H. 'Educacin, familia y participacidn econ6mica femenina en la Argentina.' DesarrolloEcon6mnco, Vol. 72, no. 18 (1979). pp. 511-537. Women in the Labor Force in BoHlla0 Partc1ipation and Earnings Katerne MaIwon Scott 1. Introdution In 1989, ten years after the United Nations approved the Agreement on the Elimination of all Forms of Discrimination against Women, the Honorable National Congress of Bolivia ratified the agreement, joining close to one hundred other countries in pledging to analyze existing laws and legislation to determine where chanes needed to be made to bring the legal codes into alignment with the United Nations' Agreement In the same year dt the Agreement was ratified, weaidy earnings for women in Bolivia wera, on average, only 63 percent of male earnings, female-headed households were more likely to be below the poverty line than male-headed households und the female illiteracy rate was twice the male rate. It is not at all clear that existing labor law is responsible for the -der-based differential in earnings and living standards observed in Bolivia, especially when there is some agreement that existing laws are not always enforced. It appears that there are other factors which affect the eamings of men and women in Bolivia. This study focuses on the determinants of earnings and those fors which exlain the observed wage differential between genders in Bolivia. By decomposing the earings functions of the two groups it will be possible to identify that part of earnings which is explained by different end'jwment levels and that part which is due to different market valaus being placed on male 2a female labor. A better understnding of labor markets in Bolivia will assist in the fornulation of policies which can complement the legislative changes being envisioned and serve to include women in the economic activities of the country on a more equal footing. The following section of the study presents background information on Bolivia, its eCLo2ny, and labor force charteristcs. Information on the data used in the study is provided in Section 3. Several limitations on the degree to which the data can be used to extrapolate to the countr a wh-:le are discussed in that section. The fourth section contains the labor force participation function "or women. This participation function provides the means by which the female earnings function can be corrected for selectivity. Section 5 contains the description and results of the male and female earnings functions and the decomposition of these is carried out in Section 6. In the final section the overall results of the analysis are discussed and recommendations for improving the situation of working women in Bolivia are presented. 21 22 Women's Enpkiyment and Pay in Latin America 2. The Bolivian Economy and the labor Market Bolivia is a landlocked country with three distinct geographic regions (highlands, valleys and tropical plains), several ethnic groups speaking different lar.guages (Aymara, Quechua, Guarani, Spanish) and an economy heavily dependent on the export of primary products. The country is poor, especially compared to the other countries in the Latin American region. Per capita Gross National Product was US$570 in 1988 (World Bank, 1990b). Various social indicators also reveal the degree of poverty in the country. In 1988, adult illiteracy was 25 percent (World Bank, 1990b). Rural illiteracy rates were much higher than urban ones (31.3 percent versus 7.7 percent) and, of the illiterate population, 65 percent were women (Wor!d Bank, 1990a). Life expectancy at birth, in 1988, was 53 years, infant mortality was 108 per 1000 live births, and the maternal mortality rate (per 100,000 live births) was 480 in 1980 (World Bank, 1990b). The people most likely to live in poverty in Bolivia are those who live in rural areas, own little land, are female, are of Indian origin, are from the central Andean Region, and work in agriculture or household industries. The physical characteristics of Bolivia (low population densiry, distinct geographical regions) which have .-ed the integration of the economy (Horton, 1989), and the dependence of the er,onomy on - iry products has made Bolivia very vulnerable to external shocks. During the economic crisis, which began in the late 1970s and continued tiimugh the first half of the 19&0s, inflation reached 24,000 percent, GDP fell by 15 percent between 1976 and 1985 and per capita GDP fell by 30 percent in the same period (Arteaga, 1987'. In 1976 (the year of the last national census), 80 percent of the population was considered to be poor and 20 percent was considered to be extremely poor'. Concentations of poverty were found in the rural areas, especially in the Altiplano. GDP per capita has fallen steadily throughout the 1980s and, in 1988, stood at only 72.7 percent of its 1980 level (World Bank, 1990c). Recent data collected from a variety of sourc: lead to the zonclusion that the state of poverty in the country has not improved and, in fact, may be worse in rural areas (World Bank, 1990a). Te laborforc. The economic crisis has led to a substantial decline in real wages and salaries in Bolivia. (See Table 2.1 for estimates of the fall in real earnings.) While there is some discrepancy in the data, both sources used in Table 2.1 show wages in commerce to have been particularly hard hit by the economic crisis. Wages in the service and maufacuring sectors were also among those which lost more of their real value. Wages in the financial sector also fell although there is some discrepancy in the data about the extent of the decline. Of specific interest to the present study is the fact that the commercial and service sectors, where the real value of wages eroded most, are the two sectors where female employment is concentrated: 81 ' A household was considered to be poor if its income covered 70 prait or less of the b2sic needa basket developed by PREALC. It was extremely poor if its income covered only 30 peret or les of the bssic needs basket 2 ho Nazional Istituw of Sttistcs of Bolivia bh carried out, in th, 1980s, a carie3 of mzrveys, some of the most important of which are, Encuesta Pemanente de Esblecimientos Economicc3 (1983 and otaUiv ;rs), Encuesta Nacional de Poblacion y Vivienda (1988), Encuesta Permanente de Hogure (annul since 1980), and the Encuest Integrada de Hogares (EIH) which combines the old lr force survey with parts of a vming Standards Measurement Survey. The EIH was started in 1989. W0men is he Labor Force in Bo&ia: Pnicoanon and Earnings 23 percent of the female economically active population (EAP) was in these sectors in 1985 (Ateaga, 19M. The distnbution of employm.4'it by sector of economic activity and occupational type has also changed during the economic crisis. Prior to 1980, the share of the EAP in agriculture was declining and th" share in manufacuring was increasing. These standard development trends have been reversed i! the 1980s as labor has shifted out of industry and into agriculture (Horton, 1989). (See Table 2.2). Tabe 2.1 Boivia: Evoiti,on of Real Salarics A.. B. GOwth Rats of Rh i Sakilo By Scctor Bvo1mi of Real Wag" By S.ctw 197145 192-1987 uvb 19S-100 Dcc. Dcc. Dec. Dec. S.cta 171-76 197645 197145 Sct5 1932 1934 1"6 1987 A.ricu- 0 -39 -39 1Ainiog -1 -26 -40 Mining 162 117 100 68 Petgelem s5 -30 29 Rydro-zo- 74 2Q0 66 - IlDsUTY 0 -39 -39 SU a Il 136 59 43 Come, -7 .60 -63 c 52 93 74 8S TrnnPoe -2 -6 -37 Utllhic S3 89 74 64 FWaDnC & Txanqt 57 56 58 71 Insanee -57 -51 -79 C 69 76 63 44 ServiC -25 -53 -64 Finncil 90 109 10 65 TOTAL -13 -44 -51 ServicMs 63 104 72 - Saute: Cazo de Pn de] Lado, 1936. S5wc: Ha, 1989. To 6, p. 35. Tabe 2.2 Idszrial Disttioa of foua Erploymcca, 17-M1986 inm peavn) Secr 1970 1976 1980 1936 Arluhar 50.6 48.1 46.5 49.9 Mllz 4.0 33 4.0 3.1 Hydtecmbm 03 03 0.4 0.5 Mam6cmdng 9.7 10.1 10.3 9 Cotto 3.7 5.7 5.5 2.6 Lildtie 0.2 0.2 0.4 0.5 TmnpogL Countnic. 4.0 3.9 5.4 5.6 C 4rnering 7.2 7.4 7.4 8.2 F ;nance 0.6 0.6 0.6 0.8 Services 19.7 19.6 193 20.0 Sour=: HmtDa, 19M, Table 4, p. 33 The other majr chnge in employment distribution has been the growth of the informal sector. This sector has alway been large, but the early 1980s saw a rapid increase in the mmber of self- employed and unpaid family workes. A 1983 study of the self-employed showed that these 24 Wmen's Employm and P,.1 fii Lana Amerca workerv as a percent of the urban labor force had increased from 29 percent in 1976 to 34 percent m 1983. The annual rate of growth of errployment in the self-employed sector averaged 5.95 percent and that of unpaid family workers grew by 7.66 percent. In contrast, the salaried labor force grew only 2.25 percent annually during the same period. (Casanovas, 1984). A recent World Bank report (1990a) indicates that another cause of the fall in income has been the movement of large segments of the working force out of the formal sector and into the informal one. Unemployment rose in the 1980s, with the highest rates found in areas whiere mining had been a significant industry (Potosi and Oruro). While figures provided by various sources differ, there is consensus that the increase was quite high; the lowest estimate is an increase of 26 percent between 1980 and 1987 (Horton, 1989). It is argued however, that unemployment rates really reflect the fall in formal sector employment and those that appear as unemployed are actually working in the informal sector (Horton, 1989; and World Bank, 1990a). Women in the laborforce. Female participation in the labor force in 1988 was estimated at almost 29 percent (i.e., 29 percent of all women aged 10 and up participated in the labor force). In the eje central', women's labor force participation was cstimated at 35 percent in 1987, up from the 1976 level of 20 percent (World Bank, 1989; and Horton, ;989). The rate of participation of women has not, however, changed significantly since 1980. Female unemployment rates have been lower than male rates in the 1980s although female underempioyment is higher.' As has been noted above, women have lower l1vels of education than men, are more heavily concentrated in the unpaid and family businesses, and are found primarily in commerce and service industries and the informal sector (World Bank, i990a; and Horton, 1989). Legally, there are still various statutes in Bolivia limiting women's full participation in the labor force. First, with some exceptions (nurses, domestic servants) women are not permitted to work at night. Second, the 1942 legal code limits the work week of women to only 40 hours, in contrast to men's legal work week of 48 hours. Third, except in cases where 'the work requires a higher proportion,' women are only allowed to make up 45 percent of the wage and/or salary earners in any given establishment (United States Bureau of Labor Statistics, 1962). Moreover, the labor code bars women from carrying o't jobs considered to be dangerous, unhealthy or hard- labor.5 The signifizance of these laws is not clear since it appears that they are not enforced (World Bank, 1989). Since the National Congress of Bolivia ratified the United Nations' 'Convencion sobre la eliminacion de todas formas de discriminacion contra la mujerQ in 1988, several commissions have been formed to review the existing legislation and recommend changes in those statutes which discriminate against women. 3 Includwa the cities of La Paz, Santa Cruz, Cochabmba md Oru". 4 Underemployment is defined as worinig less than 12 ho%= in tbh. -ence week (Horton, 1989). 3 See: World Bank, 1989; and Romero de Aliaga, 1975. Wom,e in the Labor Force in Boliva. ParticipatOn and Earnings 25 3. The lData Tle data for the analysis come from the second round of the 1989 Integrated Household Survey f1H), a bi-annual survey carried out by the National Statistical Institute of Bolivia (INE). hle survey is essentially a living standards measurement survey. Unlike the 1988 survey, The EIH only covers the capital cities of eight departments of the country (Cobija, the capital of Pando was not included), the city of El Alto, and all other cities with populations greater than 10,000. Little or no information on the rural population of the country is contained in the dat2 It should be remembered that the results of the analysis contained in the present study are applicable only t the urban labor force. The way in which labor markets function on a national level and in rural areas mnay be quite different. The EIH data used here were collected in November of 1989. Included in the survey are 7,267 households with data on 36,126 individuals. Of these, a total of 13,842 cases were used. Tlis sample included all people of prime working age range (15 to 54) for whom relevant dat were available.6 Table 2.3 shows some of the characteristics of the total sample used as w ll as the characteristics of the working population. Working men and women are defined as all those people who worked !or more than one hour in the reference week for pay. This definition excludes unpaid family workers but includes the self-employed.' Excluding unpaid family workers from the definition of the employed will underestimate both male and female participation rates. Female oarticipation rates will be underestimated more than male rates as more women are unpaid family workers than men. Excluding domestic servants will also lower participation rates of women more than men. Of the total of 13,842 individuals inciuded in the sample, 7,786 people, or 56 percent, are classified as working, i.e., receive pay. Participation rates for women (44 percent) are significantly lower than for men (65 percent). The average age of the sample is almost 32 years. Working women and men are more likely to be married than the sample as a whole. Eighteen percent of working women are heads of households as compared to 75 percent of working men. Working women have, on average, one-half year less schooling than their male counterparts. While this overall gap in education is not large, the distnbution of men and women by highest level of education completed shows some sharp differences. Only one quarter of working men ended their schooling at the primary school level compared to more than one-third of working women. The largest gap in educ.itional achievement occurs at the university ievel; only 7 percent of working women have a university degree compared to 15 percent of the men. Proportionally more women than men have atteed a teacher taining or normal school and greatea percentages of women have atended some form of technical school, especially at secondary level. 6 While the prime worcing age is usuaDy considered to be between ags 20 md 60, the limits used here reflect mor accuately the reality of Bolivia. Th mirvey itself colDect labor data for ages 10 and up. School attendance rates, however, are stil fairly high until age 15. Afer age 15, attendance is low and the percentage of woring individuals increases. Thus the lower bound of the working age populatior has been set at IS in thi audy. 7 Domestic senrants have been eliminated from the anysis d, in par, to the difficulties of detern' -ng ctal inomand also bece of the limited data in this muvey on such worke. 26 Women 's Employww wad Pay in Lain Amerca TabL- 2.3 Bolivia: Mans (and Standard Deviatons) of Sample Variable Total Working Workdng Variabl- Sales Men Women Age 31.70 33.94 35.46 (10.42) (9.78) (8.41) Married 0.66 0.76 0.73 (0.46) (0.43) (0.44) Head of Household 0.36 0.75 0.18 (0.48) (0.43) (0.38) Year of Schooling 9.28 9.50 8.97 (4.34) (4.45) (5.05) No Education 0.00 0.00 0.00 (0.01) (0.01) (0.01) Incompet PrinMay 0.16 0.14 0.23 (037) (0.35) (0.42) Fuished Primay 0.09 0.10 0.11 (0.29) (030) (0.32) Incompldc bdle School 0.10 0.11 0.10 (0.30) (0.31) (0.29) Finished Middle School 0.06 0.06 0.06 (0.24) (0.24) (0.23) Ircoplec Scondary 0.21 0.19 0.12 (0.41) (0.39) (033) Finished Seoary 0.13 0.14 0.10 (033) (0.35) (030) Sconiry echnical 0.04 0.03 0.06 (0.19) (0.18) (0.24) Highea Techical 0.02 0.02 0.03 (0.14) (0.15) (0.16) Normal School 0.05 0.04 0.12 (0.22) (0.20) (033) Univciy 0.14 0.15 0.07 (035) (036) (0.26) Home Ownership 0.61 0.58 0.59 (0.49) (0.49) (0.49) Languagg Spai only 0.71 0.67 0.68 (0.45) (0.47) (0.47) AymaraQuechusiGuarani 0.00 0.00 0.00 (0.06) (0.06) (0.06) Bilingual: Span. & Amerind. 0.28 031 0.31 (0.45) (0.46) (0.46) Bllingualk Span. & other 0.01 0.01 0.01 (0.09) (0.10) (0.09) Experience 17.30 19.88 21.96 (11.66) (10.78) (10.00) Hours Worked Per Week 51-30 44.12 (18.18) (23.00) Weckly Earnie - 110.51 68.89 (181.52) (8538) N 13,S42 5,314 2,472 L Total mnpl rdef to al people aged 15 to 54. Woddng populi coansu of al th,e no*gg for pay (aged 15 to 54). Eadudes unpaid famly workr. b. In acm Boliv Note: Labo force paiticiption rnt wu .65 for mn, .44 for vAxen, and .56 overall. Sandad deriatio in parehe1es. Source: Bolivia: Intead Household Survey, 1989. Women in the Laor Force in Bolivia: Panico n and Farnings 27 Only two people in the sample (both males) h3ve no education of any sort. This is somewhat surprising given the levels of illiteracy found in Bolivia. Urban iliteracy was calculated by INE in 1988 to be almost 11 percent with 13.8 percent of the population having no formal sdooling. Given the exclusion of people over 55 from tle sample, one would expect illiteracy rates to be lower in the sample than the population. This cannot be the full explanation, however, as 11 percent of those aged 20-55 years have had no formal schooling (Instituto Nacional de Estadistica, 1989). It is not clear why people with no formal schooling are so underrepresented in this sample. The underrepresentation of people with no formal schooling biases the level of female educational achievement more than men as a greater proportion of women are illiterate. Slightly nore than 70 percent of the sample speak only Spanish while 28 perment speak both Spanish and an Amerindian language (Aymara, Quechua or Guarani). One percent speaks Spanish and another non-indigenous language. The most striking difference in the sample is the large gap between the earnings of working men and women. Average weekly earnings for men are Bs. 110.5 and Bs.68.9 for females. These figures are consist nt with previous estimates.2 On average, women earn slighdy more than 60 percent of men's earaings on a weekly basis. Women also work fewer hours, on average 7 hours a week less than raen. If women worked the same number of hours per week as their male counterparis they wo.ld earn 80.1 bolivianos, less than average male earnings. 4. ne Determinants of Female Labor Force Partidpation Whether women participate or not depends on their reservation wage - i.e., the value of their labor in the home. When this reservation wage is below the market wage women will participate in the labor force. If a woman's reservation wage is higher than that found in the market, she will not participate. This unobserved reservation wage means that female earnings functions esimates using ordinary least squares (OLS) will be biased. Only the market wage is observed ard thus the OLS earnings function will suffer from the problems inherent in using censored samples. Heckman (1979) provides an estimation technique to correct for this selectivity bias. First, a labor force participation finction for women is estimated. The inverse Mill's ratio (Lambda) from this equation is then entered on the right-hand side ef the earnings function equationL This corrects for selectivity. Thus, the first step in estimating female earnings function is to specify a model of female labor force participation. Labor force participation among women in Bolivia is low. Only 44 percent of the present sample of women work for pay. As noted above, the definition of working women may underesimate the real female work force as unpaid family workers are excluded from the definition. Also excluded from this definition are the unemployed. This is not a significant mnmber of people (Istituto Nacional de Estadistica in 1988, calculated urban unemployment rates for women to be I Horton (1989) provides estimates of male and female labor earnings as a percw of the average eanings for all workers. In 1982, male earninp wmr 102.6 perent of the avaage and female earninps wer 9S perent. The equivalent percentages in 1987 were 120.2 and 68.6 percent for men nd women respecively. Similar calcuaons based on the data usedhe aow menearning 1.4 peentofaverage _amings and wome 70.S percent 28 Women's Employmenz and Pay in Latin Ameria less than 2 percent of the labor force). The advantage of excluding the unemployed is that there are definitional problems involved in the measurement of this group.9 The dependent variable in the participation function is a dichotomous variable which takes the value of one if the woman is working for pay and zero otherwise. A probit function is used to estimate participation rates. The regressors measure personal characteristics of the individual woman, her family and socio-economic characteristics, and area of residence. Personal eharacteristics inelude educational level, age, health and fertility patterns. Schooling is entered as a series of dummy variables for each level of schooling. Note that for secondary technical, post-secondary technical, teacher's college, and university, no data were available for the number of years of the level compleW by individuals who are still students. Hence the dummy variable for these four education categories takes on the value of one if a person has completed the level or is presently a student in that level of education. The education coefficients are expected to be positive as it has been shown that, while increased schooling increases both the asking and the offered wage, the latter increases more (Heckman, 1979) . Age is entered in the participation function as a series of dummy variables (in five year ranges) to take into ac=ount any non-linearity in the relationship of age to participation. It is not a clear a priori what the signs of the coefficients of these age variables will be. Some evidence exists showing that younger women are more likely to be unemployed (Isstitww Nacional de Estadistica, 1988) which would dcrease the probability of this group's participation. Where enrollment in higher education is high, labor force participation will also be lower. Given the extremely low rate of higher education enrollment among women this will have little effet. Ethnic origin may be an important variable since d`fenant cultures have different attitudes towards paid employment for women and have differwd opporunities in the society." A proxy variable for ethnicity, language(s) spoken, is used here. Four dummy variables are used for language indicating whether the person speaks; (1) only Spanish; (2) only Amerindian languages; (3) both Spanish and one or more Amerindian languages and; (4) Spanish and another language. It is expected that those who do not speak Spanish will have a lower probability of participation. Health will affect a woman's decision to participate in the labor market (Behrma.n and Wolfe, 1984). The proxy for health status used here is number of days (out of the previo4s month) that a person has been incapacitated, i.e., physically unable to carry ou her regular civities. Fertility is especially important in the Bolivian context. Given the high infant mortality rate in Bolivia, the number of pregnancies a women has had will not be strictly correlated with the number of children she is raising. Hence, fertility is measured by two variables, the number of childcen that the woman has given birth to, and whether the woman has been pregnant in the last year (measured by a dummy variable). It is expected that women who are, or have been, pregnant recenly wil be less likely to be in the labor force. It is argued that womentend to leave dhe labor force in order to have children, retLrning when their children are grown or at least able to take care of themselves. Thus, the number of children under age 6 (preschool age) 9 For exan=*L One mesurement of the unemployed counts sa unemployed only toe actively erhing for work (McFalane, 1988; and ILO, 1982', whilo anotber includas al unemployed all thos who say they want to work. Obviously, the definidion used wifl affec one's results. to A recent World Bank study (1990s) argues tat therp is evidence of ethmcally based discrimination in Bolivia. Women f the Labor Force in BoRvia. Paricoadon ad Ear7nigs 29 is expected to lower a woman's probability of particip ating. 'rhis especially true where diere are inadequate childcare facilities, as is the case in Bolivia (World Bank, 1989). Also included in the equation is a variable measuring munber of children between the ages of 6 and 14 a woman has who are living in her household. Given the lack of universal partizipation in primary school in Bolivia (net primary enrollment in 1988 was only 83 percent (World '3ank, 1990b)) the number of children a woman has to care for in this group is also expeced to lowzr her probability of participating. Total family size has also been included in the equation. The sign of the coefficient on this variable may be either positive or negative. On the one band, a larger household may have a positive impact on participation because of greater demands for income or the presence of non-working adults who can provide chOldcare. On the other hand, the size of the household may raise the value of the womar.'s household production activities and reduce the probability that she will paiticipate in the labor force. A woman who heads a household is more likely to work for pay. Married women will be expected to have a lower probabOlity of participation than unmarried wemen. The income of a married woman's spouse is also included in the equation. It is expected to have a negative impact on participation. The final regressors in the equation are those having to do with geographic location and socio-economic status. It is expected that different areas of the country (highlands, valleys and tropical p!ains) are associated with different probabilities of participation. The tropical plains area of the country is experiencing the greatest growth in population though immigration and participation rates will be higher in this region. To capture the effect of socio-economic status, variables measuring home ownership (a proxy for family wealth), total family income and acces to public water and sewage disposal are included. (The latter two are, to some extent, proxies for the value of the home.) The effect of wealth on participation can be either positive or negative. There is less need to earn income if one's Emily has a certain level of wealth. But, also, it has been shown in other countries tha relatively bete- off women are more likely to work outside the home than lower-income women. The results of the participation function are presented in Table 2.4. The age category omiaed from the equation is from 15 to 19 years. As can be su in the table, ai other age groups are more likely to participate than this youngest group. All of the age variables are significanL As can be seen from the simulations presented in Table 2.5, the participation rate for women peaks in the 35 to 39 age range. The 40 to 44 age group has similar participation rames but the probability of participation declines for older women, possibly because retremt for women covered by social security is age 50. Other personal characteristics are also significant. Married women are less likely to work than non-married women and those who are heads of housebolds are more likely to work. Intestingly, the number of children under six that a woman has does not have a significant impact on labor force participation. This may be becmas women in the informal sector are able to either arrange their work scbedule to fit childcare needs or are able to take their children with as them they work. A woman's health status does have a significant negative impact on the probability of participation. The effect, however, is quite small. As expected, women who are presently students are less likely to participate in the labor force than non-students. Unexpectedly, however, educational attainment at the primary, middle and secondary school levels does not have a significant impact on participation. There are two 30 Wwmin 'r EAIOYvw and Pay i. LAR Ameica Table 2.4 Probit Estimatc for Female P cipation Variable Coefficiot t-rtio Partial Derivative Contant -0.360 -2.37 Age 20 to 24 0.290 2.35 0.11C Age 25 to 29 0.506 4.10 0.199 Age 30 to 34 0.650 5.12 0.256 Age 35 to 39 0.756 5.81 0.298 Age 40 to 44 0.742 V.43 0.293 Age 45 to 49 0.600 4.22 0.237 Age 50 to 54 0.391 2.54 0.150 Married -0.390 -6.15 -0.154 Studzt -0.249 -2.92 -0.098 Household Size -0.019 -1.55 -0.008 Finid Primy 0.017 0.27 0.007 Some Middle Scbool -0.002 -0.04 -0.001 Finished Middle School 0.062 0.75 0.246 Some Seconday -0.09 -1.57 -0.038 Finished Secondary -0.094 -1.41 -0.037 Seondary Technical 0.225 2.55 0.089 Higher Technical 0.424 3.27 0.167 Teacher College 0.729 9.13 0.288 Univerity 0.279 3.24 0.110 Incapcity -0.007 -2.01 -0.003 Live Births -0.001 -0.18 -.001 Home owned -0.003 -0.08 -0.001 Head of Household 0.660 7.80 0.260 Valley 0.025 0.57 0.010 Tropical 0.068 1.35 0.027 Public Water Supply -0.080 -1.68 -0.032 Public Sewer 0.044 1.01 0.018 Pregnant -0.178 -3.65 -0.070 AyumsQuedma -0.570 -2.31 -0.225 Bilingoal 0.156 3.47 0.062 Bilingual Other -0.074 -0.35 -0.029 MU= umder 6 -0.028 -1.18 -0.011 Cbldrn 6 to 14 0.026 1.06 0.010 Income of spomP -0.000 -1.81 -0.0002 Family mcom 0.000 0.53 0.00004 Houseld Dz -0.019 -1.55 -0.008 Notes: Sampk n women aSed lS to 55 yers Particii Rate: 44.0% Lo-LftMood - -3488.2 N = 5624 Womwn in d Labor Force in Bovi. Parioa wad Ew4igs 31 Tabe 2.5 Clarcterisc Pdicted Probiba ity Age: 15 to 19 23.6 20 to 24 33.4 25 to 29 41.5 30 to 34 47.2 35 to 39 51.4 40 to 44 50.9 45 to 49 45.2 50 to 54 36.7 Family nat±us Married 41.0 Unmarried 56.5 Head of bousebold 67.0 Not head of household 41.3 Language Spokaz. Only Spanish 42.2 Only Indigeoua 22.2 BiinRtal Span/indig 48.4 Bilin, ual Span/other 39.4 Scho'Iing Sonday T.chnical 50.0 Higher Technical School 57.8 Teah College 69.2 University 52.1 Stad,f Suts Is a studat 34.8 Not a stden 44.4 Pegancy: Pregni this ye 38.5 I.bt pregnant this year 45.4 Days ihqacitWted Zero dqs icapcitaed 44.3 Number of days = .5ezn 44.1 Number of days maa 43.9 Number of days - 1.S*mean 43.7 Number of days - 20men 43.5 Notes: Mea Pnrtic4auio RaPe 44.0% Sinulatio based on probi reauk in Table 2.4. 32 Women's Eirplykmen wJ Pay in LinA Arm=ca possible explanations for this. First, women are concentrated in the informal sector where the value of education credentials are limited. Second, and perhaps more importantly, Bolivia maintains two separate school systems, one rural and one urban. The curriculum and the quality or the two systems are very different (World Bank, 1989b). Hence, women who have been educated in rural areas will have received very different training. It is no possible to determine the type of schooling a person received and the education variables for these levels may reflect differences in quality wbich confound the analysis. In contrast, attendance in, or completion of, technical school (either secondary or post- secondary), teachers' colleges. or university has a highly significant, positive effect on probability of participation. The type of schooling with the greatest impact is that of teacher training. As can be seen in Tab'e 2.5, all else being equal, a woman who attends a teachers' college has a probability of participating close to 70 percent, over 25 percentage points higher than that for a woman with the average level of education. Interestingly, technical education at the post- secondary level has a greater impact on participation than does a university education (57.2 percent prob2hkility versus 52.1 percent, respectively). Pregnancy has Y:ie expectod negative impact on participation. Women dho were pregnant in the given year had a probability of participation of 38.5 percent. Women who hzd not been pregnant in the preceding twelv4 months had a probability of participating of 45.4 percent The number of children a woman has given birth to has an insignificant impact on participation rates. Et nicity also preves to have a significant impact on labor force participation with those women who speak no Spanish having the lowest participation rates (22.2 percent, all else held constant). Bilingual womea participate at a higher rate than women who speak only Spanish. It should be noted that edtnicity and income levels are highly currelated in Bolivia (Worid Bank, 199-)a) and the effects of et"nic origin are probably partly reflecting economic stts. lhs variables measuring family wealth and socio-economc status (own home, public water and public sewage) have insignificant effects on the probabiity of labor force participation. On the one hand, this may reflec. the contradictory effects of wealth on participaion indicated above. On the other hand, home ownership may not be a good proxy for wealth in Bolivia and access to public water and sewage may be more a ftinztion of urban 'iying than socio-xcnomic statos. Neither the total filmily income nor the spouse's income hive a significant impact on participation. Contrary to expectations, the regional variables are not significant Like the variables for source of water and sewage disposal, the fact that the sax,;e is urban may account for this lack of impact of geography on labor force participation. In summary, the most important effects on female labor force participation in Bolivia are schooling, student status, age, language spoken, marital stats, pregnancy, and number of children between ages 6 and 14. Women between the ages of 35 and 39, those who are unmarried, andlor bilingual, andlor highly educated are the most likely to participate in the labor market. 5. Eanrinps Functions The standard earnings function is the Mincerian (1974) formulation of schooling, experience and experience squared regressed on the natural log of eamings: W=m in & Ubor Forco in s '4 PaP *Oad= and F1rn Vg 33 LnY= b+ SS +b2EX + b (I) V/here: LnY = the na±rral log of weekly earnings S = years o' schooling EX = experience EX2 = experience squared To standardize for hour; worked, the natural log of hous (LNH) was entered on the right hand side of the equation. It should be kept in mind duhs the experience vart^le used here is actually potential expeience (I.e., age - years of scoloking - 5). Po*tal experience will closely approximate real experiene for any peson who ha woried .edily since s/he left school. However, as women are mcre likely to move in and out of the labor force, the potential experence variable may overesimate women's ctal experiece. For female earnings, can the model answer the questions abou the effet of huma capital on earnings for women in the labor force? The sample adinthe eringfi. tions includes the entire male and female working populaion (i.e., those working for pay). Three s,Arate specifications were used: One for ;ales, one for fanales not correcn for selivity and one for females correcting for the selectivity bias. The results of the three earnings fuctions are shown in Table 2.6. All of the variables have n significant effeia on eamrnip for both men and women. Tbe impna of schooling is less for women tha men aldtough de differece is ot great. Experience and experience squred have much less of an imt on eaninp for women than n'.. The number of hours worke 1, however, has a greater impact on womea's. Tne returns to schooling are ipifican, but no as high as lthose fond in other Latin American countries". To some ext this refecs the size of the informal secor in the economy where education has less of an impad thn in the privae formal ad/or public sectwr. Returns to experience and exerienme squared are also low, perhaps for the same reasor. Only hours has a high reur (relative to men) impa-t The inve Mill's ratio is no significant 6. Discimination As the sample characristics pr in Table 2.3 demonsre, there is a substantial gender- based difference in average weekly earnings. By the two eanings equations (the male and female (unorected) equatiotS) it is posle to doemine what percent of this diffence is due to the ifferet endowme levels of the two groups and what is due to the way in which each groups' are valoW! in the market plac (Oaxaca, 13). The initial difference between the earin of men and woen can be e as: LnY, - LnY, = Xb. - Xt (2) Algebraic m lations give the following equato LnY, - LnYf = b,(X.-:) + X(b.-b) ' See .& dtspta8 ind di v 34 Wo.wnEn 's 3EsE a anid Pay b 'Lai AmAr5a Tible 2.6 Earnings Functions Woring Working Women Women Vrinbie Workdng (uncorrected (conrretd for Men for electivity) selectivity) Constant 1.575 1.346 1. 235 (14.36) (10.31) (7.69) Sdlibn (years) 0.071 0.063 0.065 (26.23) (17.08) (16.16) ENgPaielse 0.050 O.028 0.031 (13.10) (4.06) (4.22) EinF3de˘s4 Squarled -0.00! -O.-0. 3 -0.0004 (-7.24) (-2.39) (-2.61) Ln(esmkby hours) 0.354 0.424 0.424 (13.91) (17.18) (17.17) V.ag- 0.069 - - (1.19) R-sq=sn>U 0.179 0.176 0.176 114I 5,314 2,472 2,472 Not= Numb in parertAe= ar t-m:ios Depcsdum variAble is Ln(weckly eumings) An index problem arises here. There is no theoretical reason to prefer the above specification to tha fbllowing Ln-Y - LanY = bX-X) + X.(b.-b) (2b) Bo:1t specificntiou re used anJ the results are presented in Table 2.7. The first term e s the actual difference in endowment levels between the two groups. The second term on the right-hand side measures the difference in the market evaluation of cndowmets. Typically, the value of the second term is considered to be a measure of discrimiaioDi La fe market place. Tne firsi term is considered to be the explained difference in wages due to unequal endowments. The unexpained dihfence in wages calculated here is the upper bound of labor market discriminat In other words, this is tie maximum level of discrimination. Controlling for unobserved productve characteristics of the two groups could lower tiis 'uneplained' segment of the equatio It shoud be remembered, however, tha the upper bound calculated here may u It scoiU be notd tfrz tbe. diffuence in cndowuments ry, in fsct, reflect pre-muark diwTiminioaL 12 Schuhz (119) egues tha the inobsved caractercs wl create substsntial unainty intergroup Of emkings. womten in Lfte Labor Force in Bolivia: Partici4atiox and EarRings 35 underestimate discrimination if pre-market discrimination keeps women from obtaining human capital. As can be seen in Table 2.7, of the total earnings differential, approxi,nately 15 percent is due to women having lower levels of human capital endowments than men. The remainder of the observed wage differential is due to men's endow-nents receiving a higher price in the labor market. This is the upper bound of discrizinatioti. Depending on the specification used, around four-fifths of the observed wage differential between men and women is due to unexplained market pricing mechanisms. Table 2.7 Decomposition of the Male-Female Eamings Differential Percentage Due to Endowments Rewards Pquation 2a 14.9 85.1 Equation 2b 24.1 75.9 (WageM,/Wagef = 160.4%) 7. Conclusions and Recommendations The results of this study show that there are many personal and family characteristics which affect .he probability of a woman participating in the labor force. Schooling, age, marital status, status as head of houschold, pregnancy, and ethnic origin are some of the characteristics which mOSt affect the decision to participate. Proportionally fewer women than men participate in the labor force. Those women who do participate earn substantially less per week than their male counterparts, even when hours worked are equal. Part of this difference in earnings is attributable to lower endowments of human capital among women. Yet most of the observed differential is due to unexplained differences in the way in which the market values the two genders' labor. Clearly, the legal efforts underway in Bolivia to eliminate gender discrimination are a necessary step to improve women's position in the labor market. Tne results of the labor force participation funaion and the earnings functions provide information on futher areas whcre governme policies can have an impact on the labor market activities of women. Tle obvious first step to decreasing the earnings gap is to increase women's access to humann capital formation. Increasing female education levels will have a twofold effect - it will tend to increase the labor participation rates of woman and it will raise the incomes of women who work ouitside the home. It should be remembered here that increasing women's earnings will not only a&sist individual women but also their families. As evidence exists that single-headed households are becoming more common this takes cn ad; d importance. 36 Women 's &pbyzm and Pay 63 Latin Arweka Further increases in female erninp could oDme about from policies designed to increase productivity in the informal sect=. lbis sectr typically has access to very little capital and is constrained by the low levels of technology used. Programs designed to increase credit availabUity for smal businesses in the informal sector and provide technical assistance would benefit the economy as a whole. In the proes, the specifc beefits to women could be large since a sizeable percentage of women working in urban areas are in this sector. Without further information about the reasons why male labor is valued so much more highly than female labor it wilL be difficult to affect he rm er of the observed wage differential in Dolivia. But polies aimed at improvin the legal environment in which women work, increasing their access to education, and assisting the Informal sector can all have a significant impact on women's participaton rat and earnings. References Arteaga, V. "La Crisis Economica y sUs efectos sobre la mujer en Bolivia.' 'aper presented at the United Nations' Childrens' Fund Simposio Nacional sobre Mujer y Necesidades Basicas. Bolivia, June 1987. Behrman, J. R. and B. Wolfe. 'Labor Force participation and Earnings Determinants for Women in the Special Conditions of Developing Countries, Journzl of Development Economics, Vol 46 (1984). pp. 259-288. Casanovas, R. 'Los Trabajadores pro Cuenta Propla en el mercado de Trabajo: El Caso de la ciudad de La Paz, Boliva.' Paper presented at the International Seminar "E Sector Informal Urbano en America Latina y el Ecuador'. Quito: ILDIS/CEPESIU, 1984. Hecknan, J. 'Sample Selection Bias as a Specification Error.' Economerrica, Vol.47, no. 1 (1979). pp. 153-161. Horton, S. "Labour Markets in an era of AdjustL .X: Bolivia.' (Incomplete draft). University of Toronto, 1989. Mincer, J. Schooling, Exrience, and Earnings. New York: Columbia University Press, 1974. Oaxaca, R.L. *Male-female Wage Differentials in Urban Labor Markets.' Internauinal Economic Riew, Vol. 14, no. 1 (1973). pp. 693-709. Romero de Aliaga, N. 'Me Legal Situaton of the Bolivian Woman.' Fletcher School of Law and Diplomacy, La Paz, Bolivia: International Advisory Committee on Population and Law, 1975. United States" Bureau of Labor Statistics, 'Labor Law and Practice in Bolivia." 1962. Report No. 218. Washington, D.C.: United States Department of Labor, 1962. World Bank. 'Bolivia: Povery Report.* Report No. 8643-BO. Latin America and Caribbean Region, Country Operations Division I. Washingto.a, D.C.: World Bank, 1990a. World Bank. World Dewlopment Report, 1990. Washington, D.C.: World Bank, 1990b. World Bank. *Social Spending in Latin America The Story of the 1980s.' World Bank: Discussion Pape Series No. 88. Washington D.C.: World Bank, 1990c. 37 38 Wmwn ts Z*oym" and Pay in Lavn Americ World Bank. "Country Assessment of Women's Role in Development: Proposed Bank Approach and Plan of Action.' R.N. 8064-BO. Latin America and Caribbean Region, Country Operatios Division 1. Washington, D.C.: The World Bank, 1989. 3 Labor Force Behavior and Earnings of Brazilian Women and Men, 1980 Morton SJelcner, J. Barry Smith, Jon A. Breslaw and Georges AMonere' I. Introduction Interest in the role of women in the development process of Tlird World countries increased during the last decade. A major concern is the status of women in what Behrman and Wolfe (1984) term the 'special conditions' of labor markets in developing countries.2 Increasing attention is being given to analyzing the labor force behavior of women and to their returns to human capital, especially education. However, the volume of research on these women is considerably less than that for men in developing countries and for women in industrialized countries. This study reduces this gap by examining labor force patterns among married and single women, and compares these with their male counterparts. We present an analysis of the labor force behavior of Brazilian women and men using a sample of 53,000 drawn from the 1980 Census. The specific concerns of the study are the determinants cf labor force status (employee, self-employed, or no market work) and earnings of workers. We pay particular attention to the impact of education on labor market outcomes. Four population groups are considered: wives and husbands, single (never married) women and men. 2. Brazil: Economic Background An analysis of the labor force behavior of Brazilian men and women is important because it is representative of a developing country experiencing severe economic problems. As for most of its neighbors, the 1980s for Brazil, the largest and most populous country in Latin America, were a 'lost decade in terms of the severity and duration of the economic progress in the 1970s wthen per capita GDP grew on average by 1.4 percent per year in 1970-1973 and by 7.1 percent in 1974-1980. Brazil started the 1980s with much economic compro.nise and many successes, but a The authors thank Ana-Maria Arriagada for her invaluable help in providing the data base and clanfying its contents, and Linda Bonin for her editorial advice. Brenda Butler, a cotnputer science student at Concordia University, provided competent progrmming scsistance. We also thank Daniel M. Shapifo, Department of Economics, Concordia University, for providing helpful comment; and criticisms that shaped the final version of tb^- paper. 2 As Behrman and Wolfe (1984, p. 260) aptly state, the special conditions incluAe 'regional and swtoral pluralism, the relevance of human capital investments in health and nutrition, and distinctive determinants of opportunity costs for labor force participation.' 39 40 Women's Fpkrpnl and Pay in Latn Amsria early in the decade its economic performance began to deteriorate rapidly and dramatically, and has not yet ended. In 1981 per capita GDP fell by 4.3 percent, while for the rest of the decade the average annual (erratic) growth rate was about 1 percent. The economic prospects for the 1990s are not encouraging: in 1990 gross domestic product shrank by 4.3 percent, reflecting a decline of 8 percent in industrial production. A question that has been receiving increasing attention is: How have the worsening economic conditions, on top of persistent poverty, affected peoples' welfare in Latin America? Since market work is the most important income source, an analysis of the labor force behavior of Brazilian men and women, just prior to the onset of the economic difficuties, should shed light on how the economic crisis will affect structural changes in the labor markets and provide insights about the role of factors such as education in this process. This chapter examines the situation in 1980, while the next chapter considers the conditions in 1989. The vastness and diversity of Brazil and its well-documented substaatial regional economic disparities led us to perform the analysis for six distinct economic regions as well as the entire country. We thus reexamine the issue of "geographical aggregation bias' or 'regional pluralism' in assessing returns to human capital, and how they differ by types of employment and by gender.3 A novelty of this study is that it examines the determinants of wages' by explicitly incorporating the selection (self or otherwise) of persons into three types of labor force status. As cogently argued by Schultz (1988), it is important to assess whether work-status effects bias the parameter estimates of wage functions. Few studies of labor markets in developing countries analyze the impact of sample selectivity and adjustments in earnings regressions, especially when three types of labor force status are involved. Tle structure of the paper is as follows. Section 3 discusses the data used in the study. Sections 4 and 5 summarize the model of labor force status and wage deerminants, and include discussions of the theoretical characterizations and econometric specifications. Section 6 reports ; empirical findings and their interpretation. The final section contains the conclusions. 3. Data and Stylized Facts The models used in this study are applied to data originally taken from a public use sample tape (PUST) of the 1980 Brazilian Census. The tape, which represents a 3 perct sample, contai 3,526,000 individuals and about 800,000 households. The PUST was used to extract a subsample of 200,000 individuals and about 40,000 households. From these data, we culled samples of The issue of regional pluralism iexamined for Bazil by Birdall and Behrm (1984) with 1970 Ceasus data, for Nicaragua by Behrman end Birdsall (1984), and for Panama by Heckman and Hotz (1986). Unfortunately, the are very few sAdies that examine nme-feanle differences in labor mauet outcomes for developing counxtri especially in the context of regional pluralism 4 We use the terms 'wages' and 'euarings intehangeably throughout the paper. Also, 'hourly eaniings, *wage or wagoe rat are considered to be synonymous, as are "employee" and wage earner.0 Labor Force Behavior and Earnings of Braziian Women and men. 19S0 41 individuals who were generally between the ages of 15 and 65 years.5 This resulted in a sample of 28,926 wives and husbands, 11,'Z25 single (never married) women, and 12,974 single men.6 Although the Census data provides much useful information, it has a principal d;sadvantage: Incomplete information on the number of hours worked.7 Data are provided on monthly earnings, but not on monthly hours worked, nor hourly earnings. In.'lre3d. we are given data ou weekly hours of work in intervals: 0 hours (non-workers), 1-14 hours, 15-29 hours, 30-39 hours, 40-48 hours and 49 or more hours. An index of continuous hours worked is desirable before applying standard statistical techniques in estimating the wage function. The study uses hourly earnings as the principal measure of remuneration because, as discussed in detail by Behrman (1990), Blinder (1976), and Schultz (1988), it is inappropriate (especially for women and the self-employed) to use monthly earnings (monthly hours times hourly earnings) as the dependent variable. To do so may confound wage rate effects andi labor supply effects which, in turn, may bias the parameter estimates of the returns to wage-determining characteristics. The direction of the bias depends on whether the labor supply curve is normal or backward-bending. Moreover, as stated by Blinder (1976), the expedient of adding the log of hours as a regressor leads to 'strained interpretations' of the parameter estimates of the earnings function. The hourly earnings measure we use, however, is marred by incomplete information on bours worked. The data provided no direct measure of hourly earnings, but we could calculate the hourly rate from monthly earnings and hours worked during the weed of enumeration. As mentioned above, there was, however, a problem - information on weekly hours wcrked was available in intervals only, and no information was provided on the number of weeks per month that a person worked. In earlier studies of Brazilian labor markets, Bii Isall and Fox (1985) and Dabos and Psacharopoulos (1991) used the 1970 and 1980 census data, respectively, to construct a continuous hours variable by assigning valucs (usually mid-point) to each work interval.' This procedure is likely to create a measurement error and the problem can be thought of as an errors in variables problem in which parameter estimates are likely to be inconsistent and biased. The direction of this bias is not known. Nevertheless, we use their mid-point values, but note the caveat in interpreting the results below.' 5 An exception is made for married n. Tlere were 1,441 married men over the age of 65; about 40 perccnt of them worked, usLaly over 30 hours per week. ' The single (never married) m-f and woman in the sample are relatively young. The averange age of single mn and wo- is 25 yews and 21 years, respectively. About 35 percent of single men are under 20 years, and 40 percent ar between 20 and 35 years of age; the corresxoding proportios for women are 58 percent and 30 percent. Also, 38 perct of women reported that they were still ttending school on a part- or full-time basis, as did 28 percent of the m. ' Unfortunaely, the daa base also omitted information on migration for marnied and single men. In addition, for single m- it failed to include information on household charateristics, non-labor income, and whether the peron worked in the public sector. I Birdsall and Behrman (1984), who also used the 1970 Census data in their study of Brzilian men, make no mention of this isue. They do not indicate whether they use hourly or monthly earnings. As pointod out by Ham and Hsiao (1984), labor economists and econometricians have given insufficien attention this important question. 42 Women 's Emnploy,ment and Pay in Lxin America Imponfan Chuuadenistics of the Da.a To provide a background for the study, we present detailed descriptive statistics in Appendix Tables 3A.1 to 3A.6. There are substantial regional disparities, gender and work-status differences, and differences between married and single people in the summary information. These features, as well as results from other studies of Brazil, strongly indicated that the study should treat each of the economic regions separately: the Northwest (North plus Central-West), Northeast, South, the Southeast (the states of Rio de Janeiro and Guanabara, Sao Paulo), and Other Southeast states (Minas Gerias and Espirito Santo).'° For each region, we estimate the models of lalor force status and wage determination for four demographic groups, married women, marr.ed men, single women and single men. This approach allows one not only to detect regional differences, but also to draw comparisons among the demographic groups. While a complete discussion of the information in the Appendix tables would be too lengthy to present here, we do examine the more important stylized (i.e., abstracred for the purpose of the analysis) facs about Brazilian labor markets. These stylized facts provide a useful background fcr the empirical analysis. Regional d4prities.'1 There are acute economic disparities in Brazil which had a population of about 121.3 million in 1980. There are sharp differences among the highly industrialized and modem southern regions (Rio de Janeiro, Sao Paulo, Other Southeast, and the South), tht Northwest (where the capital city is located), and the Northeast, which is heavily dependent on agricultural activities. Tle southern regions have about 60 percent of the population, and 17 percent of the land mass; the Northeast accounts for 30 percent of the population and 19 percent of total area, while the vast Northwest has about 10 percent of the population. Although there are pockets of poverty in the southern regions (m Rio de Janeiro, Minas Gerias, Espirito Santo, and Santa Catarina), they are the relatively prosperous parts of Brazil, while the Northeast is the poorest area, whatever indicators one wishes to use. Looking at Appendix Tables 3A.1 and 3A.2 (data for married people), we obtain a good indication of the intensity of these regional disparities. All monetary units are in 1980 cruzeiros.'2 Total monthly employment earnings of husbands in the Northea:t is $8,020 compared to $12,790 in the Northwest, $11,730 in Other Southeast, $18,970 in Rio, $20,170 in Sao Paulo, and $13,020 in the South. The pattern for monthly asset and transfer income is similar - $2,650 in the Northeast, $3,620 in the Northwest, S4,440 in Other Southeast, $',,9M0 l0 The stat*s in each region are: North: Rondonia, Acre, Amazonas, Rorain, Pera, Amapa Centrnl-West: Mato Grosso, Goias, Distnto Federal Northeast: Maranhao, Piaui, Ceara, Rio Grmnde do Norte, Paaiba, Pernambuco, Alagoas, Sergipe and Bahia South: Parana, Rio Grnnde do Sul, Santa Catarina Southeast Rio de Janeiro and Guanabam, Sao Paulo, Minws Gernis and Espinto Santo § We cautioa the reader that we did not take spatial price variaions into account when ma1ing regional comparison because we were unable to obtain geographical price indices. For fuiher discussion of this issue see Birdsall and 3ehrman (1984) and Thomas (1980). '2 In 1980 the official excbange rate was about 40 cnzeiros = $ (US). The monetary units used are crueir expressed with the symbol S. Labor Force Behavior aAd Earnings of Brazilian Women and Men, :980 43 in Rio, $8,760 in Sao Paulo, $4,280 in the South. In brief, family income is much lower in the Northeast than in the rest of Brazil. A similar pattern is revealed when education is examined. In the Northeast, the average years of schooling for husbands and wives are 2.2 and 2.3 years, respectively. The corresponding values for the other regions are: Northvwest: 3.5 and 3.4 years; Other Southeast: 3.4 and 3.4; Rio: 6.1 and 5.4 years; Sao Paulo: 5.0 and 4.5 years; and South: 4.1 and 3.9 years. There are also some noteworthy regional differences in labor force participation patterns. In the Northeast, Northwest, and Other Southeast, wives' labor force participation rates are in the 82-85 percent range, while in the remaining regions they are in the 76-78 percent range. As regards market work activities, in the Northeast just over one-half of working wives are self-employed. In the Northwest, the fraction is one-third, and it is about 37 percent in the Other Southeast and the South. In the urbanized regions of Rio and Sao Paulo, only 20 percent of working wives are self-employed. This pattern is similar for husbands. Over 60 percent of working husbands in the Northeast are self-employed. The proportion drops to 53 percent in the Northwest, to 45 percent in the Other Southeast and in the South, to 29 percent in Sao Paulo, and to 23 percent in Rio. The above regional configuration for earnings, education and labor force participation also prevails among singles. For example, consider the data on single women (Appendix 3A.4). In the Northeast, the average monthly earnings of employees and the (paid) self-employed are about $4,900 and $2,000, respectively. The corresponding values for the other regions: Northwest: $5,200 and $4,400; Other Southeast: $4,600 and $4,100; Rio: $8,300 and $12,200; Sao Paulo $7,800 and $11,100; and South: $9,200 and $4,500. As regards education, the average years of schooling among single women is 4.1 years in the Northeast; 5.2 years in the Northwest; 5.4 years in the Other Southeast; 7.4 years in Rio; 7.1 years in Sao Paulo; and 5.9 years in the South. Finally, we note regional differences in labor force participation. In the Northeast, cnly 30 percent of single women are in the labor force. This compares with 33 percent in the Northwest, 37 percent in the Other Southeast, 41 percent in Rio, 59 percent in Sao Paulo, and 44 percent in the South. The types of market work performed by single women also differs among regions. In the Northeast, 63 percent of working women are employees. This proportion rises considerably in the other regions -87 percent in the Northwest, 90 percent in the Other Southeast, 93 percent in Rio and in Sao Paulo, and 77 percent in the South. Much the same story can be told about single men (Appendix Table 3A.5). Gender and marital status dAferences. It should come as no surprise to find that there are notable gender and marital status differences in labor force activities and outcomes. Frst, as noted above, the labor force participation rate of men is much higher than that of women. The participation rate of single women exceeds that of married women. Tnere are also interesting differences in the types of jobe, and, among workers, in earnings, hours of work, and wage- determining characteristics (education, experience). Bearing in mind regional differences, the main featres are as follows. Job composition.. Te job composition of men and women differs. The proportion of women workers who are employees generally exceeds that of men. Nationally, about 65 percent of working wives and 83 percent of single women are employees compared to 43 percent of husbands and 75 percent of single men. A regional breakdown shows that this pattern is largely sustained. In the Northeast, 39 percent of working husbands and 44 percent of single men are employees compared to 48 percent of working wives and 63 percent of single women. Tne proportions for the Northwest are: husbands, 47 percent; single men, 66 percent; wives, 67 44 Women ' Empyloymena and Pay ir Loan America percent; and single women, 87 Dercent. The male-female differences are somewhat smaller in the remaining regions. In be Other Southeast, 64 percent of working wives and 90 percent of single women are employee while 56 percent of husbands and 77 of single men are employees. In Rio, the proportions for A -es and husbands are about the same, about 80 percent, while for single men and women, the proportions are 86 percent and 93 percent, respectively. In Sao Paulo, about 80 percent of wives and 70 percent of husbiuds are employees, as are 93 percent of single women and 87 percent of single men. In the South, 63 percent of working wives and 54 percent of husbands are wage earners, while the proportions for sirinle men and women are about the same at 75 percent. Earnings. Looking at Appendix Table 3A.3 (for married people) and Appendix Table 3A.6 (for singles", we see that wcmen, on average, earn less than men, and also work fewer hours. Nationally, the average monthly wage of husbands who are employees is $15, 100 and $9,200 for wives; the values for single men and women are S7,800 and $6,400, respectively. Thus, the wife-husband earnings ratio is 0.61, while the single female-Aale ratio is 0.82. Among the (paid) self-employed, monthly earnings of husbands and wives are $14,600 and $7,900, respectively, yielding a ratio of 0.54. Single self-employed men earn $7,500, while their female counterparts earn $4,500, for a ratio of 0.60. These indicators of female-male earnings disparities rmLst be treated with caution because women tend to work fewer hours than men. Accordingly, it would be useful to take labor supply differences into account by examining differences in hourly earnings between men and women. This changes the story somewhat. The average hourly wage for employee husbands and wives is $34 and $63, respectively, yielding a ratio of 0.75. Tle corresponding values for self- employed workers is $84 for husbands and $55 for wives, for a ratio of 0.65. The hourly wage for single male and female employees is almost the same, $45 for men and $40 for women, so that the ratio is 0.89. Paid self-employed women earn about two-thirds the hourly earnings of 'elf-employed men (S43). Tlhese summary statistics on the male-female hourly earnings gap suggest thit working wives (employees or self-employed) are, on average, worse off than their huwbands. However, the gap between single men and women employees is much smaller, but the earnings differential among the self-employed remains large. We also note that the observed wage may differ from the wage offer, which takes into account labor force status decisions. In the econometric analysis we concentrate on male-female differences in wage offe:s. One should also note the regional differences in the male-female hourly earnings gap. For employees, the wife-husband wage ratio ranges from a low of 0.65 in Sao Paulo to a high of 0.92 in the Other Southeast. The ratio is 0.72 in the Northeast, 0.76 in the South, 0.80 in Rio, and 0.85 in the Northwest. Regional differences in the female-male earnings ratio of roughly 0.60 among self-employed spouses are much less pronounced. As shown in Appendix Table 3A.6 for singles, the average iaurly wage of women employees is about equal to that of men. In most regions, the female-male wage ratio is similar to the natioral average (0.89). In the Other Southeast, the ratio is 0.77, while in Sao Paulo and the Northwest, it is 0.81 and 0.85, respectively. In the Northeast and Rio, the average wage of single women exceeds thaL of single men. Worst off are single self-employed women whose hourly earnings are about 70 percent those of men. Labor Force Behavor and Earings of Braziliaa Women and Men, 1980 45 Educaion. We now turn to a brief summary of the differences in the principal variables that are likely to affect labor force status and wages. Nationally, as w. 11 as within specific regions and labor force status, education levels of working women are gentrally higher than those of men. For Brazil as a whole, the mean years of schooling for wives who are employees is 7.0, while for husbands it is 4.9 years. Single women employees have 7.4 years, while single men have 5.7 years. Similarly, self-employed wives have 3.4 years of schooling and their husband counterparts 2.9 years. The means for self-employed single women anit men are 3.3 years and 3.1 years, respectively. In each region, with few exceptions, the education levels of women employees exceed those of men employees by 2 to 3 years of school. Among the self-employed, women have an education advantage of less than a year. The relationship between education and labor force status is as expected. Generally, the educational attainnent of employees is higher than that of the self-employed, and the latter have more years of schooling than those not working. For example, the average years of school of non-working wives is 3.2 years compared to 3.4 years for self-employed wives and 7.0 years for employees. With minor variations this pattern prevails for men and single women in all regions. Age. As regards age, regional differences are relatively small. However, in all regions, wives and married and single men who are employees tend to be younger than their self-employed counterparts. For examprle, the average age of a married male employee is 38 years, and that of the self-employed man, 42 years. The corresponling values for wives are 36 years and 38 years, while for single men they are 24 years and 28 years. lu. pattern for single women differn. Nationally, the zverage ages of employees and the self-employed are about the same, 23 years and 22 years, respectively. However, in some regions (Northwest, Sao Paulo) self-er.ployed single women are older (25 years) than employees (22 years). In the Northeast, the Other Southeast, and Rio, the average ages are the same (22-23 years), while in the South self- employed single women are younger than employees (19 years and 21 years). It should also be noted that single women and men are mitch younger than married women and men. Feeriity and cild composidon. We next consider past fertility and the child composition of the household, which are particularly relevant to the analysis of married women. First we note that well over 80 percent of the married households in the sample are comprised of nuclear families that consist only of a father, a mother, and children. Ihe proportions are slightly lower for employee wives than for non-working or self-employed wives. The two exceptions are in the Northeast and the Northwest, where the proportion of nuclear families is about 70 percent. A Brazilian wife in our sample, on average, gave birth to about 4 oive or dead) children. The average is lowest among employee wives (3.0) and, surprisingly, highest among self-employed wives (4.7); for non-working wives it is 4.2 births. As expected, these numbers mask important regional differences. Fertility is highest in the Northeast and lowest in the three southern regions. In the Northeast, the averages for non-working, self-employed, and employee wives are 5.2, 6. 1, and 3.9, respectively. The corresponding values for the Northwest are 4.2, 4.6, and 3.3, while for the Other Southeast they are 4.4, 4.9, and 3.3. Fertility in the southern regions is lower. In Rio and Sao Paulo, the averages are almost the same: 3.4 for non-working wives, 3.2 for self- employed wives, and 2.5 for employees. The values in the South are 3.9 for non-working and self-employed wives, a .J 2.8 for wives who are employees. Now, we comment on child composition. The average number of each type of child - babies, toddlers, and school-age children - in the household is highest in the Northwest and Northeast. In these regions the average number of babies is about 0.56, which is much higher than that in 46 Women 's Emplo)wNt and Pay in Laun Amwrca the other regions-0.33 in Rio, 0.38 in Sao Paulo and the South, and 0.46 in Other Southeast. This pattern generally repeats itself for each child type and labor force staias. It is noteworthy that the average number of babies ainong non-working wives is much. higher than among workers. For each type of working wife the average number of babies is roughly the same in most regions. The average number of toddlers in the household (0.40) is about the same for non-working and self-employed wives, but lower for employees (0.30). Of course, there are regional differences. FJr example, in the Northeast and Northwest, the average number of toddlers is 0.50, while in the remaining regions it is about 0.34. Tlb profile of school-age children is different. The national average of self-employed wives (1.30) exceeds that of employees (0.93) and non-workers (1.04). As expected, there are regional differences. In Rio, Sao Paulo, and the South, the average for self-employed wives is 0.84, 1.10, and 1.19, respectively; for employees, 0.68, 0.88, and 0.88, respectively, and for non-working wives, 0.79, 0.85, and 0.96. The corresponding averages for the remaining three regions are: Northeast, 1.24, 1.52, and 1.14; Northwest, 1.16, 1.59, 1.11; and the Other Southeast, 1.14, 1.21, and 1.00. In contrast to married women, Appendix Table 3A.4 shows that the child composition for single women is diffent Single women may live alone and may have children, or live in households with chbldre%L However, in our sample of .elatively young women, over X10 percent are daughters or daughters-in-law of a male household head. The average number of babies is about one-half that for married womern (O.2J). With a few exceptions (Northwest, Other Southeast, and Rio), there is little discerniile difference in this magnitude between no.-working and self- employed worke-s. However, the value for emp!oyees tends to be smaller than that for self- employed, especially in the Northeast, the Other Southeast, and Rio. As regards toddlers, th,> national average is 0.24, and the mean is lower for employees (0.19) than for non-working (0.25) or self-en:!oyed (0.32) women. However, in the Northeast, Rio, Sao Paulo, and the South, the average is higher for self-employed women than that for non-working *romen, while in the remaining regions, the average is the same. The average number of school-age children present in the housebold is lower for married women (1.04) than for single women (1.35). There are fewer such children among women who are employees (1.17) than among non-working (1.43) or self-employed women (1.17). As expected, these values are higher in the Northeast, the Northwest, and the Other Southeast than in the remaining three regions. Also, with the exception of the Northeast, the average for self-employed women are 1.02 and 1.15, respectively; in the South, the corresponding values are 1.23 and 1.69. OYe"l!, the saiaet features presented above and the details providei ;n the tables indicate that there are important regional, labor force status, gender and marital status Xferences in the data This suggests a sratfication of the sample by region, work status, gender :nd marital status. Such an approach provides the opportunity to explore the extent to which the descriptive profiles are reflected in the empirical analysis of labor market outcomes. Labor Force Behavio, and Earawgs of Drwz.Han Women and Men, 1980 47 4. Femnale labor Force Parficipation Before proceeding with the rmodels, it should be noted that only 20 percent of married women and 41 percent of single women performed market work (i.e. reported positive hours of work).3 Of the working women a negligible fraction (1.5 percent) were employers, and about 8 percent of married and 10 percent of single women workers reported that they did not receive remuneration for their market wcrk (unpaid family workers). Among working wives, 65 percen were employees and 35 percent were self-employed; the corresponding figures for single women were 83 percent and 17 percenL As regards men, 90 percent of married men were etployed, and of this group, 4 percent were employers, 40 oercent were self-employed, and 56 percent were employees. All single men in our sample were in the labor force; 75 percvat were employees and 25 percent were self-employed. A negligible fraction of men were unpaid family workers. This information on Brazil, discussed more fully latcr, indicates that the labor for'e status choice model should be characterized by a three-way choice (employee, self-employment, and non-work) for women and married mn, while a two-way choice (wage work and self-erployment) shouLd suffice for single men. Thi.3 is the approach we adopt. The econometric specifications of the odels were carefully cnsidered given thle data ; our disposal. Table 3.1 below displays the definitions and measurement of the variables used in the analysis. In specifying the models we tried to maintain uniformity as much as possible across regions, gender, marital and labor force status." However, because of small cell counts or insufficient information (especially for single men) it was necessary to be flexible. While we analyze the data for single (never married) women ard men, most of our atention is given to the labor force behavior of married women. Wives are usually secondary workers who also bear the responsibility for household maintenance and -hild raLiing. We note that, for analysis and policy purposes, the labor force behavior of e:ves Is of considerable current interest The explanatory variables in the labor force status model are straightforward and standard in the literature. For single and married women, these are age, age squared (divided by 100), years of education, child composition (the number of babies, toddlers, and school-age children),' household :ize, asset and transfer income of the household, home ownership, number of rooms I The low parficipation mtes of Brazilian women should not be urprising. We note that at the tun of the centry mme comprisd 45 perct of the repotzd labo force, minly in agrizul hme ad textiles, and domsic-ervant activities. Two decade Lat the proporto fell to 20 perceat, and has since remained in the 20-25 percent range. A possible reason for the relatively bew pricipaion rte. is the manner in vhich statinical information on women's actvitia is collected. Most cens and survey practices in developing countries tpically exclude unpad fimily and intermittet workera (espocialuy in home-based market actitis) from the labor force. For furw discumin, sme Bolding (19S1 l. 14 Cmidemble effort and resourcs were devoted to ensuring tt data inconsistencies weav removed ad to expeimenting with alternative specifications. Generally, the estimates were not sensitive to altrnative specific2ticn '3 Since over RO 'ercent of murried households are ulem &miliea, the children variables generaly eflect the fertility- 'M wife. This ir certainy the case for the sample of relatively yong single womn. 90 percent of whrzm me daughters or daughktei-in-law of a male bouseLd heald Table 3.1 Definition of Variabks Used in the Analysia Maried Single VARIABLE DPSCRqPTON WOMEN MEN WOMEN MEN LABOR FORCE STATUS ANALYSES KIDS 0- 2 Number of childrn in household undcr 3 (Babis) x x KIDS 5- 5 Number ef hiren in hotiscbold 3- 5 (Toddle) x x KIDS.: Number of chl4ren in household 6-14 A HSM(ZNFAM Number of perons in hoehoAld X X AGE AgV i yewau x AGESQ2 Agt Qu0arod/10 x x a YRSEDUC Yes of education x HSLEVNW7 1 if no stcooling oompetod. 0 oiterwise tiEMPLYE Iif husb1A is an employce, 0 ouo xw WYESWRK 1 if wie work,, 0 ow s ASSEnC Ast + Transfer bt of ouscbold/100,0000 z OTHEkINC ASSE`lNC + wife's total canungs/I00,000 z HUSPJAN Hr*tnd's &Al eainp/100,000 X OWN HOME Iif bornwwn, 0 othawis x x x ROOMS Nmber of moms in the bome Ur!.BAN - Iif uban eidsew. 0otherwise x U23 =~~~ I if . Xiv inMnw .0o1vwCdu x n NCrH%&ArT I if Ev* iuNo iesOb c:erws Ke K z MARANE - I ifBr." inMarasb ocsrl, od.iwe K z z a RAHIANE - I f5f liuehbiaorS orgFi o, o win s z z a 0THERNfl - I if Ewa is O&e Not¶hza , 0 oaax'.o (aree K x K 3uO -l I if B3 iaw ofRio6Jmim, 0 O o K x SAO PAULO I if Svcs insRa of Sao PsaoOoduwiso a K K s OTRSR a I if s in OdLar Soisis3Owes 0oarimo .1 K a x SOU'H lI fE liSin&.Oodrwi z z WAGE ANALYSES LNWAOE NzSAI b of hy whmia job z K x MLISI,2 bvas ofMni(Impleem,olfw x K SECPUB = I itgove a kvso, 0 otherwi a K SOCSEC - I if socid b.Rty WatibIi. 0 odift x K x R EXPR Palwwok ipesio =(ege - .s - s o Zscboin. R K x R EXSQ2 BXp.ieu &soud a x x K YRSEDUC Yen, o educO K R K K BmToT NIsb of chDda e.C bom K a URBAN -mI if, Imdsa.O od isa z. K it REGION VARIAJLFS NORTIh-Br -I if Evr in NwUzwet, 0 oaise (rfmt) K K 2 R NORTHEA71 - I fbtE i NoN a, Oodwiz K x x x MARANE -I if Evrsndrszibor lPi,Oedwrwuno K K x K BAHLANE - I if va inl b or Spe, 0 oiw K K K OTHERNE - I if Evu in Oder Norts s. 0 Odas4s (newdma R K x x RIO -1 ifE vn inMofRIodeJ ro., 0 e wise x K s K SAO P.; ULO -I if E in of Sao PeO 0 cui K K K xK 7Tk=SE - I if Svsin d Soudw a mus, Oodarwis DS K K K SOULTH -Ilif E lshSouh&,Oiadwwisea K x K R in the house, and urban residence. In the analysis of wives, we also include the husband's monthly earnings, and a dummy variable if he is a wage and salary worker. For married men, the variables are age and its square, years of education, home ownership, asset and transfer income plus the wife's earnings, whether the wife worked. and urban residence. For single men, dte data base did not provide information on household characteristics, and the following variables were used: age and its square, years of education, urban residence, and a dummy variable indicating whether the person wompleted any schooling. For the entire Brazil sample we also estimated the mod.l %" '^d without regional dummy variables (the excluded category is Northi xst). This w?- 'is. .;n for the Northeast, for which the state dummy variables are bhhia/Sergipe and MararL o-;: (the excluded category is Other ?Tortheast states).'6 The results did not change in any significant way. Labor Force StaS ClhowL. Modl Our analysis uses the standard one-period static labor supply framework in which preferences are defined by a utility function whose arguments are the Ilicksian composite of all goods, non- market time, and a vector of exogenous variables that affect labor force decisions."7 Rational decicion making is reflected in the maximization of the utility function subject to time and budget constraints.' One of an individual's decisions is to select amongst three mutually exclusive alternatives: employee, self-employment, and no work.'9 These choices are indexed by 1, 2, and 3, and choices 1 and 2 have reported hours of work. The alternative chosen is the one that yields the highest utility. In other words, the individual compares the pecuniary and nonpecuniary costs and b2nefits of each labor force status and chooses the one that yields the largest gain. More formally, let Vj be the maximnum utility attainable for an individual if alterntive j = 1, 2, 3 is chosen. Assuming that this indirect utlity function is linear for estimation purposes: Vj = xyj + ep, where x is a (row) vector of observed explanatory variables (the non-stochastic component or measured individual characteristics), yj is a vector of unknown parameters, and eg is a random disturbance corresponding to unobserved individual differences in tastes. Tnus, the probability that alternative j is chosen is just the probability that the individual characteristics (x) 'pay off' more in the jth alternative than in any other choice: I" We aLso eaimated the model with ste dummy variables for te remaining reeions, anDd found t the results did not change in any significant way. The rmt are available fiom the =nthors 7 la keeping with the standard asi:mptions, the labor force behavior of other household m=be is assumed to be exogenous. For instac, wives make lbor upply decisions withoit referenc to thos of the husband, and likewise husbad. A futm research topic is the i=ue of joint decision-maing id &-usebold labor supply. 'i For fbler dicussiiaa of the mod wee Gree (19), Hi (1988) d Tr" md Le (1984). 19 This assumption, of coure, precludes courrt muliplejob holdings, bu this is of litle coner in this s!udy. Almost no women in the sumplo reported tat they held a ccd job. Abow 5 percent of the men reported seond jobs. .)( rVn S 1eypWy Una ray , L.AUId AMlfrw Pj = prob[V; > VJ for k X j, j, k = 1, 2, 3 = prob[x-j - xyk > k - fj] for k O j, j, k = 1, 2, 3 If disturbances are independently and identically randomly distributed, the difference b:ween the error terms (and hence between payoffs) follows a logisfic listribution in what is commonly :eferred to as the multinomial choice model of McFadden (1973): p7 exp(yj) j1, 2, 3; y-3 O(normalizahion). F exp(x-y) a-1 The estimates of the multinomial logit model can be used to obtain the partial derivative or marginal effect of an explanatory variable (m) on the probabilities of being in a given labor force status: j ,._ PI-yj, j=1.2,3 ax. a-1 't should be noted that the signs of the partial derivatives of the p -obabilities need not correspond to the partial derivatives of utility. That is, aPjadx may differ frc m N/axV.. = -y,. Even though, for exampic Vj increases as x,, increases, Pj may decline because the increase in xD, raises the payoff in another alternative (say V') by more. The estimates can also te used to evaluate the impact of a change in one variable, holding the remaining ones constant at their mean values, on the probability of choosing among the types of labor force status. We consider these simulation exercises below. After estimating thb choice models for women and men, we construct the selectivity correction variable (the inverse of Mills' ratio, X) for each type of worker, and include it as regressor in the ordinary least squares (OLS) estimations of the earnings functions. Since the selectivity coLTection procedure for the two-choice model developed by Heckman (1987) is well-known, we do not review the details here.' Instead, we summarize the salient features of Fbe less familiar three-choice model and review the procedure of Maddala (1983) and Lee (1983) for obtaining the selectivity bias correction terms. This is presented when we consider the wage equations below. Emjied Estmales - Logi Estimates of Labor Force Stauw Mamed women (Taoles 3.2 and 3.3). Turning first to the effect of children, we see that young children reduce tbh. propeasity to perform market work, a result that is commonly found in participation studies of married women. Children effects are slightly stronger for employees than for the self-employed. As expected, babies and toddlers have the strongest negative effects, while the impact diminishes for school-age children. The derivatives and the simulations also show that children are stronger deterrents in the southern than the northern regions. A larger household (the average size is 7-8 people) has a strong positive effect on labor force participation. A reasonable interpreation of this finding is that household members other than the 30 See aLso Dubin and Rivers (1989). Labor Force Behavior and Earnings of Brazilian Women and Men, 1980 51 *'ife serve as childcare alternatives and share house-keeping chores, thereby releasing the wife's time for labor market activities. Household size effects appear to be stronger in Rio, Sao Paulo and the South than in the remaining regions. As regards the housing characteristics variables, in most insunces home ownership (a proxy for wealth) does not have a statistically significant impact on labor force status. In only two cases is the coefficient significant and negative - for employees in Rio and Sao Paulo. The coefficient is unexpectedly positive and significant for self-employed wives in the Northeast. This may simply reflect the poor quality of housing in the region. The number of rooms, which proxies the burden of housework, has the expected negative impact in all regions, except Rio, where the coefficient is not statistically significant.' If the husband is an employee, this increases the probability that the wife will work as an employee, but reduces the probability that she will be self-employed. This may reflect selective mating in which relatively better educated men marry more highly educated women. We note that wives who are employces tend to have husbands who are wage earners, and that employees have more schooling than the self-employed. Income effects are captured by two variables: husband's earnings and asset plus transfer income. In most cases, the higher the husband's earnings the lower the probability that a wife will work as an employee, but this variable has litle effect on the self-employment choice. Overall, husband's earnings has a small significant negative effect. Tre other variable used to approximate income effects (asset plus transfer income, but not husband's earnings) also has the expected negative effect on the propensity to work, particularly on the probability of working as an employee. However, with the exception of Rio and Sao Paulo, it generally has a small impact The dummy variable for urbanization presumably reflects a mix of demand-side and taste effects, which are likely to work in the same direction. An urban area may provide more plentiful job opportunities and a more congenial eavironment for a wife to perform market work, and thus encourage greater participation than in rural areas. We see that living in an urban area has a positive effect on labor force participation, especially as an employee. In most regions, however, the urban variablz has no significant impact on self-employment. In the Northeast and South, the urban variable has a negative effect on being self-employed, while in the Other Southeast and Rio (which also encompasses rural areas) the impact is positive. The age of ih.e wife is significantly related to labor fo!ce status, but in a aon4irLear fashion. This is readily seen by glancing ai the coefficients, the marginal effects, ani the predicted -wbabilities. Younger wives (under 25 years) and older wives (over 40 years) are less likely to participate than wives in the 25-40 years age range, yielding an inversely U-shaped age ciadtional profile, which accords with prior expectations and other empirical results for developing countries. This pattern is generally repeated for both employees and the self- employed. a The number of rooms ina house may also be a mrrojate !or weelth, de on o the quality of the dwelling, which is not known. As regards the finding for Rio, one hyothesis. though zwot testable with thee data, is that te is better ce to domesws4 in Rio than in the other regions. Te usu sam oning can be used in interpretng the effict of regional d'unmy variables. See, forinstance, P poos and T=mnto (1989), Mohan (1988), and Standing and Sheehnm (1.78) 52 Womnm E=sp&Trwm xd Pay in Lazia Amsrica Table 3.2 MiKastMic33J Lc&i E.as4ow of iho Labor 1r000 PSi i,3443 - Marnicd Wowco C ___AM S 333 I= FAMILY AC =0 T% MIXMA40 1 I3Q U=M&ICIII 0433 MD== u304 .3741 02-~~~~~O40 4333 CUM13 (3330.. 44 a0GtA 1 G73 141.1) (ILIZI, 434.7 p333 o.LM 0.414 0&341 a4 -4 -XWO3 .3.27 3.4 L.4 -L=3 L.14 3.4 -luso -1.73 -.43= -0.13 C6,01 *no 43t -2.43 -L.33 .4.44 SAM4 4.4 -1.1 033 -.4 -7332 033 0.33) -4.1 -L E-Onval, -am -.14 ..4 1.60 41.25' 4-4.3 11 " 133 .3 .1.4 443 Ill 0 an 133 ..01, 0-4.4 .033 4.37) 431 ICQ fl7 0.74 01 48 MID 4KM -4.03 44W L L 0.37772411 a- 71 co3.1 0.13:3 -LI0 O L 13O (7.13 -t343 1.33 144 1440 (1.0 04 a 0.1 0. a uff- .43l &a -313, .3. .3."I IL=S QL24 -4.7 0.4 41 4.3 - , 7 6a a."?3.60 an3 -.13 N0l1VWO3T *U.IZ'4fl.L ~~~-3432 -0.3.333J11 it"343.4 E-VUI- ~-40.44,4 4J3 &4.3 -cm 23 0. 11-0.34 4 a4 013 0.4* -L33 4013 0la 444 44 a'mi 01.32 Wm.0 0,3 040 (?.M a 114 0:.30 0106) = a =300 0.32 r1.03 a43 44. to.4 -2.3 -3.43 -LI3 L.3 I.J113 .1.10 3 l433 4.8W -43L60 -3.43 0.11 4.33 n.143 SavI_-;Sd 40 023 -* .4.2131 di"4 3.23 433 0.42 4.37 -433 -61~ -0041 -o u3 Q.33 ..Lil CLQ(J)3 0.43 0.41 CL.S) 0.14 w4d.4 CUM2 14.3 0.4)0 lm c 41.30 0 m 113 0.40 a4.- "O . CM -3.3 4.6o -4.32 1 = L.OO 4..= 0.1 -l= 32 .433 41 4.33 4130 1 0T0(00T.AW -33..3743.-3410 -1.4.3311 33A -40.47 41j3 4.37 44A0 033 0-.13 -.A'= 0M CLU.34 -L.413 -$Ai4 4.034 .053 0.43i ..3,7 (3180 0?.3 1440 0.10 (lato 0.133 MY"1 (M344 C0 0.73) 0.1 CS30 0* a.0 .0. &.44.t.4 43 -.43= -3.13 -2.4 1.33 LAO4 -3.4 1.44 6.3 -44.3W 7A -1a3 Kg -.-3 2.1 Sau-nnl"d -(.4144 -0.725 4.379 4.13 CLAN lLW 4332 0.6 4.43Ca -1,13 0.341 4.l44 4000 1,4 .NY WM~(373 44.13 0.73 ,0 -o A CC) 0.3 384 .1 .41 414 0.44) 0.4 4:143 (13 am3 L0. . to ~~.14= -4.P1 -3.43 LAX) L.4 -6.43 0.32 -3.44 ..4.73 440 4.43c 430= 1.5 320011A4120 *3.D'JU. 313. .33337143143. -60 (3.14 013312 Q0.11 44L34 41433 0.01 0. 01323 a-,* Kau41 0.40 (4.13 10.4) 0.= .43~JC -a3.3 -737) 433 3444 2.019 -3.13 0.0 3.71 -31.23 -0.43 -2.40 -4.13 2.403 -..,a 0.43 81.3 0.84 033 SLIM738 0.44 33 033 mm.7 0.3D (4.30 4.9,3, 41.344 cL 4460. E -3.4 -4.1)7 -1.00 Q0.43 0.43 -Lrn 0,010 1.4.4 LIM3 3.420 -4 .4 -43 no 6Lo PDAUL -0AXLUJL-701. -112LJI * 044 0 a-SWm finvi44 -7.238 -&An4 -GA"4 .4.3 LAO . .3 4.33 0.43 . 02 -3.7 .4.3 -4.33 4.43 up4 -1114 04.84 04.44 0.30 0.L34 a7-'3 0.413 0a S11013) (LBO MAO 440 0.701 00 a- 443no -SSLm -4.4 -3.f3) 4470 33 -3.83 3.2 1CO .niw -430 -2.0 -L.40 2.3 &dF..pbPpd-7.41,7 .4.1 -4.0 -4.In 0.3841 CLIO4 -4.3 41.114 40.738 4.103 4 mm -1.33 43 07 033 -m (~~~~~~~~0311S 030 0.313 0.11 (O.3 86A0 SLIM 0.3 0.333 037 03 MASS3 (1.13) .3 43 4~~~~~~~~-.010 .314 4.430 0.130 I.33 4.33 80 -3.13 -1.A3 0CM -I,.*43 41 0.43 SAO LOW -34.Lfl3 * 343. .U.Lwrli - OI -ml0.3 -0.414 -4,331 -4432 8.46 0.331 4.A3 0.37 0.23 -433 SW a .440 .02 .4 1.W -0?. (13~~~~~a.=1 51.4 47.70 (L414 (11.87 a3.3 0.43 04,6) 0.461) 0.44 s413 040 .134= mm 1x3 -1.43 .440 -3.3 33a I.04 -3.1 40 SAM -20=1 -47.4 0.332 4330 1.4 -7.M7 440 41.831 4.1 0a" 0.313 4a1 alto 4.713 - - -am3 4 a 0.43 -cm0 -a 441 am3 8.3111 -330 -1.10 3.43 SA4W -430 0.4 0.3 -343 01 40 -3.313 The Mob bo do ar s dopu 4440 ubn a IM *sb" a Os m:q a.......*....o oos so=o . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 t **o A V a tl ta o3 o . . . _. :n- . . . . . . . . . . . . . _ I %a *it a I a aw a sa a a a a a a a a a a Is a a a a a a a a i a a I 0 e a It 4 a a a A 4 a a 3 a a u a a a a 3 a s a a a a 3 a 0 a 0 a a tf a 10 t . . . . . . .< . . . . . . . . . . .o to 410atUCLCe__ 13-. Wi-V OI- C=SIS Dlif-c v en SW) Wi 6100- ilio Liii- 1ciOl SO- cWI- 0181- OKI- ti- WVS 090- 01i 54- mi-WI two a" o- two im- (ING' 110 9- (A en's bro en cUel 453-10 0- owl -P 100 swi- o- 090 900- Wi Wei 011 01019 1330- W-o Cin - into sri am- cui-rn - ma w-Cii-ol muVV dO-i till GO- am.~ o"- 9 wi- uvWe 971- m- lul omi0- mlo P Io Mi 0.WI - on s- am.- art ace- wo- Cm'o-$61V GKV- tcro "Oa am- -a LIUIX GU Ii- IbIYfl- a0O GLxi- OWi- 31) WC09 ciii Q. f!- ciiI Cc"- Coi - 0'al 51J0am o (SIt) (Sit' list) (LIe (35.0 ~~~~~(UV Or Wit l091 Sit (two) UvLs) In-.. Wi-e Iwo- WVi 1600- 11100 cai- Wix $It" Wi-. 410 two 1IC)- 6L1 ak,= W- Wt. citl- CR"1 owi- (U ti Culp CaI M I09 OK - ax.~ ob 15)0 430 tIixO it)O 1331 00 ) two lad 11t Urt) 113tI 0051) 94 likO no 0- WOv- 001. tt30 6190- MVi a"O =SO fat pli- to - bl lIlt V.6 9-6 Ull,? itvSINITt-IM Ai flhl-Uici, WUzlU cutL- tWGO- Oa" 050el a(- WV- ari Wi-a. Iwo 0100 ari in . z:-o 0- 49t) 0 949 ls bIt 11(0 (550 a" - 091 tax-I) f a14 UrIt) "a - cwi- "co 101 0 100- 14(0 0C0 Wi 6100 ll- Ł100- Its-i 11169 C14- 051'- 0141!- till mill- Cill 0191 Sr1 (Oil- toli)- VV:I( sit) (5,91 '(0 till (Wio (Witl) 01Wo C9-VI be-) lit (U I)tt WW 450D n.. crn- LI-i- .g 0- rn-i- LS9( n-iV- PC-" a"o Mo tm-i- 06110- WLa-i 291004N 990M4nO9CM100H101 I B= It rI 1 WOfIM 4 1100 lMOS 03M(1D MHA b1Z--v Sm ASO(Yd I= 0=11 9= NviJN1I _______ -A%0 qft!S - -WjN& Lq am3w?s~ '0 "oqvl jo I02flW ~ i Mo11w cupp, DVO1UJY UPV7 uJ L(vtd P1W vw(0o7dt Jr. U"uuhM g w~~~~~~~~~~~~~~ . . .X . . . . . . . . . . . . . . . . . . . . . . . . .1 . . . . . . . . . . _ 2 ? § u R " R R n t R St a a a 2 D a la A a 2 x a4 9 a : A I a a Al aa n I 9 I a a R a P a a Aa a a R a a It a a aR I I * a n R4 a a 2 9 SA t x R X 5 e B X t D 4 ta r Ca A I a a a DR a2 I ta c a r. a Ce I R a a 2 RIRt a a a R a a na a a 5 R nq n it aa a a R, q I a a a a a .; . . . .. . ., . . . . . ,, . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . -- . -- . . . ..... g~ ~ ~ ~ ~ ~ ~~ ~~~~~~~~t a a PP L Qa %|t%sW D| a X. C; F | t aa a F e ata I R a I it t. a a aa| I a a 0 a x it aP a a 8 a a I 9 A a a I l~~~~~~~~ . . . . . . ., . . . . . . . . . . . . . . . . . . . . . . . . . . . . ., . . . . A P Q It X A R It t R 2 'g C R R R g a a a = ; R Ri 2! I ^ 2 I I 5 t R a R X a; x J I gI R 7 R R it n R X R it C 1 t A 21 a a 3 I t a #. a * ta a 3t a a a; IF Is e r R lr t a a a e e r a la Ia * ID t a *a a It itP P a F e e a a It 9 n. a e a I I I cA Ra a I i it it a P P r a a a I a i~~~~ ~ ~~ .} |^ , n M A F t a P °t0 a a I a it -1 ' a Rr 71 a a a I E; t it Ft i 0c : a Jr i t sg a t F a 3 -f - |- - - - - - - - - - - - . . . . . . . . . . . . . . .- . ' : . me aT It 3 'A A I S A A A D 3 3 - a A n 3 A C a 5A CS * A 1. O D A 1 R a R 7 V a S 5 R a Pk 15 aA a ti Si 4 p 13 Da a 0 a a -1 a D C 5 i it it a a It LIa 9 P It ! 0 2 F 2 It Of3 2 r 32 t ISt R 2 I# a 9 a 2 R Z t P a a a 2 3 t 7 t R 0 it P rS t a p t It pp # 8 g a g a g a a p a IC ;z91 a I ax R f a A a a a a A a Z e x aae a aa a 2 n - a a aa e r, I I a n a a a a m aa a a a R R a R n R# ap 9 n A a # A al a it x a e~~ ~~~ . . . . . . . . . -k R 3 a a an R a I a . . . . . . . . . . 11 11 11 1 I 11 . . . .e 60 Womne's E Epl ~mU d Pay in Laibi America Tahh 38 Lo& E of ALac Fame Puztk o by B- Sin Umen ALL BRAZIL NORTHEASr NORTHWES arHE RIO SAO PAULO SOUTlt VARIABLE SOrTHEA-T CONSTANT 1 2S0 -0.258 1.864 2.922 3.431 2.737 1.464 (/.31) (0.71) (3.61) (6.35) (4.S6) (6.67) (3.2 AGE -0.054 40.011 .0.080 40.134 40.090 -0.070 -04.0n (4.97) (0.AM (2.58) (4.69) (2.11) (2.81) (2.52) 40.930 0.260 -1.730 -2'220 -0.990 -0.710 -1.230 AMESQ/100 0.036 40.021 0.068 0.143 0.064 0.035 0.049 (2.34) (0.67) (1.55) (3.61) (1.09) (1.01) (1.21) 0.620 4500 1.460 2360 0.710 0.350 0.S40 YEA.RS 0.071 0.119 0.066 0.042 0.024 0.046 0.054 SCHOOL (7.52) (520) (2.83) (1.73) (0.90) (2.60) (2.25) 1.230 2.890 1.S50 0.690 0.260 0.460 0.920 NO SCHOOL 40.380 -0.110 .0.638 -0.214 4.0093 0.042 40.029 CONMPLETD (5.51) (0.65) (3.13) (.s7) (0.38) (0.28) (0.17) -6.550 -2.690 -13.810 -3-540 -1.020 0.420 40.500 URBAN 1.186 1.243 0.598 0.89/ 0.256 0.534 1.406 (24.31) (1352) (4.42) (7.00) (0.94) (3.5) (1.50) 20.460 30.310 12.930 14.820 2.820 5.400 2.4.0S0 Employrcs 9673 1609 929 1472 1147 3C2 1444 sdf-Employed 3301 1261 486 445 182 443 484 -2LLKFULL 12719.7 3333.1 1575.0 IBS8.4 1012.9 2518.1 1882.4 -2LLCKREST 14716.5 3936.4 1820.6 2C77.4 I1e5.6 2662.8 217. Noew Numbers in pArentheses are t-sta . The nurmber bdow t t-sta arc the pril deriyatives x 100 tvaluated at dte zple mcmn Labor F t Bd-aveorad Erdngso f Ara,ih Womamd M, 190 61 T i 3.9 I.ogix Siml& : Pr dabfJfu of Lab"t Fo= Pticipa ) - Si& Man E Employe-; S - !cI-fmpjoyed REGIONh ALL BRAZIL NORTHEA NORTHwEs OTHER RIO SAO PAULOA SOUTH sourrHEAST WORK 5STATUS B S E 8 E - E S E 5 a S a s AGE 15 83 17 63 37 Ty 23 & 12 93 7 93 7 .5 is 20 80 20 61 39 72 2t 82 I8 90 10 90 10 8t 19 25 Ty 23 __ 42 67 33 76 24 87 13 as 12 77 23 30 74 26 55 45 62 38 71 29 S4 16 85 15 72 28 35 71 29 S2 48 58 42 67 33 Sl 19 82 l8 68 32 40 68 32 49 51 54 46 64 36 77 23 78 22 64 36 45 65 35 46 54 51 49 62 38 74 26 74 26 61 19 50 63 37 .42 58 49 51 62 .38 71 29 71 29 58 4 55 6c 39 38 62 48 s. 64 36 69 31 67 33 55 45 60 60 40 34 66 4i 52 67 33 67 33 64 36 53 47 65 59 *1 30 70 49 Sl 72 28 66 3. 61 39 52 48 SCHOOL 0 71 29 4t 52 60 40 76 24 86 14 85 IS 73 27 YEARS 1 73 27 51 49 62 3t 77 24 S6 14 86 14 74 26 2 74 26 54 46 64 36 77 23 86 8 S7 13 75 25 3 75 25 57 43 66 34 78 22 86 14 87 13 76 24 4 77 23 60 40 68 fl 79 21 87 I3 as 12 Tn 23 5 7S 22 63 37 70 30 79 21 67 13 tS 12 7t 21 6 79 21 66 34 71 29 t0 20 87 13 8y 1 79 21 7 80 20 68 32 73 27 8t 19 tt 13 29 1 8o 20 a 21 19 71 29 75 25 81 19 88 12 0 121 0 20 9 82 28 73 27 76 24 82 l8 8 12 90 10 82 19 10 83 17 75 25 78 2 63 17 t *2 9% 10 832 IS I U 84 16 78 22 79 21 83 17 89 .1 91 9 83 17 12 35 15 80 20 at 19 84 16 89 11 91 9 S4 16 13 .36 14 St 19 32 I2 84 16 89 :I 91 9 84 16 14 n7 13 83 17 83 17 t5 15 89 11 92 a 85 I5 15 88 12 85 15 84 16 t5 15 89 it 92 8 86 14 COMIILETED SCHOOL YES 80 20 60 40 75 25 8t 19 a 12 as 12 78 22 NO 74 26 57 43 61 39 77 23 87 23 89 11 78 22 |RURAL 60 40 41 59 60 40 67 33 85 15 83 17 58 42 IURBAN u3 17 71 29 73 27 83 17 88 12 89 11 85 is ATMEANS S 78 22 _1 42 68 32 79 21 87 13 t9 11 78 22 ACTUAL % 75 25 56 44 66 34 77 23 bG 124 7 U 75 25 0BS 9673 3301 1609 1261 929 486 1472 415 1147 It2 3072 443 1444 484 T.e results show that the probability of holdirg a wage job decreases with. age, but the coefficient on age squared is generally not significant. In all regions, livirng in an urbzn area strongly increases the probabilit-y of being at. employee. The urban effect is especially strong id the Northeast and the South. Once again, the education effects are interesting. In most instaices, (the Northwest is the exception), uncompleted -lchooling is not an important factor in determining the type of job.' The amount of schooji.ng, howevez, does pl2y a role in the Northwest. the Northeast, and the 00-er Southeast. In the rcnaining regions, the coefficient of the schoolin' is not significant. 'Ihe simulations generally reflect Ltle pattern found for single wormen. Fur emumple, an extra year of schooling raises the probability of being an em-ployee by about 3 percentage points. 5. Ware leerminants The analysis of wage determinants is based on the human capital framework developed by Becker (1964) and Min:er (1974). This provides the dheoretical base for the study of wages as a function ef productivity-hancing variables. We estimate a wage function where the dependent variable is the ratural log of tne hourly earnings (including cash in-kind payments) which is obtained by dividing monhiy earnings -y weekly hours times 4.33 In doing so, we assumed that the individual worked for the entirr monlh. Tle regresscrs are the inverse of the Mil's' rdtio, poteutial work experience (age - 6 - years of schooling), experience squared (divid4d by 100), years of education, dummy variables indicating pt^lic sector employment, contributions to social security, urtan and state res.dence. Since data on ?ctual work experience are not reported, the sLt of regressors in the female wage equations also include the number of babies ever born, whizh is used to reflect interruptions in potentia wock experience. Wage Fundions - lhe I;ssue of Selectivity Bks Now, we consider the specifi ;ation of the wvage function: Let: Cl = 1, i. thc person ic an employee (1), 0 otherwise; C, = 1, if the permon Is self-employed (2), 0 otherwlse; C, = 1, if the person does not waik (3), 0 otherwi0e; The wage fuiction in the th wcrir-statu; is given by Ini'. z[ a, + Vj if Cj = 1, i = 1, 2 lO ot.erwise where InWj is natural !og cf the wage, z is a (row) vector of wage-detennining characteristics, which has some elements in common with x, cj is P. vector of estimated parameters and O is the error term. If tLe indifidual does not work a = 3), then no market wage function is observed. Traditional OLS estimation of the wage function may prcdiuce biased and inconsistent parameter estimates owing to selectivity bias because the observations on earnings by job alternative are not 2 rho dsta sowed t the proportion of single m who did t complete any schooling wg mch hiZher than ad of single wom. randomly distributed. The selectivity bias correction term (X) for the multinomial logit choice case is derived using the transformation of Maddala (1983, p.275). Tlat is, >g = 0[lj(xy)J / F(xy@, where b is the standard normal distrbution function, F is the logistic distribution function, and the transformation J = 0'F. 71fas, fer each work-status wage regression a= 1, 2) x can be included as a regressor. lnWi - zi A i In sum, because decisions about labor force status as well as wage offers influence the observed wage structure, correctionr for selectivity bias are needed to obtain consistent parameter estimates of the wage determinants. Empirical . timates Marred women. Table 3.10 shov t that the effec of the selectivity correcting variable among wives varies across regions ind work status. At the national level, both the employee and self- employment wage regressions are subje' to selectivity bias. However, a region by region comparison seems to tell a different story. In three regions, the Northast, the Northwest, and the Other Southeast, the coefficient of Uambda is not statistically significnt in either of the wage regressions. Both wage regressions for Rio are subject to sample selection bias - the coefficient on Lambda is positive and significant in the wage function of emr'vyees, and negative and significant in that of the self-employed. In Sao Paulo and the South, the coefficient of Lambda is negative and significant only in the wage regression for empluyees. TLe coefficients of the best avaHable work experience variables in the data are generally as expected, but there are regional variations. Experience effects do not appear to be present in the two wage regressions for the Northwest, and in that of the self-employed in all regions, except the Northeast where the coefficients on all experience variables are statically significant In the national wage regressions, all experience coefficients are staistically siguificant, suggesdng that there may be geographical aggregation bias' in assessing returns to work experience. Moreover, there are also some regional differences in the magnitudes cf the coefficients of the experience vas;ables (especially experience squared) of employees. Because the measure of work experience for wives is imperfect we also included the mmber of children ever-born in an awmpt to capture intertions in potential work experience. National!y, the coefficient on this variable is statically significant and negative (-0.033 for employees and -0.036 for self-employed). However, this proxy for discontinuity in work experience met with limited success in the region-specific regressions. The parameter esimates are significant only in wage regression of employees the Northeast (0.025) and in Sao Paulo (4.021). In most regions the coefficient estimates of the effects of having ajob in the public sector are not significant and, in one cue (the Northeast) it is unexpectedly signficanty negative (4.286). Only in Rio does public sectr employment have a positive effect of about 16 percent on wages. Tank 3.10 Wa. R gftaaawr- Sy kr:po- marriled Men ae Womea (rh. deponIMn variable is: LN4 WAGE) AU 33WM o730T 061033 03136 rn"ra toU73---60 W34.l 643346* Six=3 4I3 C07.4 OKI33 11343 0-41 RLU (3-42 1.33 3394 31(WI (14.73 0.33 C WI 4l,7) (344 3-1 (3441 (3333 (lot) (7.a C043 Go 03* 043 a 0Po (344 C3Dan (7. 6302304 -AM3 .31,3 a3m? a33 -04II? -93111 0313 a, 16, WI 4032* Go" -o2 -4,114 4168 433* am" -4 0t I38 43* a7 244 -0616 4-46 3173 3.0*3 .30, .0 in 3.373 4373t 1041,1 il 4 i 0.1613 3.3 31.7 a a3 0.3 0.41 1343 073 33 0.3 (04 (3.73 043 0 (1 016 an 043 (343 3.3 73* (7.3* a(4 (3*2 36.7 0.0 (3.* .4 -ago 93 4161 . 406 4 -am3a L 0SD - 3Wco 1331 40* al33 con23 04 - 4 30 337 11 .8 31 04 L all3 0.3 67 37 - 3333 i D 0001 73 - 0.3* (341 44cc m a77 34, 041* con QJM3 6.3 3.443 0103.3* 1 4 63A6 3.1) Call 0 Sll6 s 3j6 33) a0* An 6 33010 3337 a6334 3*8 0 336 0*13 DID 336 .4UI 3,41'3 (.S23 oILJ 06,63 023 01 .41 0.41 C.3 (3411 0.47 343 04a 0 33 0 1 a a0 so3 a733 041 UM al 0.33 (7. (is (166 (7.73 (67 133*L 41. 3-3* 0.' 04 3.33 3-44 6~ S0an 3.* go" 6343 3.36 alas 0073 44 19w 30*3 4003* Sam3 S0*8 QL01 33 0ow3 a3m 0332 4034 son am0 307 0C91 3016 0DA" 00*n 3U7 CM"P1 I 1341 3332 0.37 0,73 0.33 0.33 0-) 0.1 San3 0.7 161 1671 a07 I 043 all a*1a 033 (3.3 (373 (6- 033 01 0.73 0.41 164 44.3 33.6 6313333 4A3 -0433 .400 40 au 404 a" 40 6 an6 44* 4416 SL14, 4163 *W 400*3 .o044 40.8 39W .43or 4-0* -033 coo 433n -SA06 4*07 A"3 .GAM 4.4333 4OPP-031 4 911a 0l a .41 041 0.33 0.1 (33 0.3* DA. a 433 073 03 (4Al3 7 (773 ,673 436 0.33 0.41 n a.7 a3 (33 3041 0.3 0-23 1377 0* 0a33 (3.l 10604W 4,34I 631 43-31 03ISO aio aLm 3-us 3-P3*8&I O10 GM G 13 ls , 6116 0GI" 3.l1 313LU3)3ilo3 oil* 3UN 3334 0316 130 a tit302 03 337 .*3 0* 332 0Go7 134.3 0033 3374 (Wu.3 con7 OLD* (2447 0.16 a 016 033 3(333 343 a73 (67 334Gn 41 33-41 3633 a 3DS% (I3l0 10 o 00* 0CD" 373 (733) 03.43 334A4 (764 33 "to 33 633 0167 4* DA40 OM3 0264 (Lot 3-* 631 t 05 636 .03 33 64 l D6373 636 33 a 6277 0336 t si D 0663 91 60 D SO 336 3)~ CL7 a 1 0 16680 444 ton6 - 4 .4633 416 - (3 40*7 41-03 . 4030 4337 -am" 40*3 .04,134 * (7' 373 3.3 333*163* 13 133 a26 - 071 all - 063 337 (3.33 0333MWL 03*3 &39 603 6416 303 A SIM &910 3a6 *0 43413 6163 3-11 3.33 66- Sea 6.437 m 06 6316 *4 , 34 jA63 a 6W 3163 0773 006 S=3 3.80 3.50 P-4T.? 233,3, au3) Isla4 1786 301 W3 1, MA 336 3I4 Ws VML$ 73 33,14 713 362 (7a 3334 "AS. 11.1 Ws4 03.6 16A 31.7 D. 3P". 626 943 6) 16 Wm1633 6033 3333 eon 6163 066 67(1 67v" 6670 603 63*7 033 0007 0Soo 606,Ow 016 No0 43 00de 67?) 303 016 sty 031 0 733 il 616 36 03 31a 0634 Lj c 31`633 Sam 34a 33av6 3.Mg3 L.36. 13316 Set: son 33W 3.73M Son1 is" $O6 3733. 6831 01633 3I16 03316 37331 (14 J3.733 40 391 *33 Xs"6 03W Lem 3* 34A314316 00 0S3.3 333 " , . 167 43 Ł 3 S 16.0 S74 43 1* 44.3 o Is6 332 1 a3 367 3333 0*6 ibis "J =63 34 1.3 6334 US S ea OD L- 10 a1.3 (733 (37 03633 .1 a34* 316010 1633 (34, 1333 coo 06.0 (*32,33 (1641 Case3 (766 30*491, g 03( Due341 (333 331638D Us a23 03*377 C (13* 03* 33 P4.13 (logo 03(8.6 363.71 oil) V40.46341T16 336 02 1 1 36 1 104 4,6 I3 76 * 337 0.7 334 6a 069 736 344 I73 1 23 333 *80 as32 a 3I * I V$3 333 33.4 7.4 18.7 33I US 3370 (3316~~~~ila C go)3 3363 334160 I Mai7 * 143 33133 oil4 0333l D 333 3334as oil3 31"77 3333 0741 304Q t 034 341e CD333 (WU3 031" CA3 (13.41 47Jt 013) (30.3 0013913106 4)0 160 313 374~~~~~~~II 036 313 44.7 316 et7 306 16 "73 64 077 40G 34 34L 37 as3 no0 431 .33 a3 S 7Po 47.3 360 46 0 162 VI DI1673 3136 a7s 4s41 73 73 (3 lot . en1 (70.4 (v733 (it'.) 1633 (is3St (773 ofP 33341 (* (7333 13 a33 Om3 3( (777 T (331 1,77 0333 (73 33,D.3 3MM&A-s P. 3.06 - pm-.0333 - 36. Access to social security (SOCSEC) has a strong effect on hourly earnings of both types of workers in all regions. This dummy variable, as discussed in Dabos and Psacharopoulos (1991), reflects unmeasurable job quality characteri.tics and conditions of employment, including health care, oensions, and other fringe benefits. Nationally, about 80 percent of employees and 30 percent of the self-employed in the sampl; reported that they contribute to social security. Of course, there are regional variauons in these magnitudes and in the parameter estimates.' The estimated positive effects of SOCSEC on hourly earnings among the'self-employed are usually in excess of 50 percent, and range from 75 percent in the Northeast to 53 percent in the South. Only in Rio is the impact relatively small and statistically weak. In the employee wage regression, the estimated wage gain from contributing to social security is smaller. It varies from 18 percent in Rio, 27 percent in the Northwest, and 38 percent in Sao Paulo, to over 40 percent in the remaining regions. Nationally, the estimate is 33 percent. Urban residence is generally associated with higher hourly earnings especially for employees, but less so for the self-employed. The coefficient of the urban variable for self-employed wives is statistically significant and positive only in the Northeast and Sao Paulo, while in the national wage regrnssion it is strongly significant. The relationship between education and hourly earnings is indeed interesting. The parameter estimates and the low standard errors show that schooling proves to be the most consistently effective variable determining hourly earnings. Table 3.10 shows that the wage gains from schooling among both employees and the self-employed are striking. Self-employed wives have somewhat lower estimated (private) reaurns to schooling than employees. The national wage regressions imply a return to schooling of 16 percent for employees and 13 percent for the self- employed. These magr 'ides, bowever, are undoubtedly affected by regional beterogeneity.' Consider first the region-specific wage rcgressions of emp!oyees. The estimates indicate that there are regional differences. In Rio and the Northeast, the estimated return to education is about 17 percent; in the Northwest and Other Sourtheast it is 15 percent, while in Sao Paulo and the South, the returns are 14 percent and 13 perceal, respectively. There is also regional variation in the estimated returns among self-employed wives. In the Northeast and the South the return is about 10 oercent, about 14 percent in the Northwest, and about 13 percent in the remaining areas. hn surn, our estimates of returns to schooling in Brazil are much in line with those obtained in recent studies of women's earnings in other Latin American countries. 'X A high proportion of employee wives work in the public sector - over one-half in the Northeast, Northwest, and Odtr southea; just over o0e-third in Rio and tha South, and about one-quarter in Sno Paulo. We recognia that public versw private sector employment is mubjet to a selection process. This is a topic for ft're reseach. 27 Since enrollmt in a social swcuity scheme is voluntary, this vaiable also may be subject to a selection process, especially among the self-employed. 2f The incl'sion of regional dummy variables in the national wage function slightly altered the point estimates: 0. 15 for employees and 0.12 for the slf-employed. 2 See Arriatda (1990) on Peru; Behxn and Wolfe (1984) on Nicangua; Khandker (1990) and King (1990) on Peu; and TerrelU (1989) on Guatemala. evluence ol bc4Uvity uWa. iLLL AAVJbi AAt G .LL *&&kWrLM. VI Ac UV%&a U4 I.sb" '.& for employees and significantly positive for the self-employed. In the Northwest, however, we can detect no evidence of selectivity bias for either type of worker, while in the South, the coefficient of Lambda is significant only in the employee regression and in the Northeast, only in the wage function of the self-emploc cd. The effects of * perience on hourly earnings are generally as expected, though there are variations across regions and work status. All coefficients are statistically significant in the employee regressions, as are most coefficients in the self-employed regressions (the exceptions are Rio and Sao Paulo). The effect of public sector employment in most regions is either negative (Nor.'-.ast and Sao Paulo) or not statistically significant. Only in the Northwest is there a :ggm.. ' positive impact of 15 percent. In all regions, working in an urban area generally has a r ;,positive effect on hourly earnings, especially of employees, in all regions. Contributing to s: I security generally has a significantly strong positive effect on hourly earnings for both ipl( yees and the self-employed. Tle exceptions are among employees in Rio and the Northwest l.here the coefficient on this va&sable is not statistically significant. As in the regressions for wives, the coefficients of the eduration variable stand out, and in all instances there are statistically significant returns to schooling. Nationa!!y, the returns for employees and the self-employed are similar: 15 percent and 14 percent, respectively. A regional comparison shows the following pattern. Among employees, the reurnm are highest in Rio and the Northeast (about 15 percent), followed by the Northwest (14 percent), the Other Southeast and the South (about 13 percent), and Sao Paulo (12 percent). Tle regional pattern among self- employed workers is: Other Southeast, 14 percent; the Northeast, 13 percent; the Northwest and Rio, about 12 percent; and Sao Paulo and the South, just under 10 percent. Finally, we note that married women tend to have higher returns to schooling (and more schooling) than their husbands. This seems to be the case for both wage and self-employed workers. Single men and women (rable 3.11). Looking only at the regressions for which there is a sufficient sample size Cie. ignoring those for self-employed women in the Northwest, the Other Southeast, Rio, and the South), we see that selectivity effects are found in both the male and female employee equations. In all regions, the Lambda correction terms are significantly negative for female employees, but n for the self-employed. The male regressions tell a somewhat different story. For employees, the coefficient of Larnbda is significantly negative in the Northeast, and significantly positive in Sao Paulo and the South, while in the remaining regions it is not. For self-employed men, there is evidence of selectivity bias only in Rio, where the coefficient of Lambda is negative. The results for the remaining regressors are generally consistent with a priori expectations. Living in an urban area makes a smaller contnbution to expected earnings of employees ng sinles than among married people. (ne urban coefficient is not stadstically significant for the self- employed.) Moreover, urban residence has a positive imnpact among female wage earners only in the Other Southeast and in Sao Paulo, and for males only in the Northeast. Unexpectedly, in the state of Sao Paulo urban residence has a negative effect on earnings for male employees, 10 percent of whom worked in rural areas. Access to social security strongly enhanc eaminp of men and women in both work activities; the effects tend to be stronger for women than men The public sector employment variable (which is relevaut only for women be.ause of missing Information for men) shows a positive and significant effect on earnings only in the southern regions (Other Southeast, Rio, Sao Paulo, and the South). TaM 3.11 Wap Regmasion by Rcgio - Sinto bMe and Wow=n (Meo depapc e3) vwii 1 83: LN WAGE) ___________ 82.0. molE. ~~~~~~~~~~~~1lm4a? m.inmw65 ovum IIIJI65DJT 4~~~~~~~~~~~~OTM 08 3M43La3 9A83) D301)PL"712T OUMTAN? 8.5) 8394 Lo4u 3746 413 Dal 1.413 tLJ8 3.44 444 Bill 1 .74 3412 2.09 .203 IJIl 310 In T&W3 0is. Di") L.33 L107 3.394 3160 06L1) 043) (7.11 (01 a.40 (14 0.98, ~0,2 0om.3) 84)A #30431 * 44 134 a)e 41) 048 0a * at" ales3 (298 088 at) 814.13 P2 LAMUA 440A 4*04 43410 -.,133 1.3.35 403 41,"? -47166 -aim 4646 olato .410) 402In0 4133 4.04 4-46 -40) * 4 sn -01Id 013 ^DU 5 0173 4030 *IJID oh) 03) 00p 820 0.1L) 413) (.3) p98 CA. 410) p48 a 4 4 93 a, 04-1 0410 a4 * 00 oil 048 Al 83.0 14) 0,4) sagam 6 423 . . 430 . ale"90 . 00)9 or"1 3 0 .t . 434 * ( .0..- *, * . )4 * 08 PI a2l as) - - 0.3 -sco 41)97 0167 4131 0444 0aim 41434 4,1)2 ,414) 084 no 184 043) 031 US (LS 04)4 090 64 0 308*c * ,1)3 0)9 03L1) 4101) 21.4 m71) 91 0631 a648 (9o 444id1 (121) 04041 898 0.) 0~1 0.3 018 * 41 (98) KS) * (98 04C) 4401 4102 el 01) a 048 414 C)0 "IQ4 FLU1 LI"33w go"74 clan1 "C60)CAM4 0346097 (141 4140 4 4 18Wclm4 0604 0 418 007" 004) 6001 418 am 009) * 1 omit 060 0037 Glow 3041914 0024 CM"9 P 044 a 04 0.13 MOD9 I&?1) 34.9 41) CD8I 08,1) 83.9 * 39o 023 U48 f* 404 o021 oil * u 64 (33.3) 0.41 4 (l 0413) (3.44 PC0) mQ'342 43339 -41LI4 4e70 ~441 4.30,1 .41604.4)) 4306 41) LU -43) -40 *b -6310 .0133 4846 9 OID -0234 aGl OH-o.43m .41 5)-c0 -41330 -0339 -13" 4410 CM"8 39.3 8643 88.3 4L13 0413 444)l 0.13 0 48 04 V 038 "80ft 348 4) 4) GL 04 041 03 10 a04 (00 0.13 (89 p38t 0.48 0 SW TI 033 6.13 0 8329 4833 034) 063 O 644 03" 0 0M1)ti a9) no 11)t 031) 8) 06 33 0877am 0i.9)i coo03)1 0 613 ail1339GA 034I Cli4 0831 "I'6o (190) ox3.1o (748 (341 4148 8.3 0,7) (1)11, 4141 0.0) 03,3) 4333) (998 0)s 0'41- (83 is 0013 064 0.11 00) 4)7.4 108.3 0.3) 333948 4la6 all" mis 030) 0.) 013" 0461 4.0,14 -am 0600 4132 419)7 * 0 449 043 4139at" 04.33 418 000 *a )0. 0310 6.030 4208 -4)60 a03)7 311o 021 8913) 421) 0.)) on 4a4) 897P) a2n 043 (84 83.1. al0 l .3 37) * 8398 43) a3 * 03) 88) 0a 70 1341 (3.8 CD98 pi) 65001 . .4)4 . 936) 40313,16 40946 - 45ol) . *ag 4039 - 4608 403) * p1 I39 98 *i 300) . (83 . 3 0)- 4n -- a3 - - CIO P13 - 8.41 Ca034 4 Gift 0)99 0131 04)8 0)9 4,10) 0344 0)417 0115 4184 * @410 003) 4.3194 048 41140 6446 a,4147 044 419)6~0 01)98 GA14 41431 4113 HISA? 4)4 ea3 49-1 24. 2369 Os9 63.) 6.0 31.6 41" O * in) I 8lis 39) il879 41 i ) is?44)6 3194 3)1 4.) 171,4 3)20 39.1 us60 0802 03" 6734 *065 0344 o 04 1444 4181) 0360 6344 071) * ll 0)1 34 0444 0us 001 6133?$ Ow1 003 007 66ov 0410 41419 473 UWAOR 2449 21104 27 Van) to" 3.922 0104 I-VW0 3,30 311322 3A143) 9I nu a 60 1844)3600 1.33 3.004 son0 1 I 01031 )7)0 33372 27281 3013 1393a SAM8) 32.3)3 1.10 "414 39.811 493) OH3) 42013 Om8 Ox") (too) 4173) 4140) C 0)3 99 07 3)0 823)0 0400 063) 073) IS0933 084 c(884 073) 4 778 017 084 O 0)8ai 0711 80018 9) WA" m 04 29.) 43.) 394 as6 33, 3).) 8)4 S&I 32.9 4 00 2 1 37 3 )0 430dj Z0) IS$ to9 49 0 4)0 33.4 427 00 2 01 37.) 1)7 010 23 0041 8814 (384.0 41098 064 0L1- 07.)) 981) P0o 0o413) 033 48 033) 8801 S 1.) 0t33 020 983I) PA344)C 04)4 394) (39,) 8444 (3911 00 01 .3) ttc 01.61 4A M*-)00 O4 WA003414033T3 1.0 44 7.1 435 3) 4.0 2.0 toC0 a a1 03 44 00 40 6 19 431 9 4 0.)j 31 a 9 too0 13 314 38.3 418 3.4 9)3 I= PS)8 (9.41 (4.98" 0.-41 O 4183 70 0 ) OL 0C (1.9 .3 44 01) 01 3 0) 34.8 a (74 88.32 3.117 413 (it348 4 048 10 lb 302 DI0.) (7)3 CD4 30 14.30 I 048 34C1 c 029 (72~ H3USUVJ111W01 0)8 4))s 4).) 39.8 440a 484 0)6 714 46a0 4)0 49 4 d4)2 At c 403 443 416 03 46 7 410 *I a0 At - 01740 43 0. 43 0034 4)0a 47.3 42 879 804 (23) 88498 8)0) (0838) (3.00 (6418 (1.32 go0co (3.11 Il 300) 1 (v3) 3)3 (7 II o0301 (343 p!q001 () M%)1 348 (00 8)9 0 89CO (7A ) (98 Ca07) 808, 88.ao *9gm 4) 3033).30W . %.i4 A-04636 The coefficients for experience and experience squared are significant in all the employee regressions, but less so for the self-employed. The coefficient of the fertility variable in the female regressions is significantly negative (-0.06) only in Rio for employees. Education makes a large positive contribution to expected hourly earnings; the estimated schooling coefficiernts are consistently strong. Table 3.11 shows that women employees have higher returns to education than men, while among the self-employed they are abo)ut the same. Nationally, the returns to schooling among male and female employees are 13 percent and 15 percent, respectivelv. The corresponding values for the self-employed are 12 percent and 11 percent. As might be expected, there are important regional differences. In L&e Northeast, the return for female employees is 18 percent, but only 14 percent for men. In the Northwest, the return is 21 percent for women, and 13 percent for men, while in Rio the values are 18 percent and 13 percent. In the Other Southeast the female advantage is 2 percentage points (15 percent versus 13 percent), and in the South the advantage is 4 percentage points (14 percent and 10 percent). In Sao Paulo, the schooling returns are the samne for male and female employees (14 percent). As regards the self-employed, the estimated returns to education for women is bigher than that for men in the Northeast (10 percent versus 7 percent), but slightly lower in Sao Paulo (13 percent versus 12 percen). 6. Accounting For Male-remale Earnings Differentials The usual strategy in analyzing male-female wage differentials is to partition the observed wage gap between an 'endowments' component and a %aefficients' component. The latter is derived as an unexplained residual and is called discrimination.- We use the popular 'decomposition approach, first developed by Oaxaca (1973), and extend its implementation to incorporate selectivity bias (Reimers, 1985) and the approach of Cotton (1988) that addresses the 'index number' nroblem. The decomposition analysis is based on observed mean characteristics (eg. education, work experience) and the parameter estimates of the selectivity-bias corrected wage equations. These regressions yield estimated wage structures of men and women in each work status. That is, the regression coefficients indicate the way in which the labor market rewards the background attribu.es. The basic question addressed by the decomposition method is: How much would the male-female wage gap change if men and women were paid according to a common wage structure, but their work-related attributes remained as they are? We now summarize the method.' The decomposition analysis typically involves a logarithmic scale which can be transformed into monetary units. For each group of workers, the difference in the observed geometric mean wages between males (m) females (f) can be written as: 1° The limitstions of the technique in meLauring discrimination are discussed by Cain (1986), Shapiro and Stelcn (1987), and Gunderson (1989). where the Zs are the average background characteristics, and the As and the is are the estimated parameters. As discussed by Reimers (1985), the obsxrved wage differential has two components - differences in mean wage offers (based on selectivity-orrected estimates) and differences in average selectivity bias between men and women. Depending on the magnitudes and direction of selectivity bias, the differences in observed mean wages may under- or over-state the difference in mean wage offers. Hence, the decompo3ition of the male-female wage differential 'hould be based on wage offers, and not on observed wages. The difference in average wage offers is gi.en by: TrWn - nf = (# - if where: mW;5 = NW; - ,t TWf T' E - j/1 The decomposition method is straightforward and fouses on the issue of 'unequal pay for equal productivity-generating characteristics,' or wage discrimination. The decomposition of the gap in wage offers centers on differences in mean charaterscs and differences in the esimated returns to these characteristics. In other words, if the esimated returns to the chaacteristics are the same for men and women, the wage offer gap would be solely attributable to differences in productivity-generating traits. There is an index number problem in applying the technique: Which comornn wage structure (estimated coefficients) should be used as the nondiscriminating norm? Since there is no clear cut solution to this problem we perform the analysis with three norms: (1) the male coefficients, (2) the female coefficients, and (3) a weighted avrawge of the male and female coefficients based on the proportions of men and women in the sample. The difference in aver2ga wage offers can now be decmposed into two components: A portion that is attributable to differenca in regression coefficients, and a pat that can be atributed to differences in endowment. lher are at least three ways to compute these magnitudes. 1. If the male wage function is used as the non-discriminatory norm: li - -4 - A + (Z Z - ) T'he term on the right is the endowments component and that on the left the coefficients or residual component. 2. If the female wage finction is used as the non-disimtory norm: NW." - nMPf - Z X X + k - Zj) Again, the term on the right is the endowments component and that on the left the coefficients component. 3. A third alternative suggested by Cotton (1988) is to define the non-discriminatory norm as the weighted rerage of the male and female coefficients where the weights are proportions of men (P.) and women (Pf). Let ° P. Pf, ThWU nIiW - * _) + Z - + (Z - Z4) As before, the third term represents the portion attributable to differing endowments. The fist and second terms divide the 'unexplained' residual into two parts: A 'premium' or higher than expected returns for men (the first term) and a 'pealty' or lower than apected returns fhr women (the second term). In the decomposition 2nalysi8 we analyze wage differentials first between married men and women, and then between single men and women. is is done separately fr employees and the self-employed. Esdmates of DfsaLd,ation It should be noted that the decomposition aulysis is carried out using a iogarithmic scale which cEz then be tranformed into mDnetary units - cruzeiros.1 Since we are flexible regarding the choice of a non-discriminatory norm we present three estmates: using male coefficient weights, female coefficient weights, and a weighted average of the two. The endowments and coefficients components are reported in terms of logarithms, cruzeiros, and percentage of the gap in wage offers (expressed in logarithms). Before proceeding with the findings, we emphasize that there are some important limitations of the method, which in-ude missing *ariables and errors in measurement problems. lTese shortcomings ce the teruinique should be borne in mind. Manried empIokyas (rable 3.12). Looking at Table 3.12 we see that the naticnal observed wage gap (row 3) is 29 percent, but the magnitude varies across regions. It is highest in Sao Paulo (41 percent), followed by the Northeast (39 percent), Rio and the South (about 25 percent). The lowest value is in the Northwest (15 percent). After removing selecdvity bias effects, there is quite a dramatic change in the gap in average wage offers when compared to the observed differential. At the national level the gap in wage offers rises to 33 percen (row 13). However, in Sao Paulo and Rio the gap in wage offers increases to 122 and 88 percent, respectively; in the Northwest it falls to only 5 percent, and in the Northeast it remains unchanged at 38 percent In the South the mean wage offer of husbands is about 40 percent higher than that of wives, whue in the Other Southeast it is 22 percent higher. This suggests that in most cases the observed wage 31 In 1980 the official exchange rate wu about 40 cnuwnr- $ 1 US. 3 See Zabair and Arruft (1985) for a simila compqniso of howly eamingis for Bnish married m and WomOL ....lw. uii_ b uuc, ot course, to ainerences in selectivity bias effects in the wage equations of husbands and wives. The decomposition of the gap in wage offers shows that the 'endowments' component favors wives, but this is far outweighed by the 'coefficients component, which penalizes wives in favor of husbands. Using the proportional weights (rows 18-21), we see that, for the entire country, the endowments portion favors wives by 70 percent of the wage offer gap, but husbands have coefficient advantages of 170 perce.t. In monetary terms, the endowment component translates into $15/hour while the discrimination component reflects $33/hour. This pattern - .ndowments favoring wives and coefficients favoring husbands - is similar in the other regions where the wage offer gap is large, and it seems not to make much of a difference which weights are . .lerln as the non-discriminating norm. In each case, the coefficients component, as an estimate of discrimination, is in excess of 100 percent of the wage offer differential. There are several possible reasons for this large unexplained residual, and the attempt to measure 'discrimination' is faught with well-known difficulties and limitations. First, the presence of missing variables and errors-in-measurement bias may have unpredictable effects on the decomposition. Unobsved factors originating outside the labor market (e.g., household responsibilitles, quality of educadon, ability, motivation, and aspirations) and imperfectly measured observed productivity traits (e.g. work experience) have, no doubt, influenced our estimate of discriminatiou.' Second, we n3te that the decomposition of the wage gap is 3ased on the 'iverage' man and woman, so that the entire weight is placed on mean observed claracteristics which have a large dispersion. Recent work by Kuhn (1987) suggests that an improvement could be made by examining individual-specific measures of the 'unexplained component. This iznplies that it may be deairable to examine the distribution properties of the residual portion among individual women and men. Tbird, there is some evidence that the OLS estimator used by conventional analyses such as ours may result in upwardly-biased estimates of the unexplained male-female wage gap.-" Ftnally, in examining the regression coefficients and mean characteristics we see that the intercept differences account for the largest part of the unexplained wage gap.' We caution the reader, however, that no importance should be given to this. For, as Jones (1983) has shown, a firther division of the unexplained Zap between intercept and slope effects is not independeat of the arbitrary measurement of the explanatory variables, so that it is impossible to uniquely disentangle the portion of the wage gap between coefficient and intercept differences. With the above qualifications, our decomposition of the wage gap between husbands and wives suggests an interestfing conclusion: If married women and men wage earners wel -Mitl according to a common wage structure, the wages of wives would be at least as hi&i as that of husbands (compare rows II and 12). he married sdf-enployed (Table 3.13). Ihe decomposidon analysis of the earnings gap for the self-employed is shown in Table 3.13. Two versions are presented. We present the analysis of " Evidence on this poi; is given in Obsfeldt and Culler (1986) who show the a nopammetric smeanng' estimator yis a low meramo of the unexplained portion than does trdoitioal OLS. " When we ez fthe conrbution of each variable to the *discrmina±ionD component ({at reported here), we fotmd dLt, il ns cas, the reurn to background characteriscs tended to favor women rther than z, and thus con±ribod to narrowing the wage gp. Te differeace in the waie-female inlercepts far outwighed dtc coefficent differences of the explanatory variables. 9 -e e - e - e a a 8 EC, n " Q7 q< a gAos Ia lfi t t 1 , 1, 1? 0 t]~~v 0 340i ] S o 9 e e 0- 41 .- o C, * Ce o A i ~~~~~~~~~~~~~~ S I301.AS I I~~~~~~~~I19 -C the husband-wife wage gap for paid self-employed workers. Then, since some wives (and a small number of husbands) are unpaid self-employed workera, we impute a fitted wage for them and decompose the wage offer differential between all self-employed wives and husbands. The two versions yield a similar set of results. The observed hourly earnings gap for BrazD between paid self-employed husbands and wives is 48 percent (row 3, right panel). It is lowest in the South (39 percent), and highest in the Northeast, Rio and Sao Paulo (56-59 percent). The values for the Nort.west and the Other Southeast are 53 percent and 47 percent respectively. We also see that the wage offer differential (row 13) tends to be smaller than the observed gap. Nationally, the wage offer gap falls to 37 percent. Similarly, in the Northeast and Northwest it drops to about 42 percent; in the Other Southeast to 36 percent, and to only 17 and 18 percent in Sao Paulo and the South, respecively. Only in Rio does the wage offer gap (59 percent) exceed slightly the observed differential (57 percent). The decomposition of the wage offer gap yields results similar to that obtained for employees. The differential attributable to coefficient differences (the 'discrimination' component) is strongly in favor of husbands, and this portion outweighs the endowment' component, which usually favors wives. With the exception of Rio, the coefficients component is in excess of 100 percent of the gap in offers of hourly earnings. Unlike employees, an examination of the contribution of the explanatory variables to the earnings differential showed thaz the coefficient differences of the work experience variables was a consistent major contributor to widening the gap. The conclusion we rach for self-employed wives and husbands is the same as that for employees: If productivity-generating traits of self-employed wives were rewarded on the same basis as those of self-employed husbands, the hourly earnings of wives would be at least equal to that of their husbands (compare row 11 and 12). Singk enployees (Table 3.14). With a few exceptions, the observed male-female pay differential is smaller for single than for married people. Nationally, the gap is 18 percent, but the magnitude varies across regions, from 6 percent in Rio to 34 percent in the Other Southeast. However, looking at the gap in wage offers, we see that in three regions (the Northeast, the Northwest, and Rio) single women fare better than do men, while in the remaining regions the average wages cjfered to men exceed ttose offered to women. First consider tne differential in the three regions where male wage offers exceed those of females. Looking at row 18 (proportional weights), we see that the men are strongly favored in terms of the coefficients component, imp!ying that women tend to be paid less than men who are otherwise comparable !n terms of background characteristics. Tbe male advantage translates into a cruzeiro gain of $9/hour in the Oiher Southeast, $' 1/hour in the South, and $60/hour in Sao Paulo. The story is different in the Northeast, the Northwest and Rio, where female wage offers exceed those of men. In the Norheast women have average wage offers that are 77 percent higher, while in the Northwest and Rio the female advantage is about 40 percent (row 13). The decomposition of the wage differentials shows that in each region women are favored in terms of the M We also perform thc deompositon for employee based on the regressions that excluded the public sector dummy variale m the female eqution. The results did not change in any significant way. 2 22 P. ts - -E˘tl e d *' ee de o' d 4 d d a e e 2 Id _ _ e ._ _ _ . 0. ';- o" -o- -t -t - h A a --:~~~ t a j j I ql 2P ct e .4 i Is _ 1_ _ _ _ _ _ _ _ - - A 1: _ _ _ __ 1, _ _In P _ _M ! 0 - . _ _ _3_ _ . _ _ -3 . .- - A- -- S . _ _ _ _ i ne .'.ese soJ-empwoyea (i anie i. 1l). Fmally, we come to the results for the self-emptoyed shown in Table 3.15. The analysis of ths hourly earnings gap t;;at shows the observed es&iings d-ffe.eatial (rew 3, right panel) differs from the gap in offered wages (row 13). For exarnple, natiornlly ard in the Northcast the obser-ed hourly ware is about one-half that of mea. However, the offered wages of women are only 23 percent less than thGse offered to men for the entire country, and in the Northeast women have an earnings advantage of 28 percent. The de.rnposition results show that nationally men are strongly favored in terms of coefficient differceceu, wbile in the Nn'Lbeast the rverse is the case - the coefficients component strongly pen3lizs rnen. The pattern is similar when the analysis is performed for- all (paid and unpaid) Self-eraploved worKtrs (see left panei). In surn, the decoaxposition analysis for single men and w-omen yields s much lower measure of 'discrinination tban that far wives znd husbands. ltis suggests Ih- facwrs outside the labor irket (household responsibilities, fertility, and work interruptions) probably play an Jnportant re-l in the earnings disadvantage of married women. 7. DiSCuSOIs In this study we ;.ve Xusc - ":om the 1980 Census of Brzzil to investigate deter=inants of labor force stmas and earn,> mn samples of marTied and single men and woren. Uilike most stud.es on la',or markets in div.oping countries, we aralyzed labor force siatus in te.ms of a three-choice model I=P!oyee, self-vplovya, no work). Moreover, ojr analysis of earnings deterraLuants u,-ong employees and the slt'mployed explicitlv incorporates the seiection of izdi-id.als ainoZ, the three tyrs of labor force rvw. National and regionsecific selecwvi':y co.-ected wage regressions were then us"d to ex2mine the e-fects of human capital and oter w.%;e-det-:rmining characiics, and w explore the, issue v'f male-female wage eUr.e4enaials among wage and self-e:Tloyed wrk&es. Mein Thdinp 1. The reg nal diversity of Brazl is clearly reflected in the estmated rDodels of labor force staus arJ earnngs. Ihis pluralism leak us to conwmr with Pirdsall dn BDuman (1984) that much care must be taken in rechifrg conclusions from euw.x for Brazil 'as a whole.' Nation-wide estiw of Io'icy-ferdmv consid?nriovs such as returns to human cepital and ;erAer dispar;dt' may ;e ski!euding. 2. It s important to an,-lyze the labcr foxe stants, tspecially of single and married w-mne, m, terms of a three-choie auntext For example, altzhough educ aIon plzys an important role In dermining overall labor mark- participation, the principal effect is to increase dte propessity to perform wage work; ie schooling effects on self-enptoynent are generally t-i:.1. Our =apirical results suggest that o.dding labor force behavior in developing nntriwes in terms of only a 'worc or no-work choi.e is liktly to n.ask imnportant aspecs 7- +c underlying determiin faCtor.i. This paint i! rpecially releant for women, and, u;ng perods of prolbrged recessic, r mn . we!. 3. Our results retiec the w ll-dcumned findings hat zh ..presence of yong chidren ^ t>; home rd more adanced age discourages mrried wonen's labor market per~.&au-i. .'2~ 497d- d9t q .e A- ~*~ .e e*' A-E - 0 r, 8. %f 5ke,R 6 t bE,@ I a *I FE a I qR 't9t Fe8a a R, A i~srB -la'W q , P ma 8 5 iQ. la n "e.t8' & _ |8" @ is2 R e 3 ° 4 * e. g- 5 a $ ; g r tg ; .$°°-| -h " g $ @ q -2 -ItJ o It- g P -g . A= .'I-e'q~ q~ -~ e .-~ ~ .g. .~ - a -9~ . o It~~~~~~~~~~~~~~~~I -6 a Is c, Is: IC d ~ a 4. Urban reside-ice is conducive for both men and women to work as employees and also enhances their earnings. lTe effects, however, on being self-employed and on earnings are generally weak. S. a r estimates show tha^. education is ari important, perhaps the most important, determinant of labor force status and earnings. We found that education not only enhances earnings, but plays an important role in 'sorting' individuals among the alternative labor force activities. Tlese indirect 'sorting effects" of schooling strongly sugga;t the importance of incorporating them in an analysis of earnings determinants. Schooling eifects remain strong in the selectivity-corrected wage regressions for m?1es and females, especially for employees, but less so among the self-employed. We also note that onr estimated returns to schooiing are generally bracketed by those obtained for other LatiL- American countries. 6. Finally, we comment about male-female earnings differentials. The decomposition analyses revealed that the gap in wage offers between husbands and wives could not be accounted for by differences in earnings-determining traits. This residual could be attributed, with caution, to labor market 'discrimination." Our results show that f the given wage- determining characteristics of wives were rewarded on the same basis as those of husbands, the (hypothetical) average hourly earnings of married women would be at least equal to those of their spouses. What makes us1 uneasy about calling this result discrimination' is mainly that it is based on data and estimation methods that are subjec-t to well-known limitations, including husband-wife (life-cyc'e) productivity differences tiat are not easily measurable or observable. An indication that this is likely to be the case is revealed by the findings for the samples of the relatively young men and women. The 'unexplained' wage gap was generally not an issue; in fact, the decomposition analysis showed that the labor market tended to 'favor' single women. An important question, however, is will the 'premnium' persist after they marry? rolicy Implicatons A deAde has passed since the 1980 Census of Brazil and it might be asked how the findings of the study are relevant today? As stated in the introduction, after a period of relatively high growth in the 1970s, the 1980s were a 'lost decade' for Brazil. The prospects are not encouraging. The statistica for 1990 show that Brazil experienced its sharpest recession in a decade. Gross domestic product declined by 4.3 percent (US$12 billion), the monthly inflation rate was 20 percent, industrial production shrank by 8 percent, and over 250,000 workers lost their jobs in the state of Sao Paulo alone. Moreover, little p-ogress has been made in reducing the massive debt of $125 billion. The reduced household incomes and prospects for wage employment will dramatically alter the structure of labor markets. More single and married women as well as children, who previously did not perform market work, will continue to join the labor force, probably as self-employed workers. Men who held wage jobR will either become unemployed or self-employed. Large groups of young peopla are unlikely to complete their schooling and those who do will find it increasingly difficult to find wage jobs corresponding to their qualification. Both groups, no doubt, will resort to self-employment. At the same time, the sharp reductions in federal and state social expenditures on health and education in the 1980s is likely to w;ilInue. anu etrnngs, any onDQusions anm potlcy Implications are made based on our crosz-sectional (static) data. It is quite reasonable to expect that the magnitudes of the responses we have estimated may differ between 1980 and the preseat due to the business cycle. However, it is unlikely that the direction of the results would change. This is confirmed in the next chapter. Nonetheless, it is important to study with some care the dynamic (including stability) dimensions of work status and earnings determinants. One, and perhaps the only, way now available to do this would be to redo the analysis using a more rocent data seL Adopting our approach for the 1990 Census or other recent data would be particularly useful given that 1990 stands in stark contrast to 1980 in Brazil. Knowledge of the determinants of work status and earnings in both expanionary and recessionary periods suggests that policy can be conditioned on current and expected aggregate economic activity. The overall effects on the development process of the structural changes in the labor market and the budgetary squeezes on education are not yet well understood. This study provides a benchmark by which the consequences of tho Brazilian economic crisis mnay be compared. The next chapter examines the relevant issues using data for 1989. OUo 011 0i-3 3- .34 mm 40 ec00@. ag.DY 0zm skaoas0 30A o Tod 5 0ld CA4 3.40 7.31 33.1.1 mm. I" oil J.33 33.03 (1.6 I a AA3 33c.00 pA 9.50 9.0 a -313 0-133 (12.0 0.-0 O-33 9- (31.25 01.90 pan MJ23D Coal 2323dM JI O4-4 81.4 1[At 7-n3 "2 Km- 0.IS 049 3.33 13,31)2 OA5 to0 4- 4.1,11 PAX PAS) (.34? 3 (3.7) CIL5 do1 0-12 9.74 WAR14 01.31411 05,47 aos OA 145W-. 131 0. 043 OLU 0.20 24.04 1101 3 0.23 4.130 11.33 0.1 gig (Lo 491s pA" P.M (120 0-535 P3.M4 17-44 083 043 513.444 03.5135 p43 0431 u 7874r am9 . 3.IS 7.72 341 41 U 03 1.4234u 30 ai a 3.4 0* 31 4a 33.0 Al 9733 31.55? 0.132 (12.11 to3co 033 043 30,54 (19.212 01473 0-37 Py M 4? AA1 Om4 I33 7051 M43 A4." 5.73 0 34 3.43 7.13 to1 CM4 &A4 Ł3.1 70 0.3* 9.7 O.57 c2.ou 7 oill II as"3 9435 coU (7.0103 .9.47 0-33 P4 73 AM4 OA4 3.2 L.7 87.3 32 Lf 02M 2.14 6i11 11.7? 4-0 R -4 US4 0079O70 0.3) 340 0321 (it22) 0-013 0-9Wu 0423 9.33o MAIM4 950 0a-M O %i CU3 03' 024 3.34 7.00 341 31.73 a SO 03 2.43 3T0,3 a0 Jed 170 LIS0 9735 90 a (3 a.0 33370 2301 0.= 23s to 5 03.4503 71 02 44 pie To. ae IMADn ale 8.3 3.30) 7.10) 34103 PO 31U0 2.3 047 U 3 31.03 040 4-U31 0.5? 8401.Jqd 3.11Lis 045 3.10.3 803 3447 2.53 034 1430 11.2 0412 02 0m 020 0 07040v 040 044 Ma0n3 (1III 0.40) 043 gjj"Q asm 64 L3 ~~ 233 0.42~~~~~Ol 0.23 3.23 7.7 23.3 34.3 La3 @1 40 3013 03a do9 La 0.8 90 ode 0.121 a-So3 5135) 01-50 9.40 9435 9.14 06.3 9.43 0-40 0.1e b"441n 045 0am Li 1. 451 0.274 3L 30,17 2.0 a.13 4,4 ML7 Is1 8.1 0.50 Was0 e. * 0- 04 (it.9on 040 KIM0 044 03.33 MAIM3 9.43 044 Oi 833 045 -42 L.34 7A8 3.23 3445 Lo0 023 4.23 33.430 GM LA .0.4 00.1 9.1 94 .4131 0-74) a3L" to de3 0323 Was3 ca3-SOC 02.3,41 L940 0-23 20o 0.23 0.11 3.23 0.- 34.3* 14-U 3os 40.4 4.1M mm.0 an5 31 017 1.41 0-4M SAG0 OA)3U3(2. 09( 3.7 70. 0 1.03 MAIM 03.33M 904 5040 0-4 040 us.7 AI 9.A3 9. a M.0 M0. 0.81) 9-35 0-3 O.43 (32.14 07.904 A-11, 0L4 102e- 37 @3 0.23 $0. 403 7344 12.70 1.45 0.3 A;IQ $50 0.1 5.ll its 4017 W4 91 02 0. 23-OA MM4 f0.4 CIA31 9.473 Milo MAIII 940 0.33 *J - 0.go 0- 070 0.79 17.03 11.3 4.701 GAS tin1 T7.211 am0 1-23 0.31 71,011 AM a.4 to4 0.34 0-43 2.2 013o 0.73) 9.435 3)e43 0C.33 9._ Ck s b4334345 3so 0.n 0.2 &s 7.221 0.1 10.03 L.o MM1320 10.3 &A3 340 an 3.30 9.50 9.23 0.013 0-13 0.31 0235 9.44 940 90.4 04k350 OM 01-14)35 -ww 333 4.23 0*9 GM5 703 3IS5a 31.3 LCM 0.706 mm0 20.40 km3 2.1 a." 4l13 0.1 0.10 0.11z 7.03 (11.2 43 - 05 .3 1.7? 01 M 01 33.0 OM0 c.33 0. 0.3,4 OILOS 52 30 4 02.330 00.344 9.2 040 41 231 04a 0.2 am1 0.3S 3.173 34.4 4.3 0.Al 5.03 30.0,10 0.14 03 0.58 77.50 PAM0 MAN3 0.23 PAO3 mm.2 9.73 G.2 0.40 0 04.244 9.44 0-23Q Gn 02 0.1 3.30 7.33 31,14 11.= 0.3 047F 33AW3 32. 4 145 an5 am7 am1 0.n 0.00 7.82 23.42 12.3 L-ai 0-74 140b 20.13 045 04 4 27.301 9.3 92 00 005 0.3 (.7 .-3 9 02011 01.47 9.0 0.423 03 J11101 -2 57 73 M . 2744 43 1.11.0 34 1 1101 Sig am1 a"4 7.1 3.45 3 La5 2 0 415 4.31 3. am3 1.2 0. 0.01J 0.10 Ltg9 034 87.70 311 LIS 0a" 0.11 32.3 0.4 2.4 0.4 LIS3 9gin P.m 040 CL40 (31.44 00.5 0-34 9L23 0.33M a0.33 9409 0-03 lb"ARPOM on -7 0.5 0.3 7.2 4.23 ILA0 4.1 0.701 .904 .13.014 0.n 0.," 0.11 "I'm1 am4 eIWA 0.23 0.00 0128 9.3) 9.50 0.43 - 9.744 (17.943 9.44 0.40 T.W J 7ns 42.91 11,1 LS2 e1) M &6 .69 u.s azsD oxnJ oLA us s ale 0 asn m 0s.4 0 a ml% aw& 019 (L" ) O.Q 4J nt40 OAS) OJ& &*-wkp m11 7.t 41-51 LI § O 01 {11 1.X 017 04 *e0as a-9 oIL (ILu7 o141 o" o3.52 SW1 11M KM 7 it 04741 Oit I as GM $.ft ooq o.91 TW 7U .7 41.9 it" Lit" all SM &V 4-1 7m 7M O.L!7 Ai." a.7n &s PnAO 0n3 a4 IdL0b_ 3m *m J?,JX L&3 .4 H ()t I .M 0.7 an 1s1X CA ° L.Ia (11.7i C 0 P ^ 41-13 a^ 543) a 1 1ap m 71 j~ i" I OZI 10 CM 07 o"s arr (117" f"J" awo an ouJI ol. ee4 n1.7 0= 03.34 naL74 0.lt O axo 11" 41-111 aJis 0. 4 ODA Q lll Xa Tow no 7AS 4 It 143 Sa tu 5>0 _ S 7 0~~~~~~~~~~~~13.1 (IU J LD OJ U 1J,0 an O 7.13 5.b WS 0 3 O S ;7 41.7 0.4" O am 1 3rj 't- IL" a .7 4311 o 41.71 in~ ~ sa F_ n -law e o a c Tod 2rz 41.13 vto tsu ' mm t7,AD 0- 0 Med- 486 La SL4 IL a. " T UA cam . P." aow (II.O n2- P3O Win a$U0 eOA aLV nWp. "a 411 ^" Je" L3 am km lks . GLS a LX 0$4 I7)0 0A.01l) PXv o U _wM&p S 7 77w n1 11.110 a1 n lty 070 e tnbr mW 70 SLD ".l ˘ On GM U71M ZS Ws OA OO LIO Q47 0,3 IsAM "t oil TO X117 7-1 O ILn 4. D7,10 am7 SAXI O" T_ e b6 tU11 ai a771 31. LaO 0=I 14.211 0.1 OA a-gos1 OL7 OLXIM PJlE} O."I,^ OA) 4.4 bvr 3' B 7. 0 "w " 4,8 3 O.90 am0.77 4#DS I a-m OL7" 11~sa 0LS St CD1.% 0S 0X _ udomwanotae __u Amdzof Tab 3A.3 Mom and Vaiuale Uaod La Wagr Rlsmaaiom mwagi - mu mewm inimin ~~~~~~~~mm even" eam maMa m mm u, E,MMa. EMMWMM mm mwra. meiDvm mm wmm.am mume COO elUg Ises all Is LOS 41W MI6 ON I a. l ama so el" S allv I an) on SC It AlP 114 17 II 14 W OSm M 533 CD CI1,4111 SAC a CLIUS IL III am LIUem VA1 OM, CU l I I) *C II CI 4 IC as C pile *lC e *m *m ~ pi~ *u pm em *1~ cm pia .me Pas PIG cm pus Pm p111em,4114 P14: 654 6311p.m p.51 495 4111 Sailo 43141 CI 5 11 M Xl Di 1.1 1.. 15 1.1 SN 311 DI 34 2 1145 T515 31W 31.41 44.4 15.46 1444 31,46 3l 0 (ale CilIa 014 (h.P 141.51~tal alP1 (alP1 0171 i'sas co I sI 313.4 us4 UIC UIlm use4 "lPa (1s11 a ) 0441 (,I2#,3.) IS Pll 4 Il 14 US~~~I3 Sal 453 I4~~~~~~ 94od 940 &PPy 11.41 Sale 13a 491 11.11 o 14 ISOU CCI9114 CU CU i III 144 ia cia aLsa aIIa CU 1 IN 11.1) (VWM I41 as . ll i . 1C om ps m CI) c m 01 c m ii 1) 05 b 11"m as 15 0141 01l) 0445 Po 41 TWIDIIC 6,14 ICI LW LU 144 ID 331 IC? 4.4 0ll 3411) 9 (I 93 114 411 Cal Cii 14 lal0 Pk9 13 Sa 4W 154 45 3 Poayam 77 P - 1141 * a - 1 04 o 0M Gl - 0) m - 04 .6 - a (MDC 1.1 ICC S.C~~~~~~~4 III 611 L14 .44.33 UC 3.4 14, CC, LI, I, oil aC a 1 I 84 1 a i 14 13 32 - C Mi 3.1 419 119 MA~~~~~~~411,11 MC- 451 N . 14 D i l Cl 4 11C U C C 31 MC 0. I1 21.J43 1. *alC 3123 11.5 354 14 Cl lIt ICC ~~~~~~~~~~~~~~~~IC.? fl 1111 41 4 i 1. 14 CI M C M 1 34 MC PO 4 IS 7153 1. OILC (SI 0CC44 can C PC 1441 M C 0 PII" (C (iC 0ateCC , ,t Pay 41 U 31 04 4.1 144 4t 414 05 U 4iU 144 CM~~~SA 1,1, 1 144 K L4S CULa 64 141 SU a CU1 o t T 1.14 I .i a t I3 III L44 IL: M 3*"= ' J I4 II - 1 1,1 1 11,1 S ISO CASCUII StCC 144 ICSO.$ L? CUo44 145 SU SI 1WP.0 duC Us 1Ws MA CI C SM aC1.CI a CU CSLC46 at11 EM EM h9~~~~~~~~~~~1111 SA - LT : T e Si r a§eo5 e ai ed 4 oo ce to .ei toS f Z d § $ t 5 .@ e a vi o oeoe o z P e oeoe o of ;eot4i R ,o# g qn t a E A , z ; Ptn!n 3at4°t{ 6d02 =7 ittt Gr giE;E,i kg.Sl qta36fl loJO~~~~~~ Al;5_^ s&s*- *w˘t ;.s1:; 93 9i F! El 360FW2E ;t1 a5EŁ l ~~~~~~~~-i 3Ei ti' c4 4 t ri 4 eQ " a -0 a ^s-- . C v - 0c c- c c c d qEESS AEEE, a§Re SEiS. 2 , wE:1sn It; vl f-n9 aznFE' Ii o P- 0 P- ci P, e P- OSOSOP-ap- opopopot $edp~pop ci ci - c- n _-6 FF - b°d c58 1 ~ oooooooe do.eooooo oeOaO'tO' ao~O~OOO o~o°o~o2.;e oEtEd ~oEoao2 nE_ |Do ogotoIIt oRo B 6"Ri ji3Rno2it12tRa 0§;:t" I IIiIlt"5i §2 i 0 0 OL d2 0 P- 2, e P P- 0 0 a a , P- e n- o= - n e n n e e - _ 0 _ P _ c H3 X { il } I bl i i iX } t lX } 4 sX X I 3 } E III0 WI 68 C i. SW Cl 051 a W11C I as4 It" "l. Idi out1 MD tilL) ull aL tt i't liUl lUR5 a"l cut $Ms ""I "aS 61161 fil la] I wA1 Ł81 Alt "al awli U.l awl Will WWIt am&I * I I i I I5 c -L 0 i ii i i S i O It II II St VT $I it, it I it S I *Ito i's Its itr Lt -*i Sf to toK i 5 t i irn i S i S St l l t i go ii i Cl to to 'C Ii LiA l to LI it 4`5 Lt St 54 VI it VII' I it 51 Lt i S N IC,L*11 "Io. ll: LV Sip CU " 0 &M cpi lb i &dA` IS c- 5VP P1? 4I, 41W1 VL4 I., rip SC Sat II, go1 al. w oisli * il-p ilL MC Cl Al CL WI Ut WI III Cl WI Cl Ut WI 511 iii ilL PIt Ut U.S Ut ftp WI Wi CL Ut. I: Ill 'P, lo: A illr ALi) Aii f iico ss* si .m 4s n I W is K Wi 0li LI W1i `I WI tOt "It W1 Vri WVI OOI a ii'I at I * 43,hip. (US 159i 0 Ciam) 4 it .o ai) mIi Cii in iri : 5 iC). midw OKII it i) r M ) inl In mIt .L' Clr) Wild tarS Clio uit uas mid e I1 oilI IlL9 n USS N VlI68 tIl UW iri I9 TV I851 As5 I Wil WLS 11a5 I a NIl Isi Wso USo "Ii 4IU at I "I1 SUM= I Lw. Ins UPS 555 1(50 SIt Lilt itO bit CCC (155 SIC Cod biW mis ets (itS 555 mit bed sue mIiis mi,-s meI bid 655 4usdbd II-d :* tz-o ko *L 41" 4ul #"l 0* 1. lit" n Vat a LO 11 a WS 0, 0USk $ 1 file n Irl, o as tiC .t '11 fi ml ". W4I' W"s ir I" loti ts U iii mn oil LA Us1 1111 ta a iW La mi in mil Oll a"0 Wl "aS UN LW ___ .Wm iopoxA Zs mwMa mms MPMw p mosrotsqMw m mm smm spwf o pmrm sUN -opom mm ml omP wogot MR mm o m NM 035u Win- Icu0iJiW Jim aU105JM -= I4l CnA&P jiM WIM% ONAiUW dM SAOPOU CSASLoom U.AUU mOAff.. CM ONi*'Mi mlAC).U tiN WESI "Al"5 0M1)4 U. odVI so - AUVZSNAAMI AmliLlUNMWSCMA TI, -I& p- -m o1btM - -opt-le 111110 In PCSfk 191qWSA Pil MOM 9-vc Mq.j. 2w-div References Arriagada, A-M. 'Labor Market Outtomes of Non-formal Training for Male and Female Workers in Peru.' Economics of Educaion Review, Vol. 9, no. 1 (1990). pp. 331-342. Becker, G.S. Human Capfral. New York: Columbia University Press, 1964. Behrman, I.R. 'The Action of Human Resources and Poverty on One Another: What Wc have Yet to Learn.' Living Standards Measurement Study Working Paper No. 74. Washington, D.C.: World Bank, Population and Human Resources Department, 1990. Behrman, J.R. and B. Wolfe. 'Labor Force Participation and Earnings Determinants for Women in the Special Conditions of Developing Countries.' Journal of Developrewnt Economics, Vol. 46 (1984). pp. 259-288. Birdsall, J. and J.R. Behrman. 'Does Geographical Aggregation Cause Overestimates of the Returns to Schooling' Oxford Bulletin of Economics and Statniics, Vol. 46, no. 1 (1984). pp. 55-72. Birdsall, J. and L.M. Fox. 'Why Males Earn More: Location and Training of Brazilian School Teachers.' Economic Dvelopment and Cvltura! Cawnge, Vol. 33, no. 3 (1985). pp. 533- 556. Blinder, A.S. 'On Dogmatism in Human Capital Theory.' Journal of Hwnan Resources, Vol. 11, no.1 (1976). pp. 8-22. Boulding, E. 'Measurement of Women's Work in the lbird World: Problems and Suggestions' in M. Buvinic, M.A. Lycette and W.P. McGreevey (eds.). Women and Povry in dte 7hird World. Cambridge, Baltimore: Johns Hopkins University Press, 1983. Cain, G.G. *Thn Economic Analysis of Labor Market Discrimination: A Survey' in 0. Ashenfrlter and R. Layard (eds.). Handbook of Labor Economics. New York: Elsev er, 1986. ,otton, J. 'On the Decomposition of Wage Differentials.' Revew of Economics ard Sauia's, Vol. 70, no. 2 (1988). pp. 236-243. Dzbos, M. and G. Psacharopoulos. 'An Analysis of the Sources of Earnings Variation Among Brazilian Males." Eco;ucs of Education Review, 1991 (forthcoming). 86 _ _ _ _ _ * _ -_ _ _____ _ __v- - _ w _ o _ V X . C4;0. Social Science Working Paper No. 698. Pasadena,California: California Institute of Technology, Division of Humanities and Social Sciences, 1989. Greene, W.H. EconometricAnalysis. New York: Macmillan, 1990. Gunderson, M. 'Male-Female Wage Differentials and Policy Responses." Journal of Eronomic Literawure, Vol. 27 (1989). pp. 46-72. Ham, J. and C. Iisiao. 'Two-stage Estimation of Structural Labor Supply Parameters Using Interval Data from the 1971 Canadian Census.' Journal ojfEconometrcs, Vol. 24 (1984). pp. 133-158. Heckman, J.J. 'Selection Bias and Self-selection' in J. Eatwell, M. Milgate wid P. Newman (eds.). The New Palgrawv: A Dictionary of Economics. New York: Stockton Press, 1987. Heckman, J.J. and VJ. Hotz. 'An Investigation of the Labor Market Earnings of Panam an Males.' Journal of Human Resowuces, Vol. 21, no. 4 (1986). pp. 507-542. Hill, M.A. 'Female Labor Supply in Japan: Implications of the Informal Sector for Labor Force Participation and Hours of Work.' Journal of Human Resources, Vol. 24, no. 1 (1988). pp. 459-468. Jones, F.L. 'On Decomposing the Wage Gap: A Critical Comment on Blinder's Method." Joural of Human Resources, Vol. 18 (1983). pp. 126-130. King, E.M. 'Does Education Pay in the Labor Market? The Labor Force Participation, Occupation and Earnings of Peruvian Women.' Living Standards Measurement Study Working Paper No. 67. Washington, D.C.: World Bank, Population and Hutman Resources Department, 1990. Khandker, S.R. 'Labor Market Participation, Return to Education, and Male-female Wage Differences in Peru.' PRE Working Paper Series No. 451. Washington, D.C.: World Bank, Population and Human Resources Department, 1990. Kuhn, P. 'Sex Discrimination in Labor Markets: The Role of Statistical Evidence." Amerkcan Economic Revew, Vol. 77 (1967). pp. 567-583. Lee, F.L. -Generalized Econometric Models with Selectivity." Econometrica, Vol. 51, no. 2 (1983). pp. 126-130. MaddaJa, G.S. LbUnd-dependen and Qualfratve VariaNles In Econmevlca. Cambridge: Cambridge University Press, 1983. McFadden, D. 'Conditional Logit Analysis of Qualitative Choice Behavior' in P. Zarembka (ed.). Fronters In Econometrics. New York: Academic Press, 1973. Mincer, J. Schooling. Efp kence and Eaningr. New Yoric Columbia University Press, 1974. 1AVjit4, ru'.. v%'I y t.. "(6c, w"'" 'ciJsuc uI tu iCtI*uptig A iispQ.; f.uequ nces oJ Cro mI Bogota, Colomb!a. New York: Oxford University Press, 1988. Oaxaca. R. 'Male-female Wage Differentials in Urba1t Labor Markets." Inzernauional Economic Review, Vol. 14 (1973). pp. 693-709. Ohsfeldt, R.L. and S.D. Cuiler. 'Differences in Income Between Male and Female Physicians.~ Journal of Health Economics, Vol. 5 (1986). pp. 335-346. Psacharopoulos, G. and Z. Tzannatos. 'Female Labor Force Participaticn: An International Perspective." The World Bank Research Observer, Vol. 4, no. 2 (1989). pp. 187-201. Reimers, C.W. 'A Camparative Analysis v: the Wages of Hispanics, klac6s and Non-Hispanic Whites' in G.J. Borjas and M. Tienda (eds.). Hispanics in the U.S.Econorny. New York: Academic Press, 1985. Schultz, T.P. 'Education Investmszts and Returns' in H. Chenery and T.N. Srinivasan (eds.). Handbook of Development Economics. Amsterdam: Elsevier Science iPublishars, 1988. Shapiro, D.M. and M. Stelcner. 'The Persistence of the Mfale-female Earnings Gap in Canada, 1970-1980: The Impact of Equal Pay Laws and Language Policies.' Canadian Public Policy/Analyse de Polifique, Vol. 13, no. 4 (1987). pp. . 2476. Standing, G. and G. Sheenhan (eds.). Labour Force Farlcipation in Low-Income Counrris. Geneva: Interrational Labor Office, 1978. Terrell, T. 'An Analysis of the Wage Structure in Guatemala City." Jow7nal of Developing Areas, Vol.23 (1989). pp. 405-424. Thomas, V. 'Spatial Differences in the Cost of Living.' Journal of Urban Economics, Vol. 8 (1980). Trost, R. and L.F. Lee. "Technical Training and Earnings: A Polychotoraous Choice Model with Selectivity." Review of Economnki and Statistics, Vol. 66 (1984). pp. 151-156. Zabalza, A. and J.L. Arrufat. Tlhe Extent of Sex Discrimination in Great Britain- in A. Zabalza and Z. Tzanatos. Women and Equal Pay: The Effectr of Legislation on Female Employment and Wages In Britain. Cambridge: Cambridge University Press, 1985. 4 Female Labor Force Participation and Wage Determination in Brazil 1989 Jill 77rfenzrhakr 1. Introduction Over the pas. two decades, many countries have experienced dr3sic increases in female labor force particir'tion. Tow>ver, participation ratcs, in general, are sti:l lower for women than men and women's wages ae signiflcanmly iower than men-˘ in most coujntries. These differentials have spurre. an interest ii: th ! determinants cf both women's participation docisions and women's wages. White studi..s on w',men's work in developed countries ;ie ,enerafly I .sed on a single sector model, there has b' - a growing interest among developmerit econonists in the effects osr more complex labor ,Tt ~-. in developing countries on modeling and estimating wom-en's derisions to wvork and won.: n earnings func.ions The importance of arcounting for the large informal se-tor in many developing countrie. was recognized wver 30 yeas ego tv Jaffe and Azumi ('960). They observed that women engaged in informal or cottage-indtzstry" work had higher fertil ty irates than * Jmen who worked in the formal sector. Resul s from .everal vort recent studies, using more rigorous empirical analysis, have supported Jafre -nd Azumi's supposition that women's costs of participation are not equivalent across .ors.1 In this study, P multi-sector mr,odel of female labor force Daricipation and wage determination in Brazil is estimnwd. In anal.zing the Brazilian labor market it is important to distinguish bttween the formal sector and the large informal sector. However, it is also important to account for the distinction between the unregistered worker; and the self-employed within thr. informal sector. The character: t!c that distinguishes formal sector wage-earners from infortnal sector wage-earners is that formal sector employees carry a work booklet. Under Brazilian labor law, employers are obligated to sign the emlpluy;ee's wojrk booklet wnea contracting a worker. Unregistered employmeat is illegal with the except;on of self-employment. In Sect. -in 2, the distinctiorns between these three identified sectors at a more rigorously explored, the data are discussed, and some sample characteristics are presentea and discussed. Section 3 briefly outliies the theoretical polychoromous choice imdel that underlies the empirical analysis. In Section 4, the empirical model is specified. The results from estimating the multi-sector See, for example, Hiii (1980, 1983, 1`88), Smith (1981), Biau (1984), and Tiefeinthaler (1?91'. 89 briefly outlines the theoretical polychotomous choice model that underlies the empirical analysis. In Section 4, the empirical model is specified. The results from estimating the multi-sector participation equations for both single women ar- diarried women are presented in Section 5, and Section 6 contains the results from estimating the sectora! wage equations. The potential existence of sex differentials in tie earnings functions is discussel in Section 7. The conclusions of this research are outlined in Section 8. 2. Data aad S2nple Charadeistics Sector definWŁons and characterWc. Most studies of labor markets in developing (and developed) countries prior to 1980 regarded the labor market as one sector and the labor force participation decision as simply a decision to work or not to work. However, several more recent studies have attempted to construct more accurate multi-sector models of the participation decision in more complex la&,r markets. The common sectoral decomposition is to increase the participation de.zision from two choices (work or don't work) to three choices - non-participation, work in the formal sector or work in the informal sector. In these models, non-participants are considered to be those who do not work fof pay. The formal sector is defined as comprising all individuals who work for a wage while the informal sector is made up of the self-employed. In thisstuay, following Alderman and Kozel's (1989) study of multi-sector participation and wage determinaion in Pakistan, the sectoral decomposition is taken a step further. As Alderman and Kozel found in urban Pakistan, in Brazil a formal sector exists parallel to an inbrmal wage-earning sector as well as a self-employment sector. Therefore, the participation decision presents four distinct labor market alternatives: non-participation (N), worldng for a wage in the formal sector (F), working for a wage in the informal sector (1), and self-employment (S). While non-participation and self-employment continue to be defined as they are in the preceding paragraph, there is an important distinction between formal and informal sector employees. The infornal sector emiloyees are easily distinguished from their formal sector counterparts because informal workers do not carry booklets required by Brazilian labor law and, therefore, are not registered with the government Employers must sign all enployees' work booklets and then register the empioyees. Unregistered work is illegal with tbe exception of self-employment When an employer signs an individual's booklet, the employer gets access to all information on that individual's former employment because the booldet, by law, is a record of the employee's work history including wages. There are both pros and cons to being officially registered as a worker in Brazil. The benefits include eligibility for unemployment compensation, social security (27 percent of the total 37 percent is paid by the employer), protection of labor law including minimum wage legislation, benefits of labor negotiations and union membership. However, unregistered workers do not have to pay payroll taxes and their wages are not regulated by official wage indexations. In addition, people who are col:ecting various government transfers can continue to collect them while working in the informal sector. The employer faces many added costs wheh registering employees including the 27 percent social security payment and other payroll taxes, the possibilityof dismissal fines, union bargaining, and the regulation of Brazilian labor law. Employers must weigh these -tsts against the probability of being caught and fined for employing unregistered workers (s%e Table 4.1 for the sectoral decomposition of the work force for the Brazilian sample). Data. The data for this study were collected from 70,777 Brazilian households (301,088 individuals) in the fourth quarter of 1989 by the National Statistical Ser ice. Data collection was organized according to four distinct regions: Rio de Janeiro and Sao Paulo, the rest of the South (Parana, Santa Catarina, Rio Grande do Sul, Minas Gerais, and Espirito Santo), the Northeast (Maranhao, Piaui, Ceara, Rio Grande do Norte, Paraiba, Pernambuco, Alagoas, Sergipe, and Bahia), and the Northwest/Central (Distito Federal, Rondonia, Acre, Amazonas, Roraima, Para, Amapa, Mato Grosso do Sul, Mato Grosso, and Goias). These four regions comprise the stata used in a modified stratified random sampling scheme based on the 1980 demographic census. Information was collected on household demographics, individual characteristics of all household members, educational histories of all school-aged (5 years and older) household members, and labor and income detais of all household members over age nine. The final sample, used for both data and regression analyses, includes 9,973 single women, 50,452 married women, and 58,000 men. The male subsanple includes all married men whose spouse is under age 65 and all single male heads of households.? The 50,452 married women are comprised of all married women under age 65 and the subsamp.e of single women includes all female single heads of households. Sampk characteristics. Brazii experienced rapid rises in employment and productivity in the 1960s and the 1970s. However, during the 1980s, the world rLcession and debt problems contributed to a resurgece in unemployment rates and an average uinual growth rate of only one percent. Although recovery was under way by 1989, when the data employed in this study were collected, the backdrop of this study is an economy worn by a decade of recession and adjustment. In this section, statistics describing Brazilian women's labor market opportnities in 1989 are discussed. Women's participation rates, earnings, and wages are presented and compared with men's. It is often suggested that women bear a disproportionate share of the burden of adjustment. By comparing the data from this study with those from a study by Stelcner et al. (1991) which uses Brazilian data from 1980, this hypothesis is evaluated. In addition, regional disparities in participation rates and wages will also be discussed. Sex differends. As outlined in the introduction, although women are increasingly entering the labor force, men's participation rates are still higher and men earn higher wages than women. In this sample, 86.4 percent of men participate in the paid labor force compared with only 57.4 percent of women. Single women are more likely to participate in the labor force than married women. Sixty-two percent of single women categorized themselves as paid laborers while only 34 percent of married women worked for pay. Men who work make more than their female counterparts. Male workes' average earnings are 1430.36 cuzados (C) per month whle single women and married women make, on average, 811.54C and 762.53C per month, respectively. One source of the deviation in total earnings is the number of hours worked. The average man spends appr)ximately 46 hours in a primary job per week while women, on average, work around 37 h urs in a primary job. The six percent of men who hold two jobs work, on average, 20 hours per week in their second jobs while the average woman who holds two jobs (5.5 percent) spend, approximately 18 hours at her second job. I Some men over reiirement age, 65, had to be included in the male subsample in order shat husbands' wages could be predicted for all females under 65. Another source of the sex differeatial in tot earnings is a sex differential in hourly wages. Men, in this stuy. earned an average hourly v..ge of 8.34C per hour while the average employed woman made only 5.74C per hour. The result that women are making only 70 percent of the average male wage may be contributed to several factors, including diffeences in education and experience or job tenure and the sectoral composition of the work force. If education and experience are important deminants of wages and men are significanlly better educated and have accumulated more experience than women, we would expect men to earn higher wages. z vwever, the mean male has received 5.68 yeam of formal education while the mean female has received only slightly less formal education, 5.05 yes. No data on work experieaice are available in this data set. The wage differential may, in part, be due to the sectoral distribution of male and female employees within the paid labor force. As presented in the following chart, women are more likely to work as employees in the informal sector while men are more likely to work as employees in the formal sector. Women's preferences for informal sector work may be due to easier entry and exit and more flexibility in the informal sector than in the formal sector. Paws de Barros and Varandas (1987) find that there is both a higher degree of flexibility in the number of hours worked and a shorter duration in employment in the informal sector than in the formal sector in Brazil. Flexibility in work schedule is often dtemed to be more important to women than to men due to women's household responsibilities of household production and childcare. Figure 4.1 Sectoral Employment in Brazil, 1989 81f F g. . ..FrX^ al 42 ,u;f.s| All Men Married Women Nsge _orW '"s Single Women 5 tflW4 A4MVOr rOrCa rarUCIpaWn ana Wagu venmanon us 8raZil, 1JM9 93 An alternative explanation for the disproportionate number of women working as informal employees is a Jhortage of jobs in the formal sector. If such a shortage exists, formal sector emplcyers may disiminate against women who are then forced to work in the informal sector. However, Sedlacek, Paes de Barros, and Varandas (1989) conclude that no such mobility barriers between formal and informal sector em- 'vment exist. As shown in the table below, wages are, on average, higher for both men and womer in the formal sector than in the informal sectors. However, it is also important to note that the sex differential is greater amng formal employees (women earn 70.3 percent of men's wages) and the self-employed (70.5 percent) than in the informal sector (85.2 percent). Table 4.1 Sectord Wages by Sex Group in Brazil (Cruzzdslhour), 1989. Men Married Women Single Women Formal Employee 10.93 7.68 7.46 Jnfornal Employees 5.28 4.51 4.46 Self-Employed 5.96 4.17 4.37 By reviewing the i-iformation in the chart and the table, the interpreaion is that a contnbuting factor to relatively low wages for women is that women are disproportionately represented as employees in the lower wage soctor. Tnese data suggest that if the proportion of women working in the formal sector increases, the sex differential in wages would be expected to decrease. Changes in women's economic opportunitis - 1980 to 1989. It is often suggested that disadvantaged groups - the rural poor, women, children, minorities - bear a disproportional amount of the burden of economic adjustment programs (see, for example, Cornia et al. (1987)). The 1980s was a decade of adjustment for Brazil as she recovered from the world recession and began to deal with the problems of a bulging foreign debt. Although there are many measures of welfare, comparing women's economic oppori.nities in 1980 with those in 1989 will provide some insight into the effects of the adjustment progran . on the well-being of Brazilian women. Participation rates of both single women and married women increased from 1980 to 1989. According to Stelcner et al. (1991), 20 percent of the married women and 41 percent of the single women in their sample reported to be working for pay. In the sample taken in 1989, used in this study, over 34 percent of married women and 62 percent of single women were wage earners. This comparison is consistent with Edwards' (1991) study of economy-wide trends in the Brazilian labor market in the 1980s as she finds that 'labor force paricipation has continued to increase significantly during the decade of the 1980s.' Although participation has increased among both married and single women, the increases have not been proportionally distributed across sectors. In 1980, from the Steicner et al. sample, 65 percent of working maried women were employees (either with or without an employment booklet) while the remaining 35 percent were self-employed. In 1989, slightly more married women classified themselves as employees at 69 percent while the percentage of self-employed fell to 31. The opposite ransition occurred among single working women. In the 1980 sample, 83 percent said they were employees and only 17 percent were self-eployed. In 1989, the number of employees fell to 71 percent of working sing:e women while more women classified themselves as self-employed, 29 percent. ' women a cnpioynwu ama ray m LaW Amenca Although earnings are not direcly comparable across years, the female/male eanings rado is comparable and provides evidence of women's relative position in the economy. In the Stelcner et al. 1980 sample, the female/male eanings ratio was 59 percent (61 percent for married women and 54 percent for single women). This number fell slightly to 56 percent in the 1989 sample (55 percent for married women and 57 percent for single women). Single women made a relative gain in total earnings throughout the decade whfle married women lost ground to men. The female/male ratio of hourly wages is a better measure for comparing women's relatve economic strength across time. In the 1980 sample, the female/male wage ratio was 75 percent for employees while self-employed women were making only 68 percent of their male counterparts' wages. There was litde change in this statistic over the decade. In the 1989 sample, employed women were making 76 percent of employed men's wages while se3f-employed women were making 69 percent of self-employed men's wages. Women gained little ia their economic power relative to men in tbe 1980s as the sex ratio of wages improved by only one percent in both sectors over the decade. Regional dffferenditd. Brazil is a large and diverse country. Studies whi'N have accounted for its size and diversity by treating distinct regions separately and including regional dummy variables have found that regional differences should not be ignored. Stelcner et al. (1991), using data from 1980, extensively analyze regional differences in labor market conditions in Brazil. They note important differences between the highly industrialized and modern regions in the South (Rio de Janeiro, Sao Paulo, Other Southeast, and the South) and the Northeast which is heavily dependent on agriculta actvities. Their data analysis, most notably, points to much lower incomes, wages, and education in the Northeast region than in the rest of Brazil. The conclusion is that significant barricrs to migration exist which prevent the equalization of wages across regions. Table 4.2 Regional Wages in Bazil, 1989. Male Female Fenale/Mac'd (Crundow1our) (pewa) Rio *e Janeiro 10.89 8.47 77.8 Sao Paulo 11.18 8.23 73.6 South 9.00 6.68 74.2 Other Southeast 7.87 .5.15 65.4 Northeast 5.61 3.56 63.5 NorthwetCentra 9.19 6.43 70.0 a. lbe ratio of the averg wm 's wage to the avenge mans wag These regional wage differetials continue to persist in 1989. As the following table shows, mean wages are not equal across regions. (It is important to note, however, that these wages have not been adjusted to reflect any cost of living discrepancies which may exist across regions.) Wages are notably lowest in the Northeast for both men and women. The ratio of female/male wages is also lowest in the Northcast whDle this ratio is highest in Rio de Janiro. A relationship appesi. to exist between high wages and more favorable fevnale/male wage ratios. Fmalk L4bor Force Pardcpadon aid Wage DeWmhwdon i Brazl, 1989 95 There are few noteworthy regional differences in the labor force participation patterns of women, as shown In Table 4.3. Participation of married women varies by less than four percent across the seven regions while the participation of single women jumps over seven percant from a low of 59.1 percent in the Northeast to over 66 percent in the Northwest/Central. Participation rates are the lowest in the Northeast for both married women and single women followed closely by Sao Paulo. One striking figure in male participation rates is the relatively low participation of men in Rio de Janeiro. While this city definitely has the lowest male participation rate, the female participation rates for Rio are relatively moderate. Table 4.3 Regional Particpadon Rates in Brazil, 1989 Mle Female Rat Rio de Janeiro 81.0 36.8 61.2 Sao Paulo 85.2 33.3 59.4 Soutb 86.8 35.6 64.3 Southeast 86.2 33.3 61.3 Northea 85.5 33.1 59.1 Northwest/Centnd 90.5 35.8 66.7 a. Note tha the male puiticopuion rates ame lwer. '-An expoced bowe fth male sample includes s.ncm over ae 65. Mcn 65 wkh wiv under 65 wcre inchlded i the sample becaue wages had to be predicted for thce men so the women could be included in the kme sample- 3. Theoretical Model Assume that a woman must choose among the four mutually exclusive alternatives discussed in the previous sections - working in the formal sector (F), working for a wage in the informal sector (0), being self-employed in the informal sector (S), and not working in the labor force (N). The problam that the woman faces is to choose ibe labor force altrnative which maxim household utility. Assuming that the household observes the offered wages the woman could earn in each sector, the value of her time in household production, and the time and money costs of participation in each secor, the household maximizes the household utility function subject to the household time and budget constraints under each alternative. The household then compares the lcvels of indirect utility obtinable from the various choices and chooses the participation status that maximizes household indirect utility. Following Maddala (1983), the indirect utility function is decomposed into a nonstochasdc component and a stochastic component where the nonstochastic component is a linear function of the observable variables in the indirect utility functions and the stochastic component is a function of unobservables. ITe probability that individual I will participate in sector k is the probability that the indirect utlity yielded in sector k is greater than that derived from the other sectors. This implies that the probability of individual i puticipating in sector k is the probability tha; the difference betweeu the stochastic components is greater than the difference between the nonstochastic components. 96 WOm's Enployme and Pay in Zada America This analysis implies that the offered wage in sector k for individual i is observed if the individual participates in sector k and the condition for participation in k is that the difference between the stochastic components is greater than the difference in the nonstochastic components. Therefore, this is the selection rule for the multi-sector model. Consistent estimates of the three r.ector wage equations can be obtained by accounting for this selection rule in the estimations. The form of the participation equation, the method of estimation, and the calculation of the selection correction will depend upon the distributional assumption on the errors. Assume that the errors of the linear indirect utility functions are indepnently and identically distributed with the type I extreme-value distribution (also called the Weibull distribution). Given this distribution of the errors, the difference between the errors has a logistic distribution (see McFadden (1973)). Because the difference between the errors is assumed to follow a logistic distribution, the participation eqwRtion must be estimated with the multinomial logit model. The probabilides of participation in each sector, given that the nonstochastic component of iirect utility is a linear function, under the multinomial model are written as: exp(b6D) pe p , k=FS j=F,JSN. (1) The above expression requires some normalization. Using a commonly used and simple normalization, that the coefficients of the nonparticipation alternative 6, = 0, together with the three probability equations uniquely determines the selection probabilities and guarantees that they sum to one for each individual. The multinomial logit model can then be esdmated using maximum lilkelihood methods. As mentioned in the previous section, it is also interesting to estimate the sectoral wage equations. Because of the existence of selection bias (those women from whom wages are observed all have an offered wage above the reservation wage), a Heckman-type method must to used to correct for the selectivity. Hay (1980) adapted Heckman's (1979) inverse Mill's ratio correction for probit models so that it is applicable to both binary and multinomial logit models. In the multinomial logit model, the corre tion is: -.e = ( 2x-l)} ttj p -)log(P) + (L)log(Pt)) K2 j*kJ 1-Ps ( 2) where J = the total number of alternative choices, in this case J equals four. The procedure for estimating the three sector wage equations free from selection bias is then to first estimate the maximum likelihood participation equation. By using the results to calculate the probabilities of participation, Hay's inverse Mill's ratio for the multinomial logit participation model can be calculated. Then, given the inverse MIll's ratio, the wage equations can be estimated. It is important to note that the derivation of the multinomial logit model of participation did not require any assumpdon about the distribudon of the errors in the wage equations. Consequently, they can be assumed to be normally distributed and the wage equations can be estimated using ordinary least squares (OLS). Female Labor Force Parhcoain and Wage Determon i Brazid, 19 97 4. Empirical Specfiraion As pointed out in Section 3, three wage er,w;'ts will be estimated, one for each sector. The offered wages in each sector are hypothem d4 iv be a function of a vector of the individual's human capital variables and labor market cor.,!'.sns. Consequently, the wage equations are set up as standard Mincer equations (see Mincer (1974)) with formal education, predicted experience, experience-squared, regional dummies, and racial dummies included as regressors. Formal education is included to pick up wage increases resulting from human capital investment. Experience is believed to have important positive effects on both productivity and earnings but at a declining rate. The racial dummies are included to account for the possibility of discrin'.nation (white or of European descent is the omitted category) and the regional dummries are included to reflect differing employment opportunities across regions. The wage variable for each sector is measured as total income in cruzados per month divided by the average number of hours worked per month (the number of hours worked in : week multiplied by four). The reference period is the September 24 through September 30, 1989. Formal education is measured as six dunmy variables - (1) if the woman received any formal schooling, (2) if the woman completed the first four years of primary school', (3) if the women finished primary school (eight years), (4) if the woman finished secondary school, (5) if the woman Fiished college, and (6) if the womr-: did any graduate work. Ihe survey question asked was 'highest grade completed' and this variable was converted into the six dummy vaibles specified. Tle dummy variables are specified such that all six will be equal to one for a woman w.io has attained post-graduate work (the first five will be equal to one for a womnan who completed college and stopped, etc.). Therefore, the coefficients on all six variables have to added up to get the total premium paid to education for a woman who has achieved higher education (the first five added up for the total returns to finishing colege, etc.). Labor market experience is measured in years. However, because no data were collected on experience, this variable was constructed using the standard formula age - education - 6. Although this formula has worked well as a proxy for male work experience, it has not performed as well as a predictor of female labor market experience because women are more likely to take additional time off work due to childbirth and childcare. However, this proxy will be used for a lack of a bettor alternative. The regional dummy variables are specified as follow: Rio de Janeiro, Sao Paulo, Other South (Parana, Santa Catarina, Rio Grande do Sul, Minas Gerais, and Espirito Santo), Northeast (Maranbao, Piaui, Cerara, Rio Grande do Norte, Paraiba, Pernambuco, Alagoas, Sergipe, and Bahia), and Northwest/Central (Distrito Federal, Rondonia, Acre, Amazonas, Roraima, Para, Amapa, Mato Grosso do Sul, Mato Grosso, and Goias). The Other South is the omitted category in the regressions. From the theoretical derivation in Section 3, it is clear that the variables included as regressors in the participation equation should reflect the offered wages across sectors and the differing costs of employment in the three sectors as well as other factors that influence the reservation wage. Age and the formal education dummy variables are included in the participation equation to proxy for the offered wages (wages across sectors cannot be included because they are not observed for all women and because of eadogeneity) and because they will affect the reservation wage. Tle 3 A distinction was made between finishing the first four years of primary school and completing pri school because prior to 1971 grades 1-4 wam considered compulsory a endtmnce into the next levels (grades 5-8 and grades 9-11) was coolled by examinaton. QR W.., It F,inDLvmp,u aid Paw in lmi 4 e _4_ mumber of children in the household disa"eregated into four different mrouns (children uinder uoe 2, children 3 to 5, children 6 to 12, daughters over 13, and sons over 13) are included to reflect childcre costs. Tlhe tntal numher of children ir ditq,oreotitd into the P arln_ hlwmiep- the presence of some groups may increase childcare costs (children under 2) while others may decrease these ecntj (danihter over 131) A dummy for whether ar not the htlch2nd l. self-employed is included in the married women's regression to reflect other opportunity costs nf n2rtiintminv in the 2lbr forri The hnholnld hpiI hpine "If-_pinnnvPA in an artivtv, in which the woman can help is expected to raise the shadow value of the woman's time in non- mnrlrkp Qrtivitik-c and redne the probability that -he will hp a waun2 P2mrA Monthly nnPirnPnrnM incennp rtnrlulpri hnth r_2ch 2nd in-rdnA unnrn> ineinmix frnm all anim-pa) anA the husband's hourly wage are hypothesized to inacrease the reservation wage. Both of these V2f2ifhlPa f2rP nijStc,,riiv in rvss r ,nt iyad,win RPt,CA twit sll hXhtutohanA. u, blsleMw*n,4 th h...__,h.' t._v w,,u V8 - "vaW are predicted from selectivity-corrected wage equations (see Appeadix). A dummy for whether nr nnt tha hniicehnIA cume the hnme in tliMrh Ahew ora lu4na ia ;nCIt%A^A a m ,c., o -. -. - -- - - _a --&*a -.- ..w -j* -_ - * A a f.W At household wealth. Therefore, owning a home is also expected to increase the reservation wage. !be Vaci-al duImmap a t A-11d.d tn npiLr uii any "Vr4mnra in MPeas ,ohmitt a.A -,1- ar., ethnic cultures and the regional dummies are included to reflect differing labor marlcet nnrnnrhvnhtu.~ onA u2liiD.@ s&nce csanne T-ohlpAAo"nadvathaTn .w vAf regions. T.ble 4. hows the means and - _antiad dAation. (in %_ n-h-.Ies of -ll of i ..aa4al.l included in the participation and wage equations. The means and characteristics of the dependent v2ri2hlp wPre di1iiccc d in SPrtifnn 2 RPi3ie SepCaramt raefrriOnn Will ke i,nAir fnv married women and single women in the following sections, the means and standard deviatons nre presente by these _ ma, S. The D .t-ran-ft. of Y-we Tft}r Fo.ew P art-.-. -n SL'zie, uwomen Table A C nrwente the ronaiilta frnn umAnguI thu nmt,lLA-mn, enn '°. 0 ' - _~ -_ Fr- M W .___ hom ~ tgk--k A EJ 1 t equation for single women using maximum likelihood multinomlal logit methods. Statistical tests of poingia she five *wina on f;1A jn hu pr_vios ca-4n w_re a,snw fr thu cu.ha nf single women as well As for the married women and male subsanples. Consequently, the regional data were pooled A d- were i ncAled to for intercp eial X-. The partial derivatives are in bold, the logit coefficients follow and the t-ratios are the numbers in narenthueta The liog linaIIralik fnom crtima*ion otf Aa ,n,au .. ;s -IIO 101 Age is includie in the panrict iation equntion to reflp.L the effects of huimnn cpital invectmpnft on wages which will effect participation. As expected, age has a positive and significant effect on wnrkr in all three cetnre. An intPurnPt-tmnn ic thga ageaa; in"P the lm,ev nf human &anital acquired increases and the offered wage goes up. An increasing wage, holding all else constant, will ine're2ce the nmnh2hilitv nf p2rtiiahtifnn Aae.,nnarPA ;i mnthuidedA at a reartc,w ton ,i4r upn possible nonlinearities in this relationship. The significance of the squared terms in all three SMnrt sunn th the hypot of murvatureinthe effpets of ae on the prbabhilitv of participation. Education has stronger effects in increasing the participatdon of single women in the fbrmal sector than in thA IinWfrmsl anrd aef-m.nInnupe eetnra Tn FAa utiXttinal .auua-it;nn An.tuaaae. the probability of participation in the infbrmal and self-employed sectors in many instances. AttaWnng ua,.h nf the fiuet threo Ipjple nf ewuuatian mn&iea..pe tho nrI-ahflItu nf &srn,al wtne participation by between eight to nine percentage points. Finishing secondary school and college Fmal Labor Force Pardiciwon and Wage Dermiadon a BrazJI L9 9 Table 4.4 Means (and Standard Deviations) of Independent Variables Variables Married Women Single Wonmi Ago 37.09 (11.53) 44.69 (11.89) Ago-squared 1508.9 (922.3) 2138.7 (1044.5) Some Primary .820 (.386) .749 (.433) Primary - Layd 4 .617 (.486) .551 (.497) PFimay - Level 8 .291 (.454) .279 (.448) Seconday .181 (.385) .194 (.395) College .067 (.250) .091 (.87) Crduhate Work .002 (.039) .003 (.055) Own Home .667 (.471) .657 (.478) Black .042 (.201) .070 (.255) Mulato/Indian .39S (.489) .427 (.495) Asian .004 (.066) .003 (.056) Uneannd Income 173.75 (879.34) 280.31 (701.55) Husband's Wage 1.488 (6.338) Husband Self-Employed .263 (.440) # Children 0-2 .342 (.578) .089 (.317) I Children 3-S .314 (.555) .134 (.394) # Children 6-12 .831 (1.063) .465 (.83C) f Daughtas > 13 .504 (.899) .561 (.887) # SOS > 13 .428 (.794) .634 (.966) Experienca 23.728 (11.50) 29.487 (12.89) Experience-Squred 695.23 (625.7) 1035.4 (788.3) Sao Paulo .128 (.334) .1A0 (.313) Rio de Janeiro .072 (.258) .092 (.289) Norihwest/Centr .208 (.406) .205 (.404) Northeast .287 (.452) .303 (.460) each increase ft an additional 13 perntage points while attaining some higher education ineases the probability of being a formal sector employee awnther 22 percetage points. Education is included in the participation eqation to reflect the effects of w,ges on the probubXilky of participadon. Given the results, it is expected that education has the highest returns in the formal sector because women with more education are more likely to choose to work in this sector. Ihe dummy variable for owning a home and the continuous variable for unearned income are included in the estimated equation to proxy for wealth. These proxies for wealth are expeced to have ne2ative effecas on the probabDlity of participation in all three sectors. Owning a home does have a negative and significant effect ca participation in all three sectors. Unearned inome also has a negative and strongly significant effect on the probability of participation in all three sectors. Increasing income by 1000 cruzados per month decreases participation by II percenge points in the formal sector and 6 to 7 percentage points in the other two sectors. Ihe racial dummy variables are included in the partcipation equation to reflect different ethaic and cultural values about women and work across races. The reference group is white swe women. Ihe results show that black single women are more like!y to vc employees in both the formal and informal sectors than white single women but less likely to be self-employed. Indian and mulatto women are more likely to participate in all three sectors than white women. Asian women arm 19 percentage points more likely to participate as self-employed than white women. 100 Wam's Eapoymd Pay LadxAnwrka Ta,le 4.S Multi-Sector Paricipatiom Radbt, Single Wona N=9,973 Fornnl Informal Self- Employees Employees Employed Consbnt 4.012 -2.224 -3.154 (8.800) (4.629) (6.517) Ago .0338 .0042 .0109 .2511 .1546 .1892 (11.748) (6.854) (8.545) Ago-Squared -.0006 -.0001 -.0001 -.0027 -.0024 -.002S (14.989) (9.22) (10.167) Some Primary School .0823 -.022 .0143 .4943 .0626 .2767 (5.047) (.662) (3.28) Primauy - Level 4 .OS12 -.0178 -.0162 .4185 .0149 .0329 (4.864) (.156) (.387) Primary - Level 8 .094C -.0159 .0113 .5793 .1390 .2995 (5.178) (.9680) (2.371) Finished Seoodry .1312 .117 -.0449 .9977 1.210 .2716 (6.947) (7.024) (1.5) College .1319 .1484 .0002 1.214 1.641 .7368 (6.439) (8.177) (3-194) Grduate Education .224 .0352 -.0464 1.370 .7731 .2996 (1.118) (.612) (.195) Own Home .0338 -.034 -.0008 -.3034 -.3882 -.1854 (4.660) (5.517) (2.728) Blaig .0423 .07 -AW46o .4323 .7063 .2447 (3.499) (5.641) (1.974) Indian/Mulao .0291 .C291 .0095 .2828 .3551 .2314 (4.102) (4.729) (3.338) Asian .0083 -.0889 .1926 .3245 -.2481 1.369 (.5300) (.3190) (2.46) Unearned Inoms -.11 -.062 -.068 (1000s of cruzadoo) -1.4 -1.3 -1.1 # Children 0-2 -.0937 -.0209 -.0078 -.6614 -.4486 -.3649 (6.577) (4.364) (3.628) # IChldren 3-5 -.0378 .0095 .012 -.1803 .0152 .0244 (2.183) (.1080) (.2!5) # Chlddren 6-12 -.0148 .0097 .0194 -.0165 .0972 .14S6 (.411) (2.323, (3.672) (eC.128) (14.628) (11.878) -cntimned Ftamt Labor Force Pw*4amt and Wage Detemina In Bril. 198 101 Table 4.5 (ontinued) Multi-Sector Participations Results, Single Women N=9,973 Fogma informa Self- Empoye Employee Employed # Daughters over 13 -.0045 -.0035 -.0065 -.0539 -.0591 -.0748 (1.469) (1-5) (2.123) # Sons over 13 -.024 -.0102 -.0118 -.2080 -.1829 -.1871 (5.961) (4.96) (5.712) Northeas -.1095 .0064 .0405 -.5611 -.125 .0617 (6.687) (1.385) (.756) Sao Palo .02C6 .0253 -.0753 .0348 .093 -.4825 (.34) (.792) (3.874) Rio de Janeiro -.0546 .0059 .0014 -.3221 -.0883 -.1161 (2.859) (.693) (.966) Northwest -.0464 .0372 .0007 -.1895 .2051 -.0182 (2.140) (2.115) (.19 Note: Absolute t-ratiou in pashcssb. As outlined in Section 3, the varying costs of participating in the three sectors are important derminants of a woman's work decision. The number of chidren in each of five age groups are included in the participation equation to reflect the costs of working. The number of children in the three youngest age groups reflect costs associated with cbDldcare. These costs appear to be the highest in the formal sector as the mmber of children in the -2 and the 3-S age groups have significant negative effects on participation. An additional child under age two decreases the probability of formal sector participaion by nine percentage points while an additional child in the 3-5 age group decreases the participation probability by four points. Childcare costs seem to be lower in the other two sectors. Among informal employees and the self-employed, an additional child under age two decrerses the probability of partkipation by 2 poirts and less than one point, respectively. Additional chfldren in the 3-5 age group have no significant effect on participation in the informal and self-mployment sector. The mmber of daughters and the number of sons over age 13 A included in the participation equation because older children can decrease the childcare costs of labor force participation by taking care of their siblinp while mother is at work. Therefore, the number of children in each of these two groups (but especially the number of daughters) is epected tO increase participation in high childcare cost sectors. However, daughters have a significant effea only in the self-employed sector less than one percentage point) and sons have a negative and significant effect in all three sectors, decreasing the probability of participation by one to two percentage points. It appears that tenage children are Ea replacing their mother as childcaretaker but are Instead replacing mother in carning income. Older children can go to work and, therefore, their presence decreases the probability of mother having to work for pay. 102 Women's Ep1oyie ad Pay X LoW Ameria The regional dummy variables are included in the equaticw to pick up differing values and opportunities across regions in Brazil. Since a test of pooling across regions without including these dummy variables was rejected, some significance is expected. The relative region is the mother South (see the previous section). The results confirm that there are differences across regions. In Rio de Janeiro, the Northeast, and the Northwest, women are less likely to be formal sector wo.lkers than women who live in the more industral South. The only significance difference between the South and Sao Paulo is that women in Sao Paulo are approximately eight percentage points less likely to bo self-employed. Table 4.6 presents logit simulations to help to interpret the participation results. The independent variables are set equal to their sample means and a single variable is varied for each simulation exercise. The constant term has been adjusted so that the predicted probabilities of participation, evaluated at the means of the independent variables, are equal to the actual means of the dependent variable. The age simulations are adjusted to reflect both changes in age and age-sqared. As the other variables are held constant at their means, tLe probability of being a formal employee increases from 20 to 30 years but then decreases aler the age of 30. In the other two sectors, the probability of particip?xion ccntinues to increase with age through age 50. All else held constant, a 20-year-old single woman, a 30-year-old, and a 40-year-old are all most likely to be in the formal sector w; we a 50-year-6L' is most likely to be at home. The next simulation is education. As education increases, the probability of being a non- participant contimnally fills. A woman with no formas schooling is most likely not to participate while the probability of being a non-participant for a woman with a graduate education is less than two percent. As schooling increases, the probability of tz!nw a formal sector employee increases with each level. A woman with a graduate degree has a 72 perc probability of being a formal employee. The effects of education on participation in the other two sectors are not as strong. In fact, in the self-employed sector, additional levels of education generally lead to decreases in tbe participation probability. A r:oman with a graduate degree is almost as likely to not participae as she is like!y to be self-empioyed. Those women who own homes are more likely to be non-participants than those who do not cwn homes and less likely to work as either formal or informal employees. Owning a bome bas no effect on the probability of self-employment. The most stiking result from the simulations with the racial dummy variables is the relatively high probability that an Asian woman is self-employed and the relatively low probability than she is an informal employee. The most interesting results from the child variable simulations is for the youngest age groups. The probability of being a non-participant increases with each additional child 0-2. A woman with no young children has a 37 percent probability of staying home while the probability for a woman with two small children increases to over 62 percent. The probability of being in all three sector. d6creases with each additional child but the greatest decreases are in the formal sector - the probability decreases approximay nine percentage points for the first child and seven points for the second child. The number of children aged 3-3 also increases the probability of bt kg a non- participant and decreases the probability of being in the fornal sector but at mu:h smaller changes than with the 0-2 age group. Fwk Labor Force Parc*,adox and Wage Demfratiin Bianldt 1989 103 Tablop 46 Mltinmomial Logit Sirmdoati, Singl Womena N=9,973 Non- Formal InformIl Self- ParticijMnt Employee Employee Empioyed A4g-20 33.5 57.7 10.2 4.6 Age-30 22.6 S8.0 11.8 7.6 Age-40 24.6 47.2 14.7 13.4 Age-S0 36.9 23.8 17.1 22.1 No Schooling 55.4 13.7 14.4 16.3 Sobm Phriay 47.3 19.6 13.3 188 Primazy -4 43.5 .26.8 12.3 17.5 Primary -8 33.7 37.1 10.9 183 Socondasy 17.3 51.6 188 12.3 Collep 5.5 55.4 30.9 8.2 Graduate 1.8 72.3 22.2 3.7 Own Home - Yo 40.9 26.6 15. 1 17.4 own Home - No 34.1 30.0 18.S 17.4 Black 31.6 30.2 22.1 16.2 Muluao 25.4 29.2 17.4 17.9 Asi-a 26.9 23.1 7.3 42.C While 42.2 26.2 14.6 17.0 Xdi 0-2 - 0 37.4 28.7 16.4 17.5 Kids 0-2 - 1 50.0 19.8 14.0 16.2 Kids 0.2 - 2 62.2 12.7 11.0 14.0 KRd 3-S -0 38.3 28.3 16.1 17.3 Kids 3-S - 1 39.9 24.7 17.0 18.4 Kids 3-5 - 2 41.3 21.3 17.9 195 Kids 6-12 - 0 37.2 285 15.8 16.6 Kida.6-12 - 1 37.7 27.0 16.8 18.S Kids 6-12 - 2 36.2 25.5 17.7 20.5 Daughders>13-0 37.7 28.1 16.4 17.8 Daughtars>13-1 39.1 26.9 16.1 17.2 Sons > 13 - 0 35.6 29.4 16.9 1M2 Sons > 13 - 1 40.2 26.9 15.9 17 Northeat 41.7 21.8 15.4 21.1 Northwes 36.S 27.6 18.8 17.1 South 35.6 32.5 14.9 16.9 Sao Paulo 37.0 35.1 17.0 10.9 Rio 40.5 26.8 1S 17.1 Sample Maoma 38 27.6 16.4 17.9 The regional simulations show that, as in the total sample, single women in each region are most likely to be non-participants. However, women in the Northeast are the most likely to be non- participants and to be self-eaployed, women in Sao Plo are the most likely to be in the orna sector, and women in the Northwest are the most likely to be informal employees. Marred women. The mumial logit results from estiming the picipv ^n euaton using data on 50,452 married women are presented in Table 4.7. The same maximum likelihood methods used to estimate the participation equation for single women are emp!oyed here. The log likelihood fiucdon for this parti,ipation equation is -46,165. The results from estimating the participation equation for wried women are very similu to the results presented previously for single women. Age, again, has a positive and significant effect on participation in all three sectors. The education variables contine to have the stroLgest effects in the formal sector (however, all of these effects are weaker than those for single women). Unearned income continues to have a significantly negaive effect on participation in all three sectors. Children in the 0-2 age group significantly decrease the probability of part" ipation in all &ectors and children in the 3-5 age group decrease the probability of being a fonnal sector employee. Ibe regional results, again, show that women in the Northeast, the Northwest, and Rio are less likely to be in the formal sectr than women in the South and that women in Sao Paulo are less likely to work in formal sector. While most of the racial results found for single women hold true for married women, an important difference is that the high increase in the probability of an Asian single woman being self-employed does not hold for married Asian women. In fact, there is no significant difference between the probability of Asian women and white women being self-employed (this differecec was close tO 20 percentage points for single women). This result likely reflects that married Asian women are helping their husband's with tiaeir businesses (they are unpaid family workers) rather than being self-employed themselves. There are two additional arwiables included in the married women's participation equation - the husband's wage and tbh husband's setf-employment satus. The husband's wage (expected to increase the reservation wage and decrease the probability of participation) has a s3rongy significant and ne'ative effect on pn in all three sectors. If her husband Is self-employed, th;. i babWy of a woman being an employee in both the formal and info-ual sectors falls whi! the probability of being sef-employed incream by two prcntage points. The Table 4.8 presents the results from replicatng the simulations for single women using the results from estimating the married women's participation equation. The probability of participation increases until age 40 in the fiormal and informal sxtors and then begins to fall while in the sef-employ t secamr, the probability continuously inceases with age through age 50. Formal education, agai has stong negative effects on the probability of being a non-participant and strong positive effect on the probability of bein,. a formal employee. While a woman with no formal education has a 77 percent probability of b?4ng a non-participant and a six percent chance of being a formal employee, a woman with a gra ate education has a three percent probability of beirD a non-participant and a 48 percent change of working in the formal sector. Higher levels of educadon also have significant effects in increasiDg infor.nai employment. However, the effects are rdadvely negligk 'in self-enploynent. Among married women, mul women are the most likely to be Don-prticipnts. Black women have the highest probability of the four raial groups of working in both the nrmal and informal sectors while Asian women stil have the highest probability of being self-eployed (however, - - - ....-, - ..~. . we45. 4du4EflWSN~IO0 Na WOU, 1Y5Y 105 Table 4.7 P Frtp Ruhu, nMaried Womn N-SO,4S2 Forma Ifornml Self. Employee Employe Er - yed Conatat -6.387 4.853 -5.943 (31.297) (21.28) (27.583) AgV .Q2S4 .0097 .0127 .2SOS .1774 .1933 (23.029) (14.739) (17.81) Ae-a -.0004 -.0001 -.0001 -.0034 -.0023 -.0023 (24.302) (15.204) (17.258) Some Primay Scoo .38 -.0102 .0001 .5277 -.0280 .0830 (8.163) (.4960) (1.741) Priimny - Level 4 .0334 .0048 .0107 .3082 .12SS .1757 (6.676) (2.481) (4.175) Prinmy - Iee 8 .79 -.0021 06 .4992 .071S .L506 (10.753) (1.100) (2.889) Fnished Secoadasy .1091 .0944 -.0274 1.031 1.283 .0097 (20.744) (18.2S6) (.140) CAWlege .0908 .1001 .0196 .9SS2 1.397 .50S0 (17.07S) (21.277) (S.328) Graduate Educauion .1749 .1282 .1509 1.913 2.06$ 2.113 (3.583) (3.797) (3.409) Hudband Self-Emplboyd -.482 -.034 .0191 -.4331 -.4640 .0842 (11.795) (11.13) (2.481) Husbnd's Wage (pmlcbtd) s0316 -.028 -.02S -.34S6 -.4260 -.3365 (14.216) (17.515) (13.838) Own %oms -.0005 -.0163 .0061 -.0202 -.1913 .0410 (.652) (S.433) (1.193) Bieck .m .0473 .0072 .5106 .6678 .2275 (7.215) (9.002) (3.101) S=iaUdMUfO A".00 .0219 .0091 .080 .2890 .1388 (2.381) (7 S65) (4.019) Ad=an LM0104 -.O .0084 -.1568 -.6574 -.0052 (.79) (2.284) (.021) Tble 4.7 (coaninued) Particption Results, Maid Womrn N-50,452 Fond Inforul Self- EMoyees Employee Employed Unearned Irkoom -.017 -.014 -.0049 -.24 -.22 -.101 (7.626) (6.781) (4.266) ChiWrenGno-2 -.03 -.0293 -.5ZO6 -.465S4 -.3048 (16.817) (13.005) (9.077) # Children 3-5 -.0339 .0002 .0001 -.2879 -.0483 -.0494 (10.459) (1.632) (1.749) i Children 6-12 -.0209 .OOS1 .0066 -.1599 .0406 .0486 (9.913) (2428) (3.215) I Daug0htes ovea 13 -.0001 .0024 -.0008 .0014 .0281 -.0050 (.063) (1.220) (.253) Sons aover 13 -.09.000007 -.O093 -.0881 -.0114 -.0518 (4.275) (.547) (2.896 Northeast -.0154 913S .0207 -.204 -.1572 .1824 (2.965) (3.381) (4.368) Sao Paulo .0023 .0256 -.0302 -.0268 .2639 -.2957 (.583) (4.992) (S.171) Rio de Janeuro -.041 -.00:7 .0117 -.3376 -.1143 -.0571 (5.899) (1.713) (.94) Nornhwet -.0084 .0001 .002 -.0688 -.0093 .0091 (1.695) (.195) (.201) Now: Absohlt t.drb in parcabsis. while single Asian woen have 43 percet probability of being self-=nloyed, this probability is Icss than 11 percent for married Asian womev). If a woman's husband is self-employed, the probabilty of hx being - non-participa increases from 64 to 70 percentL While a self-employed husband decrases the probability of being a formal or informal sector wore (from 15 percen toD 1 percent and from 10 percent to 7 percent, respectively), it increases the probability of a woman hersdf being sef-ployed (from 10 percent to 12 percent). aiod women from Rio de Jsneiro are the most likely regional group not to participate In the labor frce; women fom the South have the highest probability of being formal sector workers; women from Sao Paulo have the highest probabiity of working in the informal secto; and women in the Northeast asre the most likely group to be self-employed. FmakaI Labor Force Parl5c4~oao wand Wage Daete7rfrtaos in Bn=i, 1989 107 T&k 4,1 Mulomial Logit Simaaionul, Married Women N=50,452 Noo- Formal Infoml Self- Pazidp&i Employee Employso Employed Ago-20 77.5. 10.2 6.8 5.5 Age-30 61.9 18.3 10.1 9.7 Agp-40 57.2 19.2 11.0 12.7 AgV-So (.2 12.5 9.3 12.9 No Schooling 77.3 6.1 7.3 9.3 Some Primary 73.7 9.9 6.8 9.6 Primy - 4 69.3 12.7 7.2 10.8 Primy - 8 62.7 18.9 7.0 11.3 Secoday 41.1 34.8 16.6 7.5 College 19.5 42.8 31.8 5.9 Gduate 3.2 47.5 41.3 8.0 Own Home - Yes 66.2 14.5 8.8 10.4 Own Home - No 65.1 14.6 10.5 9.8 Blak 56.4 19.7 13.6 10.3 Mulao 72.1 12.9 4.6 10.5 Asian 64.1 14.5 10.6 10.7 White 67.6 14.2 8.4 9.9 HuAmd Self-Pmpkyed - Yes 70.1 11.2 7.1 11.6 Husbad Self-Empklyed - No 64.1 15.9 10.3 9.8 Kids 0-2 = 0 62.7 16.3 10.3 10.7 Kids0-2 -1 72.3 11.2 7.5 9.1 Kids0-2 - 2 80.0 7.3 5.2 7.4 Kids 3-S - 0 64.7 15.8 9.3 10.2 Kids 3-5 - 1 68.0 12.4 9.4 10.2 Kids 3-5 - 2 70.9 9.7 9.3 10.2 Kids 6-12 - 0 65.0 16.4 8.9 9.7 Kids 6-12 - 1 66.0 14.2 9.4 10.4 Kids 6-12 - 2 66.8 12.2 10.0 11.0 Nordta 6S.9 13.9 8.2 12.1 Northwst 65.8 14.6 9.5 10.1 South 65.1 1S.5 9.5 9.9 Sao Paulo 65.1 15.1 12.3 7.4 Rio 68.5 11.6 8.9 11.0 Sample Means 65.8 14.3 9.3 10.6 The probability of beIng a frmal sector worker fals from 16 percent to 11 percent with the first child aged 0-2 In the bouebold and to seven percent with the second chfld. The number of children aged 0-2 also docrmasa the probabilities of informal and self-employment but at a much smaller rate. Tbe number of children aged 3-5 also decreases the probability of formal sector pudcipation (from 16 percat to 12 percent for the first child and to 10 percent for the second child) but have w significant effcts on the probability of beig in the informal sector or 108 Wom '5 Espoy^as dPay iLathx laca self-employed. While die nmnber of children aged 6-12 had little effect on singe women's employment, their presence again decreases the probability of a married woman working in the formal sector. However, chfldren in this age group also have a positive effect on informal sector participation and self-employment 6 Earnings Functons Table 4.9 presents the results from esfimang the eamings functions for single women and married women. The eanings functions are esimated as Mincer equations with the natual log of wages regressed on levels of the independent variables. The first three coltms are resuts from estmating the se.toral wage equations for single women and columns four dtrough six ;re the results from estim4tng the sectoral wage equations for married women. These resuls are corrected for selectivity using the invervo Mill's ratio for the mltlDncmila logit model p-mented in Section 3. The standard errors have been corrected for the use of an esimated inverse Mill's ratio. Tle OLS results are presented in Appendix Table 4A.3 for comparison. Becasse the selection term is significant in most of the equaions, the selecdvity coected results are used for discussion and die calculstions of discrimination to follow. Tbe selection term is strongly significant in the formal and self-employment sectors for single women and in both the formal and informal sectrs for married women. Consequently, in these cases, the selection correcion was need edto get consistent estmates of the earnings functions. Predicted experience has a positive and signUicant effect on wages in all tee sectors for both single and married women (except self-emloyed single women). For single women, the rate of return of a year of experience is approximately three percent in the formal sector, four percent in the informal sector, and one percent in self-employment. Married women enjoy slightly higber returns to expczience at five percent in the formal sector, four percent in the informal, and dtree percent in self-employment The relaionship between experience, however, is nt linear as the squared terms are negative and significant in each equation. The racial variables are also significant in many cases. Among married women, black women make 16 percent less than white women in the formal sector, eight percent less in the informal sector, and 15 percent less in self-employment. The results are similar for single black women as they make 23 percent, three percent, and 16 percent less, respectively, than white single women. These -esults suggest that there appears to be more dicrimination against black women in the formal and self-employed sectrs dtan in the inf,nal employee sector. Mulato women also make significantly less than white women in all three sm (21 percet, nine per, and 14 percent less, respectively, for married women and 15 percent, eight percent, and 10 percent less, respectively, for single women). Whie married Asian women make significantly more than white women in self-employment (56 percent), this results does not hold among single Asian women. Ihere also are regional differences in the eanins functions for both single and married women. In the Northeast, both groups make less in all dtree sectors ta in the South (th reference region). In the formal sector, married women in the Northest make 35 percent less and single women make 27 percent less and in the Informal sector, married women make 46 percent less and single women make 36 percent less. Both groups make 39 percent less in self-employment inm te Northeat than ;n the South. However, in Sao Paulo, women make more (from 23 to 47 percent more) in all three sectors than they would in the 'other South. In Rio de Janeiro and in the Northwest, there are some regional differencs but they are not across the board as In the Femak Labor Force Pankcaax and Wage Den'adadox I Brzgl, 199 109 Tale 4.9 Sectonl Wag Equatiors (Corecbt for Sd'ivity) Singl Women Marred Women Infoma Self- Informal Self- Formal Wago Employed Forma Wag. Employed Constant -.SS21 -1.146 -.6178 -1.17 -1.57 -.682 (3.33) (4.96) (2.11) (6.42) (8.01) (2.84) Selection Term (LAmbda) .3314 .1411 .6092 .3382 .3585 .0889 (3.74) (1.11) (4.32) (4.45) (4.56) (1.63) Experience (age-educ.6) .0268 .0407 .0069 .0567 .0427 .0320 (5.17) (5.56) (.669) (12.9) (10.2) (5.34) Soma Prinuar .2533 .3347 .2222 .1298 .2486 .2180 (4.21) (5.33) (3.48) (2.67) (6.10) (4.94) Primary - 4 .2294 .2779 .2737 .2928 .3240 .3559 (4.85) (4.4S) (4.41) (8.77) (9.06) (9.68) Primay - 8 S124 .5836 .5416 .4891 .5913 .4180 (10.31) (6.66) (6.17) (14.8) (12.7) (8.76) Fnished Sconday 5530 .6492 .2390 .6169 .6372 .5337 (10.64) (6.80) (2.08) (17.5) (12.7) (8.48) College .84SS .726 .6407 .7987 .7815 .7751 (17.83) (9.96) (4.36) (27.0) (20.5) (9.39) Graduate .3202 .2863 -.2158 .7144 .2779 -.1126 (2.01) (.903) (.4S4) (5.56) (1.89) (.380) Black -.M23 -.0291 -.1605 -.1563 -.0843 -.1498 (3.73) (.385) (1.79) (3.32) (1.67) (2.32) Asian .521S .2205 -.48.2 .0954 -.041 .5566 (2.42) (.477) (1.05) (.803) (.213) (2.52) Indi"n/Mulatto -.1465 -.0778 -.1043 -.2060 -.0938 -.1372 (4.47) (1.66) (2.03) (9.41) (3.73) (4.89) Northeast -.2710 -.3642 -.3887 -.3447 -.4645 -.3919 (6.50 (6.39) (6.24) (10.0) (15.1) (10.5) Notlhwest .1281 .0270 -.0367 .1178 .0431 .1A (3.18) (.464) (.S40) (4.45) (1.35) (4.62) Sao Paulo .3501 .3924 .4663 .2326 .2835 .2.541 (7.591) (4.S75) (4.806) (7.91) (8.19) (5.03) Rio de Janeiro .02726 .0473 .0012 -.0260 .0260 -.1142 (.530) (.080) (.014) t.699) (.587) (2.16) F-Statistic 190.14 139.44 51.73 408.94 397.9S 165.99 Notc: Absolut tatios in prmthesis other two secors. The regional results Indicate tht labor demand condoions dife acros regions and that bafriers to migration do exist which are preveating the equalizadon of wage rates across regions. Because of the mn in which ti; yducation .lummy variables are constructed (soe Section 4), the total effects of aducation eaninp are praeuwte in Table 4.10 for single woen and married women. 110 Won 's E3poyn"s ad Pay in Latin Ameica Table 410 Pesentage increase in earings by educazon Formal Informal Self- Employees Employees Employed Single Wo Some Primary School .253 .335 .222 Primary - Level 4 .482 .613 .496 Primay - Leved 8 .994 1.197 1.038 Finished Secondary 1.547 1.846 1.277 Finisd Colege 2.393 2.572 1.918 Graduate Work 2.713 2.858 1.702 Married Wonxe Some Primary School .13 .25 .22 Primary - Level 4 .42 57 .58 Primary - Level 8 .91 1.16 1.00 Finished Secondary 1.53 1.80 1.54 Finished Collg 2.33 2.58 2.31 Grduhat Work 3.04 2.85 2.20 Education has strong effects on earnings in all three seclors. A msartied woman who finished primary school makes 91 percent more than her uneducated counterpart in the formal sector, 116 percent more than an uneducated co-worker in the infbrmal sector, and 100 percent more than an uneducated competitor in self-employment These effects continue to increase, in most cases, through graduate work and a woman (married or single) who does graduate work can make approximately 200 to 300 percent more in each sector than if she had no education. Surprisingly, the effects of education are highest in the Informal sector. Tle incremental returns in the formal sector do not overtake those in the informal sector until the college level is reached. The total effects in the formal sector do not exceed those in the informal sector for married women until the graduate work level is reached and they never do for single women. lhis is largely due to relatively low returns to the introductory levels of education in the formal sector. There are two important points to note when comparing the effects of education across seciors. First, women who work in the formal sector are more likdy to receive benefits and social security. Consequently, some of the effects of education in the formal sector may be in the form of benefits (i.e., a promotion includes health benefits or increased vacation time) and these additional effects are not capured by the earnings functions. Secondly, wages in the formal sector are subject to govemment regulations such as minimum wage laws 2ad wage indexation. Consequently, the wage paid in the formal sector to an educated woman may still be higher thin the wage in the informal sect because the base wage is higher in the formal sector. 7, Discrimination As dismssed in Section 2, there are earings differentials between men and women in this Brazil;an sample. Married wen make 70 percent of the male wage in the formal sector, 85 percent of it in the informal sector, and 70 percent of the male wage in self-employment. Tlese differendals are similar to the racial differentals In that the highest racial differendals for black Fak Labor Forcm Pardcadon and Wage Detmhuuon in Brmu1 1989 111 and mulatto women compared with white women were in the formal and self-employment sectors. It would be interestng to explain how much of the sex eaings differential ls explained by different endowments and how much is unexplained or due to labor market struces. lhe standard Oaxaca (1973) decomposition permits us to esdmate these two components of the sex wage differendal. Ihe decomposition Is written as: ln(vwge,) - ln(%ge) = . - 1) + baX, - XJ (3) = XA.M - b) + bX - X) (4) where the Xs are the endowments and the bs are the coefficients from the esdmated earnings functions. The two equaton ae alternative represe2tations of the decomposition and neither is preferred over the other. However, becase we are dealing with index numbe, the two equations will not produce equivalent results. The first term in the decompositions is the amount of the differential arbuted to the labor market rewards or unexplained factors. lTis term is often interpreted as the amount of the differential due to discrimisation. Ite second term is the amount of the differential attrlbutable to differences in endowments. Table 4.11 presents the decompositions of the sex earnings differentals into the perentage points attributable to differences in endowments and the percentage points due to discrimination. The mnmbers in parenthesis are the percentages of the total explained by each component Discrimination appears to be slightly higher in the formal and self-employment sectors thn in the informal sector. Table 11 Decomposition of the Earnings DiffaIiala Rewds Endowmets Formal Sector Equion 3 24.3 (81%) 5.7 (19%) Equaton 4 26.7 (89%) 3.3 (11%) Iormal Sector Equanon 3 10.8 (72%) 4.2 (28%) Equation 4 11.3 (75%) 3.7 (2S%) Self-Employment Equatio 3 24.8 (83%) S.2 (17%) Equaton 4 3S.2 (84%) 4.8 (16%) (100% of diff&caZi1 8. Conclusions Labor force participation rates of both single and married women have increased since the Stelcner et al. (1991) stdy using 1980 data. In 1989, 34 percent of married women were working for pay and 62 percent of single women were wage earners. However, the increases in participation bave not been proportionally distributed across the three identified market sectors. Between 1980 and 1989, the percentage of mafried women who classified themselves as employees increased while the percentage who considered themselves to be self-employed el1. 112 W-m'a Enplymeu and Pa in La n AMMic The opposite occurred among single women The number of single woman employees fell and the mnmber of self-employed single women increased. The results from estimatin the multi-sector participation equation reinforce many of the hypotheses concering the ddeminants of the probability of participation. The theoretical model shows that the important deterninants are those variables which influence the wage, the variables which affect the reservation wage and the proxies for the costs of participation across seaors. Age and education, human capial variables xpected to inrease the offered wages, were found to have positive effects on participation in all sectors. The effccts of education are the strongest in the formal sector. lhe proxies for wealth - uneaned income, owning a home, and the husband's wage - which inrease the reservation wages ar found to have, as expected, negative effects on participation across sectors. The most important cost of participation, childcare costs, have the stongest negative effects on formal sector participaton. lhis result supports the hypothesis that formal sector work and chidcare are less compatle than work in self-employment or the informal sector. In estimating the sectoral 'wage equations, it is necessary to correct for sample selecvity. The selection correction is signifcant in many cases indicting that OLS results will be biased. Education and j.edicted experience have significant and positive effcs on eaning for men, single women, and married wuen. Although, it was expected from the participation reslts, that the highest returns would be paid to education in the formal sector, the higbest returns are paid in the informal sector especaly at lower levels of education. Despit the high returns to education in the informal sector, women with more education are more likely to participate in the formal sector. It is important to note that although the returns are higher in the informal sector, the overall wage paid to a highly educated woman may still be higher in the formal sector because the returns in the formal sector may be added to a higher startivg base wage. The base wage may be higher in the for: al sector because of the existence of labor unions, minimum wage laws, and other government regulations in the formal sector. It is also important to note that dtis discrepancy in returns may be due to non-monetary returns to education in the formal sector sudh as increases in benefits and better working conditions. As diOcussed in Section 3, the differentials between male and female wages changed litde between 1980 and 1989. In 1989, the sample indicates that women in the formal and self-employed sect'rs make 70 perr't of their male counterparts. In the informal sector, ;he dfferential is less as women make. 85 percent of the mal6 wage. The decompositions of the wage rates in Section 7 suggest that discrimination is a more important source of the earnings diferentils betwc= mea and women than differences in male and female endowments. In the formal and self-employed sectors, between 81 and 89 percent of the wage differential is attribtable to dscrimination while in the informal sector 72 to 75 percent of the differentia s due to discrimination. F~4 Lato Force POckiaton and Wage DeerLa5m i Brad, 199 113 Appendix A MaJe PFaMdcpalon EquaUon and Earnln Fumctls As dsused In the utb, a husband's wag is assued to be a determinnAg factr In a woman's participation decision. However, becaus not adl husbands In th sample work for pay and because of endogeneity problems, husbands' wages must be predicted before estimaig married women's participr Am decisions. Consisny stimating wage equations to be used for predction requires several steps: 1. Fstmate the multi-sector participation equations for males. The sectora definitions re assumed to be the same for men and women and are outied in Section 2. 2. Calculate the inverso Mil's rios - one for each nividual for each sect. See the equation fvr the invers MAill's ratio in Section 3. 3. Eimate the wage equato Inluding the appropria inverse Mill's ratios as regressors. 4. Use the resulting coefficiets to predict a wage for ea indivl In each of the three sectors. S. Use the wage of the sector in which the ndidual Is most ikely t pardcipte as that individual's wage or maket value of tim. The following tables present the reults from sdmating a mulimiam logit participation eqWon and the resuls from esdmatingd the tr sectoral wage equations for the male subsample (with and without a selectivity correction) respectively. The log likdihood from esdmathg the participation euaon for men is -61,829. 114 Wom asEiy,Iymem and Py in Lahs Ana Apedz Tu1 e 4A.1 SeciDad Pazlidpmtlm Equalica. M1 N-S8,000 Formd Informal Self- Bmpoyee Eapleyee Empl1oyed Contmt 1.458 2.9S0 1.S04 (7-.9) (14.86) (7.8) AV .1048 .0412 .0956 (13.401) (S.096) (S273) Age-S d -.0020 -.0013 -W017 (25.567) (IS.382) (I2519) Some Primay School .3652 -.2666 .0968 (7.46) (S.IS9) (2.04S) Primy - Levd 4 .133S -.2605 -.1318 ')S)7) (S.060) (2.848) Fmind Soooaday .4433 .3936 .0121 (5.297) (4.161) (.13) Caller 1.029 1.249 .8338 (9.84S) (10.976) (7.219) Graduate Edocatkio 3.396 3.335 2.SS3 (3.817) (3.673) (2.675) Urban .4657 -1.163 -1.4S6 (9.700) (23.687) (31.67S) Own Home -.2201 -.5645 .3131 (S.799) (13.845) (7.812) Baick -.0783 .1442 -.4906 (1.129) (1.919) (6.616) IndinUlMlto -.0089 .1392 -.07S3 (.238) (3.398) (1.975) Asian .3009 -.5276 .6385 (1.273) (1.552) (2.638) Uneed Incom -.0006 -.0006 -.0006 (25.337) (17.318) (19.753) Nordthas -.2742 .1616 .2009 (6.083) (3.266) (4.414) Soco Puo .0297 -.2411 -.2930 (.S79) (3.896) (5.244) Rio de Ja3;ro -.2836 -.1182 -.4737 (4.72) (1.684) (7.042) Nortrwes .1689 .6624 .4931 (3.333) (1;2002) (9.517) Note: Abohit "slioo in par-i.am. Fuak Labor Forc Pw*stioo ad Wq. D _OrAb.X IX &M1, 19S9 115 ApputNx TaNe 4A.2 Mult -Seco Wap Equaws Mm N-58,o00 Sectdvity Comr,cted ,O hdbf=d SUf- Inral Sedf- FPI Wags Employed Foml Wag Employed Constat -.6264 -.7638 -.15U -.0331 -.3945 -.0529 (96) (5.70) (1.33) (.994) (8.91) (1.03) Seldoti Term .5097 .302 .1571 (11.1) (2.92) (1.9) Expeoiece (ap-ed6 .04S8 .0421 .0332 .0399 .037S .0298 (7) (155) (11.1) (24-5) (16.9) (12.3) ExperienO -0006 -.0006 -.0005 -.0004 -.00S -.0004 (19.6) (13.21) (10.1) (16.3) (16.1) (12.7) Some Primay .3142 .24S4 .2821 .3177 .2467 .2826 (15.1) (10.8) (12.9) (15.3) (13.9) (13.0) Primay - 4 .3341 .3399 .2082 .3441 .3426 .2101 (2LI) (14.4) (9.78) (21.7) (14.6) (9.88 Primy - 8 .4316 .501 .4683 .4344 .4985 .4681 (25.4) (16.3) (14.3) (25.6) (16.3) (14.4) Finis Scodar .46 .4938 .3089 548 .4926 .3062 (27.8) (13.2) (6.92) (27.7) (13.2) (6.86) Collee .7457 .7539 .678 .7451 .7538 .6837 (37.1) (19.8) (128) (37.1) (19.9 (12.9) Graduate .2447 .0704 .2613 .2389 .0596 .2IS9 (2.91) (.432) (.746) (2.84) (.366) (.742) Black -.371 -.1083 -.2453 -.3078 -.1086 -.2441 (12.8) (3.4) (6.22) (12.8) (3.40) (6.19) Asiam .1625 .2634 .137 .81 .2576 .1315 (2.48) (1.35) (1.29) (2.74) (1.32) (1.24) In mASulauo -.1944 -.0731 -.1884 -.1963 -.0735 -.188S (16.3) (4.25) (10.7) (16.4) (4.28) (10.8) NordteaS -.1626 -.216 -.178 -.256 -.3207 -.1746 (10.8) (10.2) (8.4) (10.4) (9.97) (8.27) Sao PulIo .401 .2426 .3457 .2390 .2644 .3448 (15.5) (8.70) (11.5) (1S.4) (8.69) (11.4) Rio de Janeiro -.1479 .0011 -.0717 -.1446 .068 -.0699 (7.50) (.031) (1.83) (7.32) (.202) (1.71) N{orthwest .1098 .109 .1651 .1056 .1076 .1627 (7.42) (4.98) (7.23) (7.12) (4.92) (7.14) A. Stswd erros convected fr a of an coed in MiD's rtio. Not: Abohft t4atios in pnofta 116 Ubmwa hX5~UVadF~L ~iA Aptami TaMe 4A.3 OLS Biks. of Wow"e MukiSscsot Eaning Pptiom 8iu WOr, hifoma Sd!- hfmd Sdlf- FPmi Wao EBhplod Fonmal Wags E&yod Constant -.0731 -.93SS -.39S7 -.4124 -.7261 -.4540 (.694) (7.18) (225) (6.28) (11.08) (4.91) E 1priewe (ag..no6) .0206 .0376 -.0125 .0398 .0342 .0288 (4.M (S.SS) OM.9S (12.19) %9.l5) (S.SS9) Exporimce -.0002 -.0004 -.uWm -.0006 -.0004 -.0004 (2.07) (3.90) (1.31) (.72) (S.33) (4.92) Someo Primay .2330 .3278 .1922 .0824 .2034 .2051 (3.M8 (S.25) (3.01) (1.74) (S.14) (4.U) Primay - 4 .ml .273S .2621 .2708 2939 .3S76 (4.71) (4.39) (4.21) (8.19) (8.35) (9.68) rinmay - 8 .564 .5824 .53S3 .450 S418 .4048 (10.18) (6.64) (6.07) (14.08) (11.92) (8.8 Finidd SoOOday .S3 .6439 .1986 .S436 .5570 .5130 (10.44) (6.75) (1.77) (17.38) (11.79) (858 Cdlogs .830' .7622 .6644 .7S66 .7383 .7601 (17.3) (9.78) (4.47) (26.93) (19.96) (9.35) Graduate .3158 .28s4 -.3378 .6883 .2444 -.0970 (1.95) (.868) (.360) (S.35) (1.66) (.283) Black -.2439 .-.0401 -.2067 -.2067 -.144 -.1649 (4.04) (.S37) (2.31) (4.S1) (2.96) (2.62) Adi .5097 .2663 -.4771 .1101 .0084 S743 2.37) (T78) (1.43) (.85) (.044) (2.61) IndA/MWulato -.IS6 -.o8s7 -.1324 -.2156 -.1058 -.193s (4.75) (1.85) (2.59) (9.88) (4.218) (4.57) Northe -.2478 -.3517 -.3174 -.3299 -.4480 -.3871 (S.99) (6.30) (5.26) (12.53) (14.66) (10.4S) Sao PalWo .3475 .3320 .4617 .2429 .2890 .2568 (7.52) (4.61) (4.74) (L274) (8.338) (S.089) Rio de Jansro .0355 .os37 .0e 3 -.0043 -.0oos -.1079 (.690 (.688) (.318) (.116) (.11) (2.054) Nogdthwa .1401 .0321 -.0173 .1324 .0536 .1870 (3.47) (4.61) (.253) (S.03) (1.68) (4.75) F-SPstic 188.S 139.36 51.73 432.14 419.76 176.29 Not: Abeobht tŁ.dos hi p_heb References Alderman, A. and V. Kozel. 'Formal and Informal Sector Wage Determinaion in Urban Low-Income Neighborhoods in Pakistan,' Living Standards Measurement Study Working Paper No. 65, World Bank, 1989. Blau, D.M. 'A Model of Cild i utrition, Fertility, and Women's Time Allocation, Research in opuladlon Econonis 5:113-135, 1984. Cornia, G.A., R. JoUy, and F. Stewart. Adjustment wth a Human Fare. Oxford: Clarendon Press, 1987. Edwards, A. Cox. 'Braz: The Brazilian Labor Market in the 1980s,' World Bank, Country Operations Division, Country Department 1, Latin America and C'rnbbear Regional, 199 1. Office, mimeo. Hay, J.W. 'Occupational Choice and Occupational Earnings: Selectivity Bias in a Simultaneos Logit-OLS Model,' unpublished Ph.D. dissertation, Yale University, 1980. Heckmnan, JJ. Sample Selection Bias as a Specification Error,' Economeria 47,1:153-161 1979. HIll, M.A. Labor Force Participation of Married Women in Urban Japan,' Ph.D. disxertticz, Duke Univasity 1980. Hill, M.A. 'Fmale Labor Force Participation in Developing and Developed Countries - Consideration of the Informal Sector, Review of Economics and SatLs 63,3:459-468, 1983. Hill, M.A. Female Labor Supply in Japan Implications of the Informal Sector For Labor .orce Paricipation and Houras of Work,0 The Journal of Hwnan Resourcri' 24:143-161, 1988. Jaffe, AJ. and K. AzumL Mhe BirtD Rate and Cottage Indusies in Underdeveloped Countries,' Economcs Dewlopmen and CuUra Qiange 9:52-63, 1960. Lee, L.F. 'Generalized Ecowmetric Models with Selectivity,6 Ero. feirica 51:50c -512, 1983. Maddala, G.S. LJnUed-!eeder and Qualitative Variables In Econometrcs. Cambridge: Cambridge University Pres, 1983. 117 McCabe, J.L. and M.R. Rosenzweig. *Female Employment Creation and Family Size,' in R. Ridker, ed., Populafon and Deveopme: The Search for Sdecaive Interventdons. Baltimore: 'he Johns Hopkins Univunity Press, 1976. MFaddez, D. 'Conditional Logit Analysis of Qualitative Choice Behavior," in P. Zarembka (ed.), Fronlers in Econometic. New York Academic Press 1973. Mincer, J. Schooling, Erpcrience, and Earnings. New York: Colkmbia University Press, 1974. Oaxaca, R. '*Mre-female Wages Differenctians in Urban Labor Markets,' Internaional Economic Review 14:693-709, 1973. Paes de Barros, R. and S. Varmdas. 'A Carteira de Trabalho e, as Cond.coes de Trabalho e Remuneracao dos Chefes de Familia no Brasil,' IPEA, 1987. Sedlacek, G.L., R. Paes de Bafros, and S. Varandas. 'Segmentacao Mubilidade No Mercado de Trabalho Brasileiro: Uma Analise Da Regiao Metropolitana De Sao Paulo," in PerspecvMas Da Economla Braselra - 1989 IPEA/INPES, 1989. Smith, S.K. 'Deteminants of Female Labor Force Participation and Family Size in Mexico City," Econondc Devnew and Cuiua Oiange 30,1:129-152, 1981. Stelcner, M., J.B. Smith, J.A. Brelaw, and G. Monette. 'Labor Force Be'avior and Earnings of Brazilian Women and MeM," in this volume, 1991. Tiefenthaler, J.M. 'A Multi-Sector Model .. P-emale Labor Force Participation and Wage Detumirtation Empirical Evidence from Cebu Island, Philippines," unpublished Ph.D. dissertation, Duke University, 1991. Trost, R. and L.F. Lee. 'Technical Training emd Earings: A Polychotomous Choice Model with Selectivity," Review ofEcon anks and S haritia 66:151-156, 1984. 5 Is There Sex Discrimation in Chile? Evidence from the CASEN Survey Indenni S. GiM 1. Introduction The aim of this study is to determine the extent of and components of the earnings differential between men and women in Ch0le's formal sector. We besitate to call this earnings differeal 'labor market disaimination since the study does not seriously atempt to explain this gap.' But there is little doubt thaX is gap exists. Figures 5.1, 5.2 and 5.3 graph the schooling-, age- and tenure- earrings profiles of Chilean men and women in 1937. These graphs reveal the existence of significant male-female earnings dif.erentials across all schooling, age, and tenure levels. Tlese figures are a graphic but inaccurate representation of the true earnings differentias between men and women. They are inaccurate because: First, there may be interactions between the various components of human capital-schooling, job tenure and general work experience- so that simple correlations between any one of these components and earnings may be misleading, especially for purposes of comparison across sexes. Second, the numbers are based upon earnings reported by working men and women, who may not be unbiased samples of their respecive populations. As a result of this sampling bias, calculations of returns to human capital (of working men and women) cannot be generat-ed to all men and women, hence limiting the T he word 'dscriminaion' is fraught with misuderstanding. Boulding (1976) writes. 'Tho history of the word itself is a stange one as it has two almost entirely opposis meanings, e very good and one very bad. On the good side, it meas a corTect appraisal of complex issuas end valuations, as in the expression 'a discriminating taste.' A perso who has a discrimating tast is upposed to be able to reject what is meticious and to discount what is only superficiagly either attractive or repellent, and is thus able to exercw true judgmenL.. At the other end of the sale the word discrimination in a bad sense mean precisely the opposite of the dicriminaing tate, that is, a faiul to make correct judgmnts, especially of other people. The sequence of discrimination in the bad se, thin, is illegitimat differnces, dat is, differnces in the teatmet or rwards of differen individuals which are not in accord with some stndard of eqity.' hno definition of a stadd of equity is essential for any study that attempts to measure discriminaion. In the can of gender discriminaton, given the biological diffaeces between the Eves, it is jirticularly difficult to agree upon a sensible stadard. U1P AmPOY"w - ray- L 11 mm*r 51 5chooiaS and Phsy Imaom. a ib 1987 TENURE AT JOB AND PRIMARY INCOME ,, Is;s p p3 0 k Thfer Sax DieIbeatkmo 6B Wai? 21) Vqkkt .2 AV and Pdimy lawomi. fiii. 198 AGE AND PRIMARY INCOME isioi. . , _ W , the person wor . lhus, P = f (W, W) (1) where P takes on the value 1 if the psn partiipates in the labor market and 0 if the petson does noL 4 The data "wo amyzled for each of 3 pniasWk W _rps , noe a_oo8ing. I evels, ae, and houeobo chaacndtcs (awta as, arm dqmdmcy rati, a) mm to vay p wiportonatly wihthe dere of whedoioU of &a lbo fkm in h puve 126 WavW n 'r EWpkYWIant ad Pay bL Laga Awka Table 5.2 Means (and Stadad Doviatiou) of Variables: By RWgoln FEMALES MALES Variable Total Urban Rural Total Urban Rural LF P cipation Razt 0.272 0.296 0.138 0.663 0.642 0.760 (Fraction) (0.45) (0.46) (0.35) (0.47) (0.48) (0.43) Primary Income 20045 20885 11142 30673 34498 14885 (Cin S/Month) (26553) (27332) (1316900) (50684) (53646) (31363) Household Income 55548 60232 28903 53566 S86S6 30259 (Cileam S/Month) (79187) (83087) (42846) (73657) C77997) (41845) Schooling 8.811 9.272 6.190 9.050 9.711 6.074 (Years) (4.09) (3.98) (3.72) (4.20) (4.04) (3.52) Age 33.653 33.821 32.693 33.296 33.418 32.M (years) (14.17) (14.13) (14.33) (14.14) (14.14). (14.1A) Tenure 5.267 5.403 3.558 (.421 6.S92 5.733 (Yeare on Currntr Job) (7.17) (7.19) (6.44) (8.43) (8.47) (8.24) Fraction Married 0.478 0.473 0.508 O.S04 0.515 0.451 (0.50) (0.50) (0.50) (0.50) (0.50) (0.50) Fraction Cohabiting 0.036 0.035 0.043 0.036 0.035 0.039 (0.19) (0.18) (0.20) (0.19) (0.18) (0.19) FPction Separted t%.045 0.048 0.021 0.020 0.021 0.014 (0.20) (0.22) %0.14) (0.13) (0.14) (0.12) Fraction Wiridowed 0.050 0.052 0.040 0.014 0.015 0.013 (0.22) (0.22) (0.20) (0.12) '0.12) (0.11) Praction Single 0.391 0.392 0.389 0.427 0.414 0.484 (0.49) (0.49) (0.49) (OM0) (0.49) (0.50) Fraction Hshold Head 0.105 0.111 0.068 O07 0.512 0.482 (0.31) (0.31) (0.25) (0.50) (0.50) (0.50) Hoi-shold Sizs 5.107 5.054 5.607 5.222 5.117 5.584 (Numberof Member) (2.18) (2.15) (2.47) (2.19) (2.14) (2.48) Woke u R Household 1.627 1.626 1.637 1.738 1.702 1.903 (Number) (1.12 (1.11) (1.17) (1.16) (1.13) (1.29) I Childnat OS Yr. 0.626 0.601 0.764 0.589 0.572 0.668 (Numbar) (0.85 (0.82) (0.98) (0.83) (0.81) (0.92) F Childrat 6-13 Yn 0.615 0.582 0.806 0.606 0.57S 0.744 (Number) (0.84) (0.81) (0.99) (0.84) (0.80) (0-97) Note. Pimary income is mooy iWyxe firm main job. & Then Su Diwiwbiadon i _ Ide? 127 Market wage W depends upon stocks of producti 'e skms - human capital - such as schooling and job market experience, and the state of the labor market - regional and infrastructural factors - which determines the rewad to these skills. Rervation wage W* depends upon the productivity in activities other than labor maket work For both men and women, it mav depend upon the retmns to fiuther invesmntf in hum capital, which may require full time school attedance. Thus, for example, reseraton wage rates will be higher for students who have begun stdies toward a degree but have not yet obtined a diploma. W also depends upon household characteristics such as the number of (young) children and other dependent relatives, and on household wealth. Table 5.3 reports the results of probit esmaons for the work participation fimctions (equation I above) of women and men aged 14 to 65 years.s Table 5.3 also reports the pecentage change in the probability of participation (%Deriv), evaluated at the means of the right-hand-side variables. The results indicate that: I. Higher schooling levels are positively associated with the probability of participation, except when the years of schooling indicate incomplete programs ef study (8 to II years for both females and males, and 13 to !5 years for males). 2. The age proffe of partiipaion is Inverted U-Ohaped The proffle for males is more curved than for females. 3. Married and cohabiting women are less likely to work than those who are single or separated. Married and cohabiting men are more likely to work dtan separated or single men. Thi result is consistent with the theory that marriage allows males and females to specialize in task, and females have a compartive advantage over me in household production. 4. Being a household head is posively correlated widt the probabflity of participation for both men and women. 5. Higher household income (total income of other members of the housebold) increases the likelihood of work for women, but decrases it for men The result for women is somewhat puling. 6. Holding constant household size and the number of children, inreases in the number of workers increase dte probability of participaton for both women and men. Holding corstant the total number of household workes and children, incrses in household decrease the probability of participation for bath women and men. 7. The more young children there are in the household, the lower the likelihood of participation of women, but the greater the probability of men working. Older S Many fictiol form wvo fied. TM results do not diqp agficmtly, so the mad easily irprtbble form is rpoted. 128 Wman i' Empkvyns and Pay In Lad4 Amnrica children have similar but weaker effects on decisions to work." There is no evidence that older female children are subs es for adult females in household tasks, 8. Rural women participate lesm than urban women; urban men participate less than rural men. Using the probit regressions in Table 5.3 and Appendix Table 5A.2, we conducted simulations of the effects of schooling, age, marital status, and household chaaderistics on fmale participation probabilities. Results of the simulation exercises are reported in Table 5.4 below. Reasozs for not wr,ing. Table 5.5 reports th reasons for not working. Naturally, the sample consists of those iadivida,'s for whom W- > W. A short description of the table is that women who do not work cite working at home, in .chool, and retired Cm that order) as the most important reasons, and men who do not work we in school, unemployed, or are retired (n that order). Appendix Table SA. I reports the reasons for not working by sex, maritl s=atus, and age group. Married, cohabiting and widowed women overwhelmingly cite household work as the reason for not working in the market (with Retir-AV being the only other important reason for about 25 percent of non-workers aged more than 50 years). Single and separated women do nt work because they are in school (when they are young), or because of household work (for all ages). More than 10 percent of single women aged 21 to 40 yeus also list unemployment as a reason for not working. Prime aged men who are married, cohabiting and widowed and who do not work are generally unemployed (actively looking for work). Older non-workers are pensioners. Single men whb do not werk are in school (age groups 14-20 and 21-30 years), or unemployed (age groups 21-30, 31-40 and 41-50 years), or physically unable to work (age groups 31-40, 41-50 and 5t+-; years). Following Killlngswcrth and Heckman (1986), the error term in the work participation equation can be divided into three components: Tle error term due to differences in tastes (utility funcdon errors), the errors due to wuaccounted-for differences in budget constraints, and the errors due to discrepancies between observed and optimal decisions. This last errcr is more important for men, since 'Unemployed as a reason for not working is more important for men than for women. Such differences between the sexes ma it inadvisable to attach a uniform interpretation to the sample selectivity correction terms (invere of the Mill's rado) for men and women, since these are derived from the error term in their respective pardcipation fimuctions. Tlis problem is discussed further in the next section. I To futher exumine t effects of children oa the decision to work, eXmine the pro eti of men and women by maital liat'B in Table Al 1.2 in the Appsdix. For mmried and coabitng women, increase in the of children lowr the probility of padiipation, especially if hb e of children ae 0 to S yar old. For single ad seprated womn, the number of childen in the household inames the likelihood of wodcing. Th number of children has no effect o the participai dcisio of men, holding constat hovsbhold size and number of woke in the household. b The Si LU*ad= x asi? 12 Tabl 5.3 weak Paricipstk P All Ml & Fel Depmdat Vaiab: Do yo wodk (Yes-., No-0) Femuleakles Coeff. X i %Deuiv. Coeff. Error %Deiv. Scooling: 8-11 Yart -0.0145 0.0219 -0.5 -4.0159 0.0225 -2.6 Schooling: 12 Yes 0.2699 0.0235 8.4 0.1176i 0.23 4.1 Schooling: 13-15 Yars 0.2841- 0.0315 8.9 40.1885 0.0352 -6.5 Schooling: 16+ Yos 0.9990W 0.0376 31.2 0.1828 0.03b)2 -6.4 Age: 21-30 Yewt 0.0046° ° 0.0270 31.4 1.2882 J.0255 44.7 Age: 31-40 Yeas 1.27S0 0.0321 39.8 1.5186 0.0359 52.7 Age: 41-S0 Years 1.1497 0.0350 35.9 1.2059" 0.0391 41.9 Ag: 51-65 Yews 0 66" 0.0377 1F.6 0.4275 0.0382 14.8 Mafiedt -0.6920"' 0.0228 21.6 0.6077~ 0.0322 21.1 Chabitirg -0.6223 0.0468 -19.4 0.62&S 0.0548 2L8 Widowed 0.3843 0.0436 -12.0 40.1697- 0.0695 5.9 Sqeaz 0.0662 0.0408 2.1 0.4798 0.0630 16.7 Ekuaseid Head 0.3262" 0.0319 10.2 0.2503° 0.0324 8.7 Hoarhbold Iwomke 7.5-7 - 1.1e-7 0.0 -2.2o.6- 2.0*-7 0.0 I Household Members 40.0262! 0.0062 40.8 40.0441- 0.0066 -I.S # Hboushold Wokems 0.0618-" 0.0104 1.9 0.1479r 0.0107 5.1 Boys: S Yrs -0.024r 0.0157 -0.8 0.0631- 0.0177 2.2 Girls: 0-S Yrs -0.0034 0.C060 -0.1 O.050or 0.0177 1.8 Boys: 6-13 Yr 40.0053 0.0160 .0.2 0.0368" 0.0173 1.3 Gids: 6-13 Yrs 0.0016 O.O5 0.1 C.v2h9' 0.0173 1.0 Rural Dwuzmy -0.3701- 0.t61 -11.9 0.4906- 0.0245 17.0 Covotant -1.2090W 0.0363 -37.8 4.8W58"' 0.0363 -29.4 Losg Likeliood -16277.6 -14215.4 Chi-Square 5811.1 11025.2 Sample Size 32765 30887 Men of Depadet Varia 0.2722 0.663S t Theomiedsboolingg u is0-7 ya,' eulod agroupo I '14-20yos, d the oad maia mau cla is 'SbuIe. * IOies ipgnae d 10 Icmg IveL Indicakte aigiohca me S I leveL - Indicates ui.jalface ati I 1 pe loveL 130 Wamsa'sJ _ pym audPay XL AmuA*a Tablle Ł4 Prdctd Pzobuity of FaPil Labor Forc Paipos omlt of S)mlai Exau) Chmact=_stic Prdicted PrbaWU (Pecatw 0. Mm. Ptri t Ra All Womn 27.22 MUod Womu Only 22.70 1. CoWad Schf 0 to 7 Yo. 23.37 81toitYens 22.93 12 Yews 32.47 13 to iS Yews 32.90 16 ad More Yom 60.73 2. Ap 14 to 20 Yea 8.00 21 to 3( Yews 34.49 31 to 40 Yom 44.84 41 to 50 Yeas 39.95 S1 to 6S Yas 20.99 3. Mall Stan Mwied 13.91 sing 40.77 4. Fen& Head of EoUsod 37.78 S. Num of ail* Aged 0 to 5 Yoe 0 Camil 2859 1 Cild 22.97 2 CWdren 18.03 6. Nuzberof Chidrmad 6 to 13 Yoe 0 aldrm 28.59 1 Ud 2S.7S 2 ChIdrm 23.07 7. Ruda Reddenco 17.93 a. Simulation band on rmb prb rebitug nfo 9Married, Cobabkiag ad Widwed Women Wo -rmd i Appea& Tabb SA.5 No: ktuoept isadjwtedto makepvdtedpmbaiyeqatanprobabirty. A .7isru Sc Diwindto.n 0,t0? 131 Table 5.5 RraaoSa r notWlin: ily Sea d MarbiS (%) Muriod, Coaiting SqUUad Tota & AWoed _ S;" Women Mee Women Men Wome men 1. L4oking for Foe Job 0.16 0.60 3.91 8.49 1.62 636 2. Unenqtl 1.06 30M 5.17 10.35 2.66 *1S.75 3. Houebold Work - 8.38 3.78 35.24 1.57 67.71 2.17 4. Suxying 0.55 2.17 4835 64.15 19.14 47.40 5. Rctirud 0 dc) 7.36 45.00 269 2.12 5.55 13.71 6. Ratir 0.13 034 0.09 0.06 0.11 0.13 7. Unable to Wort 1.56 9.78 3.52 6.12 232 7.11 S. Tanpomilyln 0.60 6.G3 0.70 1.60 0.64 2.96 9. Other Reoam 0.21 1.3 0.33 S5. 0.25 4.41 Tobl Ob.arsdons 14,956 2,669 9,516 7,208 24,472 9,87 5. EAnInp WumdIos Te probkl of wsaik sdiiity. Earnin data are available only for working womi ad men, tha is, individuals with W > W-. For people who have the same wage, W, wordig me and women have smaller W* than doe who do not work For individuals w;th the sme resevation wa3e, W*, wodkzg women have relatively high mare wages, W. Becams of both these differences, workiag men and women are unrp entaive of dti male and fenale populations. and polky iferences derived from regressions for workers may ba Invalid both for nonworking men and wve. dnd for working men and women. The problem can be Interpreted, alternatively, : 1. sample sdeecvity bias, or 2. a problem of omlued rdevant viables in estmating he human capital eari fiwntion (eqa}vendy, of simft lw equto bias), or 3. a problem of tncated data, since we do not observe (offered wages for peoplo who do not work. Followlng Kjll;n8orth and Feckman (1986), let the i%ork participaton decision be: P = W + R + u (2) where P = I if the InAdu works, 0 odtwise, W is the markt wage, R is a set of variables that dermine the rervation wage W*, and u Is the error The human capita earuhp funcdon, which determin the market wage, is: W=Xp6+v (3) where X is the marix of observed human capital vaiables (schooling, tenure, geneal work experiece, ec), i Is the voctor of reur to these variables, and v Is the error term. Estimates of P obtned by equgfon 3 by ortinary least squares will be biased, since some t)f the facors that mak women more likely to work may also be factors that make them eraornary high or low wae earmems. is, v is li ly to be corated with X, which also dstermms wages. A way out of this problem is to atimat the wr participain equation 2, using X as an intv:ntmt for W (sin W is obseved only for works, but X - schooling, age, tenure, etc.- is obavod for everybody), and iluding R to esro Identifiability. ben, the uneplained part of P is a compclte index of unobsved chareristics ta are relevant for the decision to work. A summay measure of dth unobserved aurbutes is X. the invae of the Mill's ratio. X is then included &s a ressmor in the earulp equation, W Y XBfl + a 9 + , (4) which rovide3 meaure oe ta are free of seectvity bias. A is an inverse, monodc function of the ?robabily of partichlado. Thereore, the value of X must be great for m o tha u or worn kr.7 A nepWat valft of a in equion 4 implies at the uwbsarvtd anIxies that make woes earn nmmsualy high wages are also attributes that nake it lem lkely that the Inivid will in fita wor. A poskive value means that unusuamly high paid wcdmm ee dt whose rervatoA wag are low, ma it more lely for theo to be in the makat It is commonly argred "thsat pe sdecivity i g more serious problem for women. Th argiiment ;s that while only w ewho have wmialy lre aount of waket skils work outside the home, alms al mn work In the maukt At least In Ch's cse, this argument is seriously flawed. To see wby, tk anther look at Table S.5. The reasons for not working (other tham xmendln school) are chiefy 'UwWkzyment for men and 'Household Work' for women. While the rcoans fr not working ar different, both can ranl in the samples of workers being unreprfa of dteir resectv populato The reoi. that make (esnploy4) males highly paid relative to dir obseire human capitl levels may also be those Om make them more (or less) likely to be employed in the first plam Just as theory cannot help t predia tho sign of a fot wo it, do wt help us pre.ct 'he sig of a for me. Because ofthe bove argument, both female and male ernngs funcions were esdmateds the sample select;vity correion. Resndtr 4! cawLp ngre&iom A crippling limitadon of the CA!EN survey is tba there is -o informatim on amout of laboc suppiied (xus wodki per week or weeks worked per year). All svorken ar assumed to s ly the same mOunt of lbor per time eriod. Whie this is a pateny absurd assumpdon, tae was simply no wry out of this problem. Tables 5.6 and 5.7 report dte renat of ouman Oquations 3 and 4 for women and men repectively. Odd-numbeW coluns report resut for equation 3, while columns are resolts of fitting equ-Ioqn 4 to the da The mai reswhs are: Y Te raio of avesp X of #o & of uinmwkmud &hme be Ion than 1. Ths adi is abod 0.70 foe womn, and Adw 0.40 w ,_ 1. Returns to scdooling fDr males are about 13 percent, roughly 2 percent higher than those for females. Adjusting for sample selectivity lowers the returns to schooling for females margina!gy (by ! or 2 percent), but leavez the retmus to schiooling of males unchanged. 2. Returns to potental wirk experiece (Age-Schooling-6) are about 2 percent for women and about 4 percent for men. Tbe reason is that potential work experience is a bette proxy of job tenure for men than for women, who have more frequenly interrupted careers. 3. Tenureearnizgs profiles are oteeper but more curved for women. Evaluating the reurns to teaum for women and men at their repective means yields returns o' about S percent and 3 percent respectively. Due to higher multicollinearity between tenure and pote work experience for men, the more meaningful measure of returns to work experience is the sum of the retns to tenure and potential experience. This is about 7 percent per year for both men and women.' 4. The coefficient for Lambda is significant and negative in all regressions except those including marit satus dummies, and is slightly greater in magnitude for males. This inplks that sample sectivity is important for both men and womea, though one must be careful in intepretina this coefficient. For wome, given the information in Table 5.5, it Implies ths the unobserved charaeiD cs that make w.men high incon-eamers relative to their schooling and tenmre levels, also make them more likely to stay at home (or attend school). For men, a negative coefficient zbr Lambda implies tha unobserved atributes %Nich make them better earnms reladve to their schooling and tenure levels, also make them more likely to be unemployed (or attend school). 5. Rural dummies are insignificant for women, but negative and significant for men- Tis probably reflects the greater amounts of subsistence production in mural areas. The results remain unchanged when the ural area dummy is dropped from the regresions. 6. Finally, including marital sau in arnings regressions significandy lowers the urns to sc!ooling and to temure for women, but leaves the coefficients for males unaltered. Including marital status significantly aits the sample selectivity coefficients for both sexes: a for females tiples in magntde while maining negative, whe that for males becomes zero (and staically insignilicant at the S percent level). Married men and women eam about 20 and 30 percen more respectively than singe worken. Strictly p,ing, rwe sould nt be includod t earminp reessios since it wuno inclded in dhe work paicipszi funtons and siac it is iraneousy dtermined with ea_nsL (Team ia observed oly &.- wodm*s, o it canot - i bo ed in dte wout pacipaio resoemi.) To cotrect for both tho problem, ws u _ed ins =a vables tecnique instead of ordinagy lst q s, with quadmatc .ns of schoolng ad .go p &air itacticNa ins_umm for taIm Tbee an poor katmumw for om, qeqcially for vmm, but they wus the bts w could find. I reults do tot disp e pificaly, s we ipt ol te least aqar regames in the pper. Table 5.U Female Earnings Regresions Depmeadat Variabb Log (Primary Incme) (1) (2) (3) (4) (5) (6) Schooling 0.1263 0.1187 0.1163 0.1084 0.1120 0.083S (64.S4) (S6.S9) (48.48) (41.19) (45.53) (27.93) Age-Schoot-6 0.0292 0.0246 0.0191 0.0144 0.0141 40.0057 (16.92) (13.87) (8.37) (6.08) (5.82) (-2.14) (Age-Schoot.69 4.0003 4.0003 4.0002 -0.0002 -J.0002 0.0002 (-10.98) (-7.77) (-5.36) (-3.22) (-3.91) (3.15) Tcure 0.0552 0.0552 0.053S 0.0507 (18.34) (18.39) (17.75) (17.07) Tewura 4.0014 -0.0014 -0.0013 -0.0012 (-13.85) (-13.7S) (-13.23) (-12.05) Mafried Dummy 0.1571 0.3793 (8.03) (16.08) Chabiting Dummy 0.0403 0.2287 (0.82) (4.57) Widowed Dummy 0.0649 0.1658 (1.66) (4.-5) Sepaoede Dummy 0.0298 -0.0329 (0.97) (-1.07) Rurml Dummy 4.0889 -0.0241 0.0208 0.0359 4.0260 0.1267 (-3.42) (4.92) (4.67) (1.13) (-0.84) (3.98) Lambda (A) 4.0824 -0.068S -0.1973 (-9.85) (-7.26) (-16.;1) Constmt 7.895 8.2153 7.9706 8.2607 8.0118 8.8962 (252.86) (182.78) (216.47) (152.26) (215.46) (135.96) F-Statistic 1316.20 1081.65 759.41 662.51 466.33 461.6S Adjusted R 0.3253 0.3312 O.i10 0.3580 0.3591 0.3790 Sample Siz 10.912 10,912 8,305 8,305 8,30S 8,305 Mea of Variables LOg (Prizmiy In_urMn) 9.46S4 9.S668 9.5668 Schooling 9.3788 10.1467 10.1467 Age.Scbooling46 22.2929 18.2192 18.2192 Team 5.6325 5.6325 Mafried & Coabiting 0.4147 Rnal 0.0863 0.0738 0.0738 Lambda (Wodfeto onworitern) 0.7253 0.6928 0.6928 Notea t-stastics in pareathmes. Primy uxxxneis uafily finmne fixn main job. Odd numbered coin port rm1h fat equabo (3) nd am numb-trd columms rpt reolts for oq (4). Table 5.7 D_ V Eunp -o Dqindmg V.riabl Los (Primary hacoam) (1) (2) (3) (4) (5) (6) Schooling 0.1366 0.1338 0.1292 0.1262 0.1233 0.1231 (100.6() (97.77) (8L26) (85.13) (82.46) (82.11) Ago-School 0.0641 0.0426 0.0423 0.0317 0.0314 0.0301-6 (42.26) (27.66) (2882) (18.15) (19.55) (17.16) (Age.School.69 -0.0006 -0.0004 -0.0005 4.0003 -0.0004 4.0004 (-27.07) (-15.31) (-19.33) (-10.57) (-13.68) (-11.25) Tamcu 0.0356 0.0347 0.0334 0.0333 (20.82) (20.33) (19.62) (19.58) Teutre2 4.0006 4.0006 4.0006 4.0006 (-12.33) (-11.52) (-11.18) (-11.07) Marriod Dummy 0.2404 0.2214 (17.23) (13.05) Cohaiting Dummy 0.0976 0.0784 (3.73) (2.81) W-iowed Dummy 0.1181 0.1089 (2.58) (2.37) Sqparatd Dummy 0.0323 0.0169 (0.92) (0.47) BnMW Dummy 04.1176 -O.1581 4.1114 4.1452 40.0980 4.1067 (-8.81) (-11.59) (-8.23) (-10.50) (-7.27) (-7.52) Tambda t) -0.1020 4.0890 4.0191 (-13.35) (-11.13) (-1.96) Coastan 7.8S61 81017 7.9531 8.1m 7.9937 8.0397 (3SS.91) (282.54) (340.17) (265.20) (342.19) (243.18) F-Stidsic 3419.93 2792.49. 2343.29 2037.93 1461.12 1328.82 Adjuted R' 0.371 0.376 0.400 0.404 0.409 0.409 Sanqplo Size 23,166 23,166 21,080 21,080 21,080 21,080 Mamw of Varable LOS (Primary nboom) 9.8091 9.8191 9.8191 Schooling 8.8696 9.0006 9.0006 ApScbooling-6 22.4708 21.0874 21.0874 Team 6.8908 6.8908 Married & Cohabiting 0.6872 Rural 0.1950 0.2018 3.2018 Lumbda (Woaxn/Nouwokm) 0.4144 0.3951 0.3951 Not= teaIbsticsin pamiemu. Puiinay ino as inxmIb1y ixms fiom mum job. Odd numberd cobne rqpxt tombs wr equatpi (3) md am nunbered oohls rqport emits foir eqUaio (4). To understand these results, let us first examine the rationale for including marital status in these regressions. In almost every country, married people are observed to earn more tnan similarly skilled single workers. One explanation is that attributes that lead to a person marrying and staying married may be coreated with h gher unobserved job skills. Another explanation is that men who are high wage earners have greater incentives to marry, due to increased specialization afforded by marriage.e Women who are married and nonethless work may be those with very high market wage, W, relative to their reservation wage, W*. The inporlac of marital ta. In any case, it seems that the interaction betw;en marital status and sample selectivity is not fully captured by marital status dummies. To further exine this issue, we computed separate probit work participation regressions and earnings fiuctions for four groups: Married, cohabiting and widowed ('married') females, and males; and single and separated ('single') females, and males. Tlese results are reported in the Appendix as Table 5A.2 (probit participation results), and Tables 5A.3 and 5A.4 (female and male earnings regressions). The main results ot these earnings regressions are that: 1. While the returns to shooling, potential work experience, and tenure are roughly the same for married and single women, the reurns to all three forms of human capital are higher for married mer. than for single men. 2. Selectivity coefficients for women are negative for both groups, but sample selectivity is greater for married women. That is, the same unobserved attributes that make women bonaer ar tha their schooling and tenure levels warrant, are more likely to make niarried women stay away from work than single women. 3. The selectivity coefficient for married men is negative and significant, while that for single men is positive and small. lhis implies that the unobserved charteristics that make men higher income-earners than their schooling and tenure would suggest, also make it less likely that married men in fact work, while making it more likely that single men worL These maZniudes can easily be explained: While married men who do not work are either retired or unemployed, single male non-workers are mostly in school. 6. Accounting for the Earning Differental The Oamca decomposion. Table 5.8 presents the Oaxaca decomposition of the earnings gap between men and women. The priav result is that all of the higher earnings of men can b explained in tms of higher rates of return to human capital. Women appear to have higher endowments of human capital than m, than their earnings would indicate. Market structure rewards male skills more than seemingly comparable female skills. Some analysts would interpret this as evidence that the 'purest' form of discrimination - that the market rewards females less than males for the same sills - explains all of the earnings diffrential. This reswlt is robust to the choice of index type. I See Comwoal and Rupat (1990) for a tea of tre altenaive hyothese using United State dta, and the implicatio of thir roWlt for the trst of marital tas in eanings igreasion - ~~~~~~~- t.t 6.M J W OII A t i Table 5.1 Accounting for Earp Diffe ila TMe Oaxaa Decmposition (AU Numbe. am Prcent:ges) Norm: Male Wagr Functin Fund Wage Function Components: Coefficients EBdowment Coefficents Endowments W/W. I. All Women and Men Using Equatiom 114.9 -14.9 113.7 -13.7 71.3 without Teamu Using Equatios 116.8 -16.8 118.2 -18.2 77.6 with Tenure Using Equations 102.6 -2.6 93.0 7.0 77.6 with Teaure & Marital Stas H. MarTied Women and Man Using Equations 107.7 -7.7 103.2 -3.2 65.3 wiLiout Tenuw Usng Equations 112.5 -12.S lll.S -11.S 77.3 with Tenumr m. Single Wom and Men Using Equaton 177.8 -77.8 1SS.7 -SS.7 96.3 without Tenure Using Equtiou 152.1 -S2.1 136.3 -36.3 98.7 with Tenure Notes: MAk Wae fmetion u D _Viniri t Norn: Cocfflicrfa CoMpoct_ - X( Pc Pj. Endawn Coapoo3t Panale Wage ti u D icrinaoy Nonrm Coefficients Compoat - X,.( ".- ED) Undowna. Cmponet Table 5.9 Labor Fore and Earnmigs by Scx and ndustry Percwtsag IAb" Force Eamninp Industy Females Mals F/M Femles Males F/M Agricultwe&F & Fhing 8.02 32.37 0.2S 15.56 18.11 0.86 Mining 0.40 4.28 0.09 41.07 43.9S 0.93 Maubicuring 13.00 14.38 0.90 16.88 32.84 0.51 Coaatructon 0.98 9.57 0.10 28.53 23.36 1.22 Commerce 20.06 11.66 1.72 20.20 32.81 0.62 Serices: GovtL & Finncial 8.04 7.71 1.04 32.56 50.52 0.64 Srvice: bhold & Pewnooal 25.S7 5.27 4.58 9.79 21.5S 0.45 Servie: Social & Comiant 21.08 5.44 3.88 31.57 48.03 0.66 Tranaportado* &Utilitis 2.17 8.63 0.25 35.26 36.19 0.97 Not Elsewhere Clamified 0.67 0.68 1.00 27.73 43.49 0.64 Total Number in Sample 9,152 22,985 10,912 23,166 Not4 Eanings a Mondry Eairnip in Tousands of 1987 Chikan Dolan (USS1 apprxima equad to C$219). Table 5.10 Avenge Schooling, Tamre and Age by Induty Industy Schooling Tenure Age Femae Male Female Made Female Malo Agricthue & Fishing 7.49 6.20 2.27 5.78 29.84 34.05 Minrg 11.91 9.09 8.39 7.95 35.27 36.58 ManufuchzrSig 9.70 9.56 5.63 6.51 34.45 35.00 Camstruction 11.66 7.90 6.12 5.21 31.78 37.96 Commerce 9.93 9.67 4.98 6.5S 35.40 35.82 Seonsee: Govt. & Fmcial 12.14 11.36 4.89 6.82 32.82 35.94 Serices: Bhold & PeuYool 7.33 8.70 3.78 7.36 32.57 36.78 Servies: Social & Comnmnity 13.55 13.04 8.20 8.77 35.02 37.16 Transpoton & Utility 11.68 9.69 7.48 7.06 33.50 37.02 Not Elwewhte Clusified 10.3, 10.89 5.12 S.3S 32.62 32.94 All Workers 9.38 8.87 5.63 6.89 34.37 36.09 All Non-Wore 8.24 9.26 0.76 0.68 33.37 26.66 Total Number in Saml 32,676 32,078 9,249 23,035 32,67S 32,078 Now: Schoolng is H ;ghe Ckads Attaid Cm Year). Tenure i Numbr of Year. Wo*kd at Cumat lob. Age is in Year. &a Airf A,mw1n S4Ut 1 JY Table S.1l LAor Forc and Eanins by Sex and Ocwapat Pecetg IAbor Force Euninp industry Femal MAl F/M Females Males P/M Professional & Tehnical 17.07 7.44 2.29 39.73 73.18 0.54 Managrs & Proprtors 0.14 0.66 0.21 180.42 168.94 1.07 Administadve Persannl 15.68 9.14 1.72 32.30 49.47 0.6S Traders & Vewdors 1S.24 8.62 1.77 20.91 31.58 0.66 Sevicoe Workers 32.92 7.14 4.61 11.16 17.87 0.62 Tnsport Works 1.04 11.03 0.10 28.12 23.50 1.20 Non-Precision Workers 14.60 35.98 0.41 1243 14.80 0.84 Armed Fors Persand 0.02 0.99 0.04 51.56 44.66 1.IS Not Elsewboh Classified 0.39 0.38 1.00 26.54 23.41 1.13 Total Number in Saqle 9,152 22,985 10,912 23,166 Notes: Earnings arc Monthly Earutinp in Thousands of 1987 Chklan Dollars (US$I approximey equal to CS219). Table 5.12 Averag Schooling, Tenure and Age by OcuWpo Occupation Schooling Teaure Age Pamlo Mle Femle MPo Femb!e Male Profesional & Technical 14.78 14.8S 8.62 8.68 34.68 37.08 Manager & Proprietos 14.42 12.59 11.55 15.2S 40.26 45.32 Adminitaivo Pwoonel 12.6S 11.27 5.89 7.80 32.37 35.99 Tradens & Vendors 9.86 9.56 S.1: 7.02 35.23 36.62 Service Workers 7.57 7.95 3.98 4.92 33.96 35.78 Trnsport Wodres 9.87 8.31 S.10 6.25 31.31 36.04 Non-Precision Wodrer 8.20 6.22 4.88 5.86 34.37 34.16 Precision Workes 899 8.11 3.22 S.91 33.49 36.03 'Arwed Forces Pesnnel 1.OO 11.09 3.53 1211 34.43 33.93 Not Elwswher Classfied 9.5S 9.85 5.60 7.11 31.92 32.06 Non-Workers 8.24 8.88 0.76 0.65 33.48 26.51 Toa Number in Saple 32,676 32,078 92,49 23,035 32,675 32,078 Nokts: Schoorng is Hghoe Gra&d Attind cm Ycar). Teanur is Number of Yar Wod at Curent Job. Ag is in Yea. However, the wage differential s endrely due to the diff ial b ew the earnings of maried men and women. For singl men and women, the gender earnig diffntial i insignificat Tbhs b a stronger flnding than other studies which find that 'darimlnatlon' (Iffce In coefficients) is-as expeed-mch less for single than for married workers. In this paper, because nformation on hous worked is not available, maritl staus Is liely to pick up some of the effect of differences in bhors worked by men and women; It is Ubdy that hours worked in J40 Wam 's ZpymeM d Pa, ba rLatA A a the mrket differ significantly for marfied men and women, but are roughly equal across sexes for single workers. PosiMe explansr. BelDre this earg differential between marred men and women (of about 6000 Chilean dollas per month) is deemed due to discrimination, it is usefil to examine the labor market fiuther. he main argument against the discrimination view is that having the &ame amount of market skis is not enou to ensure that earningp are the same: It is the use to which these skills are put dLat detemines the returns. To exam this question, a logical first step is to str.dy gender differen in the industrial and occupational composition of employmesL Table 5.9 lists the industrial compositko of female and male employment and their mean earings. Females are conenrat in non-financial services, commerce and madcturing, out of which only household and personal services is an extraordinarily low paying sector. Males are relatively more dispersed, though about one third of all males work in agriculture and fishing. Except In construction, femal earn less dun their male cnteparts in all sectrs, with the largest differentials existing in housebold services and manufactring (one dominated by women and the other by men). Table 5.10 lists industry means of schooling, tenure at current job and age. While working women are generally more schooled dtan working males, they have less tenure, and are younge. Given these levels of human capital across sectors, the earning differentials in manfacturing, social and commanity serve, and commerce seem particularly unjustfied. The divese nature of these three sectors also rules out production function explanations of gender gaps in earnings. The occupational distibution of female and male employment ratios and earnings (rable 5.11) clearly shows that occupation in which women are concentrated (Professionals, Administrative Personnel, Traders, and Service Workers) are also those that have the lowest femaleto-male earnings ratios. Table 5.12 shows that these occupations have the smwllest differences in schooling and the largest male-female differences in tumre and/or age. There is no occumpzion where the schooling advantg, of women is not offset by tenure and age advange of me These tables suggest that there are large positve interactions between schooling al work experience (both job-specific and general). The nature of female human capital seems to prevent it from obtaing markt returns that are sepdar y due to either scholoing or temnre, since average tenure is generay maller for women, both across secos and occupations. 7. Discussion This paper analyzed the detrinats of work participaon and earings for women and men in Chile. Cbile is an intesting case study since male-female differences in health and the levels and t3pes of schooling are not large, but gender differentials in eamings remain significant ITe average female worker earns about 30 percent less than the average male. Tne main results are: 1. Provincial variaion in key labor market variables is explained almost endrey by the extent of urbaniztion of the province. 2. Sample selectivity bias is importan for married and single women and married men. The inteipretaion of the samle selectvity coefficient in the earnings regressions is, h There sk DLrcrbinbatm i aCk? 141 however, different for these three groups, since this depends upon the reason for not participating in work. For married, cchabitating and widowed women, the reason for non-participation is household work, for single and separ-ted women it is both household work and schooling, and for mrried men it is unemployment or (premature) retirement. 3. Most of the eanings differental is due to lower retums to the observed componenAs of human capial of women, esptqcally schooling. The rate of return to schooling is about 2 to 4 percent lower than that of males. 4. These reslts can be intepreted to mean that the Chilean labor market discriminates againt women Tbe occupational and indusral composition of female and male human capital and earnip suggests, however, that the returns to schooling are an increasing function of geneal and job-specific work experience. Since work histories of females are likely to be different from those of men, the results can also be explained in terms of intrinsic diffeence between female and male human capital. This study suggests that in studying labor market phenomena, especially for women, it would be more fruitfdl to concenae on the collection and analys of more refined work experience data, than on the issue of sampling bias of observed work histories. 142 Wcu's D*4ua l t ad y h Ztis A*ua App4di Tabb SAl Ream for 1 It Woddrkig Fiunlal by Marita Stata ad Age (pecoent) Age Group (Yeas) 14-20 21-30 31.40 41.50 51-6 All Married CdbmMdzo & Widowed Woum 1. LookingforFiztJob 1.04 0.34 0.03 0.07 0.03 0.16 2. Uneaployed 0.69 1.33 1.70 083 0.38 1.06 3. Houshold Wok 94.10 95.33 94.69 90.93 71.46 88.38 4. Studying 3.82 1.14 0.14 0.23 0.05 0.55 S. RItirfd (Puaonm, dge) 0.00 0.31 1.53 S.SS 23.36 7.36 6. Reader 0.00 0.02 0.06 0.37 013 0.13 7. Unable to Work 0.00 0.43 0.6R 1.26 4.12 1.56 8. Temprary mdiv 0.17 0.05 0.28 0.4C 0.16 0.21 9. Otbr Ressons 0.17 0.05 0.28 0.40 0.16 0.21 Totl Obtevations 576 4,138 3,521 3,010 3,711 14,9S6 Single Sqwated Wonm 1. Looking for Pirl Job 3.19 8L1S 0.51 0.00 0.00 3.91 2. Unemloyed 1.60 11.66 15.62 6.20 3.42 5.17 3. Housebold Work 20.41 55.20 69.27 66.67 49.05 35.24 4. Studying 73.26 17.5S 0.00 0.00 0.00 48.3S S. Rered (Pesiocsc.) 0.05 0.57 4.24 13.18 31.18 2.69 6. Raitiar 0.W0 0.04 0.17 0.26 L14 0.09 7. Unable to Work 1.17 4.78 7.98 10.59 13.50 3.52 8. Temporaiy Inctiv 0.14 1.32 1.70 2.84 1.S2 0.70 9. Other Rtaso 0.17 0.70 0.51 0.26 0.19 0.33 Total Obsevan 5,733 2,281 S89 387 526 9,516 1 71iser Sex D lwxuar i n. 143 Appeadx Ti SAL1 (coat wd) P.Reasos w Not Work Male, by Maia S& tus and Ap Omani) Age Group (Years) 14.20 21-30 31-40 41-SO 51-65 All Tied, Cobs & Widowed W 1. Looking for Fin Job 5.71 2.99 0.S7 0.00 0.00 0.60 2. Unemployed 20.00 55.97 62.29 37.23 13.24 30.35 3. Household Work S.70 6.47 7.14 4.11 2.04 3.78 4. Stdying 45.71 10.2D 0.29 0.00 0.00 2.17 S. Retird (Peasw, dc.) 5.71 249 9.43 396 68.73 45.00 6. Reatier 0.00 0.00 O.S7 0.00 0.49 0.34 7. Unabl to Wofk 2.86 5.97 6.29 10.82 115S 9.78 8. Temporrily Inative 8.51 12.94 12.29 8.44 2.82 6.63 9. Other Resons 5.71 2.99 1.14 0.43 1.13 1.3S To'AlObseratioa 35 402 350 462 1,420 2,669 Sigle & Spate Mec 1. Iooldng for Funi b 7.80 13.56 1.50 0.00 0.00 8.49 2. Unemployed 3.59 27.33 39.50 25.38 16.41 10.35 3. Ho=ehold Work 1.20 2.03 3.S0 5.38 3.G8 157 4. Studig 80.2 31.59 1.50 0.00 051 64.15 S. Retird (Pa cm, dc.) 0.02 1.31 7.50 21.54 4S.64 2.12 6. Reatis 0.00 0.00 0WA 1.54 1.03 0.06 7. Umble to Work 2.00 11.14 35.00 3S.38 26.67 6.12 8. Temporaily Ina1e 0.62 3.60 5.00 8.46 3.59 1.60 9. Other Reaom 454 9.44 6.50 2.31 3.08 5.55 Total Obserao S,1S7 1,526 200 130 195 7,208 144 Wear's Eppxent md Pay ib Ladx A,aLc Ap4awir Tabl SA.2 Work Participaton Prei Eimate. By Marital Stb DepindentVai&.: Work? (Yes-I, No-0) FewAles Male MarrId WVickowed Sing & Mried, Widowed Single & & Cobabitating Seprtd & Cah _ting Sepwted Schooling: 8-11 Yeas 0.4224 -0.11897 0.00808 0.20683 Schooling: 12 Year 0.368U- 0.16961- 0.1406Z" 0.01207 Schooling: 13-15 Yoe 0.79838 -0.10329- 0.14190 0.49657 Scbooling: 16+ Yoer 1.442S9' 0.45249 0.42136" -0.22878- Agp 21-30 Years 0.33964- l.lS493- 0.44959 1.37922- Age: 31-40 Year. O.S7143- 1.481S0 O.S9082' 1.68564!n Age: 41-50 Yeoa 0.44493- 1.22799' 0.31934 1.26741- Age: S1-6S Year -0.02334 0.53017- -0.46811 0.871864 Houshold Head 0.51807- 0.262&r 0.3631S 0.24247 Household Income -7.40.7& 1.53o-6 1.60.6 *3.01o6 I Household Membes 0.02613 -0.08746 0.00976 -0.0645f U Housebold WorkesD 0.0269r 0.1083S 0.02105 Q22930' Boys: 0-5 Yrs -0.1773g9 0.10984- 0.03208 0.0288 Girs: 0-5 Yrs 4.17139- 0.17180- 0.02820 004274 Boys: 6-13 Yr. 0.07520- 0.07610 0.0387S 0.02510 Girls: 6-13 Yrs 40.05S93- 0.11423'- 0.02089 001749 Rural Dummy 4.44907i -0.29074- 0.30Q21- 0.S6S17 Constat -1.28960- -1.06S93 0.39812n -Q7S933- Log Likelihood -892 -7030.7 -6836.9 -7169.4 Chi-Squuo 2376.4 36128 170S.S 4701.3 Sample size 18,844 13,921 17,063 13,824 Mea of Vaables Pazicipuzo Raw 0.2270 0.3309 0.8333 0.4S27 Schooling 8.2)36 9.S988 8.8207 9.3343 AV 39.9215 255275 41.2932 23.3692 Household Head 0.0989 0.1125 0.8667 0.0602 I Household Membegs 4.8804 S.4678 4.9116 5-5600 # Houshold Wodker. 1.3132 1.4105 0.7449 1.4844 # Childrui O-S Yr. 0.7219 0.6004 0.7482 0.392D I Chiidnm: 6-14 Yr. 0.6413 0.5812 %;. '608 0.5369 Runr Dummy 0.164 0.1406 O.lo.'5 0.2002 Indicat 4pihcance at 1O paIe"t lsvd. - Indiate significoan at 5 p net lee. - Indicatc agnificance at I p a LveL k TAMe Sgg Dbffk*XbdM hN OA? iC - ~ ~~~ TU L FmI. Nwm RWsu6=c by Msrib &Am Dw.sh Vadbi Log Q%my Jasm) mP Re Vald 8si (LI) (14) (7 (1 (1) (18) Schooing 0.1291 0.1024 0.0942 0.1149 0.0SO 0.909 (45.84) (3.16) (19.43) (41.05). (35.06) (25.62) AgScbod 0.0161 0.0112 0.0015 0.0282 .0130 .0.000 (S.63) (3.95) (.380 (11-" (4.38 (4 w08 (Ag"dbol.) *. ,.0002 4.0001 000 4004 4.0001 '1.00 (-3.56) (-1.6Z (163) t 7.1 (-1^n (J.93 Tamno 0.0564 0.0463 (13.13) (ILCS) Tanrv,- 4.0014 0D0011 (-490 (4.28) Ruml Dummy -0.0394 0.0613 015OS *Q1036 -0.0207 .0882 (-239) (1.59) (2.3 (-2.91) (457) (L71) Luabda (X) 420=2 4.1658 41137 4XV* (-13."5) (4.61) (4.91) (-9.14) Cousua L1103 9.0013 8916 79659 3.4284 L6360 (149.02) (106.11) (75.10) (191.M (127.01) (115.1) F-Sas& 810.8 706.32 fl2.55 489.14 413.06 26843 Aomd R2 0.3633 038 0.4014 0.2717 0.5 023 Sam&l Sim 5,768 5, 3,37 5.234 5,234 4,426 Mom of VYAabim Log (Piuy Ina.) 9S496 9.7 9.3757 9.4291 Schooding 9.05Q 10.227 9.7254 10.0739 Ago- S dff 27.S991 22.6173 16.3219 14-M1 Tamo. 7.1827 4251 RIw 0.0665 0.0638 0.0860 0.0628 Lmbda %WogwWon_wogw) .80S4 0.7S52 0.6038 05956 Noalm i-iaiks ni puresum. Pkiq kam= a moly hasm Av omh job. Ma linns Rapm, by Ma3t St*= Dyumh Vadabh Lo# (Pun.,7 hoc.) Mwrie Widcd Sile & Cc6itiag & Sqem&*d (19) 20 (21) (22) (23) (24) ISsAwling 0.13 0.W67 0.1283 0.i1S" 0.1161 0.1166 (82.40) (64.16) C73.8) ('5.0 (44.51) (40.73) Age-School-6 0.0436 0.0334 0.0174 '3.0350 0.0377 0.034 (23.48) (17.71) (8.23) '40 (13.06) (11.9 (A,1Scb,-C) 4.0005 -0.0C01 0.0001 ).0004 40.000 4.00&5 (-15.66) (.2.23) (2.73) (I.21) (-872) (-8S1) Taut. 0.035 O.C113 (1.73) (3.03) Tawzuut 4.0005 -0.0002 (-1057) (-1.76) Rura Dummy .0.1567 4X310C 4O.24 4.0231 .00112 Q0.0lS (.9.15) (-16.97) (-1S.87) (-1.) (05) (-5 Lu, Q)404 057 0.0174 0.0278 (-21.8S) (-21.23) (1.62) (2.47) 8.0727 53634 4987 8.076 Lo0147 7.91 (249.75) t2423 (7.92) (213 ( 17) (139.00) F-Stitic 2268.77 1964.6S 1490.19 63S.94 59.39 34S.42 Adjusd ;;! 0.3604 03789 0.41?0 0.2643 0.2645 02704 Sample Si= 15,096 16,096 14,574. 7,070 7,070 6.506 Mam Of VaiaMlu Log (rizzy n ;ms) 9.26 9.9804 9.4144 9.441 Schooling 8S660 9.0152 S. &9661 Ag-SchoolingS 26.438 24.8246 13.0072 12.3514 Tam,. 8.4187 3.3192 Rwl 0.1647 0.G638 0.2672 02159 Lambda (Worbn/Noawc;wk-) .691', 0.6388 0.4922 0.4828 Not.: t-etai in pumb.a. P-.w:Amfq -.'I immthy ro -. fivm n.ad job. Referenres Bouldin;, K. 'Towaid a Theory of Discrimination.' in P. Wallace (ed.). Equal Empkymen Oppotuniiy and the AT&T Case. Cambridge, Massachusetts: The MIT Press, 1976. Cornwell, C. and F. RLpet. 'Unobservable Individual Effects, Marriage and the EarniDgs of Young Men.' Working Paper. State University 'f New York at Buffalo, 1990. GUi, I.S. and S.S. Bhalla 'Incame Growth and Improvement in Living Standards. Mimeograph. Washington, D.C.: World Bank, 1f90. Heckman, J.J. 'Sample Selection Bias as a Specification Error.' Economarica, Vol. 47 (1979). pp. 153-161. Killinigsworth, M.R. and JJ. Heckman. 'Female Labor Supply: A Survey' in 0. Ashenfelte and K. Layard (eds.). Handbook of Labor Economics. Vol. 1. New York: Elsevier Science Publishers, 1986. Mincer, J. Schooling, E#prlnce and Earnings. New York.: Columbia University Press, 1974. Oaxaca, R. 'Male-female Wage Differential' 'Trban Labor Markets.' InrnatoonalEcononic Review. Vol. 14, no. 1 (1973). pp. 6', 1. Stelcner, M.J., B. Smith, J. Breslaw and G. Monette. 'Labor Force Behavior and Earnings of Brazilian Wonmen ard Men, 1980." This volume, 1992. United Nations Development Programme. H=tan Developneo Report. New York- Oxford University Press, 1990. Willis, R.J. and S. Rosen. 'Educadon and Self-selection.' Journal of PolJdcal Economy (supplement). Vol. 87 (1979). pp. S7-S36. World Bank. World Dewl ema Report. Washington, D.C.: Oxford University Press, 1990. 147 6 Labor Markets, the Wage Gap and Gender Discrimination: The Case of Colombia Jabne Te,Jo' 1. Introduction Earnings differentials between men and women have bcen dunumented in a large nmber of studies both for developed and developing countries. However, unlike the case of more advanced countries2, very little work has been done in developing countries to ex4lain why men are systematically paid more than women. In many earnings finctions, for eAample, the inroduction of a dummy variable for sex has produced statistically significant coefficients that indicate sizable differences in pay between sen and womena but very litle effort has been made to investigate more thoroughly the source of this differential'. Rather, researchers have been more concerned with explanations of differentials between regions (rural-urban), between sectors (moder-radi- tional or formal-irformal) and between industries (agricuta&re versus manufacture, etc). As explained below, gender wage differenals can be associated with a numbe of factors mtcii as differences in productvity, working conditions, discrimination, etc. This paper is an attempt to identify the degree to which these and other factors conbue to the explanation of the wage gap between men and women in Colombia. The structure of the paper is the following: Section 2 presents a discLssion of the reasons for the existence of wage diffaentials in the labor m3rket and the meaning of various forms of The author wifh to wknowledge fte useful commnts and vwpport received from A. Beny, P. Bowla, J. Newtc,n and to paricipants in the Developmt Workshop of the University of Toronto. The enoi remaining ar3 iine alon. 2 A smmay ef the work doa in developed countries in this ar can be found in Guntdeo (1989). 3 For a summary of txe mod important findings of this type of alysi in the cas of Developing Countries swe Fields (1980). 4 Importt exceOO ar the ooCtbUtion by Chapma snd Harding (1985), by Gannicott (1986), and SchuUz (1989). discrimination. Section 3 intoduces a model to measure the eaning gap between men and women and its compositon Sections 4 and 5 preent the results of the estimation of the Lodel with Colombian data. In Section 6 a comparison with other country studies is made. Secon 7 investigates some aspects of women's access to high ^-ylng occupations. Section 8 summarizes and concludes. 2. Wage Differendals and Discrimination As explained below, two types of discrimination can be identified: market discrimination and non- market discrimination. Both manifest themselves in wage differentials but are generated in very different ways. A brief discussion of these two forms of discrimination and the way in which they are generated is important at this point. Compidve maW . It is difficult to explain discrimination in the labor market under competitive conditions. In competitive markets wage diffenes are either temporary or Icompnsating differendals' that reflect differences in working conditions, workens' charaterisdcs and preferes, or human capital endowmen. If women prefer, say, safer jobs, or have lesser amounts of human capital dt men, they will, of course, earn lower wages. However, in this case there is no dicriminadon In the labor market but compensations for differences in productvity or working conditions'. However, discrimination can exist outside the maket and the result of it be reflected by the market in the form of wage diffeala in favor of men For example, women could come to the market with smaller human capital thdwm than men because they do not have equal access to the educationa system. Or it could be that the double role of being home-makers and workers make women's invement in education less profitable than men's. Women can also be vicdms of social praices and prejudic that crowd them Into 'feminiDe- occupations (maids, secretaries, teacbers, etc). Wages in these occupaons would be lower than if women could compete with men in other occupations. Whatever the case, the competitive market reflects discriminatory practices in the society, but is not the source of them. Discrimination takes place outside of the market sphere. Non-compecrve mane. When non-competitive elements are recognized, the possibility of discrimination in the market arises. lhe following are som exaamples of market generated wage discrimination: I. Market ion (formal-informal, modn-trditional, etc.) can occur along sex lines. In this case wage diffeentals between segments of the market coincides with sex differeaials. D Becker's (19S7) thoy of dimcimination aflowa foc the pcuaiblity of discrminatio in compeative conditioms by introducing eo asuipton tLqt podce wauxim their utility wh;ch is a function of profitb and the sx composiiu of his labor fc. tn this aswupmo te wag diferential against wom is the amount necsmy to comparto tdo enloyw (or gp of infloe*AIa workes) for the disulity produced by hirng oe addowt l wm, rar than a u. a In most developing oomri whool eaz met d atios san the levoe of educationl atmment am lower for womn dt for m Ete for exame Nqjafindeh and Maerck (1988). 2. Non-competitive firm may pay wages above the market wa&3 rate in order to lower labor turnover and have a continuous queue of workers available fbr work. Ihe rationing of these (well paying) jobs can be done on the basis of sex (favoring men over women). 3. Discrimination can be generated by pa2riarchal auitudes of non-competitive employers who can decide to pay higher wages to male employees because they adhere to the belief that men have more economic responsibiities than women, or that women are secondary earners. 4. Information problems may be responsible for discriminadon against women. For example, job evaluaon procedures can be gender biased or rely heavily on the subjective opinions of male supervisors. S. Systemic discrimination against women can exist in the form of gender biased job requirements (height or physical stegth7) or access to the necessary networks tha facilitate the enurance into high paying jobs in the economy. On the basis of the arguments above, two forms of discrimination can be identified: Market and non-market discrimination. Market discrimination exists when there is a systematic wage differential that cannot be explained In terms of oDmpensafing differtials. Non-market discrimination exist when social practices result in women entering the market in conditions of disadvantage relative to men. In both caes discrimination manifests itself as wage differentials but in the case of non-market dicimination the market only reflects the diriminatory treatment of women in other areas of the society. In other words, market discrimination produces a situation in which men and women are remunerated according to different rules while non-market discrimination is more likely to manifest itself in the form of different 'productivity factors' (endowments) between sexes. 3. A Model to Measure the Size and Composition of the Wage Differentias The methodology applied here !6 well established in the literature8. The departure point is the earnings function of the human capital theory. According to this theory the wage rate (W) of a person i can be expressed in the folowing way: ln(Wi) = XiB+Ei (1) where Xi is a vector of explanatory human capital variables represeniDng relevant characteiscs of individual i, as well as a set of dummy variables that reflect differences in working conditions and characteristcs of occupations; B is a vector of associated parameters and Ei is the error term which is assumed to have mean zero and constant variance across the population. The model can 7 In some cu thes requireami meflect legiote needs of the job. However, in the modern technological world pur physc a*lmt have becm low important than attrbute that can be acquired thuough raining. ITh retno of hee requirements can sve th purpose of limidn the number of job applicants in oder to reduco tho comt of screinS and hiring. The prblem is that these procedures, intentonally or not, moy ietrict the access of women to cm-min occupations. Se Gunderson, op ciL be expanded with the incusion of other variables (ntiuions for ex aple) to make it genal enough to test the importmwe of different theoredcal explanatons of wage rafts. As Is well own, e.estimtion of e ion I by on inary least squares (OLS) is likely to produce biased parameter esdmaes due t the fact that the samples are not random9. This problem can be solved by addng to equation I the inverse of the Mill's ratio, which is a monotone decreasing function of the probabilty that an observation is selected inDo the sample". Equatlon 1 then becomes Ln(WI) = XiB + bLi + Ui (2) where Li is the Mill's ratio and U the error m (assume to have zero mean and constant variance). One way of proceeding is to sep t men and women and define dfferent wage equations for each group. In this case on has: ln(Wmi) = XmiBm + bmLlI + Umi i = l,...,Nm (3a) ln(Wo = X jBf +bf + U j 1,...,Nf (3b) where the subscripts m and f refer to males and females respcv ely. By doing the necesary transformadons, the *corrected' average wage differential between men and women can be expressed in the following way: In(Wmz)n(WO4)bmLm-bfJ) = (X-X)Bm + Xf(Bm-Bf) (4) where ln(Wm) and ln(W) are the predicted average wage rates; and X 'Xf LM4 and If ar mean values of the respectivo variables. Equation 4 decomposes the wage diff al between men and women in two ways: Ile fir term on the right measures the differential due to differences in the amount of human capital and other variables (endowmets), whie the seoond term measures the differendal due to the application of differe remuneration rles (market discrimination). In a competiive market the latter term should be equa to zero and the toDl diffe ial would reflact only diffences in the amount of human capital between men and women or compensating differen between occupatons. This difference in h_ , bowever, can be the result of non-market dicrimi- nadon. For example, women can have smaller human capital endowments becuse somehow their acess to the educational system Is resicted. f maret disination aginst women exists, then Xf(Bm-Bf) will be positive". 9 See for ample Groom (1974). 'o Se Hackman (197. "I Th. implicit aawptica h1e is dat if no -ket d imincrim existd wo wud be pmid accorSding to the am uo thot apply to m (namely Bm). It is poub4o to arge that this I _eutmat the diciminatim comanet beco t limo y the pa_mtr of a situton (call tbem Be) au bewe Bm mod BŁ.E In s con the model in the tet hu to be modified lighly but the emow of ft ueysis rmois to wage sp a.Y tLia lh,na,r Ofvmwao 7he Case of Colmwbia 153 4. Wage Equations and Sex Differentials In Colombia Tne methodology presented above requires two steps: First the estimation of separate wage aquations for men and women, and second, the decomposition of the total differential along the lines of equation 4. This part of the paper deals with the first step while the second step is discussed in Section 5. The data used in the estimation is a household sample for BogotA (Colombia) collected by the Colombian Department of Statisdcs (DANE) in December 1979. The sample includes both labor force participants and non-participants. For the estimation of the wage equations only the sub- sample of wage earners was used, but the selection bias correction is done on the basis of all the sample. Table 6.1 presents the sample gross earnings per hour differentials between men and women by occupations awJ types of jobs. According to this table men earn on average 32.7 percent (28 percent in geometic terms) more than women for each hour of work. The differential ranges from 3.23 percent in favor of women in transportation occupations, to 127.8 percent in favor of men among supervisors. In terms of the types of jobs, the highest differendal is among white collar workers (32.2 percent in favor of men) while the smallest one is among domestic servants (4.9 percent in favor of women). The special case of domestic servants is analyzed in Section 7. Tbis is a very pardcular group composed almost exclusively of women (only 2 men versus 349 women in the sample are classified as domestc servants). Serious problems with the quality of the data for this particular group exist For the estimation of the model in equations 3a and 3b the following variables were dsed: 1. Human Capital Variables EDUCATON = Years of Schooling EXPERIENCE = Number of consecutive years in the same economic sector. A quadratic form of this variable was used to allow for the pos- sibility of decreasing returns to experience. SENIORITY. = Number of consecutive years in the same firm. TRAINING = I if the worker received training in the firm and 0 otherwise. 2. Occupations: a set of dummy variables equal to 1 if the worker is in the respective occupation and 0 otherwise. The occupations are DIRECTOR, SUPERVISOR, PROFESSIONAL, CLERICAL, SALES PERSONAL, and (personal) SERVICE OCCUPATIONS. 3. Other Variables: Another set of dummy variables that are equal to I if the worker has the characteristic descnrbed. UNION = There is a union in the firm. MARRJZ = T-Lhe worker is married or lives in a common law situation. R-MIf6R = Tle worker has lived in Bogota for less than three years. nS The geometric mean redce the weigt of the extreme positive values, thus ptoducing a lower mean than the arithmetic aveage. The geometric differntl is the diffence between the logarithm of the gometric mes ad is more comparble with te resuls of the model in sections 3, 4 and S than the sritumetic moan. Table 6l Avemo4 Ibnom pe EHour by Occupaton and Sex Bhoo White Domestic Ohet Mean Geonmetic COBlar Colar Sevnnt (Anithm) Mean DIRECTORS m- 89.28 89.28 71.90 women 76.28 76.28 61.39 % diffnsue. -17.04 -17.04 -15.80 SU'ERVISORS mm 45.39 68.49 59.80 40.41 women 17.52 29.7S 26.2S 32.22 % differce -159.08 -130.22 -127.81 -55.41 PROFESSIONALS mmn 80.42 107.60 15S.98 190.03 79.73 wornm 77.20 755.81 81.05 59.96 % difference -39.38 79.36 -34.52 -28.50 CLERICAL WORIS mmal 14.70 37.53 37.46 24.65 wnorn 38.70 38.70 30.69 % difference 3.02 3.20 1.26 SALES PERSONNEL mm 37.09 36.73 24.65 woen= 11.21 19.44 19.08 15.99 % difference -90.79 -92S1 -43.29 SERVICE WORCERS DlltXI 19.20 11.76 19.13 1S.99 women 18.24 17.82 12.36 14.51 11.75 * difl -7.74 4.85 -31.84 -30.84 MACHINE OPERATORS M-A 22.78 96.56 16.36 25.22 20.40 women 18.64 17.84 18.61 15.64 % differaece -22.21 -441.26 -35.52 -26.56 CONST. OPERATORS men 19.74 13.47 19.70 16.73 womn 12.27 12.27 11.89 16 diffeco .60.88 -60.55 -34.13 TRANSPORT WORKERS mem 18.19 22.93 21.59 19.15 wome 19.30 43.40 22.31 18.23 % differene 5.7S 47.17 3.23 -4.90 OTHER OCCUPATIONS men 293 37.57 26.38 2357 women 18.70 1S.79 12.36 17.SS 16.13 % difference -22.62 -137.94 -50.31 -37.89 TOTAL mml 22.92 51.17 11.75 119.72 39.73 26.3S women 18.60 38.70 12.36 160.14 29.94 19.92 U difference -23.23 -32.22 4.94 25.24 -32.70 -27.95 t_ _ -onx-nf 4. LAMBDA = The iverse of the Mill's ratio. As indicated above, this variable is introduced to correct the selecton bias problem. The details of the esfimation of this variable are in the Appendix. In dii mo-d the parAmeters of the variables in (I) can be interpreted as the retu to various forms of measmred human capital, the parameters in (2) measure a combination of compensating differentals and the effect of some (possible) barriers to entry in some occupations. Finally the parameters of variables in (3), except Lambda, measure the (prcntage) wage differendal atributable to those variables. Separate equations were estimated for men, women including servants, and women excluding servants' and the results are presented in Table 6.2A and Table 6.2B. For comparison purposes the (selection bias) corrected and uncorrected estmates are included. The statistics are in general good: Tle adjusted R-squared are above 0.5 (high for cross-ction analysis), and most of the parameters are significantly different from zero. Even the parmeters with low t-values are interesting because they shed light onr important aspects of the fiunconing of the labor mark -s. It is interesting to note that the t-values of Lambda are below the 5 percent significance level fhr men but not in the case of women, indicating that there is not a serious selection bias problem in the sample of women. As explained in Section 5, the conclusions about the compositic a of the wage differential betweea men and women are not very sensitive to this type of correction. In spite of the low t-values of Lambda in the case of woime, the effect of the correction is the ex- pected one: It decreases the values of the coefficients associaed with the human capital variables such as Education. The general results of the estimation are consiswit with those obtained in a large number of other stuW&es in developing countries". The returns to education are slighdy lower than 10 percent, which is the value usually obtained, but this is not unreasonable given the large number of variables included in the model, not avaiable in other cases. What is more important for the purposes of this paper is that there is a clear difference in the estimated coefficients (corrected and uncorrected) of men and women and that the esdmates for women are very sensitive to the inclusion or exclusion of domestic servants. Tbe exclusion of domestic servants makes the es- timates for women closer to those for men. In general men have higher returns to education than women and the quadratxe; form of experience indicates that the returns to additional years of experience decline faster for woram than for men. Women receive higher return to ses .ay and on-the-job training. Some occupation premiuxs are higher fior men than for women. Such is the case of Directors and professional occupations. Unions bnefit both men and women but the exclusion of servnts decreases the contrbution of this variable. The premium to variables like recent migration (R- migr) and service occupations are not significandy different from zero. 13 Ther ar two rsons to excude domestic svantse One is that tir mesumed wage rat is lm rliable than in the can of other wxkers, because a prportion of this p at is in kind (food and hlter). Two, that domeAic sean constitute a vay particular group of individuals (young, neducated, female, immign) that ae not omparable with the rb of the wome in the sample e caew of domnstic servants will be analyzed in Soction 7. 14 Good evies of thd tudio can be found in Fields (1980) and PacharpoWo. (1981). 5S Size and Compositio of the Wage Gap lbe average wage gap and its composition, as expressed ia equation 4 can be computed directly from the information given In Table 6.2A. Endowments are measured by the mean values of the variables in the model and discrimination by the difference in the coefficients. A summary of these results is presented in Table 6.3. The most obvious imnplicadon of these results is that a large part of the wage differential between men and women is due to the exwnce of domestic servants and to the fact that they are particularly poorly endowed: When domestic servants are excluded, not only does the total differential drop by about 64 percent, but the endowment component also is significantly reduced (by about 90 percent). Contribution of hwan ay$ variabes. It is Important to notice that when endowments are measured by the mean values of the variables in the model (Table 6.2A), women (excluding servants) are not at a great disadvantage relative to men: On average they havt higher levels of education than men but smaller amounts of other human capital variables. On balance this results in a small negative contribution of human capital endowments to the gross differential. The greatest disadvantage for women In terms of their humar. capital eudowments is their short amount of expezience, a result of the double role that they play as homemakers and workers. As indicated above, tbis can be a result of non-market discrimination, a reflection of individual pre- ferences, or an (endogeows) response to the existence of market discrimination. The lack of better data makes it difficult to carry the analysis any further. The effect of correcting the selection bias is to increase the total wage differential and the market discrimination component, leaving the endowment component almost unchanged. Cerainly a closer look at the factors that produce these results is of great intest. Table 6.4 presents the contribution of each variable in the model to the total differential in hourly earnings estimated on the basis of the corrected coefficientsu. The rest of this section considers only the results of the estimations excluding domesdc servants oast three columns of Table 6.4). Tle particular case of domestic servants b analyzed below under 'Some International Comparisons.' In contrast with the above results, the contribution of human capital variables to the discrimination component is shockingly large: Tle fact that the returns to women's education are lower than those of men is the single largest discrimina ory element found in the analysis and has a magnitude larger than the total wage differential between the two groups (10.36 percent versus 9.85 percent). It is important to notice, bowever, that given the kind of data used here (the only ones available), not all the measured discrimination component of education can be attributed to actual market discrimination. It is possible that at least one part of it reflects differences in productivity created by differences in the quality and typo of education that women *aceive. If this is the case, the actual market discrimination component is overestimated by the results in Table 6.4, and the non- market discrinination elenrent is underestimated. s Mmhe exerc with the uconcd coefficients was also done eA the rmlts do not chnp dramaiically. Table 12.2A Corrccd and Unoeorrcted Regr uion Esimates of Wage per Hour Equatio. M E N W M EN WOMIM EXECLUD10 SUVANYS CORRECTED C0aRECD CORRBC'D MEAN COEFF COEFF MEAN COEFF COEFP MEAN COEPF COEFF Imago 2. 2.2922 2.167S 2.2712 2.1355 2.2164 (59.1) (45.0) (41.4) (24.7) (41.5) (30-9) Bdoeas 7.47 0.0S34 0.0613 7.141 0.0711 0.0661 8.161 0.071t 0.0686 (19.7) (193) (12.0) (9.59) (12.0) (11.0) Dq.dm. 7.038 0.0246 0.0226 5.557 0.0221 0.0216 5.412 0.0413 0.0402 0.32) 0.33) (4.21) (4.09) (6.19) (6.00) ExB m' 115.901 4.76B-4 -3.6%54 76.606 -4.9244 4-.065,4 6739 *0.0011 *0.0011 (4.21) (.2) (3.36) (3.25) (4.3 (4.72) Jait 4m72 0.0107 0.0111 3.S15 0.0148 0.0CW 3.975 0.0138 0.0143 (4.00) (4.17) C97 (3. (3.48) (3.53) Ta 0.22U 0.0940 0.0926 0.141 0.1473 0.1397 0.179 0.1390 0.1295 (3.13) (3.1) (3.33) (3.11) (335) (3.06) Dess Q011 0.628 0.6400 0.016 0.935 0.9"17 0.006 1.0112 1.0075 (5.29) (5.37) (3.49) (3a.47) (3.33) (3. snlaa, 0.045 0.342n 03540 0.028 0.177 056 0.036 0.1914 0.2115 0-44) (55) (1-99) (l16) (217) (234) k1skmal 0.128 03679 0.574 0.122 0.6916 0.69t3 0.154 0.6773 0A900 (10.9) (10.9) (936) (937) (6) (a.73) a cd 0.163 0.1317 0.1475 0.239 03054 07 0302 02942 0.300 3.55) (3.32) (5.74) 5.73) (5.79) (S36) SaImpnee 0.076 0.0910 0.0615 0.071 .0.0466 -0.0540 0.096 4.0485 0.0553 (I.3)JI (1.71)sI D.73&/ (0.83)W (0.81)bf (0.91)w 8mies Wok 0.090 .02740 .0.275S 0.35S 4.1206 4.1257 0.1U 0.0454 4.06 (6.12) (6.25) (a46) (2.75) (0.94)W (0.75V U" 0.93 0.1306 0.1244 0.229 0.1233 0.1348 0.2 0.093 0.1074 (4.45) (426) (3.-6) (.41) (270) (2.9 UffIlad 0.627 02209 0.1875 0o31 0.0798 0.0926 0.39 0a0356 0.0609 (7.65) (6.13) (2.39) 0) (1.I0)w (1.62)1 R.~qr 0.109 4.0214 .0.0237 0.169 4.0454 -0.0470 0.077 4.0473 4.0461 (DSft (O-W (1.07) (1.09)b (0.8 I!/ (0.78 La1da 0.498 4.1426 0.S31 4.0652 0.76 .0.092 (2.73) (131)?y (1.49)& R' (A4usad) 0J325 0.5316 0348 03.417 0.5411 0.5339 N. Omn t2060 2,060 1.454 1,454 1,151 1,151 Rpm In Bgckio mul .vs a. BiaiSmelwd bg_ s 1 &05 S. 8 0 #awWvaIabmI0% - -'-.-- Tabl 6.2B Ramdb of dho Estnti of a SimPle Enm Capita Model (D edt vrie - loganb of houly we) Wome Men Women Ezcluding Servnt Comtd Crd Coef C Cooff oeff Cooff CasE! Coeff Intercept 2.0734 1.9445 L.8653 1.8514 1.9317 1.9523 (62.3) (43.S) (S2.6) (49.6) (45.9) (39.8) Edw-aution 0.1294 0.1263 0.1338 0.1318 0.123S 0.1237 (40.6) (39.6) (3S.3) (31.0) (29-8) (27.9) Expwiec 0.1469 0.0432 0.0444 0.0440 0.0630 0.0662 (12.8) (11.7) (9.36) (9.20) (10.4) (10.1) Expeftien2 -0.0008 4.0006 .0.0009 -0.0009 .0.0014 .0.0014 (7.62) (6.23) (5.83) (S.74) (6.O ) (6.09) Lambda 0.2422 0.0430 .0.0423 (4.76) (0.78) (0.5S) R2 (adjusted) 0.453 0.4SS 0.479 0.473 0.476 0.471 N. Obeerv. 2,144 2,144 1,489 1,489 1,177 1,177 Figurs In bmikeu ae gt-zatio Contwiudon of occupaons. In this cae dowments me;sures dr; effect of the distriution of men and women across dffeet occupations. The mea values in Table 6.2 indicate dut close to two-thirds of the women are concentated in hee occupations: Claical, Seice occpations, and Professlonals, awd in all tareu of these occpadons women an over-represented. The regression results irdicate tht clerical an professioal occupations receive a premium relative to ote occupations, but servi are penaized. This indicaes that high particiation in savice occupatdons is a disadvamage. If women were paid acwrding to the same rules that apply to men, the advantages that women derive from having higher participation than mn inl derical and professional occupations is offis by thdir higher patcpation in service occupations. This explains why the contbtion of *endowmenws to die total differential so small: Two-tenths of one percent in favor of womeo. lhe discrimiam-on ceponent of ti grop of variables is large and im favor of women (negative sign) IndicaOng that, on averzge, women receive lager percentage premiums than men in the sam occupations. The magniude of this compone is slightly smaler than the total wage differental between men and women which means that if occupatonal premiums were the same for men and women ie total wage diffental would roughly double. The large size of dte component Is due mainly to dte difference in pmum in the three occupations where women conceate. LAaOe maxa, mw wag *V da(knndr DtsrmWwT heCasiofCoba Tahe .3 D _opsiiuof Sam Wor Gap Eedownab Discrimination Tota (Xm-Xf)Bm Xf(Bm-Bf) Uncorreted Eostimatoa Gap Including Servanft 0.2241 O.OS32 0.2774 Composition (80.8%) (19.2%) (100%) Gap Excluding Soranrs 0.0716 0.0241 0.0957 Composition (74.8%) (25.2%) (100%) Corrected Esti a Gap Including Scamnt 0.2038 0.0636 0.2774 Composition (77.1%) (2.9%) (100%) apV Excluding Servant 0.0184 0.0801 0.098S Compostion (18.7%) (81.3%) (100%) m Th correcton rcfurd to heci the sdcion bin cots io estimates are based in resb presne in Table 6.2. Although much more reearch is necessary in this area, this result is consistent with a siuation in which women have (other dtii s equal) higher reseation wages tha men and search until they find the job that pays what they expect or drop out of the marke If successful in finding a job, women are likely to receive higher wages than men in similar crcumstances. This can be important in occupations such as dericsl and service jobs that do not require age amounts of educadon, and wfhere the acul wage is more closely related to the worker's resarvation wage and the amount of experience hai to other form of hman capital. Contribudon of odur vadaMks. The contrbu'ko of these variales (Union, Marital Stats and Migration) is positive both in terms of endowments and disiminaon. The largest effect is that of marital status, indicating tha being muried reresents a higher premium for men than for women, and the proportion of married men is 60 percent higher than the propoation of married women. It should be noted that the variable R-migr is not significantly different from zero in any of the equations. 6. Some Intrtioa Comprso Although the ptupose of this papr is wt to make a complete review of the studies on gender disaimination, it is interesting to compare the results of this paper with those for other countries. Table 6.5 presents a summary of slected country studes. The low mumber of developing coun- tries for which information was fmnd reflects the ck .)f attention that gender wage differentials have received in these are". It slould also be noted Ct the methodologies of the studies presented are too different to make them properly comparable and therefore one should be careful not to read too much into these results. Table C4 Contnibntien of de Vaiae in the Modd to the Wag Difforial hnlding Swrat Excluding Servants EBdowmat Diacriminstm T. ' Endowments Discr-ninaio Tota Education 0 0330 0.108S 0.141S -0.0499 0.1036 00537 Expainoo 0.0187 0.0144 0.0331 0.0186 -0.0458 4.0272 Seiority 0.009S -0.0123 -v.002 0.0044 40.0127 4.060 Training 0.0080 -0.A66 0.0014 0.0045 4.0066 -0.0021 TOTAL HUMAN CAPTAL 0.0693 0.1040 0.1733 -0.0223 0.0386 0.0163 Director 0.0051 4.0010 0.0041 0.0051 4.0011 0.0040 Superviso 0.0050 0.0042 0.0102 0.003?. 0.0051 0.0083 Profesicoals 0.0034 -.0100l 4.0116 -0.C149 -0.0177 -0.0327 Clerial 4.0112 -0.0381 4.0493 4.020M 40.0461 4.0666 Sales Pero 4.002 0.0O06 0.0104 4.0018 0.0134 0.0116 Service 0.0739 4.0537 0.0202 0.0270 4.0449 4.0179 LOTrAL OCCUPATIONS 0.0771 4.0931 -0.0160 4.0019 40.0913 4.0932 Union 0.000 4.0024 0.0056 0.0005 C.O0S 0.005S Married 0.0S7g 0.0302 0.0881 0.0429 0.0504 0.0933 R.migr 0.0014 0.0039 0.0053 4.0008 0.0017 0.000p TOTAL TrHER VAR 0.0673 0.0317 0.O990 0.0426 0.0571 o.0997 Inecepvt 0.0210 0.0210 0.07'8 0.0758 lAmbda 0.047S 4.0477 4.0002 0.03ff 4.0414 -0.0018 TOTAL 0.2613 0.01S9 0.22 0.0580 0.0387 0.0967 Note: FoDowing ,qoo 4, d 'onecrd' we gppinTable6 63 can be obained by sobbacting from doc tadl in tbis dtble vak es Lamb Sourc: Tablo 6.2. One of the most aresftag observations in dbc table is that even if donemdc servants ar inluoded in the compaison, the wap esWdmated for Colombia is one of the smallest by national standards, only t gap in te savice indosty in the Unite Stat is unler. If san are excluded, the gap in ti pqe is by far the smallest The gap for Malaysa Is also lower than most of the difala for developed counuiu aldho sightg bihe dta the Colombian onL Tp' #van presn wag diffeoals and composition closr to dtose found In developed wunuies A simllar thing hap wi the d.crimlnadon compone Apin devloping countries have smaler levels of dicrmio,n dtm developed oncs, whether in absolute or In rdstive lIrmk. lhic part of !he cfarAer deals with two quesins: (1) Is there evidence of the existence of dead- end, low paying occupations ii which women tend to concate ("Female Ghettos')? (2) Do women have equal access to h gh payingjob? The answer to the first question is yes and to the sacond one is no. Ihe cse of dom.nItc serw . 'The best e le of female dead-eni occupations that are poorly ?aid and require long and unregulated hours o. work is domestic sntice. As Tablo 6.6 shows, domestic servants are almost exclusively women, represent absut 18 percent of female employment (23 percent of female wage earners), aDd almost 60 percent of female employment in service occupations (Table 6.2A). Domestic servants outnumber female blue collar workers. Only c erical occupations employ morc women thn the occupation of domestic service. A large proPortiori of domestic servants live witi the fmilies fot wvhich tliy work, which m.ans that part of their riary is in ninthe form of room and board'. Accordingt one of the few sties of domestic servaras done in ColombiaP, !his group not only receives the lowest salaries in the market:' but also has very poor working conditions. Particularly serious is the situation of live- in servants because m prctce they do not have a worling schedule (basically they are on call all the time) &nd frequently becomc rictims of physicA, psychological, ard sexual harassmenL The vast majority of womzo in this occupaiMn ar young (frequently in their early tcens), uneducated (average schooling 3.3 years) rural immigrants who have little iuman capital of any type." For many this is their fis job and the first time they experience life in an urban locWion. These characteristics make domestic servants a special case of fenale occupations th", as explained abe"e, accounts for more dtan 50 percect of the total wage differntal between men and women. '5 Tois cream smious interpretation pr 'blean with the ucome figcre avilabls, siuce an arbitay monetuy value is givn to the in-kind compoomt ot the remunertion of this peasos by the finies that employ &Am. Given Jhat . is tw employer who estimawe the value of the in-kid aLay -wd dec s the number of hours tha sM wtr, dh upicion is dtAt dhe wap of domestic seants is ov-estimd. This is an additional resoa to preset reslts including and excudin L'uiS - of women. " see M&uex (L355). is Th minimm leal wag for dcsbc saevants is lower that of uorkerws and thLy are edte4d to fewer '.eC finge brofits (for mLnWlc averance payments and vaca6ion) than the rea of the ab force. i9 Even the o.=W educational levels at zideading becas in many cases they attded Wchool in nul areas whe the quality of edocution is Epficandy worse than in cities. 8 To wy knowledge, no sud ha ben doe to invesiat whether domesic sevie mcoupabo are permanent cr Iccotry. May people believe that domstic srvce is a temorry ocapacn tha allows young immigra wommn tim o conie th necemry miket kowledge to mo to otbs oc- caupata ThiMs howevur is rc conssut with the fact tht domestic svnts hav higber lev of exrafa thrn other women (6.11 versu 5.41 yerspectively). Unt bettar inhxmatr bKwoo available th best hypotSesis to work with is tha domestc sevice oocupazionam 'ar m or permant- ohm is believed, or thyt thoy ar a nr infcicett way of acquiring markt knowledge Ramlt from Selected Wae Difwf t Studie Study En/iwnxwaf Diacrim Tota USA Msnufict 1970' 0.320 0.389 0.709 (45.0%) (,S.0%) (100.0%) USA Manufict 1981b 0.12) 0.283 0.412 (31 5S (68.9%) (100.0%) UFA Service 1981b 0.100 0.152 0.252 (39.7%) (60.3%) (100.) Canada 1970' C.187 0.323 0.510 (36.7%) (63.3%) (1looD) Doe to: Hui Capital 0.051 0.34S 0.396 Edato (-0.001) (0.242) (0.24) OCcupado.o 3.055 0.044 0.099 Other VatiAbks 0.081 -0.066 0.015 Canads 198M' 0.223 0.409 0.632 (35.3%) (64.7%) (1OOm) Mlaysia 1979' 0.280 0.046 0.326 (85.9%) ;14.1%) (1000%) Taiwxa 1982' 3.177 0.264 0.441 (40.1%) (59.9) (1000%) L See Hod,on and EPnlagd (M96. b. sce Moalgoe2cry and Waseer (198). o. GOundao (1979). d. Educao is pad of the Hun Capia1 vaiables in tho line bcn ThU infbrmo in brcka is p d only fr comparion pwpom. e. Mflle (i957. 7Vwis s the only Mudy that -or-es for selcdvky bias and does it wkh a mashodologyaim to the one used bere. S Chapman and Harding (M9). g. Gannicot (1 ). Tablefc Num.ex of Employed Wordbe in the Sample by Sax and Type of Job Men Womn White Colar Wodkes 1,239 914 Bhue CoLar Wodres 890 254 Unpaid Family Helpes 22 so Einployets/owncsm 204 33 adependant Wodkes 673 346 Domkisc Serants 2 349 Totd 3,030 1,946 The inclvsion of domestic servants in the decomposition of the wage gap has the effect of increasing the toal contribution of endowments in a significant manner (from 0.018 to 0.224, Tphle 6.3) and lowering that of disaimination by less than 2 percentage points. Although these clwages are not unexpected, the small decrease in the discrimination component is intereting frxause it indicates a high degree of stability in this component. Almost all the change in the gross wage gap Is due to changes in the endowment component. The contrnbuon of the human capital variables to the two components of the wage gap (Table 6.4) is also affectd by the indclusion of servants. Two points are worth mentioning: One, an incease occurs in the endowments component of human capiul variables, almost all of it explained by an Increase in the endowments component of education. Two, there is also a large increase in the disimination component of the same group of variables (about 6.5 perCe ta- points), but In this case the explanatory factor is the increase in the discriminaton component of experience (about 6 percentage points). Other occq?a*iow. As mendoned above, clerical occupations zonstiute over 30 percent of the female employment and represents the largest concentration of women in the Bogoti labor market. Tle definition of ths ocapation is very broad, includi. g a large number of occupations such as typist, secretaries, receptionists, etc. As in the case af developed countries these are traditional women's occupation that do not offer many advancement opportnites. Something similar can be said about service occupations. On the other hand, the proportion of women in professional occupations (15.4 percent of the sample excluding srvants) is larger than that of men, indicating a significant presence of career oriented occupations avaiable to women. This is consisent h wi the relative high levels of educational attainmen of women and with the relatively low leves of discriminaion found in this paper. One question that arises is whether women have the same access as men do to the high paying jobs within a particular occupation. An answer to this question can be found by estimatng the probability that a person receives a wage rate above the average of the occupation as a fiuction of a series of relevant variables, including gender. Thi was done by estimating a 'opit equation of the following form: Pr(Wij > WJ) = h(Education, Experience, M-exper, Seniority, Sex) (5) where Pr(..) represents probability, Wij is the houry wage of worker i in occupation j, Wj i the average wage rate in occupation3, and M-exper is market experience measured as Age -Education- S. The results of this estimation sre presented in Tabl 6.7. In geneal they have the expected sign and are dearly significant from zero. The most relevan conclusion for the purposes of this paper is the negative ..l of the coefficient of sex, which indicates that, other things being equal, women have a !vver probability of receiving wages above cie average In their occupation than men do. In other words, there is a sysematic tendency for women to be employed at the bottom of the wage scale in each occupation regardless of their human -capital-endowments. lhis implies that although the access of women to high pay occupadons is not completely blociced, still women find it harder to nove to the top of each occupation. Table 6.7 Logit Esimates of dth Problty of receiving Hourly Wages Above Xt Occupation Avetug Intrcept -1.3831 (11.0) Educadon 0.0998 (9.79) Market Experiene 0.0251 (5.85) Experience 0.0077 (1.20)' Seoionty 0.0725 (8.48) Sex -0.572S (7.85) a. Significwn levcl abow I pe ceat Note: Figures in brackca am t1-aiu 8. Summary and Conchusions The analysis above has to be tken s a first approximation to the undanding of the wage differences between men and women in the labor markets of developing cuntries. The results are interesting and provocative but have to be regarded as tentative until more research can be done. A list of the most imporn findinp is the following: I1. It wa a. 5mnted that the total wage gap between men anrt women is below 30 percent, lo-wer that wbat ha been found in developed countries. If domestic serans (which constio a very special gxr'p) are excluded from the comparison, the differential drops to only 10 percent, indicating a surprising degree of equality between men and women. 2. As expected, the compositon of the gender gap is senitive to the inclusion of domestic savants. If they are hicluded 77 percent of the gap can be explained by differences in endowments and occupations and only 23 percent by market discrimi- nation. If they are excluded, the ark discrimination component consDtue about four-fifths of the total gap, but given that the toal gap is only 10 percent, one should conclude that maret discrimination is rather small. There is however the possibility that there is a large non-market dscrimination component responsible for the iarge endowment differential. 3. Although there is evidence of the existence of 'femwle ghetmos- such as domestc seranm and some clerical occupations, a significant proportion of women have access to occupations that offwr advaw'ment opportuities. Tbis is consistent with the relatively small levels of market diaimination and the relatively high education- al levels that women have. However, evidence was found that within each occupation men sti have better advancement opportities than women. Appendi 6A Sdecdon Bia Comcdon A typical problem faced in the atmation of wage equations such as those in equadons 3a and 3b in the text vlds cross section data is the fact that the sample used Is not random. The in- dividuals choose to work as wag-earners on the basis of a muber of fators, some of which can be identified. Those indviduals who choose not to work as wageearners do not repot wages. In this cue, the use of ordinary least squr to estimate the wage equation produces biased est;ates. an (1979) developed a simple methodology to correct the problem which conists of the i(oI% action of the inverse of the Mill's rado (Lambda) associated with the probability of bdng a td sa mle as an explanatory variable in the mDdd. The steps to follow are the following: 1. Estimate the parameters of the probability that an Obsevation is in the sumple using a probit function applied to the whole sample. 2. Estimate Z according to die folowing expression: Z = -Yk (Al) where Y is the vector of explanatory vaiables in the probit in point 1, and k is the associated vector of parameters. 3. Estimate Lambda as: Lambda = f(Z)AP(-Z) (A2) where fo is the normal density and FO the cumulative normal distribution. 4. Use Lambda ns a regressor in the wage equations. Esthson of probitfiauxow. Probit e es of die probability that an obsavation is in the sanple were estimated for each one of the three groups that were compared in the tt (men, women, and women xcludg sants). Given the diffeece in the factos that dermine the pardcipation of men and womcn the equatons are ligty different For men: Zm = Gm(finc, age, age, ducatinm th) (A3) For women: Zf = Gf(finc, fSn, age, oducaon, mr, atsn) (A4) where lfinc is the narl logaithm of finc, and finc is the income of the rest of the family esdmated as the difference between the totl famDy income mimn the income of the in- dividual. Atsh Is a dummy variable equll to 1 If the ndidual is aending school ad 0 otherwise. On the basis of the results of the probit eutim equalon A2 above becomes: For men: Zm = u.9824 + 0.OSS1Sflnc - 0.0902ao + 0.0014ge, (14.8) (10.9) (12.7) - 0.0379educaton + 1.0248ah (A4) (7.05) (17.0) For women (1n1udung servants): Zf = 0.8285 + 0.0089finc - 0.0002flnc2 + 0.0102age (6.39) (4.84) (6.42) - 0.1270education + 1.1Slauh + 0.4390marr (AS) (21.3) (19.5) (9.72) For women excludig servsin: Zfs = C1156 - 0.0043finc + 0.OOOOlfinP + 0.Ol9&go (3.65) (1-99) (13.1) -0.0684 ducation + 1.3224arhJ + 0.7897marr (A) (12.4) (26.2) (!8.1) Lambda can now be calculated and used as a regrsor in the wag equtions. References Becker, G.S. The Economics of Discrimiwaon. Chicago: University of Chicago Press, 1957. Chapman, B.J. and J.R. Harding. SeX Differentials in Earnings: An Analysis of Malaysian Wage Data." Jounal of Devdopment Studks., Vol. 21 (April, 1985). pp. 362-376. Fields, F. 'Education and Income Distributian in Developing Countries: A Review of Literature in T. King (ed.). Educadn and Income. Washingtcn D.C.: World Bank Staff Paper No 402r World Bank, 1980. Gannicot, K. 'Women, Woges and Discrimination: Some Evidence from Taiwan." *conomic Dewlopment and Cultua Change, Vol. 34, no. 4 (July, 1986). pp. 721-730. Gronar, R. "Wage Comparsons - A Selectivity Bias." Jtounal of PolUtecl Economy, Vd . 82, no. 6 (1974). pp. 1119-1143. Gunderson, M. "Male-Female Wage Differendal and Policy Responses." Journal of Econcmic Literxture, Vol. 27, o. 1 (March, 1989). pp. 46-72. -'Decomposition of the Male/Female Earnings Differential: Canada 1970." CnadianJo urn of Economics, Vol. 12, no. 3 (August, 1979). pp. 479-485. Heckman, J. 'Sample Selection Bias as a Specification Error.' Economnrica, Vol. 47, no. I (January, 1979). pp. 153-161. Hodson, R. and P. England. *Industrial Structure and Sex Differences in Earnings.' Industi Reladons, Vol. 25, no.1 (Wlnter, 1986). pp. 16-32. Mendez, M. and R. Askevw. The Colombian 'Muchacha dd Servicio'". Unpublished mimeograph. Bogota, 1985. Miller, P.W. 'Gender Differenc in Observed and Offered Wages in Canada, 1980." Canadian Journal of Economics, Vol. 20, no. 2 (May, 1987). pp. 225-244. Montgomery, E. and W. Wascber. 'RaCe and Gender Wage Inequality in Services and Manuftacuring." Industrial Rd3rons, Vol. 26, no. 3 (Fall, 1987). pp. 284-290. 167 hau. iropolos, G. -Retou 3D Educadon: An Updaed ntnonal Compairon. rCaradw Educadoi, Vol. 17, no. 3 (1981). pp. 321-341. Najafizadeh, M. and L.A. Mmurck Woiddwlde Educaonal Expinsion firom 190 to 1980: lhe Failure of the Eaion of Schooling in Devdoping Countrie.0 ourndl o Dewloping Arm, Vol. 22 (April, 1988). pp. 333-358. Schultz, T.P. Rum to Womau's EAci.on Wuhligton D.C.: World Bank, 1989. 7 Female Labor Market Participation and Wages in Colombia Thkrry Magnac' 1. Introdudcion Tbis paper describes the estimation results of a microeconometric ziodel of female participation in Colombia. Tle samples used are drawn from urban household surveys tetween 1980 and 1985. The set of exogenous variables includes individual as well as household-related characteristics. The results are used to assess the power of the modeling ststegy in explaining the dramatic increase in female participation raes. Wage equations are estimated correcting for selectivity biases and compared for males am' females. 2. The Colombian Labor Market The industrialization process in Colombia dates back to the 193Qs but a significant Increase in the industrial labor force took place in the late 1940s. Colombia was in those days mainly an agricultural country: The agricultural sector employed 55.6 percent of the whole labor force in 1951 (Bourguignon, 1986). GDP growth between 1950 and 1980 had been sizeable despite tbe 'violencia' period in the 1950s and had averaged 3.5 percent in the late 1960s and early 1970s (Sarmiento Palacio, 1984). This period matked the end of the import-substitution policies, the industrial sector producing 85 percent of consumption goods, 50 percent of intermediate goods and 20 percent of capital goods, consumed in the country (Sarmiento Palacio, 1984). Expont- promoting policies for manufactured goods were then set up but Colombia mostly remained a single-ood exporting country where coffee still accounted for 70 percent of exports in 1980, and 50 percent in 1984 (DANE, 1986). Tle Colombian economy underwent a crisis in 1974 but quickly recovered becas of the booming ooffee prices and exports in 1976-86 ('bonanza cafea"). Inflation soared to 30 percent yearly in 1980. In this disequDlibrium context, the inenatonal crisis of 1979-80 hit Colombia very hard. A severe recession began and lasted at least unt 1984 when production got back to its 1980 level although t!La employment level lagged behind. I want to tak 1. A. Ocampo, Diector of Fodesamlo, Bogots, where I began this wo*rk a thak the DANE in Bogot for hving given me acces to det and computers while I wu (her. To be also acknowleded is the belp of people in the Laborstoim d'Ecooomie Polique in Paris and eqoecialy P. Bourguigwon who follow4d and enouged this workw Th ulal disclaimer applies 169 170 Wovaa 's EmplyMard hPa L Awrca Nevertheless, the history of last twenty years shows that the Colombian labor market has adjusted very quickly. In fact, despite catastrophic predictions by the ILO in 1970 (Misidn de Empleo, ILO, 1970) of unemployment rates in the 1980s, the economy and particularly the modem sector succeeded In creating a huge number of jobs in absolute and relative terms. The ILO report correctly predicted the r-Amatic increase in labor supply but underestimated the labor demand growth. Even if labor demand and supply growth are hardly distinguisbable using the aggregate data, some points are worth mentioning. Several factors caused the growth in labor supply. First, the demographic growth rate, though decreasing, was still large because both fecundity and mortality were decreasing. Another facor feeding urban labor supply growth was the continuous flow of migrants from rural to urban areas. Migration only stabilized in the early 1980s when 65 percent of the population was living in towns of 20,000 and above (DANE, 1986). Fmnally, female labor force participation rates had been increing in the 1970s and 1980s. The ratio of female to male workers went jp from 20 percent in 1964 to 26 percent in 1973 and 31 percent in 1978 (Bourguignon, 1986). However, the international crisis changed the labor market situaton. If the analysis is restricted to the urban labor market (Coyuntura Econdmica, 1985), it is dear that global participation rate decreased during 1981-82 but went up again in 1983. It shows that the unemployment rme dramatically increased (7.5 percent in 1980 to 7.5 percent in 1985). Labor demand was stgna or decreasing, particularly in the modem sector, and the number of self-employed soared. The modern sector's wages begmn to decrease at the end of 1985. High inflation prevailed during this period (20 to 25 percec). The focus of this study is married women's participation rates, becauLe they seem the most responsive to charging economic conditions. The standard participation microeconomic model Killingsworth, 1983) is estimated using household data for six consecutive yeas. Exogenaus variables used are individual as well as household-related characteistcs. Tbe usual effecs are found. Human capital variables positively affect the probability of paricipation, while the number of children and other household member's incomes act negatively on iL The mnmber of other women in the household have a sizeable impact and capture the likely substitutions in domestic work in the housebold. The results show that the coefficients are generally stable over time. The predictions of this model can then be used to assess the power uf this participation model in explaining the huge increase in female participation which took place during the period. Even if it explains only 30 percent of the increase, it relates the increase to changes in education and fecundity over the period which corroborates a finding derived from tracroeccinomic studies. Tbe residual (70 percent) might be attributed either to missing variables aetermining labor supply or to labor demand factors. Demand and supply are obriously connected to wages. To informally evaluate the impact of the demand factors, some esdmation results related to the wage equations during the period are proposed. They are compared to equations for the male household's heads and other members in order to gauge discrimination effects. Some evidence is produced showing that the difference between male and female wages might be biased if the work experience of females is not correctly measured, as Is the case in tbe&e data Tle plan of the paper is as follows. Section 3 presents a descriptive analysis of female labor participation during the period, a very brief survey of previous stdies and the estimation result of the paricipation models. It concludes by assessing their explanatory power. Secdon 4 discusses and presents the results of the esdmations for the wage equations. -_._. - ' a.- ~.~ *_* _ 4A S UJWg PY4 AJ J/X 3. Fenale Labor Force Participation Descfdptve Analysis. The data analyzed in this section and the estimation results given in the subsequent secticns concern yearly subsamples extracted from urban households surveys undertaken between March 1980 and March 1985 (Encuestas de Hogares, DANE). These surveys are briefly disaLussed in Appendix 7A along with the method of construction of subsamples which include households composed of a man and a woman aged between 18 and 60, reported as being married or living together. The labor participation rate of married women, defined as the ratio of those working or currently searching for a job population to the whole population was 22 percent in 1975, 30.5 percent in 1980, decreased to 28.1 percent in 1981 and increased again in 1985 to 35.7 percent (Table 7.1). The decade 1975-85 can dearly be split into two periods, the frt being a period of continuous growth (1975-80), the second showing a fall followed by a dramatic increase (1981-85). The married women's pard6ciption rate increased much more than the whole population's which went up from 52.5 percent to 57.3 percent between 1981 and 1985. It must also be noted that the married women's unemployment rate, after having decreased between 1975 and 1980, shot up between 1981 and 1985 from 7.5 percent to 15.1 percent. Unemployment expectations apparently did not put off labor market entries between 1981 and 1985. In particular, 1984 saw the largest increase in the two irdicators, unemployment and participation, which corresponds with a partial recovery of the Colombian economy after the crisis. TsUe 7.1 Participation and Unemployment Rates of Marnied Women (pervua) Towns 1975 1980 1981 1982 1983 1984 1985 Baffanquila P 15.9 21.0 22.8 21.0 25.5 26.5 24.4 C - 5.2 7.5 7.6 6.7 9.1 10.7 BogotA P 25.4 34.4 27.6 33.8 32.6 40.0 41.6 C - 9.3 6.5 8.6 8.3 15.3 17.1 Moddlin P 17.4 28.3 25.7 25.1 26.2 26.4 29.2 - 12.4 8.9 11.2 16.4 17.4 14.4 Call P 19.4 35.3 29.5 30.9 30.2 32.7 33.0 C - 5.4 8.1 11.8 8.6 11.9 16.7 Mcdiumnsized towns P 26.5 37.8 28.6 38.1 41.1 40.7 c 4.5 7.7 8.0 10.0 10.5 9.8 Total P 22.0 30.5 28.1 29.2 30.5 34.4 35.7 C 9.4 7.5 9.5 9.5 14.5 15.4 Notes: P paricipation ratw (iucL uneaVp7mat) C - unenploymbt rta (umpla7mbY AC/puzia) Medium-sized to: Bscanmn, uMizale, Pao. Firt, differendating by owns, after a uniform increase of participation rates in every tou bwn 1975 and 1980, the 1981 reduction is quite tong, espeially in the tee major towns, Bogoti, Cali and MedellfiL2 oughout the second period, Taie 7.1 shows that die pricipation rate, which sens unstable in th3 thro medium-sized towns, increased between 2 and 4 percent in Banqula, Medefn ad Call and dramatically In BogotA. Thus, BogotA seems to have been a large generator of employment for married women, even if the 1981 crisis had a strong imp t nmus be recalled that migrations to BogotA came to a stop in 1975-80. It might be possible that fim, in particular in the service Industry, no longer being able to recruk migrants into their labor force, tried to employ some workers from lower participation-groups. After 1980, labor demand growth is less marked in other large towns because those underwent severe cris, as was the cme with the textile indutry in Medeilfn. Bogoti, the commercial, finincl and administrative center of Colombia is dearly the one with the greatest potenti for growth in the tery sector. PadcIpation tates are dearly positively associated wih education level (Figure 7.1). The two periods are again disnct. Between 1975 and 1980, labor participation increased uniformly in all groups and the universitydree group's paticipion rate had risen to 65 paecet AfRr 1981, the greatest increase in participation rates was among low-level education groups. Conse quendy, the participation crve, as a function of education, took on a parabolic form in 1985. Vage 7.1 Maried Womm's Partca oy Eduation Pa ticipation rate () 80 1985 70 ,1981 60 / r 30--- 10 0 S School Yers 0-1 2-3 4-6 6-7 8-9 1-11 12. Primary Secondary University 7 Te March 1981 surey w replaoed in BoSota by a much laer mple (Eaudios do Pablacida). Somn iplinS bin night bavn onrred Participatio ratcc ars ined quit low in Boga is March 1981 but the ubo urvey thr monts later Ov a remt only mildly greater than ian Mar During the second period, th labor supply of diffrent agegroups changed, above all for cohorts aged between 35 and 45 years (in 1985) (Fgue 7.2). Arrows in Figure 7.2 show the evoludon of participation by age cohorts. if the participationi wve as a function of age in 1981 Is uwed to predict the entry-ext flow for cohorts over 30 years, that flow should be negative. In fact, it was positive, as was the case for cohorts aged less dtan 45 years in 1985. Married women stayed In the labor market longer tha previqs genaations. Tlis evolution was quite similar between 1975 and 1980 when the ncee had maiy been nodceable for cohorts betwoen 2S and 35 years. Figure 7.3 clearly shows that the 1980 crisis madc young women widtdraw from the labor force, but they reentered the mar as won as the crisis waned. The models esdmat in the fo lowing secto make use of some exogenous vaiables relaed to individual and houehold charac s of nuried women. It is usefil to describe the evolton of these variables a presentod in TableF 7.2 snd 7.3. The most worthy points in Table 7.3 ar the following: Fht, the mumber of persons per .bousehold decreased, mainly becas of a diminishing number of children and teeog. Thhi I a consequence of the decre in fecundity in Colombia since 1970. At the same tim te nmeber of adult men and women remained stable. }be 7.2 PArticiptia RPae by Age G.p and Cobot .40- 30- 20 - 20.25 25-29 30-34 35-39 40-44 45-50 50-54 55. g Arlo liie 7.3 Polaitial md Rceem Wage W.W-W Potential wage W A B D Participation condition W-W 30 45 Age Table 7i2 Vaiable 1981 1982 1983 1984 198S EDLUC Z0.9 11.4 11.6 11.7 11.8 (3.6) (3.0) (2.9) (2.9) (3.3) AGE 3S.4 3S.4 35.4 35.6 35.8 (11.1) (8.3) (8.1) 8. (9.2) AG 40.9 41.0 41.1 41.3 41.2 (12.7) (9.S) (9.4) (9.3) (10-S EDUM 11.5 12.0 12.3 12.2 123 (4.3) (3.S) (34) (3.4) (3.) NOdr: Snda am, an in pa_L EDUC. EDUtM - yX of edvafim (wik aL buban). AGE, A tI (wife'. n huab&ndas ap) fl:, the unemployment mrate dcained in this ca x th5 ratio of unemployed to ac&e ashts in the rwusehold, weat up from 6 percent in 19E1 to 9.5 pec in 1985 for men, and from 10.5 percent to 14.5 percent for womn. -ie unemloyment rate in those households reached 10.7 perctut in 1J85, in comparivn to a global unemployment rate in the popuWation -.al to 7.1 percert. Thus, these households were seemingly less affected by unemployment than houseOlds headed by iitoGe person. This confirms the flndinp of praious studies ef unemployed people (Ayala, 198!). Hoveve', this effect may a!sc be pretd as a selectiv;ty efrect on household size, hcause :he sub.Anles under sudy comprise larger bo.sseholds. Also, other variables may be reda:d to the sample sviecio rue and be cornatd to the unelnplrym2nn probabBity. Concludi:g, it is clear thA the number of active pesons remined stah.e while the nrnemployment rate had increr.d, that is to say the la)or .upply of thu famNiies j witeut more members being employed. NevertrIeae, thee is also some iubqtitudon betweme active womten and ative ten. Tbis ha. as usual, two possible meanings. Fuit, a supply effect If women bcgin uo work beause men we mwe affeed tb une;npoym (the adcikioDw. worker effect). Second, a 1aaaad effect if firms sys;rdal1y replace men by wouen or if fims employing women (e.g. :r the sLrice indlary) L- --winL fise than othwr (e.g. manafacwringj. I might also be a conseuence of 'iscmL ' jf womn tam less than men with equd producivAy. In the s;bsmuent sections we wia rav the c?portunity to corfirra the demand effect dWad on offemdk AIbrC pa.idpidoii . s. Few stuies of female labor supply in CO.orrobia in the literature have bev conducted. Tese iMae stdies by de rGdme et*-. (198 1). ataeda (1981)E B .xguip (1981) and (Cz'l2vet (1981). Ibe piper by de G-.5xa e: al. (1981) used Jaa *om the 1973 C-.,vs and data fron. one of tha urban houebold nrveys (December 1980). (Dible 7.3) '..he estimatecd odel consists of jcint labur supply equatos for husband and wife with wags cs regressors? The study by Casieda LC cince-nel by the relatoship between fe andity and - artici:tation. The .12ta used come f,=m oxwe of aw, urar. bousehold surveys (Jau 1977). he- in2ol use. is a probit ecii aon. Tba third study (Sourggwor, 1981) deals with a simultaneous estimaion of ried wom's particmwon and u.-e of servant at home. Some regressicra results of pardcipak zn uogeis railables am givei albough they are not the ,ocus of the pape. Te pq^ by Calavet (1$8I) giv.n the esuls of the estimation of pn.ticipatfn fa;zt1nn fior Ious6-Jd headk m3 their spises usiq data from the urban ousebold :urveys in 19t5 and 1978. Sanaiing up the results acrca these papers, which isvesdgate ilfWere points related to !abor supp.y, it is genrally agreed duit thel wage effoct on labor suply is positive for women white, Nuband's wage has a neptive impa Tis conforms wit the usua income effect and iaq;l4 nat leisure is a wtmt good. Addidcnally, the usual effects aae found. Huwa capita! vai,.) ' -: bave a positive imppc on i0or supply while the effect of dldren is dearly negative. ir#; .' . itk ust be noted tat. cxeept iu the C-asafeda t1981) study, these r.sWiltt are plaguel y *.vity ble. W Sar rogrmd an 11 bso cshw A ?a vsAA. 1981 1982 1983 1984 198S Children 0-3 0.SS 0.53 0.52 0.50 0.48 4-12 1.13 1.12 1.03 1.05 1.03 Wom Active 0.38 0.39 0.40 0.41 0.42 Unemployod 0.040 0.03S 0.039 0.058 0.061 Stdats 13-18 0.38 0.37 0.3S 0.36 0.34 19+ 0.019 0.027 0.025 0.029 0.026 Tractive 13-18 0.057 0.060 0.048 0.048 0.046 19-60 0.90 0.89 0.88 0.85 0.84 60+ 0.042 0.041 0.041 0.04S 0.043 Mean Active 1.23 1.21 1.18 ' 16 1.18 Unemployed 0.073 0.077 0.096 0.10 0.11 Stuet 13-18 0.34 0.32 0.32 0.30 0.29 18+ 0.026 1.031 0.030 0.030 0.030 ioative 13-18 0.022 O.n70 0.020 0.020 0.018 19-60 0.11 0.12 0.11 0.115 0.11 60+ 0.012 0.OJC 0.009 0.010 0.010 Total 5.32 5.26 5.10 5.09 S.04 N 5,SS 10W50S 11,53S 11,526 9,741 Notw N I -u3dxsof cbm0natm. To clarify dting, the mariet women's participation moddel is briefly presnted here. It must be borne in mini4 that ths is a reduced modd that can b derved from asumptions on preferece of the agents. Mm aim he is not to esdnmat the elasdciy of labor supply with respect tD wages, but to permit a descripdve analysis of femae labor force padcatiop . Some aueqM have boos am& k% etimt hus of wuck eqimaoms (Mgacw, 1987) bua the reast nmirly poist cut dat t6 "i4it it nom_ gLi&eamy differwt firm 0 as fr a woem we co_cued leo main ficxbiqty he aw f. puzlcptm and non-ticipAim Participation if w - w > 0 Non-rarticipation if w - w < 0 where w is the potential or mark3t hourly wage and w' the resenration or asked wage. Two additional equations are specified: log(w) = XB + u log(w) = ZT + v The first equation, the wage equation, relates tue logaritnm of the hourly wage rate to human capital variables (X). The second equatin, the reservation wage equ%ion, is derived from the preferences of the married women and depends either on personal or household characriscs.' If u and v, mnnditional on X and Z, are upposed binormally dLstbuted, the esfimation segy to get consistent estimators is Probit and the reduced form ciodel comprises variables X and Z. This model was estmaed using six rubsequent years (19W845) keeping the same set of exogenous variables. The households' chaacteristics retained in the present specification have been chosen for their significance among a large nzmber of explicadve variables (Appendix 7B). These exogenous variables are the following4 Human capial %ablks.' EDUC completed years of education EXP work experience which is not repoted in the survey and which is approximated he:e by age minus education EXP2 work experience squared Household's characterlrdd3 IMAR husband's income in thousands of pesos (co.stant 1981) DM dummy variable equal to I in case o! non-reported husband's income IOTR other household members' income in 1,000 pesos D number of other household members' non-reported incomes ' This set-up is called fth mb chnavnis6c model beca: it sme th the married womn is the last o choose in the family. Or simia y t se hod well-defined pefaewe condition on all other vaiables 'he interrviorpa we am going to p hoseo asume at household organuiat ad decisions to woik by oher i_s sven and that it is not made sinzltaneously with the married woman's decision to -*wL This is a dispube asmption but an instrumts viable procedure to corct these biaes iu out of th cpu of the prst pwp. uInfluene th potential wage funtion CX) end possby pfance.s well. Inoonm we real ieome ooquted by ddlating _minal byn aggregate cow_ption price index (DANE, 1986). El number of children between 0 and 1 year E2 number of children between I and 3 years HSTI number of male studen between 12 and 18 HST2 number of male students over 18 FST2 number of female students ovec 18 HDE number of unemployed male adults MJI number of inactive women between 12 and 18 MAI number of inactive women between 18 and 60 MYI number of inacive women over 60 in the specification, dummy variables for towns were added to this list in order to correct firs for some sampling effecs In different years and, secondly, to take into account the fact that ferhale participation depends on the unique economic development and labor demand determinants in each town. All estimation results are given in Table 7.4. The coefficients of the human capital variables (EDUC, EXP, EXP2) are significant and stable across time. Differences are not significant across years. From these esdmates ca be deduced that the highest point of the predicted pardcipation cuvre as a funcion of age is around 30 years although it varies slightly across time in the interval 28 to 33. It is largely before the highest point in the wage equation finction. This implies that the reservation wage increases strongy after 30 as represented in Figure 7.3, since AB > CD. This can be aritod to changes in the domestic production fimction when the woman ages and acquires a lager relative proe .ivity in comparison to the other members. Specialization in domestic work is impotant at this age. This can also be related to genration or cohort effects if there are systematic differences in the divisiou' ef domestic work across generatirns. Unfortunately, a cross-section analysis does not pamit us to ditnguish between the two intexpretations. The husband's hucome effect is signiJicant (except in 1981) and negative as expected. However it is unstable (1981-82) and the differences are significant. Strong interactions with other composition variables might be responsible for this instability. Similarly, the other members' iancome effect has the expected negative signL DuWmy variables for non-reporting errors ars significant and negative only for other members. Ths confirms that other members' income effect is much larger than the husband's. As the estimted model is a red 'ced form, this income effect might take into account subsiutin effects between the maffied woman's laboi supply and other potential workers in the household. The substitution with the household head is a prio*I smaller. Nevertheless, variables such as the mmber of active members in the household have proved to be nonsignificant (Appendix 7B). Among the household's compositin vaiables, the effect of chi'drz aged between 0 and 3 Is largely significant as expected. The opportnity cost of the married woman's dme i-creases when she has young children. Additionally, the influence of the umbers of inaCtiVe women on the participation probabiity is pcidve and significanL This is clearly elated to subsdtuions in domestic work within the household between women. If the married vwman works outside the * no muied woma is cleary so inluded in d_ coi. TAbl 7.4 Pabit Edmn_ioc Rwlot of tte LAbor P cipaion Modd 1980 1981 1982 1983 1984 1985 hitecqt -1.09 -0.94 -1.15 -0.74 -0.68 -0.90 Ehu 0.061 0.065 0.06S 0.064 0.057 0.062 (9.2) (9.9) (9-9) (15.1) (7.6) (14.0) Exp 0.020 0.022 0.021 0.019 0.017 0.026 (2.5) (3.0) (3.9) (3.9) (3.S) (4.9) Hipx -0.00058 -0.00051 .0.00055 -0.00051 .0.00054 -0.00059 (4.2) (3.9) (5.7) (5.8) (6.1) (6.2) Ima -0.020 40.0050 -0.012 -0.023 -0.022 -0.024 (1.9) (0.5) (2.9) (2.5) (3.5) (4.6) Dm 40.09 0.0026 -0.020 40.043 -0.13 0.004 (1.9) (0.05) (0.6) (1.2) (2.9) (0.0) Iotr -0.063 -0.049 -0.113 -0.102 40.082 -0.080 (2.9) (1.9) (S.8) (6.1) (4.4) (S.2) d - -013 -0.10 -0.077 -0.11 -0.15 (2.2) (2.9) (2.1) (2.2) (2.3) El -0.23 -0.22 40.27 -0.24 40.25 -0.21 (S.3) (5.0) (8.2) (7.5) (8.0) (6.0) E2 -0.11 -0.11 -0.13 -0.12 -0.14 -0.10 (2.7) (25) (4.4) (4.4) (S.0) (3.3) HSTI -0.026 40.077 -0.026 -0.039 4.026 -0.016 (0.8) (2.6) (1.2) (1.9) (1.3) (1.7) HS12 -0.03 -Q21 -0.17 -0.21 -0.22 -0.17 (0.4) (1.8) (2.4) (3.2) (3.2) (2.4) MS77 40.10 -Q19 40.23 40.18 -O.lS O0.OS (1.6) (1.3) (2.7) (26) (2.1) (0.6) HDE -0.020 J 024 40.104 -0.11 -0.047 -0.17 (0.3) (. a) (1.8) (2.4) (1.1) (3.4) MJi 0.26 G.22 0.16 0.29 0.29 0.2S (4.9) (3.0) (3.3) (5.6) (5.7) (4.3) mmI 0.36 0.10 0.15 0.15 0.11 0.17 (5.0) (1.9) (3.9) (4.3) (2.9) (4.3) MV! 0.29 014 0.37 0.21 0.22 0.14 (3.3) (L7) (5.9) (3.4) (3.8) (2.1) BARRANQ. -0.17 4.0G0 .1.40 40.64 4-49 40.61 (1.5) (5.7) 6.0) (10.2) (7.8) (9.6) BUCARAM. 0.01 -0ll -0.16 -0.44 -.034 -0.37 (0.1) (1.0) (1.9) (5.7) (4.4) (5.1) BpSOgL -0.04 -.43 -0.037 -0.48 4Q.17 -0.15 (0.3) (4-5 (0.7) (8.9) (3.1) (2.8) Tabde 7.4 (coninued) Probit Etimation RwzUs of the Labor Participation Mo"ACJ 1980 1981 1982 1983 1984 1985 Manizales -O.S6 40.S6 -0.43 -0.78 -0.28 -0.43 (3.9) (4.4) (4.6) (8.8) (3.4) (5.4) Medelln -0.17 -0.S2 -0.27 -0.62 -0.50 -0.47 (1.S) (5-0) (4.2) (10.1) (7.9) (7.4) Cali -0.43 -0.39 -0.90 -0.50 -0.35 -0.36 (3.9) (3.9) (1.3) (8.3) (5.8) (S.9) N S,21S 5,58s lo050 11,53S 11,526 9,741 LOGV -3014.3 -3167.8 -6003.3 46735.S -7040.5 -6004.5 LOGVIN 40.578 O.S67 -0.571 -0.584 -0.610 -0.616 SRV 386.2 328.0 681.9 717.9 756.4 686.8 Note: Ib lcor pritipon Max (1,0) is the dependet vaiabe. Exogeaous vaiabke* am defined in the tocL N number of obsaovaio LOGV = lg4kihood LOGV/N - mean lo&-hlbood SRV keElhood ratio iatsuo (Ho: aD purame4ls (22) are qual to 0 eOCep the inktcrpL SRV disrbutioni i symptodealy il(22) under Ho. T-Staiistic arm ibw in pamxbeaea. house, then other women take charge of the lomestic work. The effect of a young or old woman is much stronger.9 The negative effect of the munber of unemployA mnen ir. the housebold, though unstable, looks as if an expected income itArpreta- would De needed instead of the usual additional or discouraged workeL effects. Since it is negative, the additional worker interpretation can be discarded. However, as the presce of an unemployed woman does not matter much, the discouraged worker hypothesis sould imply that the married woman assesses her opportunities to get a job by looking at the men's unemployment rae and not at woman's. This is unliky. Moreover, the differenco betwe the coefficients of th; dummy variable for non-reported other members' Income and of the number of unemployed male aduts is no stgnificant This effet could then be related to a minin inome. However, this coefficient is the result of these three effects. Going back to the hpoesis proposed previously to explain the increseo in tfr' ratio of active women to active men, tho demand effect seem to be the most likely inepretation siux the suwpp:y effect (additional worke) soems to be hardly notdcable. It is more difficult to intpret the edoct of the presence of students in the family (HSTI, HST2, MSI2). These effects we nepdve and significant in most cases. It belies the thesis ttit married effec of afctive - =s t signi (A aix 13.2) wh cofiru th findin f CaiQvet (1981) about rn'S n tcilas in domeic woar women work in order to pay for their children's education.Y These effects can have two economic meanings." An incomo-effect (of children), expected in the short run, is real since many students work. It can be noticed that the presence of girls going to school is not significant although it is for boys. A tentative explanation would be that for girls two effects are combined: The first one similar to boys which is negative, and the second, a positive substition effect similar to young inactive women's (MW!). But the Instability of the result makes this interpretation shaky. The last group of variables are the geographic dummy variables.' Several groups can be distinguished. Barranquila, Maniales ad Medeliln have low paricipation rates. Bucanmaga, Bogoti and Cali belong to the medium range below Pasto where participstion rates are the highest. The parallel evolution of global participation rates by towns and the coefficients of dummy variables must be noted. Table 7.5 presents the results for BogotL Table 7.5 Diffaences in Participion Rawe ad Dumwy Variable Coeffidmt for l3ogo(t 1980 1981 1982 1983 1984 19NS Differce in 1.1 -15.4 1 -14.5 -3.6 -4.4 participation rtes with Paao (percent) Coefficient of Bogo4i -0.04 -0.43 40.04 -0.48 -0.17 -0.1S These results show that the estmat model explains very litto of the difference between participation rates in different town. These variables puaty control for sampling biases, above all in a small sample like Pasta, but these effects are very unstable, They might also show different evolutions of labor demands in the different towns. For example, the labor market in Pasto depends heavily on the economic relatlor&ip with the neighbor ccntry, Ec9ador, which tended to deteriorate in 1984 and 1985 aftr the Andes pact was called off. To Mum up these results, the predicted pmdcation variation as a fimction of factors can be computed at the mean sample point (Fable 7.6). Ttis table is just another way of presentng Table 7.4 and does not need further comment In conclusion, if this model sbows some classical and expcted effects, such as the influence of human capital variables, children or incomes, it reveals also strong substiton effects within the household and sets forth the importance of th, household organization on the probabilky of the married woman's participaton. 30 In survoys wbere muaed womm ars asd their tenons fi wos the mt co answw is ao pay for my children's educdon' (Gutidrez de Pireds (1975)). " No doubt dh siLaned of edon deciim and tho decsion o wt i likely to play an iporlant role Feramplethepmma of sWftd in th sehobldt o pood to ages te maried woman less the labor muket '3 De miming dummy is reae to aM wher do particiption rate is high. Table 7.6 Participatio ProbAbiity Vriaion . a PVunioa of Variables (for aaddtionl uni with initial probability equal to dth 1M parti3paion ate) 1981 1982 1983 1984 1985 Educ +8- +r +8- +- +r Imar -0.6 -14- -2.6 -2.S- -2.8- laIr -6 -lo" -9- -9- El -26 -31- -2r -29- -24- E2 -13r -IS- -14-* -16- -12- HST1 -9 .3 -4 -3 -2 HSP2 -2S -19 -24- -25- -20' MfSl7 -22 -26- -21- -17- -5 HDB -3 -12 -13- -S -2(Q MMI +26 +t8- +33- +33- MAI +12 +Ir +ir +13- +20- MVI 16 +4r +24- +2S- +16' Note: Si&;aicsfcc of octfiatS 5%) or15-) 'he data and the estim cannot be used to asses the explant power of this model. As we could not estimate this model on sucked data for 1980 to 1985, the following subsection proposes a simple decomposidon of the various effects of those variables taring U,-, period 1981485. It will allow us to distinguish more deady thtd supply and deman effects. Expanafory power of Oha d Jbhr p mc i ut: lhe probit reslts have shown that the model was rathr stole dwimn the period. Despita this fact, the explanatory power of these models are usually low (llnvgorth, 1983). However, these results can be used for shortuenn predictions of patcipton rats. As our purpose Is to test the predicdve ability of the model in a context of very larg increases in femaJe labor participation, we computod forcasts of the pardcipation rue in every yar using the coefficients related to another year. Table 7.7 shows the following vAb3es:- . 1 ~~~n pi=E(FMC~) EV FNXt s mlyss howr is r c t 1981 to 1985 bscmAu of the non-uimultaneon avaiabiity of lb M0 and o6hr wzvsys. where I Is an index of the year for whk:h the predicton is ;omputed, j s an index of the set of the esdmated coefficients, b,, relatd to yearJ, A9 is the m uber of obseratons in each yea, 4 are the exogenus vaiables for year i and obsavation K, and F(.) s the cumulative distribution function of a standard normal varie. It clearly ppears that changes in the ezoge variables explain little of the global increa In participation rates. The predicted increase belongs to the interval 1.8 percent (1981) and 2.9 percent (1982) much less than eo l Ionrea of 7.8 percent. It implies that mare than 60 percent of this increae may be arted t changes in the esmated coeficients. Stern and Gomulka (1990) proposed to disdtnguih the comparative effects upon pardcipaon rates of chages In exogenous varfibles md cages in the coefficient. Tho difference beteen dt estimated participation rates cam be writtew Pi - pim 3 B (FX,,b,,a))- E(F(X,,b,s)) + B ((X,,,bj)) - E (F ,b,,Hu)) = chaes in variables + chages ir. oefflclets using dte same notations as befor It can be computed by groups of variables and the results appear in Table 7.8.14 It appears din changes in variables decermining the hicrase in participation are mainly due to in in men education and to the decreases of the number of children below 3 years. The income effects are small. On the other hand, clges In the ceefflcients mainly come fom the coef-ficient for BogotA This barely explains the sizable hirease of labor participation In that town. Table 7.7 Pdiss of Poutcipim Rate Usng Diffea Stu of Coeffcias Coefficient of: 1981 198 1 1984 19 Da: 1981 0.230 0.283 0.288 0.325 0.340 1982 0.2W 0.298 0.307 0 337 0.360 1983 0.297 0.301 0.310 0.342 0.361 1984 0.302 0.306 0.309 0.350 0.361 1985 0.307 0.312 0.309 0.350 0.3S8 Noma At letanction rLi) the actuaI retur am yuqiotcafy clow to the mfedratm. " Anodh (symetric) doouqestim eusts. Cluoge in coofflce atm moted usig the mu in 1981 homed of 1985, and chag ia vables vsing coefficits of 198S ine1 of 1981. It wuid Sive appozimdely de sun mesults (M*lb 1987). Table 7.8 Deoomposito of CImae in Labor Paricipatio RBto over 198148S by Vaiables mad Coofficimkt Chi in variables in cocfficiaits Variabe gruS Eduattn + 1.7% -1.3 (Edu) Exeiec 0 +1.4 (Exp. oxp' 1nowe +OS -1.3 ([ar. Dm Jee,r D) Childrze +0.4 +0.2 (El, E2) lnctive Womn .0.1 +O.S Other menbers 0.1 +0.4 (HSTI-2, MM HDB) bnter and lowms -0.2 +1.0 BOgoI +0.5 +4.1 (BOG) Total 2.7 5.1 Note Compua&ies wes doe wka eo pemnted i to tI fr ich0 roap. variabkes and onw-i-a of ta i hawe beb }kqtoo an d a M lomir briiid vale Concluding if 25 peceat of the hams in parcption ra Is corecy prodictod oy the model, oDnfirmed by the effects of educaion and fecundity, this is clearly a modificaion in the *center of gravity of the model, Cmterapt and dummy varibles) which is implied by the participatio evolution between 1981 and 9MS.s It should be recalled that if costs of access to the labor market are signifcant, the pidtcpon model does not pemit their dentfication from the reservadon wage (Cgn, 1981). Only an estmatin of an equatio n of hours of work woWd permit this. Additinally, productivity gn during the peod andor exog u grwth of real wages are no t Identi fia om dem d variatio. The absece of variables in the model desarbg labor dmand or peculiar onomic conditioa migt explain the resuals of the predions, here calld ch In die co Icients. Thae cha in 1to, demand can be desaibed alternatively by studyg the dermnioa of wa over the paod since those are reated to changes in spply nd dmand. I Tis mek;od licb h nls ot vcaobyvaiagroups Anoth er procede coold hav bemi used by sKkig t do cbmisdos. JIomw , hat method wuld pssbly give diffenot emalto becaus of the cnc4inewr mu of the mnod 4. Wage Functions Tables 7.9 and 7.10 show the resuts of such estimates for differen groups in the population drawn from different survey studies (Mohan, 1981. Carrizosa, 1982). The dependent variable in those regressions is generally the logarithm of total inome or labor income. Results are rather scattered. Tbe education coefficient, that is to say its mean yield, varies between .14 and .20. These variations can be explained mainly by different survey coverages or by the different sets of variables used. These results imply that income doubles for every 5 additional years of education which roughly correspond to an entire cycle in a primary or secondary school. lTese yields are much higher than in develuped countries, as generally the case is in less developed countries (l)Cs) (Psacharopoulos, 1973) but even seem to be quite large by LDC8 standards. The usual long-term interpretation of Micer (1962) would relate these yields to the interest rates but they may also be related to high costs of education (Magnac, 1987). Given the limited access to the financial markets of many households, the second reason may be the most important. However, In the 1970s large increases in average education took place since primary education became compulsory in the 1960s. Education yields should have shown large short-un variations. This indeed shows up in these results. The qeestion of whether these fluctuations are related to demand or supply shocks remain unsolved. Maried womeAn' wage cquadoas. In order to correct for the seletdvity bias In the wage equations (Killingsworth, 1983), a Heckman (1979) procedure is used here. The inverse Mill's ratio associaied with the participadon equations estimaed in the previous section Is InMluded as a regressor. Tbis method wil give consistent estimates of the coefficients but inconsistent estimates of the standard errors." Table 7.11 shows the estmation results for the following equation, with and withoat selectivity corrections: log(w) = Xa + u where w is the hourly wage rate, defined as the rado of labor income to normal hours of work. The R2 Is relatively high given the number of observations and the coefficients are largely significant, especially for educaion. The usual positive effect of educadon and the positive but decreasing effect of experience is corroborated. Witho selecivity corrections, an additional year of education increases wages on average by 14 percent and the maximum point for the wage profile as a function of age is reached for experience equal to 30, that is to say around age 40. Coefficientw have a positive trend between 1981 and 1984 but decrease again to their 1981 levels in 1985. On the other hand, 1980 coefficients appear o be very large. This decrease can be partiaily explained by the evolution of labor participatica over the period. In 1983, and above all in 1984, there is a large increase In partcipation rate and women, with low education and litle experience, enter the labor marke lhis group, with low earnings, make the wage function steeper (Figure 7.4). This phenomenon could be attributed to the modification of the selection rule aross time. oTs ame ported for nfoDnztn rposx For 1980, a tdard comploet likelihood me ws used but td numerical mlt sw not any dffere from dhs given here. Table 7.9 Incomn Fuoctioa Etmate Variables in. Edac Exp Exe SEX YP R2 N Sbtdies (1) 4.8 0.173 0.121 -0.0018 0.881 47 (7.4) (8.8) (7.3) (2) nd 0.151 0.135 no 0.127 0.70 n.d (17.8) (7.2) (-0.9) (4.3) (3s) 5.08 0.167 0.078 0.0011 0.63 1016 (38.9) (17.6) (12.6) (3b) S.88 0.151 0.068 -0.009 0.51 3640 (59.2) (25.8) (19.3) (4a) 4.85 0.201 0.068 .0.001 0.32 rLd. (50.0) (21.6) (17.3) (4b) 4.86 0.219 0.066 -0.001 0.32 m.d. Notes: it Ltomept; EDUC - Ed9esto; EXP - ExpuieDCP EXP2 - E& ice squune SEX - Scx YP - Pahes hcom. N - Nu of obarnaio Sqdat teat in bradkata Sourmc Mobla (1981), Canizo (192. Od&n Sourac: (1) Scbut (1961): Mm, Biogc in 47 V (agogatd) 196S: Dqiadmt vuibi - lS ( baat iWOeO). CZ) uag (1974): Roa sad urban popuidow dcpadrAt var6bl - kg of hbor iunom. 1970. (3) Doupgipno. (1960) bMci bogui Depsadot w'aabh - g of moKti o a) in 1971 b) i 1974 (4) FSadU (1977 DCpmoda vairbh - lg of totl ioKnn in 1973. a) Salarwed b) Sdflmphyd Sdectivity bias corections ham an hmportan influeae on esmates. ducation yield ne s by 15 percent and expeienc yield by 40 percent The inverse Mill's ratio is posit and significanL But hi comecon docs not change the concusions made above on the temporal evohldon of the coeftiients." Coraai6oa with wge equaiowfor odw membn In the howhoUko Wage equatons for the yeus between 1981 and 1985 have been esmw:eed fr husb and resul are pres d in Tae 7.12. Sinilar results are gvon in Tablf .13 for other mienibers in the housebold in two subamples, sualarie workers and sef-emplo. -dl n~h mlal bypod_b Iw do lHe*~ meiod is go binormll ofdie do iMnmm If Uds hypodS is not vified di d inm g _ ive Iomeo estm Ths might be c of do Fue 7.4 Influence of Now Entants on Education Yieds in the Wap EquAtions Yield Wage functon in 1984 Wage function in 1983 New entrans. 1983-84 Education Table 7.10 War Equtiow in Bogota lotaapt Edno Exp Bxp2 R2 N Men 4.26 0.119 0.068 -0.0010 ).329 2Z16 (28.9) (18.5) (13.4) Woma 4.29 0.099 0.055 40.0012 0229 1047 (15.8) (7.6) (5.5) ToWl 4.23 0.114 0.067 .0.0010 0.323 3264 (M a)l2.7) (14.S) Noda: - SiIar ooaveado than Tabl 7.9 now 2. Dcodem varble - kg of mctb1y howes. Swudo: 1es in b(1k97. Source: Knglw gL AL (1979 Table 7.11 Mafied Women's Potential Wage Equaoms, 1905 lahac Exp Exp2 Minl R2 Mea 1963 ^ 24 0.17 0.039 -.90058 - 0.387 4.81 (N=- ;17) (24C0) (4.7) (3.6) 1.49 Cc174 0.046 -0.00077 O.4A 0.393 (B).9) (5.4) (4.5) (3.4) 1981 3.09 0.135 0.023 -0.00033 - 0.330 5.01 0q=116S) ~~~~(20A) (2.7) (2.0) 2.20 0155 0.031 -0.00052 0.50 0.337 (17-6) (3.5) (3.11 (3.5) 1982 3.30 0Q132 0.030 -0.00046 - 0.380 5.36 (N=210Q)(1) (S.l) (4.1) 2.45 0.152 0.039 .0.00.=7 0.44 0.387 (.CJ) (6.4) (5.7) (5.0) 1983 .3.3 0.Ql}0 0.030 -0.00045 - 0.371 5.54 (N-252S) (D33) (6.2) (5.1) 2.40 0.161 0.u42 -0.00072 0.57 0.381 (30.4) (7.2) (U.19 (6.7) 1984 3.44 0.144 0.033 -0.00050 - 0.35S s.72 (N=3 ,r - 0'JO (7.2) (6.1) 2.34 0.169 0.046 -O.O.O03 0.72 0.378 .tRI;[} (8.1) (7.5) iO 1985 3.76 Q136 0.029 -0.00040 - 0.378 5.91 (N=2571) (343) (6.0) (4.9) 3.14 0.151 0.032 -4 ' 047 040 0.333 (29-7) (5.6) (4J) (4.5) Note.: fendent vazi1bie -g (map), Mom - mn(loE wage). St tes ddc is bmdz. The uvimation of husbands? wz equaions show that the educazion yield is less important for men than women but, on avezrg, wage rates seem to be higher. These difflrenves would seem to indicate that Odisc..r1m tin bwtwtzn nmel and women is iower in hiqhcr educa:icn groups. However, it seLs foith tho proUn of the appruximadtion of the variable work expei-ce by the difference (age-educatloa). Woin have not only a lower probabiity of participation bta also lowea levels of experleece t (ago-education). In order to show the impotance of bis approximation, we deveop a vcy simple model. Assume that the true model is Om by the wage equation where EXP2 was deleud for the sake of simplicity. ~o() = a + b.EDUC + eXP + v Table 7.12 Husbands Wage Equatios 1981 1982 1983 1984 1985 line=ept 3.33 3.43 3.S4 3.68 3.72 EAuc 0.121 0.126 0.129 0.132 0.133 (41.0) (S6.3) (66.7) (73.5) (67.4) Exp 0.036 0.040 0.04S 0.040) 0.044 (9.2) (12.7) (16.7) (15.3) (15.3) Exp2 -0.00045 4.0C048 -0.00055 -0.00045 -0.00050 (9.2) (12.3) (10.5) (10.4) R. 20.298 0.320 O.S53 0.3S6 0.357 N 4048 6781 7698 9208 7771 Meao S.32 5.61 5.86 6.00 6.14 Notes: Dqpco&zA ioWbk - WS of hourly wag ra. StAdew tets iai brakw A priori we have b > 0° c > 0. Esdmated wage equations are however given by the approximated model: log(w) = d + e.EDUC + fX + u srith X=AGE-EDUC but u=c(EXP-X) + v is dearly correlated to EDUC and X. This endogeneity problem should imply that the estimates of d, e, and f are A,iased estmates of a,b,c. So as to estimate the direction of the bias, let us write the auxilialy regression EXP-X = C + A.EDUC 4 B.X + t lhen if E(v I X) = O is assmed, the estimates of d, e, and f are unbiased esdmates of (a+c.C, b+c.A, c+c.B) because the true model can be rewritten: Wn(W)= (a+c.C) + (b+cA) EDUC + (c+c.B).X + (v+ct) Finally, an heristic argumet shows that A > O and B < 0. As a maer of fact, if X is fixed, the negative difference (EXP-X) describing the oppoite of the time spent out of the market ,acreases as education increases because labor participation increases with education, with fixed 'K. Hence: Cov(EXP-X,EDUC IX) > 0 and A > 0. Moreover, if EDUC is fixed, the negative difference (EXP-X) decrease as X increases since the participation rate is less than one. Then Cov(EXP-X,XIEDUC) < 0 and B < 0. Concluding, the esdmator of the education yield in the regressions we used is an upward biased estimator of the true yield and the esdmator of the rienc-. field is a downward biased esdmator of the true experience yield. These biases go in the same direction as the differenes between malo and female wag functions. These differecs mlght thus be a staically uriou Tabk 7.13 Other Members' Wage Equations. Itrcept Edne Exp Ex9 Sex R2 N Meow Wage earnes 1981 3.03 0.151 0.050 -0.00071 -0.392 0.490 2,951 4.71 (47.8) (15.5) (11.6) (18.8) 1982 3.27 0.151 0.053 -0.000-76 0.40 0507 4,835 4.97 (63.7) (20.7) (15.2) (24.6) 1983 3.36 0.158 0.052 -0.00068 40.40 0.52ti 5,395 5.16 (70.0) (19.8) (12.8) (24.8) 1984 3.75 0.147 0.04S -0.00062 -0.4C- 0.495 5,711 5.37 (67.0) (17.4) (11.8) (26.0) 1985 3.67 0.156 0.052 -0.00075 -0.38 O.S43 5,062 S.52 (70.0) (19.9) (14.1) (24.2' Sdf-employed 1981 3.42 0.120 0.050 -0.00065 -0.39 0.238 377 4.81 (9.3) (5.1) (4.4) (3.9) 1982 3.53 0.129 0.035 -0.00041 -0.23 0.222 651 5.14 (7.2) (4.2) (3.2) (2.8) 1983 3.67 0.131 0.033 -0.00040 -0.20 0.282 724 5.39 (16.2) (4.5) (3.6) (3.0) 1984 3.74 0.129 0.028 -0.00034 -0.11 0.250 845 5.52 (IS.9) (3.7) (2.8) (1.6) 1985 3.94 0.110 0.042 -0.00OS6 -0.17 0.119 n.d. 5.36 (10.1) (4.4) (3.6) (1-9) Notes: Dependet vfible - bg of bowy wag ratr. StWedt tcku in brackt. arfct related to the bed measure of the true market experience. Ihe education yield for women may in fut be leu than 15 percen In order to correct for this experience bias several methods are possible but it is ncmary to have panel data The analysis of the wag funtons for other members in the family show that no significant differences appem In the human capital yields. In contast, the sex variable is significant and for salaried workr; women earn 40 percent less than men, all other dtinp being equal. Discrimiaon thus seems to oe very Important. But it must be nodced that these results are valid for sdariod workers but less so for the self-employed. Among .ie latt, wages are explaind loss by human capital variables. It is possible that the sef-employed wages have larger varations across ime than salaried workers' wages. S. Conlusions In this paper, the results lead to some firm conclusions but also pose some questions about the basic hypothesis of the model. First, the results of the participation model seem robust and stable across time. The results show the importance of the classical effects of human capital variables or incomes on the labor participation of married women. They also permit us to measure the subsdtution effacts within the hczseDold. Nevertheless, even if 25 percent of the increase in participation is pzedicted by supply effects, such as increasing average education of decreasing fecundity, its explanatory power remains quite small in cross-section aid rather mild in time- series. The estimation of wage equaions is usual but omits variables related to occupations or to the demand side of the labor market The latter seem to be an important determinant of the evolution of the labor market (Magnac, 1991). In particular, the segmentation hypothesis of the labor market should be considered. However, in this case the estimation of wagf equ3tions by rerely including occupational status variables is plagued by major biases since those variables are endogenously determined. Occupational stats is chosen at the same time as participation. It is thus necessary to use a more complete model so as to treat it in a more rigorous way. -- -J-. - . - -. - Appendix 7A Ho.sehold Composition Variables Left Out of the Analysis of Female Participation Number of persons in the fan;' Number of children of the houenold's head aged between 3 and 12. Number of children of other members. Number of acdve adult men and women. Number of students aged between 12 and 18. Number of unemployed women. These variables were left out of the analysis because they were not significant in all samples except the number of persons and the number of active women which were left out because of a strong colinearity with other members' income (IOTR). Appendix 7B Presentation of the Data The data used in the estimations come from the yearly Encuestas de Hogares (EH) from March 1980 to 1985 (EH26 to EH46) collected by the DANE in the four major cities of Colombia, BogotA, Medellfn, Binranquilla, Cali, and in the smaller towns Bucaramanga, Manizales and Pasto. The DANE in recent years aims to include suburbs in the surveys but it was not the case in 1980 to 1985 surveys. The survey methods are homogeneous in the period under study, with the exceptions of March 1981 in BogotA where the sample is much larger and of March 1982 for the three smaller towns for the same reason (Estudios de Poblacidn). Generally speaking, the main questionnaire consists of questions related to individual characteristics on work, income, etc., but it is easy to construct household variables from the survey. These surveys or similar ones have been studied by Ayala (1981) who compared the results to a survey undertaken by the CEDE (Universidad de les Andes, Empleo y Pobreza, 1978) in BogotL Differences are rather mild, but the Encuestas de Hogares seems to underesimate the number of children in the family and domestic services as well. Similarly, it seems that partial work is underreported, in particular by unemployed people. Another possible criticism is the sampling strategy based on the 1973 census. The latter is not renowned for its coverage. However, the DANE reactualises these predictions by cartographic methods. Nevertheless, as suburbs are left out, no coverage exists for the districts called barrios de invasion setting up very quicldy. The poorest families are surely missed. The sample that we selected retains the following criteria: The family must be composed of a male household head and his wife or com;anion, the latter being aged between 18 an4 60. The number of households varies between 5,030 and 10,000 (Table 7.6). Generally left out are 20 to 25 percent of the households present in the whole sanple. All household's variatNes have been constructed from individual observations by counting methods. References Agular, N. 'La mujer en la fuena de trabajo en la America Latina un resumen intoductorio.* DeJarrol/o y Socddod, Vol. 13, January (1984). Angulo Novoa, A. and LApez de Rodrfguez, C. Trabjo yfeavididad de la mzer coamblan& Fedesarrollo: Bogoid, April 1975. Ayala, U. Comparaciones nertemporals de estadlfcas jobre ferza laboral. Bogott Universidad de los Andes, CEDE, 1981. . El emplko en las grandes cludades colombfanas. BogotA: Univetsidad de los Andes, CEDE, 1981. Bayona, A. 'El descenso de la fecundidad y su impacto sobre la participacidn de la mujer en Is atividad en Colombia3 in Impllcaclones socdoeconOinlas y demogr i de dercenso de lafecwiddad en Colombia, Vol. 18, April (1982). Berry, A. and M. Urrutia, Income Dstrbution In Colomba. New Haven, London: Yale Univasity Press, 1976. Bonilla de Ramnos, E. La Madre 7Taoadora. Document 66. Bogott Universidad de los Andes, 1981. Bourguignon, F. Participation, emploi et travai domestiques des femmes mariees.' Consommation, Vol. 2 (1981). pp. 75-98. .'he Labor Market in Colombia.' Washington D.C.: The World Bank, Report No. DRD1S7, 1986. Bourguignon, P., P. Gagey and T. Magnac. On Esdmating Female Labor Supply Behavior in Developing Countries.' Doc. LEP, Vol. 103, Jamury (1985). pp.41. Ckcers, 1. Algunos aspecsos de la skuaddn de la m4iJer lJadora en ColombIa. Unpublished Dlssertation. Bogoti Universidad de los Andes, 1977. Callavet, P. Allocation du tcnps des menages a Bogotd, CoJombt. Unpublishbd Disettaton. UnlvmityofParWs, 1981. 194, Wornw 'J E.VIymM and Pay bi Labi AerasC Castafieda, IT. "Ddenin del cambio poblacional en Colombia. Desarroflo y SoCedad, Vol. 4, July (1980). -. 'La participacidn de las madres en el mercado urbano en Colombia. DesaroUo y Socledad, (1981). Departamento Administrativo Nacional de Estadfstia (DANE). Boletines de estadstica. Bogota. -. Mewodolog(a de las anuestas de hogares. Bogott, 1986. -. Colombia Estad(stica 86. BogotA, 1986. Fields, G. and T.P. Schultz. 'In )me Generating Functions in a Low Income Country: Colombia.' Reviw ofl come and Wealth, Vol. 28, no. 1 (1982). pp. 7187. Gomez de, M. I., B. Kugler and A. Reyes. Dternwantes econ6mlos y denogr4ficos de la partlc;act6n laboral en Colombia Bogoti CCRP, 1981. Gourieroux, C. Economer des variabls qualftaves. Paris: Economica, 1984. GutiErrez de Pineda, V. Esuctura, fincdn y camblo de la fizmilla en ColombiL Bogoti ACFM, 1975. Heckman, J. Sample SeleLtion Bias as a Specification Erro9r.- Economkrica, Vol. 47, no. 1 (1979). pp.153-161. Kugler, B., A. Reyes, and M. I. de Gdmez. Educacldn y Mercado de TRabjo Urbano en Colombia. Bogot± Monogrffas de la CCPR. Vol. 10, 1979. Killingsworth, M. Labor Sqpply. New Jersey: Cambridge University Press, 1983. Kugler, B. Influencia de la educacion en los ingresos de trabajo: el caso colombiano.' Rev. de Planeaci6n y DesaroUl, (1971). Leon de Leal, M. La mujer y el desarrollo en Colombia. BogotA: ACEP, 1977. Magnac, T. Analyse de l'offre de travaU sw un march concuwwzdl ou segmente. Unpublished Dissertation. Paris: EBESS, 1987. '. Competitive or Segmented Labour Markets?' Economevica, Vol 59 (1): 165-187, 1991. Mohan, R. lhe Dsennlnww of Labor Eanings In Dewloplng Metropolis: Estimate fiom Bogotd and Call, Colombim Washington D.C.: The World Bank, 1981. Munoz, C. and M. Palacios. El niflo trabajador. Bogota: Carlos Valencia Editores, 1980. RLnis, 0. 'Distribucidn del ingreso y crecimiento en Colombia. Desarrollo y Sodedad, Vol. 1, January (1980). Rey de Marulanda, N. El W o de la mx4er. BogotA: Universidad de los Andes, CEDE, 1981. Fmak Labor Maw*ra Park*adou and Wagus LCombia 195 Rey de Maunanda, N., U. Ayala, M.C. Nio and P. DurnL Ehpleo y pobra Bogot: Universidad de los Andes, CEDE, 1978. Reyna, J. V., H. 0. Buendf and C. C. Argaez. Desarroio socia en la d&ada del 70. Bogotd: UNICEF, 1984. Sen, N. and J. Gomul}a *lbe Employment of Married Women in the U.K. 197083.- Econflca, Vol. 571 (1990). pp. 171-200. Urrutia, M. Wnners and lawr In Colmbia's Econonic Growth of dke 70's. London: Oxford University Press, 1984. Chapter 8 Women's Labor Force Participation and Earnings in Colombia Edurdo Velez and Caroyn Wiaer 1. Introduction This chapter contributes io the rather small literature on factors influencing female labor force participation and earnings in Colombia. The movement of women from the home to the workplace is generally seen to be an indicator of increasing sex equality in society since it implies improved access to education by women and reduced fertility rates. While women's labor force pan sipation rates have increased substantially in Colombia (from 19 percent in 1951 to 39 percent in 1985)' relatively little is known about women's work experience, their occupations, or their earnings relative to men. In this chapter we address the following questions: What factors influence a women's decision to participate in the labor market? Are human capital indicators lower for women than men? and, What accounts for the earnings differential between the sexes? The following section briefly describes the Colombian labor mar'-et. Section 3 descnbes the characteristics of the sample used in this analysis and Section 4 identifies the most important determinants of women's labor force participation. Section 5 presents earnings function estimates for male and female workers respectively, allowing us to examine earnings differences while controlling for human capital endowment3. In Section 6 we decompose the earnings differential into the portion attributable to differences in productivity related variables and the portion attributable to 'unexplained factors Oargely differences in the way employers reward male and female workers). A discussion of these findings and their implications for policy formulation is presented in the final section. 2. The Colombian Labor Market A wealth of resources, extensive industrial diversification, and prudent fiscal management has led to sustained economic growth, averaging close to 5 percent per anmun since the 1960s, and the continuing real growth in real incomes. In the last few decades the country has experienced a rapid social transformation that has affected the structure of the labor force and labor-supply behavior. In fact, the urban share of the population increased from 31 percent in 1938 to almost 70 percent in 1985; total fertility rates ILO (1990). 197 198 Womuam t enF ymum and Pay i Latn Anwka declined by about 45 pert from the early 1960s and are cumently estimated at about 3.5 percent; maternal mortality that was 254 per 100,000 live births in 1964 was 107 in 1984; primary education enrollment more than doubled, and secondary education enrollment increased six-fold since the 1960s; and a substantial modification of the sectoral distnbution of the labor force occurred - the agriculural sector acounted for 57.2 cnt of the labor force in 1950 and 34.3 percent in 1980, the industrial sector for 17.9 percent in 1950 and 23.5 percent in 1980, and the service sector for 24.9 percent in 1950 and 42.3 percent in 1980. A significant change in the Colombian labor market over the last few decades has been the increase in women's labor force participadon from 19 percent in 1950 to 39 percent in 1985. Women contne, however, to be heavily represented in the informal sector and it is estimated (Federico de Alonso, 1990) that 64 percent of working women were in the informal sector in 1990. In terms of educational achievement, gross enrollment ratios at primary and secondary education are about the same for boys and girls. Even in higher education women show good standing reladve to men; in 1986 the enrollment ratio for higher education as a whole was 13.1, and was 12.6 for women. Since the end of the 1970s more women than men have been attending higher education (DANE, 1985). However, field of study varies significantly by gender, with women being found in educational tracks that lead to low-paying careers. The average education of labor force participants has inreased substantially over the past 30 years; more than 40 percent had no education in 1951, only 8 percent had gone beyond pimary education, and the illiteray rate was around 10 percent. he average educational level of the labor force has more than doubled since the 1960s; an impressive change. 3. Sample Characteristcs The data used in this malysis are from the June 1988 Natonal Household Survey conducted by the Departamento Administrativo Nacional de Estadftica (DANE) in the largest Colombian cities.2 The survey covers about 75,000 individuals aged twelve -,pars or older in more than 20,000 households and provides detailed data on indivil socio voonomic and labor status. A 10 percent random sanwle of households was selected for use in this analysis. As we were nrimarily interested in prime-age workers, we retained in our dbsample individuals aged 15 to t0 years. Table 8.1 shows the mean diracteraistcs of the saziple by gender and, for women, by work status. Individuals were clLisified as working if they were employed in the formal sector, reported positive earngs and worked more dtan 2 hours a week. Within the sample of working males and females, individaals who reported earning less ta 10 percent of the mean hourly wage or more dtan 15 times the mean hourly wage for their sex were excluded. Tlis procedure resulted in our dropping nine cases from the sample in which reported eamings were over three standard deviations from the mean. The sample used in the analysis was composed of 3,163 working males, 1,748 working females and 5,735 non-working females. The female participation rate in the sample wbb 25 percent Th sample is repreivtave of ColomLia's urban populatn and the socio-ecnomic composition of each city. The citie and metopolitan areas included in the smple ar BogotL, Medellfn, Cali, Bam ,qilla, Buc_aFang C Cwtagis Cucuta. Manizles, Paso, Ibagne, Pereir and Villavricencio. Won 's Labor Force Pardc and FAmbtgs Ix Cok/1 ba 199 Table Ll Colombo Mos (and Stndad Devato) of Sample Vibles Working Woking Non-Wordng ia ctAstlc. Males FemalP Femaes Age 34.2 32.7 31.4 (11.16) (9.59) (11.39) Married () 64.2 38 4 47.4 (0.48) (0.48) (0.49) I Chi] Jr under 6 yea .63 .S1 .59 (0.88) (0.84) (0.85) Head of Houshold 65.3 19.7 6.8 (0.47) (0.39) (0.25) kftion Yeas of Schooling 7.6 8.7 7.1 (4.08) (4.22) (3.61) Level of Educaion (%): No formal educabon 2.6 1.8 4.4 Incomplde prmry 16.3 13.3 18.6 Priary 20.5 16.5 19.5 Incomplete econda 30.9 26.0 36.3 Seondry 17.0 24.0 13.7 Incomplet universty 5.3 9.1 5.4 University 7.5 9.5 2.1 Weekly Eaminp (po) 10,727 9,078 (13,114) (9,766) Years of Expevence 20.5 18.0 (1247) (11.25) Hours worked (weekly) 49.9 46.1 (12.07) (11.44) N 3,163 1,748 5,735 Notes: Figure puaen m sndad deviacos. Female p o Ra =2 p- Sounce: ol Houold Survey, 1988. Working women bave, on average, one and a half years more schooling than non-working women and approximatly one year more schooling than woring males. Working women are also more likely than working men tD have completed seondary schoolig and ewr ae d or completed higher educadon. Despite this, working women's weeldy earning are, on average, only 84.6 percent of working men's (9,078 pesos compared to 10,727 pesos). This earnings differential is not comploety explained by !ie slighy fewer hour worked per week by women; if we estmate average hourly income women ar approxmately 9 percent less than men.' It is pose that gender differnce in labor mart expeiece nay acoownt for om pat of his eaning differential. However, a variable 'years of boW ma rk experienoo' -s boe coasduct by surting an individual's yw of education plus dix frm his/her age, u per Reckmn (1979) and is callquenty notn cab i of eperience. It u likely to overediab wom's experiene since they withdraw from the lbot Ikt more feuntly thn - nd for onger periods becuse of childbewing. 200 Women'S E'niy g 5ph wd P4y 6n LdAt Amhica Table 8.2 Occupatidc DistnrmtioDn of Workr by Gendgr, Formal Setor, 1988 Mesa Mean Aean Weekly Weekly Weekly War Maio Wage Femles Wag pess) Occupation (5) (peso0) (%) (pesos) A Workers Professional/technical 7.7 24,75 12.5 17,418 21,293 Administrtive 1.3 38,440 0.9 25,751 34,979 Clerical 10.5 9,4S 22.9 9,549 9,507 Sales 20.4 11,556 20.7 7,959 10,264 Service 9.7 8,829 21.1 6,744 7,682 Agricultual 2.1 14,991 0.7 13,643 14,784 Laborer/Opr Ov 48.5 7,893 21.4 6,270 7,575 All Occupations 10,726 9,077 10,139 It is interesting to note in Table 8.2 that almost half of all male workers in the formal sector ae employed in the lowest paid occupation Oaborer/operatve). Women are, however, more heavily represented than men in the next two lowest paying categories, service 'nd clerical. Women's average earnings are lower than men's in all occupational categories except clerical. 4. The Determinants of Women's Labor Force Parddpaidon Given that female workers average one more year of scbooling than working men but that they earn, on average, only 84.6 percent of men's wages, we are interested in deermining what part of the earnings differental is actually due to differences in buman capi2al endowments and what part is 'unexplained by these factors. Tis 'unexplained' component will largely reflect differences in the way employers reward male and fema'e workers.' However, we are faced with special problems in estimating earning functions for female workers. The problem arises becase a woman's decision to participate in the formal labor market is influenced not only by her market wage, but also by the value she accords her work in the home (i.e., hGr reservation wage). In general, a woman's reservation wage is likely to be the highest (and hence her probability of participation in the labor market, lowest) when she has young children for whom to care.5 If we estimate earnings functions for working women we wil be using a self-selected sampla (women whose market wage exceeds the value of their time in the home) and our esimates wUi yield biased results. To correct for his selctivity we esdmat a probit model in which the 4 This unexplaid' c om wt is U ney tak e to reprlt the *Uppt bouzd' to di since other factors arm also lika y to ocmtnbt to this 'unexplained' comoten If, for insance, we omit explanatory variables frow ~nb arninps equatis the estinute of disciriminaton will be biased upwards 5 ln this stdy we asume tat p_img males do not have the eame options regarding labor for participation as females Males hav traditioilly been viewed as provide fr the humily. Females, except where they ae haods tf househlds, have had the option of withdrawing fiom the labor market to undertako childrearing and ho_cma activide. Wm 'j Labor Force PariJ49 and Earnigs in Cokonbia 201 probability that a woman will pardcipate is esdmate given her paental status, age, educa- tioal level, the size of the household in which she lives, and her status as head of household or otherwise.7 These probit estimates are presented in Table 8.3. To illustrate the magnitude of the probit coefficients we estimate simulations predicting female participation rates for each coT3'Lkin while holding al other variables at the value of their sample mean (see Table 8.4). Table 8.3 Probit Pemtes for Female Participaton Pau Variable Coefficicat t-ratic Mean Derivative C nt -1.927 -15.96 1.000 Age 20-25 U.685 11.55 0.234 O.z07 Age 26-30 0.862 13.55 0.151 0.261 Age 31-35 0.842 12.18 0.103 0.255 Age 36-40 0.884 12.09 0.103 0.252 Age 41-45 0.705 8.92 0.067 0.213 Age 46-50 0.562 7.01 0.068 0.170 Age 51-55 0.382 4.14 0.051 0.115 Children (0-6 yrs) -0.134 -3.53 0.393 -0.040 Household Size 0.018 2.46 4.418 0.005 Fewale Household Head 0.753 14.07 0.;ll 0.231 Incomrplete Pimary 0.260 2.32 0.172 0.078 PrimarY 0.400 3.61 0.187 0.121 Incomplete Secondary 0.382 3.50 0.336 0.115 Secondary 0.833 7.43 0.162 0.252 Incomplete University 0.767 6.20 0.063 0.232 University 1.309 9.95 0.039 0.396 Notes: Dependet Variabis Labor POrce Pticipaion Sample Women Sged 15 to 60 yers Mean Particuption Ratc 25% Log-Likbifhood = 3476.3 Schooling is ened as a series of dummy variables for each le el of schooling. The probit coefficients in Table 8.3 show tha the probability of particpating increase steadily with each successive level of education smccessfully completed. The lxtent to which additional education increases the probability of participation is evident in Table o.4. A woman with the mean values of all other characteistics and completed secondary schooling has a r-edicted probability of labor force participation 7 percent higher than a woman with completed pimary school (probability = .34 versus .20). A woman with completed university has a predicted probabiity of participation I It should be noted dat oe data only provide information on wumber of children aged 0 to 6 yws by household. Whor there is moe (in one wom in s household, it is not possible to determine to which woman the childrn belong. We tbseform lose some of the explanatory power of this variable. I mid metbod was dpeloped by HOcmln (1976) and has been widely used. See, for example, Gronau (1988). 202 WcmruI E'A pr, and iy Ladfn Ama 56 percent higher than a woman with completed secondary sc!ooling (probability = .53 versus .34). Two variables controlling for hmusehold effects are included in the probit model, household size (a continuous variable) and whether the woman is the head of the household (entered as a dummy variable). Lauger household size is shown to have a positi:e, but very small effect, on a woman's participation deciion. By contrast, being a household head has a substantial impact. A woman with the mean vaues of the other chrateristics but who is a housebold head has a predicted probability of participating of .47 compared to .21 for a woman who is not a household head. Many studies have shown a womaz's pardcipation decision to be strongly influenced by family stucture, particularly if she is the mother of young children.' Ihis is also found to be true in Colombia where the prece of young child-en (aged 0 to 6 years) is shown to reduce the probability that a woman wi participate. A wvoman has a predicted probability of partk.ipa!ing of .20 if there are young children in the household and .25 if no young children are present Tatle 84 Predicted P-¶icipation Probabilities by Chamcteri;tic cactumutic Predicted Probability 18Y0 pdmuy f.11 No Educatim 0.16 prinry 0.20 Inc=Vlde Secoedary 0.19 Secesday 0.34 Icomlet University 0.32 Ux'%erzity 0.53 Presec of Children (0-6 vearsl No 0.25 ys 0.20 :.mlyHdd Hm2d NO 0.21 Yes 0.47 Overl Men Paticipaton Rat 0.25 Note Pubabiky of paricii while holmng other variabks eoa at thir sample mca. S ee, for example, te on Xador, Ven=el (1989), and Argetina in this volum Woax,, Lbo'r Force PW*4mmo and Eaxahp ln Coe 2Q S. Earnings Funcln We estmated earnig% fanctions for me and the 1,748 women in our sample who were labor force participants. 'Me regression estimates based on the standard human capital modl where the dependent variable is the log of weekly eanings and the idependent variables are expdence, years of schooling and log weekly hous worked.9 The experece proxy is entmred as a squared term to test if the earnings function is parabolic in the experience term Table 8.5 Eanin P fi Womm Woar for for Variable Men Selectwiity) sehlvity) (1) (2) (3) Constant 5.662 6.115 5.66 (31.432) (2456, (26.93) Schooling (years) .120 .099 .112 (35.41S. (17.15) (2.37) Log Hours .426 .447 .457 (9.515) (8.71) (8.88) Experience .046 .027 .035 (11.941) (5.12) (7.58) Experience -ured -.000 -.000 -.000 (-6.270) (-2.64) (-4.49) T mbdac -.206 (-3.29) R2 .304 .299 .294 N 3,161 1,748 1,748 a. Conltd for bctivybax usingprobk e fDrpobabity of bor ptrt work in Tsbk.3. Enrors corfected hr the use of an mverc Mils rtio. b. Not corrected for sectvivy bias OLS using the subsapl of wotking women C. mInvcer Mils Ratio calcultd uming probi u for the psobabliy of workag in Table 8.3. Notes: Dependent varbiab - log (weekly earing). t-values ae in padhess The first column of Table 8.5 presents the results for the male sample. The rate of re to schooling is estimated to be 12 percent which is consistent wimn previous research on Colombian r$ an labor markets. The log emrin increase with experience but at a decreasing rate, as is expected in a normal age-earnings profile. 9 See Mincet (1974). Io See Mohan (1986) mad Pn&ocWo and Velez (1991). TabLe 8.6 Decouposition of the Male/Female Eanings Differential Pecntage of Male Pay Advantage Due to Diffecs in Male Pay Specification Eldowmets Wage Structmu Advantae Correted for Seectivity Evaluat at Male Mens 14.81 (2-28) 85.19 (13.12) 100 (15.4) Evaluated at Female Means 8.02 (1.23) 91.98 (14.16) 100 (15.4) Uncorrected for Slectivity Evaluated at Male Mewas 22.14 (3.41) 77.e6 (11.99) 100 (15.4) Evaluated at Femle Meam 12.31 (1.89) 87.68 (13.5) 100 (15.4) Notes: WnJWt = 118% Figures in parebsesm am pa ges sbowiag the male pay sdvarsage. We estimate two earnings functions for women. One uses the standard Mncerian model and the other 'corrects for potetil selectivity bias by including the Lambda from the probit equation. Ihe selectivity corrected estimates in column 2 of Table 8.5 show the rate of return to schooling to be about 9 percent, less dtm the 11 percent from the uncorrected esdmates. Hence, if we omit the selection term from the earnings function estimates we would be biasing the marginal rate of return upward. 'Te significant and negative Iambda indicates that dt: -e is a strong positive correlation between due unobserved chaaceistics which are likely to mnake wcmen highly productive in both the ma and the home. These unobservables are, however, the characteristics likely to ience women to remain in the home. 6. Discrimination As was noted in Section 2, working women in Colombia earn, on average, 15 percent less per week than working men. Using the Oaxaca decomposition method we are aWle t- decompose this into a component due to differences in human capital endowments and a component due to .unexplained factors' (which principally includes differences in the labor mFAet structurp for men and women, i.e., discimination)." The standard Oaxaca decomposition method expresses the difference between the mean log) wage rates of males and females as: LnY, - LnY, = Xt(b - B) + bm(X. - Xr) (la) or, alternatively as: LnY, - LnYf = X,9, - b) + b?(X. - Xe (lb) See Om" (1973). There is an index number 2,:bo!:uL here but there is no advantage to choosing one equation over the other. Consequently wa pyesent the results of both in Table 8.6. The first term in both equations is the part of the log earnings differential attnbutable to differences in the w2ge structure between the sexes and the second term that is part of the log earnings differential attributable to differences in human capital endowments. Although we estimate the decomposition for both the selectivity corrected and uncorrected samples, the former yields the more reliable estimate since it essentially uses women's offered wage (being estimated from the entire sample of women) rather than the paid wage (estimated using the sanple of working women only). The male pay advantage is 15.4 percent. Using the selectivity corrected estimates evaluated at the male means in Table 8.5, approximately 14.8 percent of this pay advantage is explained by observable factors, or differences in human capital endowments. The rest of the difference (approximately 85 percent) is due to differences in thte way males and females are rewarded in the labor market. 7. Disaussion Female labor force participation rates are shown in our study to be positively influenced by education. However, women are largely concentrated in occupations which are lower paying and have fewcr opportunities for advancements. Prior studies in Colombia show that women pursue educational tracks which lead them to these occupations (Velez and Rodriguez, 1989). The findings also show that being the head of a household greatly increases the probability that a wom!an will participate. The earnings differential of 15.4 found in our sample is surprisingly low, even when compared with those in many in industrialized countries.Y2 This may be partly explained by the exc:usion of non-formal workers from our sample. Tenjo (1990), in an anal3 sis of Bogota's labor force in 1979, reported the wage differential to be closer to 30 percent wL?n informal workers were included in the sample. The presence of minimuml wage legislation, firmly enforced in the formal sector, may prc-!ide another explanation for this small earnings gap. When e:-plaining eaning, we found evidence of selectivity bias in the determination of weekly wages, pointing out that traditional ordinary least squares (OLS) coefficient estimates are biased upwards for women. Although human capital chactiscs are relevant in explaining earnings, the Oax.a decomposition suggests that differences in the labor market structure are more important than differences in human capital endowments ia explaning male-female wage differentials. Hence much of the eanings differental can be atuted to discrimination. Future research should study issues influencing women's choice of education field as this appears t n imporant factor affecting their income levels and occupational opportunities. Another aspect that should be considered is the sitation of female heads of household as they face more constraints to increase their participatien. t2 Eaings diffemntsls aro typially arornd 25-30 pemen. See Guadersoa (1989), Tznato (1987, Zabala and Tzanate (1985) and Cloty and Duncan (1982). Refernces Departamento Administrativo Nacional de Estadfstica (DANE). SO A7os de Estad(stlcas Educatfvas. Bogota Departamento Administrativo Nacional de Estadfstica, 1985. Gregory, R.G. and RPC. Duncan. "Segmented Labour Market Theories and the Australian Experience of Equal Pay for Women.' Journal of Post-Keynesian Economics, Vol. 3 (1982), pp. 403-428. Gronau, R. 'Sex-Related Wage Differentials and Women's Interrupted Labor Careers: The Chicken and Egg Question.' Journal c,'LaborEconomics, Vol. 6, no. 1 (1988), pp. 277- 301. Gunderson, M. 'Male-Female Wage Differentials and Policy Responses." Journal of Ecoromic Literature, Vol. 27, no. 1 (1989), pp. 46-117. Heckman, J. "Mhe Common Structure of Statistical Model Truncation, Sample Selection and Lin,ited Dependent Variables and a Simple Estmator for such Models." Annals of Economic and Social Measurement, Vol. 5, no. 4 (1976), pp. 679-694. Heclman, J. *Sample Selection Bias as a Spechicaiion Error." Econometrica, Vol. 47, no. 1 (1979), pp. 153-161. International Labor Office. Yearbook of Labor Staistics: Retrospectie Editon, 1950-1990. Geneva: International Labor Office, 1990. Mincer, J. Schooling, Eperience and Earnings. New York: Columbia University Press, 1974. Mohan, R. Work, Wages, and Welfare In a Deveoping Metropolis. Consequences of Growth in Bogota, Colombia. New York: Oxford University Press, 1986. OAxaca, R. "Male-female Wage Differentials in Urban Labor Markets.' Internsdonal Economics Review, Vol. 14, no. 1 (1973), pp. 693-709. Psacharopoulos, G. and E. Velez. 'Schooling, Ability and Earnings in Colombia, 1988." Economic Development and Cultural Ozange. forthcoming, 1991. 206 - -.- - -. -, -.- 6 - - Rico de Alonso, N. 'Caraterfsticas y Condiciones de la Participaci6n Laboral Femenina a Nivel Urbano en Colcmbia.' Paper presented at the Workshop on Mujer y Participacidn Laboral. Bogota, April 1990. Tenjo, J. 'Labor Market, Wage Gap 'd Gender Discrimination: The Case of Colombia." Mimeograph. University of Toronto: Department of Economics, 1990. Tzannatos, Z. 'Equal Pay in Greece and Britain." Industrl Relations Journal, Vol. 18, no. 4 (1987), pp. 275-283. Velez E. and P. Rodriguez, 'Mujer y Educacion en Colombia." Mimeograph. Instituto SER de Investigacion. BogotA, 1989. Zabalza, A. and Z. Tzannatos, Women and Equal Pay: Ihe Effects of Legislafton on Female Employment and Wages In Britaln. Cambridge: Cambridge University Press, 1985. 9 Female Labor Force Participation and Earnings D.fferentials in Costa Rica Honeu Yang 1. Introduction Over the last decade int in the treatment of women in the labor markets of developing countries has increased dramatically, and more attention is being paid to analyzing the labor force bebavior of women and the returns to human capital, especially education. Do women and men enjoy the same returns to human capital? Is there an earnings differential between working men and women? In the case of Costa Rica, the evidence shows that such a gap exists. Figures 9.1 and 9.2 reveal the existen of significant male-female earrdings differentials across schooling and age. What factors cause this difference? And how do these factors influence a woman's decision to participate in the labor force? In this chapter we try to answer these questions. First, we determine the earnings differetial between male and female workers in Costa Rica. Then we estimate the extent of wage dicrmination against females. In the following section Ne briefly review the economy and labor market in Costa Rica. In Section 3 we discuss the data used in this study and present the main characteristics of male and female labor force participants. In Section 4 we examine labor force participation and the factors influencing women's decio to pardcipate. In Secdon 5 we analyze the result of the male and female earnings functions and the decomposition is carried out in Section 6. In the final section we discuss these findings and their implications. 2. Tle Costa Rican Economy and the Labor Market Foi most of the last twenty-five years economic growth in Costa Rica has generated improved employment oppormuities for workers. However, during the economic recession in 1981-82, labor market conditions deerorad countrywido. Fomunately, recession was the exception rather than the rule in Costa Rica. Gross domestic product grew by 6.5 percent per annum in the 1960s and by 4.5 percent per amnmn between 1970 and 1982. Between 1963 and 1973 the Costa Rican labor force became markedly better educated. Toe proportion without education feUll from 15 percent to 10 percent, the propordon of illiterates fell by virtually the same percentage, and the proportion with only one to three years of primary- education fell from 37 percnt to 26 percent. At the same time the proportion with four to six years of pimary education increased from 37 percent to 45 percent, the proportion with secondary education from 9 percent to 16 percent, and the proportion with university education from 2 percent to 4 percent (Fields, 1988). Between 1965 and 1988 the higher education 209 e1u Womrei 's ?mpkoymenl and Pay in Latin AAtrkca m9.1 Schooling-EAnings Profiles by Gender Costa Rica 1989 (Costs Rkc Colon, monthly) 60.000 40,000 30.006 /0a ao,ooo 10.000 o . . , * . Ifl . * I I i I I 1 I U 0 1 2 a 4 6 e 7 S 9 10 II 12 13 n4 15 16 1? 17 19 Schooling (yeurs) F%Ure 9.2 A&D-Eunings Pofiles by Gedr Costa Rica 1989 (Coatn PA" Coloa mothty) 20.60 - 10,000 / 0,1000. 1I 20 as It S5 40 46 a,& (YOU) Femak Lbor Force Pancdain and Earnfings DiJfferenfials in Costa Rica 211 enrollment ratio rose from 6 percent to 24 percent and the secondary enrollment ratio Increased from 24 percent to 41 percent; primary school enrollment was virtually 100 percent (UNESCO, Statistical Yearbook). Economic growth has brought more employment opportunit.es and created enough new jobs to keep pace with the growth in the labor force. Not only are more people employed, but the mix of jobs has improved in favor of the better-paying categories: Wage-earners in place of unpaid family workers; professional, technical, managerial, and office workers rather than manual workers; manufacturing and other sectors in place of agriculture; public as opposed to private employment. 3. Data Characteristics The data used in this study come from the Encuesta de Hogares de Propositos Multiples (EHPM), a nationwide household survey that is conducted by the Statistics and Census Departnent of Costa Rica. The data were collected in July 1989. The data set contains 34,368 individual observations from 7,637 households. Information is available on personal characteristics of the population such as age, sex, education and area of residence. Employment variables include occupation, job category and hours worked per week. Informztion exists on labor income, other income and family income. From this data set, a total of 15,867 cases were used. This sample included all individuals in the prime working age range (20 to 60 years) for whom relevant data were available. Table 9.1 provides descriptive statistics for the main variables in the sample. Working men and working women are defined as those who worked for more than one hour for pay during the reference week. This definition excludes unpaid family workers. As Table 9.1 shows, the female labor force participation rate (27 percent) is significantly lower than that for men (76 percent). The marriage rate for working females is lower than for workik,g males, and is much higher among non-working females. Working women have less children than non-working women. This suggests that marriage and family have a great influence on female participation. Table 9.1 also shows great differences in the distribution of men and women by education level. Only 11 percent of working men had completed secondary school compared to 22 percent of working women. The largest gap in educational achievement occurs at the university level: Only 5 perceLt of working men have university degrees compared to 11 percent of working women. Overall, the average years of schooling for werking females is two years greater than that for their male counterparts. This was confirmed elsewhere in a recent study (Gindling, 1991) which showed that from 1980 to 1985 the average years of schooling for working women was about one and half years more than that of working men. This is typical in Latin America and Caribbean countries.' However, despite the fact that women have more schooling than men, they earn less than males. Tables 9.2A and 9.2B sbow the education and earnings differentials by occupation and working sector in more detail. Women's schooling is higher than men's in all occupational categories except two: Managers and service workers. Women's average earnings, however, are lower than men's in all occupational categories. In the public sector females have an average of I In Argendna aveage schooling for worldng feniales is 9.4 year, while for working men it is 8.8 year; and in Venezuela, average years of schooling Fie 8.5 and 6.9 for working females and males respectively. 212 Wmen r Employmed and Pay in Lagb. America Table 9.1 Mes (and Standard Deviations) of Sample Variables Variable Worling Males Working Females Non-working Females Hours worked/week 47.64 40.53 (13.33) (16.29) Workdng experence (yeas) 72.5 19.1 I'2.5) (11.4) Primary income/month' 18497.77 14942.14 (18307.45) (14525.66) Family incomelmonth' 30085.94 36826.81 23976.12 28153.41) (35964.48) (23435.96) Age (years) 35.11 33.57 35.68 (10.76) (9.68) (11.49) Head of household 0.74 0.21 0.09 (0.44) (0.41) (0.28) Household size 5.10 5.23 5.20 (2.29) (2.49) (2.23) No. of young children 1.46 1.39 1.52 (1.38) (1.35) (1.42) Urban 0.42 0.61 0.42 (0.49) (0.49) (0.49) Marrid 0.73 0.46 0.76 (0.44) (0.50) (0.43) Years of schooling 6.66 8.47 6.18 (4.02) (4.21) (3.77) No education 0.07 0.03 0.08 (0.25) (0.16) (0.27) lncomplete pr 0.16 0.09 0.17 (0.36) (0.29) (0.38) Completed primary 0.43 0.35 0.43 (0.50) (0.48) (0.50) Incomplete secondary 0.13 0.13 0.11 (0.33) (0.33) (0.32) Completed seconday 0.11 0.20 0.12 (0.31) (0.40) (0.33) University 0.05 0.11 0.05 (0.22) (0.31) (0.22) Graduate school 0.03 0.06 0.004 (0.24) (0.17) (0.06) Sample size 5,463 2,126 5,892 L In current Costa Rca Colom Notes: Labor Forc aricipato Rfe: Peanale 27%; Male = 76% Sample incld aged 20 60. Working popultion consist of all those working. Excludes uwaid fmiy workem Number in paFnthese are stios. Source: Costa Rica 1989 Houseold Survey. FOAals Znbor Force ParIc{alox and Eandngs DoTSridal In Coma Rica 213 Table 9.2A Mean Earnings and Education by Occupation and Sex Earnings Schooling Occupational Cery (Colon, monthly) (yea) Males Females Males Females Professionalecbnical w3rkers 35,309 27.591 12.8 13.6 ManageuAdmnamiistors 43,883 31,656 12.0 9.7 Office workers 23,905 19,556 10.2 11.3 Storekeperstvendors 22,176 12,763 7.7 7.9 Agricultul workers 12,112 8,506 4.S 4.7 PortcJjauitor 17,260 10,434 6.4 6.7 Senice workers 17,481 8,557 6.4 5.9 Oveall 18,459 14,941 6.7 8.5 N S,471 2,138 5,435 2,122 Sourcv: Costs Rica Hogcold Survy, 1989 Table 9.2B Mean Earnings and Education by Sector of Employment and Sex Eanings Schooling Oocupaonl Catego (Colon, mo1fly) (years) Males Females Males Females Public 27,468 24,9S4 9.5 11.7 Private 16,584 10,928 6.1 7.2 Overal Mean 18,458 14,910 6.7 8.S N 5,501 2,145 5,465 2,129 Souro: Costa Rica HoYEJold Survey. 1989 214 Women '. E;Plcymen aJd PAy ii, Lanai America two more years of schooling than males, and in private sector one year more. Nevertheless, women make 65.9 percent of men's earnings in the private sector and 90.8 percent in the public sector. 4. The Deterninants of Female Labor Force Participation As is commonly known, the major factors which influence women's labor market activity are educational attaic ent, marital status, fertility, 'need' for income (which is measured by husband's income, family income excluding female's earnings, the number of earners in a family and the household status of women) and age. Whether women participate in the labor force or not depends on those factors and their reservation wage. That is, when a woman searches in the labor market for a job, she will h3ve some idea of the wage she desires or merits, based on her value at home or her previous wage. She can thus be viewed as setting a minimum standard for jobs she will find acceptable. She will accept a job that pays above this critical value and reject offers below this value. This means that our sample of working women are self-selected. Therefore, if we use this non- random sample to estimate the earnings function for female workers, the result will be biased. Non-working women are u observed. In order to correct for this selectivity bias, we use the well-known two-step method proposed by Heckman (1979). A probit equation is used to estimate the probability of a woman being in the work force and the inverse Mill's ratio (Lambda) is computed and added to the ea-nings function as an additional regressor. In the probit work participation functions, age and schooling are entered as a series of dummy variables for each age group Cm 5 years cohorts) and each level of rchooling. This is to take into account any non-linearity in the effect of either age or schooling on paz!icipation. Other durmmy variables in this model are marital status, residential area, and being a head of household. Number of young children and household size are continuous variables. Table 9.3 presents the results of probit estimates for femalework participation. Using those results we predict the probability of labor force participation for each characteristic (Table 9.4). As expected, women with incomplete primary education are less likely to participate in the labor force. At the completed primary level, however, educational attainment does not have a significant impact on participation. This is explained by the fact that about 50 percent of women are service workers and 30 perceLt are blue collar workers. Tnis implies that the .'alue of education credentials in the informal sector is limited. A similar situation exists in Bolivia. The other education levels show a positive significant impact on participation. Secondary school graduates have an estimated participation probability that is 14 percew age points higher than incomplete primary school graduates. University graduates have a p:t!ipation probability 21 percentage points higher than the graduates with some primary education. Graduate school graduates have the highest participation probability of all (54.2 percentage). Married women participate less than unmarried women, 17.7 percent versus 40.4 percent. Tbis reflects the fact that married women are likely to withdraw from the labor force if they have young children. The variable Presence of Children shows the difference among number of children. The more children a woman has, the less likely she is to participate in the labor force. Being a household head is also associated with a higher participation probability than that of non- household head. Fgmak Labor Force Pawican ad Eanings Dfferenrats hI Caa Ria 215 Table P.3 Pichit Esimmaes for Female Labor Force Participation Variable Coefficient t-ratio Mean Partial Derivative Constant -1.241 -10.10 1.000 Age 20 to 24 0.524 ;.95 0.198 0.163 Age 25 to 29 0.668 7.60 0.134 0.208 Age 30 to 34 0.739 8.36 0.163 0.230 Age 3S to 39 0.879 10.04 0.133 0.273 Age 4 to 44 0.744 8.23 0.097 0.231 Age 4S to 49 0.622 6.67 0.081 0.193 Age 50 to 54 0.350 3.60 0.072 0.109 Incomplete primay -0.126 -1.73 0.149 -0.039 Completed pfimary 0.042 . .68 0.412 0.013 Incomplete seoondazy 0.160 2.14 0.116 0.050 Completed soondury 0.329 4.58 0.142 0.102 Universty 0.5?6 6.38 0.065 0.163 Craduate Sciiool 0.934 8.40 0.024 0.290 Marriod -0.684 -15.63 0.680 -0.213 Number of young chil0tn -0.027 -1.71 1.487 -0.009 Urban 0.289 7.63 0.473 0.090 Household head 0.340 5.67 0.120 0.106 Houwehold sai 0.019 2.10 5.200 0.006 Notes: Sample iluhxe women aged 20 to 60 year. Fenae Iabor force participaton rtxc 27%. Log-LOA*lood - 4062.8 Smple siwe 8,039 The variable for residrtial area shows a positive and significant effect on participation. Women living in urban areas 'e a 45 percent greater participation probability than those living in rural areas. This suggest; that urban areas provide more job oppornities and a more congenial environment for a woman to participate in market activities. As expected, age has a positive and significant influence on participation and the relationship between the two variables is U-shaped. Women in their early 20s have a 21 percent probablity of participating, and reach the peak of employment in their late 30s. S. Earnings flctous In order to explain the variation in earnings in the samr:e by differences in the human capita chaateristics of the indi'idual, we use the standard ?.-inGerian earnings functions (Mincer, 1974). The independent variables are years of schooling, years of working experience (Age- schooling.6), working experience squared (to account for the concavity of the earnings-experience profiles), and the log of hours worked per week. The inverse Mill's ratio, which was derived from the participation equation enters as an additional regressor to correct for sample selection bin. 216 WOmm's Emp1yxiN and Pay bs Lati Amaica Table 9.4 Predc FemPd Labor Forc Participation by Selected aracteristics aur_dstic Predicted Probability Overnll Mean Paaiicipazon RAt 27.0 Age 20 :o 24 21.3 25 to 29 25.7 30 to 34 28.0 35 to 39 32.9 40 to 44 28.2 45 to 49 24.2 ,0 to 54 16.6 Eduaion. comvplde Prizy 16.9 Incompl Seconday 25.2 Compldod Seonday 30.9 Compkts UniWViy 38.1 Compldt GrCade School 54.2 Fanale head of RWesod Yea 34.1 No 22.7 Marital Stah. Maried 17.7 Single 40.4 Pkeence of cbidla (O to 12 yeas) None 2S.2 One 24.4 TWO 235- Three 22.7 Four 21.9 Five 21.0 Six 20.2 Readdn Urbam 28.9 Rral '19.9 Notw Probabiliy ofpmtcQionboidigg otervarihblnmr 0n- edir uap Si;m-laio dono only frorivablswboSceIa ar UdAkcafly aipfficantd Band on doe mb reported in Table 9.3 In Table 9.5 we se that the Lambds variable is insignificant his can be interpreted as evidence that :here ls no self-secdon (Copn, 1980) or ta women as a group are more bomogeneo than initialy perceived. Since dhre is no statistical evidence on !hia point and the selecdvity corrected and uncofrected poin esdmates ae vimally ideraical, we wfll not diff coni between thm in disuing the -us rts below. F,= ex2ff, (5) where P., is the probability that each man, z, is found in job j, Z; is a vector of endowments for each man, n, K,4 is a vector of parameters to be estimated (for men) and k = all jobs. 6 is estimated using the logit technique and then: P = exnf , is calculated for each female worker, n, E:k expL L;. and sector, j (where Z& is a vector of personal characteristics for each womana, n.) The mean of P,, in each sector is Ja. Adding and subtracting 6'4, A, and 'XPJ from equation 3 one obtains: W -W, = (6) jF j s U> wJ) + 4'X> n 'X4r3) + 2wJ)+ @b M PF+) I TJ + JE where WE is the part of the W explained by differences in endowments, WU is th.e part of W unexplained by differensc in endowments, JE is the part of I explained by differences in endowments, and JU is the part of J unexplained by differences in endowments. WU can be considered a rough approxsmation of wage discrimination against women, while JU can be can be considered a rough approximation of job discrimination. Job discrmination is conidered to be the situation where qualified won ame kept out of higher paying jobs. Wage discrimino is considered to be the situation where womea are paid les than equally qualihed men in the same jobs. WU is a weighted aveage of the difference between what wome ean now in each job and what women would ear in each job if they were paid awording to the same wage structure as men. JU is the wage difl erential between what women earm now and what they would ean if wonen faced the same job assignm4nt structure as mn. An altemative measure, analogous to the measure discussed in the last footnote, is to construct WU a die .ifference between the wages men eam now in each job aod the wages m- would ean if they were paid according to the wage determining strcture of women, and to consruct JU as the wage differential between what men earn now and what men would eau_ if they faced the s job assignment structre as maL That is, estimate Se using the logit technique calculate.- PMT = &;'Lo for eac male worker, i, and sector, j. The mean of 4up in ech sector is A. Then add and subtract ^'ZP4 and #;O'X,4 from: Wm- Wr C 2:(nD + E -P,J) (4.) w ~~+ 3 to obtain Wm - WF ;(P[BLq -Bax4D - + (6A) 1;(R'&)PwP* + '[0n-PwP = WU + WE + JU + JE These estimates are presented in a footnote in Section S. Wh7y Women Earn Lem Tnan Men in C&sa Rica 227 4. The Data and Specification of the Variables TIhe data used to estimate the decompositions described in Section 3 are from the Household Survey of Emnployment and Unemployment conducted by the Statistics and Census Department of the Costa Rican government. The Household Survey is conducted annually and we use the most recent survey available, 1989. The survey provides reliable data on wages, some job characteristics and some personal characteristics of workers. We use data from the Central Valley of Costa Rica to Lontrol for labor market conditions which may vary across regions and affect male-female differentials. The Central Valley includes the two largest cities in the country, San Jose and Alajeula, as well as two of the next four largest cities, Heredia and Cartago. ne cities in the Central Valley are all within commuting distance of one another, and are expected to comprise a unified labor market. Over 60 percent of the Costa Rican labor force works in the Central Valley. The literature on labor market segmentation (for example, Piore, 1971) justifies dividing the labor market into sectors. In segmented labor markets some workers are in 'formal' sector jobs where working conditions are better and status and wages are higher than in the 'informal' sectors. Access to forrtal sector jobs is limited and formal sector workers are protected from competition from Informal sector workers by unions, labor protection legislation or internal labor markets. Wages in the formal sectors are higher than those in the informai sector for workers with identical human capital. Gindling (1991) argues that the labor market in the Central Valley of Costa Rica can be thought of as segmented into at least three distinct sectors. We consider two 'formal' sectors where workers are 'protected.' The public (formal) sector is the highest paid sector, workers being protected by unions (the public sector is the only heavily unionized sector in Costa Rica) and government wage and hiring policies. Private-formal sector workers are protected by legislation (primarily minimum wage laws) and mav be paid efficiency wages by large firms. On average, they earn lower wages than public sector workers. Workers in the informal sector are not protected by laws or unions (worker protection legislation is not enforced in the informal sector), and are paid the lowest wages of any sector.' (For a discussion of labor market segmentation in Costa Rica see Gindling, 1991 and 1989b.) Tenjo (1990) notes the imrportance of the domestic servant sector of the Colombian labor market in bringing daout male- female earnings differentials. In this paper, we consider domestic servants as a separate sector. Unfortunately, we do not have reliable data on the value of payments in-kind to domestic servants. Because payments in-kind (for example, room, board, transportation, etv.) can be expected to be larger for domestic servants than for other workers, reported wages for domestc servants will probably under-estimate the actual returns to labor of domestic servants relative to other workers. Work in the United States on discrimination and differential access of women to good jobs has focused on higher paying occupations (for example, Brown, Moon and Zoloth, 1980). In this paper we also measure the difference in wages between men and women due to different access to higher paying occupations within sectors. We are able to distinguish between four 7 If a worker is employed by the wtal governuest or a seni-autonomous (pam-statal) enterprise thea he/she is assigned to the public sector. If a workpr is not assigned to the public sector, and belshe either works in a firm with more hn five employees, or has more than high school education, or is classed as a professional or tochnical worker, he or she is assigned to the privat-fomal sector. If a worker is employed in A firm with five or fewer employee snd does rot have high school education, that worker i9 assigned to the informal sector. On avage, the highest wages are paid in the public sector and the lowost in the informal sector. 228 Wou5e Fmpioynu and PaY in LW America occupational class m order of descending average wages): directors and managers, profes'.ionals and technical employees, administrative personnel, and laborers. These occupational differences are usefil only for the two formal sectors.' We thus divide the labor market into ten different sectors and occupations (jobso) - an informal sector, a domestic servant sector, and four different occupations within the public and. private- formal sectors, respectively. The wage fizdion. The dependent variable in the wage funcion is the natural logarithm of hourly wages for the principal employment of the worker. The numbers presented in the results section will be the difference between the ave.age logarithm of male and female wages. The difference in average wages in the Central Valley of Costa Rica between men and women is 3.5 percent of the female wage, while the difference between the average natural log,arithms of wages is .05. There is a one-tone correspondence between the percentage average difference and the difference in logarithms.' The independent variables in the wage function include measures of human c2pital endowments. These are years of potential experience (age minus years of formal education minus five), EXP, experience squared, EXP2, and two measures of education: years of formal education (primary, secondary or university), ED, and a d-ummy variable whic!i indicates whether or not a worker has had non-formal education, EXTRAED. Tle independent variables also include an indicator of the location where the worker lives. Birdsall and Fox (1985), in a study of school teachers in Brazil, and Behrman and Wolfe (1984), in a study of women in Nicaragua, concluded that differences in the oDst of living in different regions were important determinants of women's wages and male-female wage differentials. We include a dummy variable that is one if the worker lives in a rural area, RURAL, and another dummy variable which is one if a worker lives in an urban area that is not San Jose, URBAN. We expect that cost of living will be higher in San Jose, and higher in urban than rural areas outside San Jose. We also control for two data problems. We expect the reported wages of self-employed workers will be over-estimated relative to salaried workers because they wfll include returns to capital and entrepreneurial input as well as labor. Therefore, we include a dummy variable which is one if 8 We use the Lemaional Occupational lassification (Casificacion Internazional Unifornm do Ocupaciones) to claify workers i different occupations We divide work'l into professional and technical workers (one)digit classficatio 0), directos, and manages mcluding owners, onedigit classification 1), adminisative personnel (one-digit classification 2), and laborers (one-digit classifications 3 through 9). 9 'Wage am defined u the rtio of eamings to hour worked (data on hourly wages am nw available). Earmigs- are defined a lbor eamngs fsom the principaljob. The dat on earnins and hours worked are not strictly comparable; reported eanings are 'nomnall mothly eamings while reported hours worked ar the hocr wored in the week pcor to the smrvey. We estima hourly wages by dividing monthly earing by 4.3 divided by the hous wcsked per weed Ile housclhold muveys do not sample all groups of people in the proportion dat they are found in the populaion. For exemple, because households in mul aras are yp:ed farthe apat than thos in wban areas, people in ursl us under-sampled reve to those in urban aras. T data contain weighing fctors which allow the rsearher to esimat population pramdes from the smple dat. Tae averge wages reporte i this pwer are unweighted. Why VWnlowr Ean Less l ha Men in Costa Rica 229 the worker is self-employed, SELFEMPL. A second data problem is that we do not have data the value of payments in4kin. This problem is especially important for domestic servants. We attempt to address this problem by including a dummy variable which is one if a worker receives payments in-kind, INKIND. Jot (sedor and occupation) assignment functions. The dependent variable is a qualitative variable that denotes the sector and occupation of the worker (see McFadden, 1984). In these equations we want to control for any gender differences in human capital endowments which may affect access to higher paying sectors anl .., ations. Therefore we include, as independent variables, the saue variables that wnr. included in the wage function.'" We also want to control for differena i Xn tastes for airferent sectors and occupations. For example, a worker may prefer to work in the informal sector oecause of the more flexible working hours (even though pay is less dun in the formal zecor:). Included are dummy variables indicating whether a worker is married, single or divorced, MARRiED, whether a worker is the head of a household, JEFE, and the number of children in the househo,d, CHILD. Table 10.1 presents the means and stndard deviations of the variables used in the estimation of the wage and job assignment funcions for mra and women. (Table 10A.1 in the Appendix presents the means and stndard deviations of the variables used in the estimation of these functions for each sex and each sector and occupation.) Women average more years of formal education than men (8.7 versus 7.2 years). Employed women are also more likely to have had some non-formal education (30 percent of employed women have had some non-formal education compared to 15 percent of men). lhis may be because more highly educated women select to enter the work force than women with less education. On the other hand, most men may decide to work regardless of their education level. Also, younger workers are more likely to have hbgher levels of education in Costa Rica and the female work force is, on average, younger than the male work force. On average, poteal experience is lower for employed women thar employed men (18 years compared to 22 years for men). Ihis is perhaps also a reflection of the age distributions of men and women in the work force. Employed women are less likely than men to be self-employed or to be heads of households. Part B of Table 10.1 reports me- and standard deviations for the variables used in this study using a sample whicb exdudes domestic servants. As noted before, the wages of domestic servants are likely to be un d because they do not include the value of payments in- kind. ITis may artificially drive up the measured male-female wage differential. By excluding domestic servants from the sample, we can examiae the effect of domestic servants on the male- female wage differential. If domestic servant are excluded from the sample, the average woman earns a higher wage than the average man. Also, the proportion of women reporting in-kind payments falls (from .068 to .018). This indicates that many domestic servants receive payments in-kind and that this data problem may be causing part of the observed male-female wage differential. However, care must be taken in interpreting this result. Domestic servants are almost always women. When domestic sevants are excluded from the sample, the lowest paid to SevaWl vareables tht wem icludd in the estimation of the wage function are not included in the estimation of the job asgn functions because they do not exhibit any viation within at least one saector/occupation group. The URBAN, RURAL, SELFEMPL and INKIND. Tbese variables ar also excluded from the e stm of he wage ftions for each sex and sector/occupation. In addition to disme variables, we exclude EBP2 from the esimae of the sector assignment functon because the estimate will otherwise not converge. 230 Wonrnm' EnpIymew axd Pay in Ladb Ameria Table 10.1 Means (and Swandard Dev iaio) of Sample Vaiables by Gender Incluing and Exchuding Domestic Servan- Cental Valley of Codsa Rica A. All Workeg Variable Female Male Wag 98.9 102.3 (100.3) (114.8) Log of Wage 4.29 4.34 (0.771) (0.728) URBAN 0.273 0.226 (0.446) (0.418) RURAL 0.272 0.418 (0.445) (0.493) E3P 18.0 22.1 (12.5) (14.6) EXP2 481 700 (668) (876) cl) 8.66 7.22 (3.94) (3.91) EXTRAED 0.301 0.174 (0.459) (.380) INIKIND 0.0,81 0.040 (0.252) (0.197) SELFEMPL 0.162 0.237 (0.368) (0.425) MARRFED 0.677 0.685 (0.468) (0.465) JEFE 0.161 0.653 (0.367) (0.476) CBILD 1.23 1.326 (1.30) (1.318) N 1,262 2,609 Cosud My W Ean L= han Men rIn Ctam Rica 231 Table 10.1 (continued) Means (and Standard Devriaons) of Sample Variables by Cender Imlhdlug and Excluding Domestic Servants Catra Valy of Costa Rica B. EbxSd2ft Dbowesti Smrants Variable Fenude Male Wage 108.7 102.5 (105.2) (114.8) Log of Wage 4.42 4.34 (0.705) (0.728) URBAN 0.288 0.225 (0.453) (0.420) RURAL 0.242 0.418 (0.429) (0.493) EXP 17.7 22.0 (12.5) (14.6) EXP-Z 464' 699 (653) (875) ED 9.21 7.23 (3.94) (3.91) EXTRAED 0.338 0.175 (0.473) (0.380) INKIND 0.0178 0.0400 (0.132) (0.196) SELFEMPL 0.182 0.238 (0.394) (0.426) MARRIED 0.679 0.684 (0.467) (0.465) JEftB 0.158 0.653 (0.365) (0.476) CHILD 1.18 1.33 (1.25) (1.32) N 1,065 2,599 Nots: Te defiakini of the varwbke are She in Section . Te da ued am from the 1989 Houehold Swveyof Eploywat sd U wnpaymei, cowduced by the Censu Dqanent of the Governm t of Cost Riu. Th dt co_n eibinog faor which alow the rewberto amnt poPulaoa pFainds femn the mi d&t. Th estiate ceposd in this table amrot weighted. 232 Women's Enployment and Pay in Latm Aerica women are excluded fron the sample. ThIs will artificially lower the measured male-female wage differential. This illnsaratez the iole of job segtegation in driving the wedge between male and female wages. While excluding domestic servants from the sample does not appreciabl y change the averages of the other variables used in this study for men, it has an important effect on the magnitudes of some of these variables fGr women. For example, the average level of education and the proportion of women who have had non-formal education rises (from 8.7 to 9.2 years and fron- 30 percent to 34 percent, respectively), indicating that domestic servants are among the lowest educated female workers. In the following section, we will estimate separate decompositions using samples which both inlude and exclude domestic servarts. 5. Accounting for the Male-Female Wage Differential The results of the estimation of the wage functions and the decompositions describe in equation 2 are reported in Table 10.2. In the wage fymctions for both sexes, all coefficienuts but three are significantly different from zero at the five percent level of significance. Tne coefficient on the variable which is one if the wor.er receives payments in-kind (INKIND) is not signifitnly different from zero for men. For both men ard women, the coefficients on the vaa !able which is one if a worker lives in an urban area outside of San JGse (URBAN), and the coefficient on the variable which is one if the worker is self-employed (SELFEMPL) are not significantly different from zero at the five percent level. The part of the male-female wage differential explained by differences in 'adowments, E, is negative (-0.123). This indicates that, after controlling for education, experience, location, payments in-kiid and selfr-ei yment, the average wages of women are higher than men's. lhis is largely due to the fact that employed women have higher levels of formal and non-formal education than employed m-n Differences in years of formal schooling ,ED) between men and women account for most of the ragaive E (the effect of differences in yeirs of formal schooling on E is 4.145). The differeial explained by differences in endowments, E, would be more negative if it were not for the fact that men have higher levels of potential experience an women. The impact of differences in experience (EXP plus EXP2) on E is 0.058 (0.160 p'as -0.102). Differences in average levels of the non-human canital explanatory variables (urban (URBAN) and rural (RURAL) location, self-employment (SELFEMPL) or payments ikid_ (INKIND)) are not important in driving male and female wages apart. The part of the male-female wage differential not attributable to differences in endowments, U, is almom-t three times the total male-femae wage differential (U is 0.172 while the total differential is 0.05). This is driven mostly by the ad-antage men have in reurs to experience and differences in the const tenn in the wage functions (the impact of the two experience terms on U is 0.156, while the impat of differences in the constant terms is 0.!24). Rates of returns to both formal and non-formal education are higher for women than men. (Tre impact of differences in the rates of retmn to formal education on U is -0.116.) Tle impact of all other v-iiab!es on U is relatively small. These results suggest that wjmen are discriminated agaist .al the Costa Rican labor n-arke. However, there are important qualifications. The experience variable measures potesstial eperience-what experience would be if individuals began working when they left school and never stopped. Research indicates that womon are more likely to leave the labor force (for example, to take care of children) and re-enter at a later date. This means that women's actual labor mrixt experience will be overestimated relative to men's. One cannot T;ze 10.2 _ aimmw of tho Wage Fnc.ions, by Gender And Estimaes of E and U Including and Excluding Domestic Servants Cntral Vallee of Costa Rica A. All Workm Faaeui Male ' -J Variable CONSrANT 3.02 3.15 0 0.124 (0.0738) (0.0517) UlRlkN -0.0322 -.0236 0.00111 0.00237 (0.0409) '.0309) R7JRAL -0.163 -0.183 -.0270 -0.00561 (0.0425) (.0278) EXP 0.0240 0.0396 0.160 0.280 (0.004U" (.00271) EXP2 -.000206 -.0)046S -0.102 -0.124 (0.00076) (.0000448) ED 0.114 0.101 -0.145 -0.116 ('-W0 (.00356) EXTRAEY) 0.0808 0.111 -0.0141 0.00905 (0.039% (0.0321) INKIND -.491 -0.270 6.00754 0.0150 ,0.06?8) (0.058 SELPEMPL -0.067 -.0430 -0.00323 0.0129 (O.0O05) (0.0286) R-Squared 0.405 0.364 Std. Ewor C.557 0.582 ^!f de Regressin N 1,262 2,609 ToWai E: Explained U:Une4ainWd by Endowi.ents by Endowments RMIMCN G(L)'XF -0.123 0.172 Notes: Thc d:pendet vmiable is natural bgprkhm of wagcs. Standard ermn of the coa L-c in j u Le coc dcu r Fcnder i on aowa variable. (fc.r cxzampk, the coefficicats on ED, EXP, etc,) n uML, finle). L is a the smn of eadomuz for each eonder i B, is a vector of the ooefficiet rapoted in this table. X is a vector of mcan wae detmning characteistics for each geandr. Cctnoed - Table 10.2 (continued) Esuimaes of tbh Wage Functions, by C-ender And Estm*m of B and U Inlnding and Excluding Domesic servants Cental Va&ey of Costa Rica B. ExIduding Domestic Servants Femalde Male B' ,-xJ b,WX V'ariable CONSTANT 3.10 3.15 0 0.0522 (0.076) (0.0517) URBAN -43.0430 -.0226 0.00142 0.00586 (0.0416) (.0310) RUMAL -0.113 -0.181 -.0319 -0.0166 (0.0451) (.0279) EXP 0.0256 0.0396 0.171 0.248 (0.00431) (.00272) EXP2 -.000223 -.000466 -0.109 -0.113 (0.0000808) (.0000449) ED 0.108 0.101 -0.198 -0.0744 (0.00W09) (.00356) EXTRAED 0.0681 iJ.111 -0.0181 0.0144 (0.0387) (0.0321) INKIND -.180 -0.273 -0.00606 -0.0168 (0.133) (0.0589) SELFEMPL -.0278 -.0453 -0.00209 -0.336 (0.0501) (0Q0286) R-Squared 0.3S3 a362 Std. Erior 0.569 0.582 of tho Regression N 1,065S 2,599 ToWa E: Expained U:Uneaplined by Endowments by Endwmvzoents &'MrX) @16)'WAF -0.193 0.112 Notes: The depq=dent variabe di c nat rtl lzbpr8m of wages. Smadud cmns of the coeffici cac rin L is the coefficicnts for i on _adowt vauiable z (for exampic. the coefficicrs on EED. EXP, dor) (1 = mal, fomale). L is a the mean of cadmow m g z foo each geddr i B, is a vector of tho ooffSaiet repotted im this table. KA is a voctor of mean wags dderniaing chuaractdsica for cach geader. be sure if the gender difference in the coeficients on experience is due to differences in the rate of return to experience or to unmeasured differences in labor market experience. The fact that the rest of the difference is due to differences in intercept terms indicates that much of the difference in the wages between men and women is due to factors we canrot identify." The importance of domestic sevants in driving the wedge between male and female wages has already been noted. When domestic servants are excluded from the sample used to estimate the wage equations the conclusions drawn above still hold (see Part B of Table 10.2). E is still negative (4.193) and U positive (0.112). The most i.nportant variable causing E to be negative is still education and the most important variable causing U to be positive is potential experience. The biggest difference between parts A and B of Table 10.2 is the impact of the intercept term. T7he wage differential due to differences in the intercept terms is smaller (from .124 to .0522) in the estimates of the wage equations without data on domestic servants. This indicates that a part of the unexplained difference found in the first set of numbers is caused by the impact of domestic servants. With one exception, all coefficients are similar to those reported for the estimation where the full sample was used. The exception is the coefficient for the variable wbich is one if 2be worker receives payment in-kind for women; it is not significantly different from zero at the five percent level. Table 10.3 presents the male-female wage differential for each sector and occupation, the proportion of men and women in each sector and occupation, and estimates of W (the part of the wage differential due to different wages paid to men and women in the same sectors and occupations) and J (the part of he differential attributable to different access to higher paying sectors and occupations- see equation 4). Women are over-represented in the relatively higher paying sectors and occupations; specifically among professional and tecnical workers and administrative personnel ir: the public sector (see the last column of Table 10.3-the difference between the proportion of men and the proportion of women in these two sector/occupations is -.0.080 and -0.0302, respectively). There is no significant difference in the participation of men and women in the highest paying sector/occupation in the public sector, directors and managers. Women are also over-represented as domestic servants, the lowes paying sector/occupation (the didference between th^ proportion of men and the pro,-irtion of women in the domestic servant sector is 4.152). Women are under-represented in the other three of die four lowest paying sector/occupations; the informal sector, and among laborers in the public and private-formal sectors (see the last column of Table 10.3, the differences between the proportion of men and the proportion of women in these sectors are positive).' Despite women being over-represented in the lowest payinD domestic servint sector, J, the part of the male-female wage difference attributable to different access to higner paving sectors and occupations, is a negative -0.0223, indicating that women are, on average, over-represented in the higher paying sectors and occupations. W, the part of the wage differental due to different wages paid to men and women in the same sectors and occupations, is a positive 0.0723. This indicates that, on average, woiren are paid less than men in the same sectors and occupations. Women are paid more than men in only two " Using the decompositions described in equsion 2a, the difference attributable to endowments, E, is -0.129, the difference not attributable to endowments, U, is 0.178. 12 On average, women are over-represented in the public sector id undner-rpresented irc the informal and private-formal sectors. Table 10.3 Average NatU Lgauidim of Wages by Sector, Occupation and Gender, Male-Female Diffeencs m the Natudl Loganthm of Wages by Soctor and Occpaion, Male-Femae Difernc:s in Assignmet to Sector and Occupaio, Estifn of W and I A. MAWU k Secw ane Occuation Wi Who} W, WM3W1J p__ Pe p__-p_ infora 4.06 4.05 4.08 -0.03 0.314 0.170 0.144 PrivateFoma: Profes6uonl and Tehia 5.01 5.06 4.90 0.16 0.0422 0.0372 0.005 Directors and Managers 5.18 5.23 4.90 0.33 0.0261 0.00951 0.0165 Adminie Personnel 4.49 4.52 4.47 0.05 0.0356 0.0777 -0.0420 labores 4.20 4.22 4.15 0.07 0.416 0.315 0.102 pRN: Professonal and Technical S.18 5.26 5.12 0.14 0.0448 0.124 -0.080 Directos and Managers 5.34 5.30 5.42 -0.12 0.00651 0.00634 0.00018 Administative Persound 4.80 4.8S 4.74 0.11 0.0387 0.0689 -0.0302 Laborem 4.52 4.54 4.47 0.07 0.0724 0.0357 0.0367 Dometc Sevts 3.57 3.67 3.57 0.10 0.00383 0.156 -O.IS2 ;PO(WMrWF) - W 0.0723 WF\P'4-PF-) 8 -0.0223 Table 103 (cotinued) Average Natuml Logarithm of Wages by Sector, OccQuai and GenJer, Male-Female Diffeces in the Natual Logarithm of Wages by Sector and OcUpation, Male-Female Differences in Assigprit to Sector and Occupation, Estimates of W and J B. Escdudinz Domestic Servants Sector and Occupstion Wi W,4 WFJ WyWWls Pb Pe PMrPF Ioforn 4.06 4.05 4.08 -0.03 0.314 0.201 0.113 Professional and Techical 5.01 5.06 4.90 0.16 0.0423 0.0441 -0.0018 Directors ansi Managers 5.18 5.23 4.90 0.33 0.0261 0.0113 0.0149 Administative Peramed 4.49 4.52 4.47 0.05 0.0357 0.0920 -0.0562 Labore-s 4.20 4.22 4.15 0.07 0.418 0.373 0.045 Public: Professional and Techri 1 5.18 5.26 5.12 0.14 0.0450 0.147 -0.102 Dirctor and Manages 5.34 5.30 5.42 -0.12 0.00654 0.00751 -0.00097 Administave Pe oml 4.80 4.85 4.74 0.11 0.0389 0.0817 -0.0428 Laborers 4.52 4.54 4.47 0.07 0.0727 0.0423 0.0305 EjP3(WWW,) = W 0.0670 ;W,3(PwP j= J -0.148 Notes: W; averagc natual bgrrihm of wage in sectorloompstion j. W = average naturl bhgarkn of wages for male in scdcr/oczpationj. w9 avGagc waural kgarithm of ag w rmalr in wctor/occupasioaj. P= proportion of im in sector/ocapuionj. Pi= proportion of females in achb uctooocptionj. sector/occupation groups: directors and managers in the public sector and in the informal sector (see column four in Table 10.3-the difference between the average natural logarithm of wages for men and the average natura logarithm of wages for women is -0.12 and -0.03 respectively for these sector/occupations)." The second set of estimates in Table 10.3 excludes domestic servants from the sample. When this is done, W does not change appreciably while J falls (from -0.0233 to -0.148). This is expected because, with the exclusion of the lowest paid women from the sample, women are now even more over-represented in the higher paying sectors and occupations." Estimates of the multinominal logit model by sector/occupation are presented in Table 10.4. Tables 10.5 and 10.6 present the estimates of the decompositions described in equation 6. In the last few paragraphs we concldded that women earn less than men with the same endowments, and that women are paid less than men in the same occupations and sectors. This is consistent with the decompositions presented in Table 10.6. WU, the difference between the wages of men and women within sectors and occupations that is not explained by differences in endowments, is the only one of the decompositions reported in Table 10.6 which is positive (0.114). This indicates that the primary reason why women earn less than men is because women in the same sectors and occupations as (observably) equally qualified men are paid lower wages. In particular, men are paid more than women with the same endowments in every sector/occupation except among directors and managers in the public sector (see column I in Table 10.6-only among directors and managers in the public sector is the difference in the average natual logarithm of wages for men and women unexplained by differences in endowments negative, specifically it is -0.0610). WE, the difTerence between the wages of men and women within sectors and occupations that is explained by differences in endowments is negative indicating that, on average, within each sector and occupation women have higher human capital endowments than men. Only among administradve personnel in the public sector are men paid more than women because they have higher levels of human capital endowments (see column 2 in Table 10.6-only among administratve personnel is the difference in the natural logarithm of wages between men and women due to differences in endowments positive, specifically 0.00313). 13 In the public sector as a whole, and in the infornal sector as a whole, women are paid more t-mn men. 14 Estimatne of W and J using th decompositions described in equation 4 are 0.0521 for W and -0.00211 for I when domestic servnt ae included in the sample ,i 0.0519 for W and -0.132 forI when domestic servants are excluded ftom dIe sample. u In estmating the wage fu 4) (lb) There is always an index numberproblem experienced here. Theoretically, there is no. Jvantage to estimating the results using male means or female means, so we presert both. lTe iirst term in both equations is the part of ;he log e3rnings differential that can be ascnbed to differences in the wage structres between tie sexes and the second term is that part of the log earnings differential that can be ascribed to differences in human capital endowments. 3 See Groma (1988), pp 277-301. See so Ng, Scott, and Velez and Wnme in this volume. Erigs Funmcion Women Women ilim (Coffecte1 (LUnzorected Variabl9 Unc,ffet fct Selectivity) for Selectivity) Co=taA 6.665 7.285 6.585 (127.85) (40.88) (65.50) Yeara of .132 .109 .147 schooling (36.76) (8.85) (19.85) Experie-- .086 .056 .067 (24.4) (5.56) (9.71) Expe3ienc- -.001 -.001 -.001 squared (-19.78) (-2.52) (4.49) Self- -1.217 -1.313 employed (-19.84) (-4. (-5.32) I e A~da -1.487 (-6.70) R1 2.391 .362 .303 N 3,334 1,217 1,217 NoUs: T-mtios am in pa.-ewbes. DeFcU4.mt vuiabi - lo eckly waes). 'rable 15.5 presents the results of the decornonsition using the selecivity corzeced sample sinc it yields a more credlbt. estimate. Tabie 15.S Dc :a of the MdelIFen:ele Eaiags Difetenti- Patc of Eanings Diffwetial Due to Diffeenm in Specification Endowments Wago StnInro Eva1=:.ed at 2D.0 80n0 Male Mea Evabuted at 2S.1 71.9 Fe: mIW M- Iis Ns W,,/Wqsl7S due to differences in human capital endowments, and 80 percent is due to unobservable factors. lhe equation evaluated at the female means shows that approximately 28 percent of the differential is due to differences in hunman capi 1 endowments and 72 percent due to differences in the way men and women are rewarded in the labor market. 7. DIscussIon As has been shown, women's participation raws are posidvel) nfluenced by the amount of educatdon. Urban residence has a positive effect on partcipa` r md the presence of teenaged children in the household does not appeal to influence ,-m!'; J& ucision to work. A wage differential of ;4 percent is low in comparison with other Latin American countries, and with some industrialized counties.' Tlis may be explained in part by our inability to identify several of the important factors in the decision making process, such as hours worked, public versus private sector and formal versus informal sector. The small differen ial may also be due to several factors tat we can identify in the existing data. Working women have, on average, one and a third more years of schooling than working men. Wotking women are also more likely to complete higher levels of education than worldng men. Twenty-eight percent of the working women attempted Junior High with 23 percent completing, compared to only 20 percent of working men who attempted Junior High with 14 percent completing. Similar!y, 15 percent of working women attempted High School with 12 percent completing, compared to 8 percent of working men who attempted with 4 percent completing. Despite the low earnings differential between men and women, this study has shown that rlJy a small proportion of the differential can be explained by differences in human capital endowment3. The 'upper bound' estimate of discrimination is 72 or 80 percent, depending on the equation used. Given the nature of the survey data used in these analyses, males may have hat endowments which were superior to women's but of which we are not aware. If this is the case, the lack of information will bias the esti if the component due to wage discriminaiion upwards. Clearly, further research into the factors influencing women's participation decisions needs to be done. F%:rther research should include those human capital factors that were missing or had to be estimated in this study. Especially important are hours worked per week and tenure in the job market. 4 hi Biitain, women eam 74 pement of men's wages. Khandker reports wome's wags being about two-thirds of mn's in Pe while Ow Ns found w.ram's wma in Argentina to be 65 percent of men's (both in this volume). See also Gmuwnd (1989), Tzsnnstos (1987), Zabalz and Tmnnatos (1985), iAd Grgory and Dun (1982). References Carlson, S. and J. Prawda. Basic Education in Mexico: lrends, Issues and Poliy Recommendations. Washington, D.C.: World Bank, June 1991. Grugory, R.G. and R.C. Duncan. 'Segmented Labour Market Tleories and the Australian Experience of Equal Pay for Women." Journal of Post-Keynesian Economcs. Vol.3 (1982). pp. 403-428. Gronau, R. 'Sex-Related Wage Differentials and Women's Interrupted Labor Careers: The Chicken and Egg Question." Journal of Labor Economics. Vol. 6, no. 1 (1988). pp 277- 301. Gunderson, M. 'Male-Female Wage Diiferentials and Policy Responses." Journal of Economic Literature. Vol. 21, no.1 (1989). pp 46-72. Heckman, J. 'Sample Selection Bias as a Specification Error." Econometrica. Vol. 47, no. 1 (1979). pp. 152-161. Tzannatos, Z. 'Equal Pay in Greece and Britain." Industial Relations Journal. Vol. 18, no. 4 (1987). pp. 275-283. World Bank. Mexico, Selected Policy Papers. Washington, D.C.: World Bank, June 20, 1989. World Bank. Staff Appraisal Report, Mexico, Water, Women and Delwopmen Project. Washington, D.C.: World Bank, May 24, 1989. Zabalza, A. and Z. Tzannatos. Women and Equal Pay: The Effects of Legislation on Female Empioynen and Wages in Britain. Carnbriage: Cambridge University Press, 1985. 348 16 Female Labor Force Participation and Wages: A Case Study of Panama Mar Arends l. Introduction This chapter exaniines the differential in earnings between males and females in Panama. In the sample, female wages are 85 percent of male wages, a very high percentage for Latin America. What are the rcasons for this high percentage? How does Panama's low participation rate for women affect female earnings? Tne country is also interesting because of its extensive Labor Code, which has many sections pertaining specifically to women. Because of long standing structural problems in the labor market, Panama has a high unemployment rate, especially for women. In Section 2, the economic and labor market situation in Panama are discussed. Section 3 pertains to the data and includes descriptive statistics. Section 4 presents the results of a muivariate probit model for both men and women that aitempts to determine which characteristics make an individual likely to be observed in the work force. The results of the probit are used to correct for selectivity in earnings regressions. Section 5 discusses the results of earnings regressions for men and women, both correcting for selectivity and without correcting for selectivity. A decomposition of the earnings differential is calculated to estimate how much of te differential is due to differences in endowments and how much coulu be attributed to labor market discrimination. Lastly, there is a discussion of the policy implicaticns of the findings. The Panamanian Eonomy and Labor MarkG TLi? population growth rate was 2.2 percent from 1980 to 1989, which is average for Latin Amet-ica as a whole. Thirty-five percent of the population was aged from 0 to 14 years in 1989, while 60 percent of the populai3n was aged from 15 to 64 years, about average for Latin America. The labor force growth rate was 2.9 percent in the 1980s, averagirg 3.3 percent for women.' Panama had a GNP of $1,760 pe- capita in 1989. Growth was high from 1965 to 1980 at 5.5 percent, but the economic problems of Latin America in the 1980s affected Panama also, and Ec=nomist IntelligWnce Unit (1991), p.55. V"VA _a VI,y .J VIAV;Au "V u *:OV w :0. AuC eAJ1Uy wa vCry neavuy or,eneou wwaras services, which accounted for 74 percent of GDP in 1989, while ind,istry accounted for 15 percent and agriculture for 11 percent.2 Recent political problems strongly impacted the economy. In 1988, United States' sanctions and massive capital flight led to 16 percent contraction of GDP. In 1989, the economy did not recover and GDP fell again by .9 percent.3 However, the most chronic problem in Panama is persistent unemployment. A 1985 World Bank country study stated that unemployment was, without a doubt, the gravest economic and social problem.' In the study, the 1983 unemployment rate was estimated at 9.5 percent. The situation steadily worsened du'ring the 1980s. Unemployment estimates for 1989 ranged from 16.0 percent to 20.1 percent.5 In the sample used in this study, the male and female unemployment rates were 14 percent and 22 percent, respectively. To confront Panama's unemployment problem, one strategy was to increase the public sector. The Torrijos regime, which governed from 1969 to 1981, used this strategy throughout its tenure, and enacted the Emergency Employment Program in 1977. When the Emergency Program ended in 1980, 25 percent of workers were in the public sector.6 Because of budget constraints, the public sector could not continue to provide employment, and the percentage of workers in the public sector steadily declined tnroughout the 1980s. However, the percentage of workers in the public sector remained high at 21.9 percent in 1989.7 The participation rate declined from over 60 percent in early 1970s to just over 50 percent in 1982 and 1983. This happened because of greatly increased enrollment in secondary and tertiary education, a reduction in the voluntary retirement age from 62 to 55, and a falling female participation rate.' However, according to one official source, female participation rates rose during the 1980s from 17.8 percent in 1980 to 20.8 percent in 1989.9 The total labor force participation rate was estimated at 58 percent in 1989.'° An important contributing factor to unemployment was Panama's labor code. Instituted in 1972, it substantial!y increased the cost of hiring labor for employers. Workers were given more job security, benefits, and bargaining power. It required employers to pay high severance pay which increases with the length of s^rvice, discouraging temporary hires. The total burden on 2 World Bank (1991), Tables I and 3. ' Economist Intelligence Unit (1991), p. 54. 4 World Bank (1985). p.9. I Tlhe estimates come from the Economist Intelligeace Unit and the World Bank Panama Operations Desk respectively. 6 World Bank (1985), p. 19. 7 World Bank, Latin America and the Caribbean, Country Department II, unpublished table, (1991). I Worid Bank (1985), p. II. I World Bank (1990), p. 238-239. '° The Economist Country Profile, 1991-92. sjiuAyQb 1"%;.uuuw a wLrwen-inoW nwouS ano palO vacaaons was estimated by the World Bank to be 40 percent."1 Employers could not reduce a worker's salary, so piecework could not be paid on the basis of productivity. As a result of these regulations, Panama's labor costs were among the highest in the Caribbean Basin. The WVorld Bank recommended changes in 1985 as conditions for a structural adjustment loan. The reforms were finally accepted in March 1986, despite a ten-day work stoppage by the unions and fierce political opposition. Tle modifications permitted piecework, encouraged rewards for productivity, and rationalized overtime provisions for some firms.'2 Whether these changes have had a great impact remains to be seen. Many aspects of Panamanian law apply directly to women in the labor force. There are provisions in Panama's Constitution (Article 62) and in the Labor Code (Section 10) that employers mu;'s provide equal pay for equal work. Legal redress is available, but the burden of proof is on the employee. No real attempt has been made in Panama to address the issue through the courts. Low female labor force participation rates and women's willingness to be self- employed are two explanations why this is so. 3 There are provisions in the Labor Code which may add to the female unemployment problem. Employers are required to provide 14 weeks of maternity leave, with the employer making up the difference between regular pay and social security payments. It is unlawfiu to dismiss a woman during pregnancy and five months thereafter without judicial approval. New mothers are entitled to a paid hour break each day in order to breast feed their babies. If a company employs over 20 females, it is required to provide a nursery. Also, the Code forbids women from working in dangerous occupations, such as mines and civil construction. In a Labor Code survey, Spinanger interviewed employers in various sectors of the economy. Two-tirds said that matemity protection laws discouraged them from hiring women. Employers were willing to increase wages by 25 percent to have more flexible maternity arrangements." Such laws may encourage employers to discriminate against all women, including those who have no intention to have a child or older women who do not plan to have more children. Another characteristic of Panama's labor market is regional disparity in earnings. Heckman and Hotz (1986) found evidence that the Panamanian labor market was segmented by regions, with less developed regions showing higher rates of reurn to education than developed regions. Rural regions such as Darien, Veraguas, and Cocle had high rates of return, while the Canal Zone and Panama City showed rates of return comparable to the United States. Also, in the Canal Zone, wages are about three times higher than in the rest of the country." Only 2.6 percent of workers are employed in the Canal Zone, but their salaries, which are raised in real terms in accordance with United States cost of living changes, may have prevented other wages in the " World Bank (1985), p. 17. t2 Tollefson (1989), p. 135. 13 Spinanger (1984), p. 21. "4 Spinanger (1984), p. 29. is Heckman and Hotz {1986), p. 540. " Because of the Panama Canal teaty, former Canal Zone employee who became employeea in Panaa were guaranteed wages and condidons similar to those their position had commanded when employed by the U.S. (see Tollefson, p. 142). Canal areas of Colon and Panama from falling, which would help solve the unemployment problem."7 Panama has a long-standing, strong commitment to education. Adult illiteracy was 12 percent in 1989. Schooling is compulsory for 9 years, and begins at age 5. Enrollments were higher than average for Latin America, given Panama's per capita income. Primary school enrollments as a percentage of the relevant age group were 102 percent in 1965 and 106 percent in 1988. Secondary school enrollments were 34 percent in 1965 and 59 percent in 1988, compared to an average in Latin America of 48 percent in 1988. Twenty-eight percent of the relevant age group was enrolled in tertiary education in that year, a percentage that was only surpassed by three Latin American countries, all with higher per capita income-Argentina, Venezuela, and Uruguay.18 3. Data Characteristics The data for this study were taken from the Encuesta de Hogares-Mano de Obra of August 1989, by the Office of Statistics and Census of Panama (DEC). The survey consisted of 8,817 households, comprised of 38,416 individuals. Out of this sample, the individuals of economically active age (ages 15 to 65) were selected, giving a sample of 23,196 individuals. The survey covered both urban and rural households, and the data were weighted to give an accurate representation of the population. One limitation of the data was that about 30 percent of employed males had no hours and/or no income reported. Over 90 percent of the males that were recorded as 'employed' but had no hours or no income were either family workers or self-employed workers in agriulure. Table 16.1 summarizes the problem. Labor force participation includes employed and unemployed workers. Woa: force participation includes only those who were recorded as 'employed' in the survey. The third column labelled '+Hours, +Income' consists only of workers who reported positive hours and positive income. The table breaks down these rates by province. Overall, a wage rate could be calculated for only 71 percent of the male workers. The problem is severest for the rural provinces of Darien and Bocas del Toro. Because of the low percentages of male workers with positive hours and positive income, there is a selectivity problem when examining the male wage functions. The males for whom hours and income are available are a special subset of the male workers. Therefore, in Section 4, the results of separate probit equations for both the males and females are presented. For females, most rural women were classified as 'housewives' and therefore inactive. It is likely that many of them are actually unremunerated family workers. Unfortunately, there is no way to determine the kind of work these women do. In Table 16.2, the means of the sample variables for the working and non-working samples of males and females are shown. For the table and the subsequent regressions, working was defined as having positive hours and positive income. About 50 percent of the men and 30 percent of t'ie women in the s: nvle of individuals aged 15 to 65 were classified as working. Schooling was calculated by caking the number of years completed at the highest level and adding the number of years required to finish preceding levels. Because it was not known if the individual 7 World Bank (1985), p. 19. 18 World Bank, (1990), pp. 238-239, and World Bank (1991), Table 29. Table 16.1 Participation Ratew Male Female Labor Work +Rours Labor Work +Hours Province Force Force +Incoms Force Force +Income Bocas del Toro .88 .81 .78 .21 .19 .19 Cocle .88 .79 .36 .36 .29 .23 Colon .80 .64 .54 .40 .30 .29 Chiriqui .84 .72 .52 .34 .26 .25 Darien .94 .94 .19 .32 .31 .18 Heffern .84 .80 .40 .33 .29 .28 Los Santos .89 .86 .43 .30 .24 .23 Panam .79 .65 .57 .48 .37 .37 Veraguas .89 .85 .29 .28 .23 .20 Country Wide .83 .47 .4i .28 Notea: Labor Force includca individuals reported a cnaployed and unemployed. Work Force includes only individuab reported an employed. + Hours +Lomc includes only employed individuals reported with posiive h-urs and positive inom. It xachudca moat self-employed and amily workrs in agricutu. completed his or her degree, or how many years were repeated, the measure of schooling is subject to bias. For both men and women, individuals who worked had higher education and were older than those of tle same gender who were not working. Working females had one more year of education than working men on the average, and were less than a year younger. Five percent more working women had 4 to 6 years of university level education than working men. Sixty-one percent of working women lived ia the province of Panama, compared to 47 percent of the non-worldng women and 56 percent of the working men. Fifty-six percent of the non- working men lived in a rural area, compared to only 37 percent of the working men. Only 24 percent of the working women lived in a rural area, while 47 percent of the non-working women did. Working men had higher monthly earnings, more weekly hours and higher tenure than working women. The overall ratio of female to male hourly wage was .85. Working women were much more likely to work in the public sector (36 percent versus 28 percent), less likely to be self- employed (15 percent versus 21 percent), and more likely to work in a small firm (38 percent versus 34 percent) compared to working men. WoBd~ Wo$-Wo~fri~g workng NeWagw j A$a 35.57 31.17 34.81 32.95 (19.79) (15.46 (10.74) (14.p B=adon (Yet) 9.21 7.33 10.45 7.76 (4-40) (4.02) (4.35) (4.03) No Educa6om .03 .07 .02 .07 (.17) (.25) (.13) (.25) Inoompl6el Ptinzy .10 .19 .07 .15 (.31) (.39) (.25) (.36) PriM7 .25 .26 .20 .24 (.43) (.44) (.40) (.42) IcouplOt SeoWQY .23 28 .21 .31 (.42) (.45) (.40) (.46) seboanday .10 .11 .22 .13 (.39) (.32) (.42) (.34) L than 4 ym. C .05 .nn .05 (.24) (.22) (.31) (.22) 4 yr .dn b o .11 .03 .16 .03 (.31) (.16) (.36) (.17) Tchbnical .03 .02 .03 .02 (.13 0.4) (.16) ( 4) Bocaa d1 Toro .04 .01 .01 .02 (.19) 0.1) (-E) (.15) Cool. .05 .10 .05 .08 (.23) (.30) (.22) (.27) Colon .07 .06 .07 .08 (.26) (m) (.26) (.27) firuiq .1$ .14 (.2 .36 (.36) (.35) (.32) (.37) . . .~~~~~~~~~~~~~~~~~ciim TAble 16.2 Means (and Standard Deviatbo) of Sample Varibls Working Ncm-Working Working Non-Working mawes Miles Females Femal Daren .00 .02 .00 .01 (.06) (.13) (.07) (.10) Herrer .04 .06 .04 .05 (.19) (.23) (.20) (.22) Los Santos .03 .05 .03 .04 (.18) (.21) (.17) (.20) Panama .56 .43 .61 .47 (.50) (.50) (.49) (.50) vean-pas .05 .14 .05 .09 (.22) (.34) (.22) (.29) Ruri .37 .56 .24 .47 (.48) (.50) (.43) (.50) Prmary Monthy 341.86 274.15 EAmings (Blboss) (395.94) C266.96) Weekly Hours 42.76 40.14 (12.66) (12.39) Primazy Wage 1.97 1.67 (Balboas/Hour) (2.43) (1.60) Self Employed .21 .29 .15 .00 (.41) (.45) (.35) (.06) rublic Sector .28 .36 (.45) (.48) Private Sector Employee .49 .48 (.40) (.40) Employer .02 .01 (.15) (.09) Small Firm .34 .46 .38 .07 (.47) (.50) (.49) (.25) TenLre 7.92 7.30 (8.01) (7.15) . cortinued Tab! 16.2 Means (and Standard Deviations) of Sample Variables (continued) Working Non-Working Working Non-Working Males Malec Females Females N People in 5.10 5.52 4.96 5.5S Household (2.45) (2.63) (2.35) (2.52) N of Children .72 .56 .64 .79 Aged 0 to 6 (.98) (.93) (.91) (1.05) # ,f Children .64 .68 .67 .73 Aged 7 to 12 (.92) (.91) (.98) (.99) Household Head .65 .36 .22 .10 (.48) (.48) (.42) (.3 1) Housebold Monthly Pnmary 581.98 235.70 710.95 310.27 Income (Balboas) (603.39) (422.08) (709.71) (445.45) Total Monthly Household 657.62 340.51 819.33 413.20 Income (Balboas) (695.60) (509.54) (814.20) (535.24) # of Employed 1.96 1.70 2.11 1.45 in Household (1.06) (1.27) (.99) (1.09) N 5,446 6,205 3,190 8,355 Notes: Partcipation rate is .83 for men, .41 for women. Forty-seven percent of men and 28 percen of women hwve positive hours ad posive inome and a define as working.' Regarding household charactestics, both working women and men came from smaller households than non-working individuals. In the sample, there was not enough information to determine which adults were the actual parents of the children in the household. Therefore, as a proxy, the number of children in the household was used. Working men had more children aged 0 to 6 than non-working men, while working women had fewer children than non-working women. Working men and women were more likely to be household heads than non-wcking men and women. Also, working women had significantly higher household primary income and household income than working men. Non-working women had higher household income than non-working men. There was no variable in the survey for mital status. Table 16.3A presents the wage differentials between men and women, broken down by sector and level of education. In every case except in the employer Eetor for workers with primary education (which includes only 9 women), the male wage rate exceeds the female wage rate. Ile employers' wage rates tend to be highest for both men and women. For men and women, the public sector is better paid than both the private sector and the self-employed sector. In every sector except employers, the ratio tends to be low for both primary md less educated and for university educated. It is intersting to note that sdf-employed women with some secondary or completed secondary educadon do well compared with men with similar education; the ratio ic .89. For the public sector and self-employed categories, the university-educated women acually have the lowest wage ratio In the sector. Aeaw LatOr Force Part7icadon anad Wages: A Case SMsdy of Panara 357 Table 16.3A Wage by Schooling Level (Balboas per hour) hM2lic Private Employee F/M F/M Male Female Ratio Male Female Ratio Some Primary 1.57 1.14 .73 .94 .60 .63 Primary 1.66 1.28 .77 1.10 .62 .56 Some Secondary 2.03 1.80 .87 1.34 .96 .64 Secondary 2.75 2.13 .77 1.79 1.52 .85 l'Jiversity 4.57 3.22 .70 3.34 2.43 .73 fechnical 2.59 1.83 .71 1.73 1.03 .58 mRoyer -Self-Employed F/M F/M Male Female Ratio Male Female Ratio Some Primary 2.51 .93 .37 1.11 .72 .65 Primary 2.18 2.44 1.12 1.16 .89 .77 Some Secondary 2.49 2.41 .97 1.37 1.22 .89 Secondary 6.26 3.30 .53 1.58 1.41 .89 University 7.31 6.45 .88 3.11 1.75 .56 Technical N/A .83 1.02 .85 .83 Wages for women with little education are significantly higher if self-employed than if a private sector employee. For women with secondary school level education ane. above, private sector employees earn more than the self-employed. For men, the pattern is similar with self-employed workers earning more than private sector workers at low levels of education, and vice versa at higher levels of education, but the difference is not as great as for women. This implies that access to the formal sector is difficult for those with low education. Table 16.3B presents the -.age differentials by region and education level. The ratio of the female wage to the male wage tends to be bighest at the secondary school level across regions. Also, for all levels of schooling, the ratio is lower for Panama and Colon, the most urbanized re,;ions of the country and the two regions with the highest average hourly wage. Women have higher overall average wages than men in Bocas del Toro, Chiriqui, Los Santos, and Veraguas, which, with the exception of Veraguas, are middle income provinces.19 Darien, the poorest province, has very low femuje to male wage ratios for those with less than primary education, some secondary education, and a university education. Tle rato of female to male wages tends to be more tivoruole in middle income provinces for workers with intermediate levels of education. 19 The claafications of regicns as high, middle, and low income are from Hecknn and Hotz (1986), p. 521. 358 WomeX '5 Fmpioyuau and roy i Lanx Amerwca Table 16.3B Wages by RPkoon Sex, and Schoolug Level (Blboas per hour) Overall Lem g Primy Some Secondary University Technical Pfimay Secondaiy Bocas del Toro Mao (2.4%) 1.58 1.30 1.53 1.53 2.19 2.43 N/A Female (.5%) 1.60 1.08 1.40 1.36 1.93 1.89 N/A Ratio F/M 1.01 .83 .92 .89 .88 .78 Coce Male (3.3%) 1.23 .86 .95 1.11 1.63 3.24 1.29 Femle (2.0%) 1.06 .48 .5' 1.07 1.75 2.47 1.40 Ratio F/M .86 .56 .54 .96 1.07 .76 1.09 Colon Mile (4.6%) 2.07 1.07 1.65 1.68 2.69 2.82 1.19 Female (2.8%) 1.52 .56 .79 1.12 1.88 2.20 1.08 Ratio F/M .73 .52 .48 .67 .70 .78 .91 Chfriqui Male (9.1%) 1.27 .97 1.05 1.09 1.38 3.03 1.97 Female (4.6%) 1.44 .77 .86 .97 1.96 2.59 .87 RAtio F/M 1.13 .79 .X2 .89 1.42 .85 .44 Male (.3%) 1.88 1.43 1.59 1.96 1.86 5.54 1.22 Female (.2%) 1.21 .60 1.15 .98 1.59 1.45 N/A Ratio F/M .64 .42 .72 .50 .8S .26 Herrera Mal' (2.4%) 1.43 .79 1.08 1.15 1.83 3.54 1.19 FemAle (1.7%) 1.33 .60 .52 .93 1.73 2.87 .79 RaFio F/M .93 .76 A8 .81 .95 .81 .66 L4 Santos Mae (2.1 %) 1.26 .78 .87 1.30 1.86 4.03 1.62 Female (1.1%) 1.34 .54 .58 1.5 1.79 3.16 1.72 Ratio F/M 1.06 .69 .67 .81 .96 .,8 !.06 Panama MLhe (34.3%) 2.35 1.27 1.40 1.70 2.38 4.47 1.87 Fenale (23.6%)1.83 .83 .88 1.25 1.79 3.05 1.23 Ratio P/M .78 .65 .63 .74 .75 .63 .66 Vaeguas Male (3.3%) 1.39 .70 1.05 .95 1.78 3.17 2.37 Female (2.0%) 1.48 .54 .71 1.25 1.63 2.71 .42 Ratio F/M 1.06 .77 .68 1.32 .92 .85 .18 Nok* Pcrctagea in prccmthca nqrat peratntwe of *U worke in each group. Wage differentials between men and women by occupation are shown in Table 16.3C. Men are concentrated in ^rtisanry, agriculture, and personal services, while over half of the f;males are employed as profesa,onals and teachers or personal service providers. Education levels are lowest in agriculture, personal services, and artisans. The ratio of female to male wage is low less than 65 percent) for personal services, sales people, and artisans. These professions make up about 50 percent of the female labor force. Professions with a very high rato (agriculture and transport) make up a very small share of the female )rk force. The wage for female office workers and unclassified workers is higher than for male workers in the same categories. In each case where the female wag is higher than the male wage, females in that category also have more years of schooling than the males. In some of the categories, males make more than females, but also have higher schooling, such as pe.-onal services, sales, artisanry, and personai services, which could explain the differer.tial. -or the categories of professionals and administrators, where there is little difference in schooling or females have more schooling, the. t'ale wage rate is higher, and that is not readily explained. However, the grouping aprofessionals' inc!uats school teachers, who are low paid compared t2 others with university degrees. The differential has a large impact 'an women with university education and very low education. Table 16.3C Wage Differenfials by Occupation Male as a Female ar a percentage of perentage of Work Yus. Work Y. Ratio Few/) Profession Wage Force Ed. Wage Fore Ed. Male Wage Prcfessionals 4.46 11.3 14.84 3.05 20.8 14.80 .69 Administrtors 3.83 7.8 12.^-. 3.38 3.7 13 4 .88 Office Workers 1.79 6.2 11.29 i.89 24.2 12.32 1.06 Sales People 1.71 9.4 9.24 .99 10.0 8.52 .58 Agriculture .92 14.2 5.38 1.72 .5 5.52 1.87 Transport 1.63 9.5 8.62 2.98 .1 11.51 1.83 Artisans 1.70 18.6 8.80 .89 5.7 8.62 .52 Clothiag, Furniture Other Artisans 1.34 3.6 7.43 1.48 1.2 9.10 1.10 Unc!assifiad 1.22 6.7 7.43 1.26 1.2 8.09 1.03 Workers Persona Services 1.26 12.6 7.82 .78 32.5 7.04 .62 Overall 1.97 100 9.21 1.67 100 10.45 .85 In this section, the resutlts of a univariate probit regressinn are discussed. The probit was estimated separately for men and women. The need to estimate the probits arises because the working men and women are a seltked subset of 2l1 men and wo.men. They are people who obtained wage offers higher than their reservation wage. Reservadtin wages are affŽctexi by tastes, age, and schor'ing. For example, an individual with high education would have a higher resJ'rvat12n wage because of raised expectations and would be less willing to take a lower paying job than someone with low education. A woman who has y)ung childrmn would be less likely to work than a woman who has no children beeause her time in tha household is more valuable to her. An individual in hizr or her teens wou.d be less likely to be in t&. -work force because of opportunities fir schooling, and a long work life ahead to earn a return on schooling. A man in an urban area may have a lower reservation wage than a man in a .iiral area because in urb n areas, goods that may have been rcadily available from a smaHl plot of land in a rural area must be bought. For exawple; an urban work;er may have to buy the fruits, subsistence food or firewood which he could gathef or culdvate easily if he lived in the countryside. In order to obtain unbiased estimates fo; !he reurn to schooling, experiecce, hours, and tenure, it is necessary to correct for selectivity. lr.N is Jo:. using Heckman's (1979) twsep procedure. The probit includes as independent variables schooling levels, age levels, regions, and variables that represeut the structure of the horsehold. lbz inverse Mill's ratio (Lambda) is computed in order to acwunt for the unsee variables that affect the decision to worl. Then, La;nbda can be included as a regressor in an ordinary least squares (OLS) regression. A positive value of the Lambda coerticient implies that characteristics that make aa individual more likely to be in the work force aLso lead to higher earnings, while a negative vahJe means that characieristics associated with staying out of the work force imply higber earnings. An example of a characteristic that wouild explain a negEtive Lamb t coaefficiect is higher education, because it increases the reservation wage, decreasivg the poA ability of work force partic;pation, while higher schooling also eam-s a compenaiDg differential in earn;ngs. Because such a high percentage of men do not _ave Murs or income reported, a prubit regression is estimated for men as wedl as women. The probit for men includes the same independect vaiiablus as the probit for vjmen in order o nake comparisons .beween them. A prioi, the research!er would expect that the number of children would have a% strong negative effect cn the female particpa.3ictn decision, becaus females traditionally carry more responsibility in the bousehold for child care. living in a nrual area woulC bc likely to decrease participation ior both males and females. One would expet the age group to have less of an effect on male Darticipax:on rates than female raies becauze. male participation rates are consistently high, while femalew have more elastic labor supply. Tables 16.4A and v.o43 present the results for the probits for men and women. Table 16.5 presents a simulation where the effect of eachi characteristic on the probability of work force participation is eziamined. AUl other values are held at the satnple mean, so that the effect of only the relevant cdaracterisic can be detmined. Fist, it is evident for every characzsistic that the probability of a give male being in the work force is higher than for a given female. Looking at education levels, for females the liklihood of working increases with higher education luvels. -n Table 16.5, participation rates incease from 10 percent for dtose with no education to 48 petcent fot those with over 4 years of university edi-cation. These resvlts contrast with the results for nales, where those w ith technical education have the hi,ghest likelihood of participation 1 ante 16.4A Probit ResuIts for Male Work Force Participation Variable Coefficiea: T-Ratio Partial DMrivative Coustant -2.808 -29.08 EducatiogLevels Suma Primary .200 2.80 .079 Complete Primazy .560 8.06 .223 Some Seconday .651 8.80 .259 Secondary .737 9.47 .294 l;echaical .812 7.42 .324 less ihan 4 Y. Uni-'. .453 4.93 .181 4+ Yrs. Univ. .683 7.32 .272 # of Cildrea 0 to 6 .041 2.81 .016 J of Children 7 to 12 -.034 -2.35 -.013 Ane GrOUD Age 20 to 24 .9C,2 18.71 .360 Age 25 to 29 1.158 21.88 .462 Age 30 to 34 1.186 20.05 .473 Age 35 to 39 1.210 19.1i .482 Age 40 to 44 1.150 17.28 .458 Age 45 to 49 1.014 14.70 .404 Age S0 to S4 .826 il.44 .329 Age 55 to 59 .649 8.26 .259 Aga 60 to 65 -.030 -.39 -.012 Reon Bocas del Tom 1.358 12.81 .541 Cocie .182 2.78 .072 Colon .632 9.15 .252 Chiriqu: .67S 11.88 .269 Darien -.338 -2.28 -.13S Herrera .243 3.23 .097 Los Santos .469 6.00 .187 PParina .493 9.S4 .196 Rurl -.276 -8.32 -.110 Head of Household .900 23.30 .359 Total Household Income .XtC 13.69 .000 Numt,er of worke .270 21.13 .107 in Household Notes: D'pcndcut Varible for probit is wiatber ididual reorted posti hours and posirve ncome. Bue group is no education, age 15 to 19, Eving in Vargu. Probit Results for Femal7 Womk Force Participation Variable Coefficient T-Ritio Partial Derivative Constant -2.74', -24.57 Education Leve Some PrimLy .206 2.28 .064 Complete Pimay .609 7.10 .191 Some Secondary .489 S.56 .153 Secondary .857 9.51 .269 Tochnical .649 5.44 .204 less than 4 Yrs. Univ. .837 8.43 .263 4+ Yrs. Univ. 1.224 12.1 i .384 Nof CVjen 0 to 6 -.098 -6.L7 -.030 Iof Childr 7 to 12 -.1)47 -2.97 -.014 Aye Grwo Age 20 to 24 .460 8.28 .144 Age 25 to 29 .850 14.71 .267 Age 30 to 34 1.151 19.43 .361 Age 3S to 39 1.206 20.13 .379 Age 40 to 44 1.010 16.13 .317 Age 45 to 49 .826 12.26 .259 Age 50 to 54 .426 5.78 .134 Age SS to 59 .056 .65 .017 Ag 60 to 65 -.113 -1.24 -.035 R&in Bocas del Toro -.109 -.36 -.034 Cocle .110 1.40 .034 Colon .190 2.46 .0a0 Chiriqui .035 .52 .011 Darien .057 .31 .018 Hecnr .161 1.86 .050 Los Santos .181 1.95 .057 Panam .12 2.06 .039 Rural -.388 *10.87 -. :22 Head of Household 1.008 22.91 .316 Tota Household Incomo .000 9.94 .000 Number of workes .450 12.05 .141 in Household Notes: Dpndentvai for probit is wbhc inmivim rpofled pokive hour and poive in-oe Ba group is no education. 4e 15 to 19. living in preino of Veaguas. Prodkied Paidcipsaim Probabilities by Chracteristic Charactristic Male FnMale bAcaon No Educatio .30 .10 Some Primay .37 .14 Complete Primary .51 .25 Some Secoodauy .55 .21 Complete Secordauy .58 .34 Technical .61 .26 1 to 3 Yr Univetsity .47 .33 4 to 6 Ys Univensity .56 .48 Number of Childr= AMd 0 to 6 None .49 .27 one .SI .24 Two .53 .21 Throe .54 .18 Numbsr of Cutldren Ad 7 to 12 Ncoe .51 .26 OCo .50 .24 Two .49 .23 Threeo .47 .21 Aim IS to 19 .22 .10 20 to 24 .55 .20 25 to 29 .65 .33 30 to 34 .66 .44 35 to 39 .67 .46 40 to 44 .64 .39 45 to 49 .59 .32 50 to 54 .52 .19 55 to 59 .45 .11 60 to6S .i. .08 Bocas del Toro .82 .18 Cocde .40 .25 Colon .57 .27 Chiriqui .59 .22 Darien .22 .23 Herrera .42 .26 Los Santos .51 .27 Panama .52 .25 Venraua .33 .21 Live Rud fArea No .SS .30 Yes .4S .18 HSMehold HeOd NO .33 .20 Yea .67 .57 Number of Wodrws in HouSho4d None .31 .08 One .41 .16 TWO .52 .30 Tmuee .63 .47 ,AA VA FQA%rAiAt. A* V LAVU, LMLb %AJUIU LrL1cQ VYC4 "Irp 1Vc1viuufl wases w or Lose Wno auemeX the university due to raised epectations, vwhile women with university education have a low reservation wage due to a strong preference to work outside the home. Another contiast is the effect of the number of children on work force participation. Women with children aged 0 to 6 are le.. likely to be in the labor force than women with no children, while men with young chiE.wren are more likely to be in the labor force than men without children. r ie probability of participation for females drops from 27 percent for women with no children aged 0 to 6 to 18 percent for women with three children in that age group. For men, the probability increases as the number of young children increases from 49 percent with none to 54 percent with three small children. Women are caring for children in the home, while men --e earning outside the home to support the family financially. The number of children aged 7 2 affects men and women about equally. The effec is small; for both men and women the -o.nability drops about 4 percent when the number of older children is raised from none to three. is for the effects of age, both men and women have peak work force participation rates between 35 and 39 years of age. lhe female pattern is more concave than the male pattern, with female participation rates dropping at a younger age than male l artic&pation. The probability for women drops from 32 percent to 19 percent between the 45 to '9 and the 50 to 54 age groups. For men, a dmp of this magnitude occurs between the 55 to 59 and GsO to 65 age group, where the participation drops 14 percentage points. Tris is expected given the discussion above, and given econometric labor supply studies, which find that female labor supply is more elastic than male labor supply.' The regional variables affect male participation rates strongly, while for females, only 2 of the 8 regional variables are significant at the 5 percent level. The two significant variables, Panama and Colon, are the most urbanized provinces in Panama. For males, every regional variable is significant at the 5 percent level, with probability of participation the highest at 82 percent in Bocas del Toro, a middle income province, and the lowest at 22 percent in Darien, a low income province. This can be explained by the fict that men in the poorer regions are more likely to be self-employed agricultural workers, and therefore excluded from the sample of working men. Both male and female participation is affected by whether the individual lives in an urban or rural area; for both the coefficient is negative and significant, and the effect is larger for women. Living in an urban area implies for men an 11 percent greater probability of working and for females, a 12 percent greater probability of working than living in a rural area. This is consist with the prediction made above. Tle variables which proxy household structure, total household income, whether the individual is the head of the household, and the number of occupied people in the household, are all positive and significant determinants of both male and female participation rates. For women, the probability of working increases from 20 percent to 57 percent if she is a household head. For males, the corresponding percentages are 33 percent and 67 percent. For both men and women, with more workers in the household, the probabDlity of the individual working increases, and the increase is more for women than men. What effect do these variables representing household structure have on labor force participation? A member of a richer household is more likely to be working than a member of a poorer household. The richer have greater access to the formal labor market. Jnce again, this could s See Killingsworth and Heckman, (1986). ov uuc w u1o mLssing ua ior ptucr i&tmers aw ineir iamuines. AS tor une temanes, twere are probably interactions between high education level and high household income, since we;l- educated women tend to marry well-educated men with high earning potential. Household headship is an important determinant because household heads bear more of the financial responsibilities in the family. Lastly, it is not self-evident why the number of occupied people In the household has a positive effect. With more employed members of the household, a given individual has less need to work, but families with many workers may be poorer families that must se.ld children and women into the work force in order to maintain their living standard. It could also mean that there is a family 'work ethic' with members preferring to work outside the home. There are important differences between men and women in how characteristics affect work farce participation. The differences are greatest with respect to number of small children, education levels, and regions. In the next section, the results of these probit regressions are used in earnings functions to correct for selectivity. S. Earnings Functiors In Mincer's (1974) model to estimate earnings regress!ons, the log of eain!ngs is regressed on schooling, experience, and experience squared. Tne earnings function is concave and increasing in schooling and experiencc. In this section, the model is used to estimate separate male and female earnings functions, both correcting for selectivity and using the most basic model. In the case of Pananma, it is rossible to use tenure instead of potential experience as the regressor. Tne latter experience variable is usually calculated by taking age, subtracting the years of schooling and subtractig six, which is the age when schooling is assumed to begin. This proxy is likely to be upwardly biased, especially for women, because it does not take into account years spent out of the A srk force since the completion of schooling, nor the deterioration of experience wilen a person stops working for an extended period of time. Women are likely to have interrupted careers if they have had children. For this reason, tenure, which is the amount of time spent at the present job, is a better proxy for women of hunan capital acquired on the job than the estimate of experience. The drawback of using tenure is that if workers change jobs frequently, it discounts accumulated experiene which transfers between jobs. In Panama this presents less of a problem than in the United States where workers are mobile and change jobs readily. There is a high unemployment problem, which makes workers less likely to quit jobs and search for better ones. Log of monthly earnings is the dependent variable, while years of schooling, tenure, tenure squared, and the log of monthly hours are the independent variables. Including the log of hours on the right hand side of the equation rather than using the log of hourly wage as the left hand side variable allows the value of the elasticity of earnings to hours to differ from one. To correct for selectivity, according to the Heckxman procedure, the inverse Mill's ratio (Lambda) is included as an independent variable in the earnings regression. Table 16.6 presents the results of the regressions for males anr females, both including Lambda and excluding Lambda. From the table, it is evidert in the uncorrected regressions that females earn a higher return to schooling and tenure, while men have a higher elasticity of log earnings to log hours. Both exhibit a concave earniDgs profile, with decreasing returns to tenure. The rate of return to education for females is almost 3 pecent higher than for males, and the rate of return to tenure Table 16.6 Earning. nimcdons Males Females (Corrected for (Cofr..cted for Males Slectivity) Fenmaes Selectivity) Constant .722 1.951 .45S 1.098 (6.130) (16.771) (3.982) (9.018) Schooing .097 .072 .119 .098 (Years) (47.692) (27.312) (46.360) (30.363) Log Monthly .659 .561 .599 .589 Hours (29.187) (27.366) (26.741) (27.169) Tenure .079 .056 .103 .089 (27.688) (19.729) (25.384) (21.578) Tenure -.002 -.001 -.003 -.002 Squared (-16.866) (-11.016) (-17.131) (-14.594) Lambda -.679 - 389 (-20.586) (-12 276) Adjusted .456 .524 .605 .629 R-Squared N 5,445 5,44S 3,189 3,189 Notes: Numbcn in pirentmis are t-atios. Dependent variable is log of monhly income. is about 2.5 percent higher. A one percent increase in hours leads to a .66 percent increase in earnings for men, and a similar increase in hours leads to a .60 percent increase in earnings for females. The fit of the regression is better for females than males-the R squared is .61 for the female regression and .46 for the male regression. When Lambda is added to the male regression, the return to schooling drops from 9.7 percent to 7.2 percent. Similarly, the return to tenure decreases from 7.9 percent to 5.6 perent. The elasticity of earnings to hours also decreases. Tle coefficient on Lambda is negative and significant, which means that charaistics that earn a higher return also make a man less likely to be in the work force. This could occur because men with high educational levels are less likely to work due to a high reservation wage and lack of jobs which meet their qualifications. For females, when Lambda is included in the regression, returns to schooling drop 2 percentage points from 11.9 percent to 9.8 percent. Retrn to tenure drop from 10.3 percent to 8.9 percent. The elasticity of wages to hours worked fills marginally from .6 to .59. The coefficint on Lambda is also negative and significant, but the value of the coefficient is much less negative thdn for the men. Again, the characierstics women have which allow them to earn a higher return make it less lik-ely that they will be observed in the work force. Women with qualifications that earn high returns in the work force have a high reservation wage and prefer to stay at home. Tle negative Lambda could also be caused by high unemployment rates, which are hip'ler for women than men. In Table 16.7, unemployment rates for men and women by education levels are shown. Tnhey are based on what the person reported in the bousehold survey, and indicate the proportion of those who say they are unemployed over the total who report being employed plus the unemployed. Unemployment rates are highest for those who have secondary school level education. They are very high for women and reach 29 percent for those with a secondary school level education. Surprisingly, they are lowest for those with low education, but these workers are likely to be self-employed, or family workers, and therefore report themselves as being employed, while they may not have rerorted hours or income. Table 16.7 Unemployment Rates by Education Levels Education Level Males Females Len than Primary 6.4 13.7 Primary 9.6 15.3 Incomplete Secoar 21.6 28.3 Secondady 20.1 29.0 University 14.5 16.7 Technical 24.9 21.6 Not.: Unemploymcat rdics are baFd on resporndnts' ansive, and arm the mio of unemployed to the hbor force patcipants. Participants include both toe reported u cmpyoed and a unemployed. In Table 16.8, the results of an alternadve earnings specification including the sector of employment are p.esented. Including the variables self-employed, employer, and public sector wo: er does not have a large effect on the coefficients of the other variables. Compared with Table 16.6, returns to education fall by about one percentage point for men and women vithou correcting for selectivity, while they fall about two percentage points for women when coTrecting for selectivity. Ret:rns to teaure are about one perceatage point lower for males and two percent lower for females. For men, those who are self-employed earn between 23 and 25 percent less than private sector employees, while working in the public sector increases wages by between 27 and 26 percent. Employers earn the highest wages, earning 51 to 58 percent more than the base group. When Lambda is included in the regression, it decreases the coefficient on the sectors slightly. For women, being self-employed implies 35 to 36 percent lower earning?, while working in the public sector earns a 34 percent premium compared to pre sector employees. Again, the highest earnings are gained by employers, ranging from a 57 to 62 percent premium. In regressions both corrected and uncorrected for selectivity, the sector choice has a larger impact on women than men. Table 16.8 Alternuive Earnings Functons Males Females (Cofreted for (Corrected for) males Selecivity) Females Seectvity Constant 1.098 2.322 1.2489 1.912 (9-447) (20.374) (9.875) (9.875) Schooling .087 .062 .1011 .080 (Yeas) (42.355) (23.831) (38.880) (24.985) Log Montly .611 .513 .4828 .469 Hours (27.678) (25.783) (20.351) (20.656) Tenure .0640 .042 .0799 .066 (22.161) (14.753) (19.484) (16.043) Tenure -.001 -.001 -.0m20 -.002 Squared (-13.056) (-7.205) (-13.099) (-10.553) Public Sectoe .275 .261 .344' .339 (12.313) (12.169) (13.218) (13.281) Employee .586 .507 .6233 .566 (10.275) (8.767) (5.621) (5.167) Self-Employed" -.231 -.249 -.3455 -.358 (-zO.178) (-11.803) (-9.800) (-10.486) Lambda -.673 -.389 (-20.926) (-12.791) Adjusted .497 .564 .6463 .670 R-Squared N 5,445 5,445 3,189 3,189 a. Base group is privae wseor cmployces. Notea: Numbers in parnthcais arm t-aios. Dependent variable is bh log of moxxhly carning. 6. Disarimination The upper bound on wage discrimination can be found using Oaxaca's (1973) equatons: ln(Earnings,) - In(EarningsJ = X(b.-b + bt(X,-XJ (1) = X(b.-bj + b3(XC-Xj (2) Where X. represents the means of the dependent variables for males, X, represents the means of the dependent variables for feanles, b, is the matix of estimated coefficients for males, and bf is the matrix of estimated coefficients f,r females. Both equations give the differential between the predicted values of earnings for males and females, b,X.-b,X. Tle first term on the right hand side in equation 1 gives the part of the differential that is explained by differences in how male and female human capital endowments are rewarded in the labor market (wage structure) evaluated at the male means. The second term calculates the part of the differential due to differences in the means of the dependent variables of men and women (endowments), multiplied by the female coefficients. Equation 2 is the same breakdown, but calculated at the female means rather than the male means. Tnere is an index number problem with the two equations. However, it makes more sense to evaluate the differential at the male means, since this paper is examining potential discrimination against women. In calculating the percentage of the differential due to endowments and to wage structure, the means of the entire sample of men and women are used for schooling, and the means of working men and women are used for lo-, hours and tenure. Both working and non-working individuals have reported schooling, but onily working men and women have positive hours and years of tenure. Neither the Mill's ratio terms (LamdA7) nor their coefficients are included in the equation because the parameter of interest is the mean for the whole sample, not just working men and women.2' Table 16.9 shows the calculations of the Oaxaca decompositiou using the regression coeffic.ents correcting for selectivity. There is not a large difference between the calculations evFluated at the mnle means (eq4ation 1) and at the female means (equation 2). From 14 to 15 percent of the differcatial between male and female wages can be explained by endowments, while 85 to 36 percent are explained by the wage structure. However, it should be noted that the differential due to wage structure is an upper bound on diszrimination. If there ale attrbutes not measured here that are valuable in the labor marke;, and men have these attributes in greater quantity or quality than women, the upper bound on discrimination will be upwardly biased. However, if there are societal characteristics that are not measured here preventing women from entering the labor force or inhibiting women from acquiring human capital, the measure of discrimination .il be underestiaed. Tabwe 16.9 Decomposition of Sex Earnings Differential PerAgew of the Diffential Duo to Differences iD Specification Endowments Wage Stru;ctu Total Corrected for Selectivity Equation 1 14.7 85.3 37 Equation 2 13.9 86.1 37 Note: The decomposition is based on the results of Table 16.6. 2 See the chapter in this voluma by Psacharopoulos on Vemml. 7. Discussion Despite high educational attainments for women, Panamra's female participation rate is low. A very important impediment to equality is difficulty in finding a job indicated by the incredibly high unemployment rate women face. Ihis is especially a problem for women with secondary level education. Ile overall ratio of female to male w-.ge in Panama is favorable compared with other Latin American countries, such as Uruguay, Venezuela and Peru.' However, at the minimum 85 percent of the differential cannot be explained by differences in endowments between men and women, and can be attributed to wage stucture discrinmination. A topic of further study would be whether the Labor Code has impazted female wages favorably. The evideace presented here indicates that for women who have jobs, the wage gap is small zelative to other countries, but that the labor code also discourages cmployers from hiring women because they must provide benefits that are specific to women. Employers perceive that it is cheaper to hire men. The code could be reformed to provide more flexibility to employers for providing maternity benefits. Laws designed to help women may actually hurt them. The choice to work as an employee or to be self-employed has an important implication for earnings for both men and women. Individuals with lower levels of education seem to earn higher wages as self-employed workers than as employees. To alleviaze high unemployment, credit could be extended to women who would like to be self-employed. However, in the ling rn, sound economic growth wold provide private sector jobs. In the sample, only a small percentage of working women are self-employed (15 percent) and with a big increase in the nmmber of self-employed women, undoubtedly their wages would decrease. With high enough economic growth, employment in the private sector would increase and wages wouJd increase to reflect productivity growth. Labor market discrimination seems to be more of a fiator for women with very low educational levels and relatively high educational levels. This could be because given the high unemployment rates for women with secondary education, only very qualified or determined women can get jobs, and, therefore, their wages are a higher percentage of men's wages. Better enforcement of anti-discrimination laws would help those women that are private and public sector employees earn a wage more equal to male employees. n See Kandker. Arends, and WiLntr (in thi volume) who repot the ratio to be about .67, .75 and .78 in Peru, Uruguay, and V aemel, rpectively. lil'ferences Economist Intelligence Unit. Mcaragua, Costa Rlca, Panama: County Profile 1991-92. London: Economist Intelligence Unit Limited, 1991. Heckman, J.J. 'Sample Selection Bias as a Specification Error.0 Economerica, Vol. 47, no. I (1979). pp. 153-161. Heckman, JJ. and V.J. Hotz, 'An Investigation of the Labor Markez Earnings of Panamanian Males: Evaluating the Sources of Inequality. Journal of Human Resources, Vol. 21 (1986) pp" 507.42. Killingsworth, M.R. and J3J. Heckman, 'Female Labor Supily: A Survey' in 0. Ashenfelter and R. Layard (eds.) Handbook of Labor Economics. Amsterdam: North Holland, 1986, pp. 103-204. Mincer, I. Schooling, Experience, and Earnings. New York: Columbia University Press, 1974. Oaxaca, R. 'Male-female Wage Differentials in Urban Labor Markets.' Intenational Economic Review, Vol. 14, no. 1 (1973). pp. 693-701. Spinanger, D. 'Labor Market in Panama: an Analysis of the Employment Impact of the Labor Code. A7el Working Ppve, No. 221, December 1984. Tollefson, S. 'lTe Economy' in S. Meditz and D. Hanratty (eds.) Panama a Couwy Study. Washington, DC: United States Government, Department of the Army, 1989. pp. 123-171. World Bank. Panam: Stmcusral Qu'znge and Growth Prospecds. Washington, DC: World Bank 1985. World Bank. Second Stracsural Adjsmrem Proetea-Panama. Latin America Division, Panama Desk, 1986. World Bank. Socil Indicators of Dewlopmenm 1990. Baltimore, MD: Johns Hopkins University Press, 1990. p. 238-9. World Bank. World Developmern Repors. New York: Oxford University Press, 1991. 371 17 Women's Labor Market Participation and Male-Female Wage Differences in Peru ShahiU R. Aandkd' I0 Introduction This study uses Peruvian Living Standa&d Survey (PLSS) data to estimate women's labor market (i.e., wage) participation, and male-female differences in peoductivity (measured by wages). The purpose is to (1) identify thosa characzeristics that enable some women, though not as many men, to participate in the wage sector, (2) determine the private economic returns to education by gender, and (3) evaluate how much of the male-female wage gap is due to differences in human capital. Identifying the constraints to women's labor market participation and productivity is an important policy exercise in Peru where female participation rates are below the average for the region (Suarez-Berenguela, 1987). Results indicate that gender differenc in human capital, such as education and experience, account for some of the observed differenca in labor market participation and productivity. Estimates of the returns to education show the private rate to be generally higher fnr women, especially for secondary school level and rural areas. However, school enrcllrmert rates for females are lower than for males, indicating that parents invest less in .emale than male children. The study uses a household model framework (Becker, 1965) that provides an estimable labor market participation equation. T1is equation can help estimate the relative effect of individual, household, and market factors in influeniing an individual's labor market participation. The study uses, In addition, a human capital model as per Becker (1964) and Mincer (1974) to analyze wages in the wage sector. The focus here is on human capital variables such as education and e!perience as determinants of productivity. Wage estimates provide measures of the private rates of returns to education for men and women and can be used to identify how much variation in male-famale wages is due to gender differences in education and other job-related characteristics. The human capital model, however, may iLt satisfactorgy explain variations in wages since productivity is likely to be determined by a number of factors including, but not limited to, human capital variables. A satisfactory analysis, therefore, requires identifying potentially I Comments on on arlier draft by George Pscharopoulos, Zafiris Tmnatos, and Ind-nait Gill a grmtefiuly acinowledged. I am also indebted to Iorgp Castillo who provided excellent assista_ in analyzing the data The usa disclair pplies. 375 observable charncteristics other than human capital that can affect an individual's wage. Unfortmately, these are not clearly understood and are thus difficult to incorporate in the analysis. There are, however, ways to reduce the impac of unobserved characteristics. This study uses a sample selection correction technique to estimato the severity of sample selection bi. in the wage estimates that may arise because the analysis is restricted to wage earners. Tlis procedure determines whether sample selection correction significantly alters the wage estimates and hence the estimates of the returns to education and gender differences in productivity. The chapter is structured as follows. Section 2 brPefly describes the Peruvian economy and labor ma-ket to ilustrate women's positio.i in the overall economy. Section 3 discusses the data and highlights the differences between niales and femr.nles in terms of wagc>related characteristcs. Section 4 discusses the determinants of female labor force participation and Section 5 the wage determinants and returns to education. Section 6 discusses the extent to which male/female earnings differentials can be attributed to diffre.ces in the way the market structure rewards male arnd female workers. Policy implications are in the concluding section. 2. Pe-a: Economnic Background and Women's Status in the labor Market Peru is a mriddle income country with a pc capita income of US$1300 in 1988. The economy is heavily dependent on mineral resources which are its major export goods. It also has considerable potential for fishing and hydrocarbon resource Jevelopment. The country is geographically divided Sy the Andean msuntains into three regions-the highlands (Sierra), the rain forest (Selva) and the coast regions. About half of Peru's 20 million (1.87 estimates) population lives in the coastal regions, 40 percent in the Andean highlands, and the remainder in the Selva (i.e., Amazon) region. Table 17.1 presents data relating t.z employment, unemployment, and labor productivity in Peru during the pe.riod 1970-85. Between 1970 and 1985 the labor force increased from 4.2 million to 6.6 million, of which about 88 percrnt were employed. During this time the employed labor force increased by 46 percent with 28 percent of that intitease being employed in agriculture. Unemployment increased by a'out 7 percentage points over the same p:riod. Column 4 in Table 17.1 gives figures on adequate employment for the labor force who are employed. According to these figures, the rates of underemployment in Peru ranged between 46 and 54 percent of the labor force in 1970 and 1985, res extively. Thus, underemployment is more serious than open unemployment in Peru. Labor productivity, defined as value added per employed person, is also a problem in Peru. As colutmn 5 indicates, labor productivity declined in Peru between 1981 and 1985. Tncreasizz labor productivity and adequate employment opportnities are significant problems in Peru. In Peru women's labor force participadon rates are below hose ir. many Latin American countries (Suarez-Berenguela, 1987), but it has increased between 1970 and 1985. For example, women's labor force participation increased from 34 to 43 percent in urban Peru.2 Women's all-Peru participation rate is 57 percent for 1985 (Schafgans, 1990), tbeir participation being much higher in rural than urban areas. Of all economically active women, about 18 percent wor's in the wage sector, 55 percent are farmer, and 27 percent work in the iformal (non-farm) activities. Women are thus predominartly in agriculture and self-employed non-farm aci 'ities. Table 17.2 shows that women are predominanly employed in the informal sector; in 1985 about 2 So T&ble 23.2. The figures wre calculated as: ColuQmn 3 x Coluum 11100. Ureaqpoymcnt Emoymen TOWa Rsaw (Paroat L,bor nf dth LAhbc Noa- Adoqu&t Labor pone Por") Tot: Asicu1m Aicukw Empk-yucne ProycduW (1) (2) (3) (4) (5) 170 4, 167.3 4.7 3,971.4 1,873.6 2,097.8 2,05-.9 613.4 1975 4,617.5 4.9 44581.3 1,950,0 2,631.3 2,537.7 676.2 196W0 5,C7.2 7.0 5,21i.7 2,046.0 3,164.7 2,341.4 680.3 1'8l 5,77.0 6.? s,377. 2,2727. 3,105.3 2,613.7 684.2 198. 5,958.0 7.0 5 ,4..0 2,328.1 3,211.9 2.567.5 666.7 1983 6,136.7 9. 5,58s.7 2,355.1 3,230.6 2,306.9 S85.5 19&4 6,35 1. IO.; .5,64.4 2,374.9 3,302.S 2,242.0 603.2 )85 6,555.5 11.8 5,781.9 2,397.7 3,389.2 2,235.4 609.C t. Thou*"d of paona. t'. Mm MInwty of Labor ckLusi& a paio as umkwzmpjrpe if w%akty w'rkias bcus anr bam %han 35 andlor L%o mA= i tu than it ')rl Miw'nM %AgP dJUBC'd Ł* inuOo. c. Va)zo Addcd oaMi aof1V'' v cm pcwm Scum.c: MinisKz of ' .abor and i ... e . Saati2 lr*b, 19?').985. Tabb 17.2 Sector. Dutim ef bK Fcoauaiy.uuw PqF!at&m (EAP) An Urtan Pot by Gander, 197045 (1) (2) (3) DibmjR Distriton Women as Sador of EAP of Feowe of $Clot21 EAP LAbor Fore 1970 198S 1970 1V^ 5 li3 1985 t$& *L iM .2 46S 2. S,dPtlpyed 36.0 51.7 23.1 19.a 54.4 A4.8 Ewployeas 16.5 13.1 9.3 9.0 22L0 24.9 Dcmes6c worker 6.2 1.6 39.5 33.6 93.0 929 5rnw 4LIU W ZLA MLh 11 Al '6' i*oo 9.7 8.0 12.7 16.1 22.0 38.2 :ffiOY. . >oohr1L4 9.1 3.4 7.6 7.0 MO0 Lveam*t Uo2qo10 13.2 15.4 i4.! i3.9 2S.Q 0 2.8 Now0 S 9_lde 19psi womLS. Source: soifm.Dugpe~a 191 anW PUS d&m. informal seclor compared to 10 percent in the formal sector. Women's wages in the formal sector are lower than men's. Women's employment options end productivity perhaps reflect their low education and employment opportunities. Peruvian women have about five years of schooling or average compiared to seven years 'or men. Only 8 percent of men did not attend school compared to 25 perment of women (King and Bellew, 1990). ITis paper attempts to nccount for differences in women's labor market participation and wages using household sunrey 'ata fiom Per. 3. Data Characteristics The data are drawn from the Peruv;an Living Standard Survey (PLSS) household data collected jointly by the World Bank and the Peruvian Instituto Naciotial de Estadistica (INE). These data provide detailed socio-economic information on over 5,100 households and 26,000 individuals. Tle samDles were drawn from a self-weighted national probability sample of Peruvian households and represent aa approximate 1/100 sample of the population. The samp!mng fra.ne is based on a 1984 National Health and Nutrition Survey. About 25 percent of the households in the PLSS were in metropolitan Lima, 28 percent in other urban areas, and 47 percent in rural areas. The data were collected batween June 1985 and July 1986 (see Grootaert nd Arriagada, 1986). The s-mple includes workers aged 14 to 60. The wale earner participation equation is estimated using information fa; aI potential workers. The wage equation in Section 5, however, is estimated ordy for men and women reporting wage *i remuneration and bours worked tnd who list this as their main occupation in the week prior to data cotlection. Self-employed and unpaid family workers are thus excluied. This reduced sample wnsists of 2,255 men from !,56 households and 898 women from 783 households, drawn frorn a total of 6,429 men from 4,142 households and 6,942 women from 4,387 households. The wage labor market participation rate is 13 percent for women and 35 .ercent for mel. Table 17.3 gives the mea.ns and standard deviatiors of the variables bv geider. Women wage earners have one more year 3f educatirm on average than men wage earners. Enmployed women also have more vocational training - 52 percent of women iave training compared to 31 percem of men. Despite thi ;, wornm', receike about half of men's wages.3 This 'suggests Fliat there are significant differences in ^zge stractures between men and women. Occupational segregation may cause this n-ale-female wage gap. Women also come from relatively wealthier howmeholds In terms of ladiholding and unearned income). The data suggest that more married (or cob Lting) men participate in the labr market th3n married (or cohabiting) women. Employed women are also vounger on 2verag' tbau r'snployed men. 4. Determlnants of Female Labor Force Parlidpaston What influences women's participation il the labor market? Do women differ from men in responding to labor nmarket opportunities? Does human capital (fof instance, educadon) help woomen, more than me, to participate in the wage sector? Do women face different market ) The real Sourly wage rte, i.c., nominai Lsiry wagg deflted at 1985 consumer price indices (RHW) is defined as R!iW m -C/AA wbeme AC - anna conpe-sation = mowhly pay x months wurked in the past yer AH --. c al hous - weekly hou . months worked in the past year x 4.33. Note ala the above mr,e-femnale wage difvenco is adjusted for male-female sampl differences. Mes (an S&adad Dviions) of Sample Variables Females Males Variables Werking AUl Woridng All for wae fur wage Number of obsavations 898 6,942 2,255 6,429 Real hourly wage rate5 3.1U4 1.152 3.820 1.601 (1.684) (2.545) (2.330) (2M) Education Years of chbooli 9.013 S.606 8.212 6.991 (4.272) (4.327) (4.143) (41 Primay 0.226 0.321 0.343 0.390 (0.419) (0.467) (0.4;'5) (OA8 Seondauy 0.423 0.222 0.345 0270 (0.494) (0.416) (0.476) (OA4" Post-Socondary 0.189 0.049 0.132 0.074 (0.392) (0.215) (0.339) (0Z4 Vocational Traing O.S18 0.239 0.309 0.195 (O.S00 (0.427) (0.462) (0390 Secondary temical dipoma 0.031 0.012 0.024 0.014 (0.174) (0.109) (0.153) (0.111 Post-Scondary diploms 0.074 0.019 0.032 0.017 (0.261) (0.135) (0.177) (0O. University diplonm 0.117 0.025 0.082 0.042 (0.322) (0.156) (0.274) (O;X) Attended public school - 0.758 0.691 0.847 0.838 (0.428) (0.462) (0.350) (0.I9 Age 30.871 32.056 33.432 31.057 (9.816) (12.566) (11.354) (12.71M Married or coh3bitin3 0.408 0.SS6 0.624 0542 (0.492) (0.497) (0.485) (0AE TJneaned rel i3come (xlO0O) 2.98) 1.796 2.164 1.800 (8.555) (6.904) (11.433) (9.8n LaIdholding (becuts) 1.673 3.6S3 1.624 1.899 (35.328) (49.515) (19.770) (S1.7nr OUA reidenc, 0.313 0.305 0.324 0.302 (0.464) (0.460) (0.468) (0A4) Rural rsideanc 0.147 0.397 0.235 0.402 (0.354) (0.489) (0.424) (0A9% S. Ind at June 1985 prie Notes: Numbes in puwexa are gada dcvhdoa. Sowu Peru Living Standard SrVe, 1986 participation in the labor market. This section outlines a theoretical famework tr, address women's labor market participation and reports the results from Peru. The decision to join in the labor market, given the constraints, is basea± on an individual's income-leisure trade-off. A household model framework can help identify the constraints that affect an individual's allocation of time (Becker, 1965). Tlis model identifies individual characipristics such as education and experience, household characteristics, including landholding and unearned income, and market conditions, such as wages, which may influence an individual's allocation of time. Thus the time allocatod to different activities, including leisure, can be written as a function of individual, household, and market characteristics. The time allocation data can produce a discrete choice structure of whether or not individuals participate in the wage market. The decision can be estimated using a probability function independently for males and female: as follows: Y. =r. + Xer,,, + ZMJE + en (I) Y( = (2+ X,r,, ) Z, + el a) where: Yi(i=m,f) are binary dependent variables with I if the ith individual participates in the wage labor market and 0 otherwise; X is a vector of individual characteristics that influences an individual's time allocation; Z is a vector of houeehold and market factors which also explains why an individual participates in the labor marke., r is the vector of coefficients to be estimated, and e is an error term.4 Different reasons can justify the inclusion of individual (X), and household and market (Z) factors as explanatory variables in labor market participation equations I and 2. An individual characteristic, such as the level of education, can be treated as an explantory variable that may indicate the potential productivity of an individual at home and in market production Holding market wages constant, an increase in the level of an individual's education can increase his or her probability of labor market participation if it increases the opporunity costs of staying at home. The household's constraints include such household asset variables as landholding, which may act as a proxy for productive household assets. The productive assets exeri a price effect and an income effect on an individual's labor market participation. The price effect would raise the marginal product o; 'shadow price' of an individual's labor, while the income effect would encourage an individual to consume more of his or her leisure - even at its given opportuity cost. The household's unearned income - another household characteristic - can influence labor market participation via a pure income effect. Market factors such as market wages exert an income and a substitution effect on an individual's time allocation. These factors may also include community variables, such as the household's proximity to community services (schooling, health, and banking servic s). Tlese variables measure the impact on time z!location 4 Y, equal t zero includes individuals who are either self-employed in faimily busine ard farning or exclu.-:vely engsged in non-maket home prodi ctn. Imluding sclf-employment and home production in one category assumes that the degive of independence between participation in these two activities is almost zero (Khandker. 1987). ?:o tes is done to aess the validty of this asumption, but for simplicity we assume that these activities can bejointly underaken with low trpsactions cost or sw;tching from one job to the othier and heuce in this sense, the participation decisions are not independenL consumption.' How do we estimate the labor market participation equation? Because the dependent variable takes the value of 1 or 0 in both equations 1 and 2, the error structures yield heteroscedasticity; hence ordinary least squares produces inconsistent estimates. A maximum likelihood method such as the probit technique which tkes care of heteroscedasticity problem can produce efficient estimates (Maddala, 1983). 1 shall use this technique to estimate both equations, I and 2.6 Table 17.4 reports probit equation results that examine the probability that a woman will join the wage labor market. A similar probit equation is run for the male sample and is also reported in Table 17.4 for comparing the response pattern between men and women in Penu. Based on the Likelihood Ratio test, the hypothesis that inarital status has no effect on the participation of men o. women is rejected. Table 17.4 is then based on the preferred specification that includes marital status, landholding and unearned income as identifying variables in the labor market participation equation. Consider first a woman's decision to join the labor market. Both general and technical education affect her decision just like they affect a man's participation decision. However, the response coefficient differs between men and women. Vocational training and secondary education increase women's lahor market participation more than men's. In Peru as a whole, the probability that a woman will join the wage market is about 10 percent higher if both men and women have vocaiional training. Additionally, the probability that a woman will join the wage market is at least 5 percent higher if both women and men complete secondary school. This suggests that improving women's education can increase their labor market participation faster *han a similar increase in men's education would affect their participation. Public school attendance seems to be an irportant determinant of both women's and men's labor market participation. Both unearned income ana landholding (which measure the income effect on leisure) generally decrease the probability of being in the labor market for men and women. Landholding significantly reduces men's participation in the wage market, but only affects women's participation in rural areas. Labor market participation for both genders is lower outside Lima, 32 -nd 53 percent lowar for women and 33 and 74 percent lower for men, respectively, in other urban areas and i *al areas. Ilere is a higher probability that women will work for wages than men in rural Per. Using the above probit results, we predict the effect of changing certain characteristics on -women's labor market pa-ticipation. Two types of predictions are made, one using the women's prr,bit equation and the other using the men's probit equation. The second predicted category S No information on any of ths mui;et factors available except for rur.l areas. Thus Z variables include only household-level variabke I A single probi: which esimates sepamtely 1 and 2 may produce inefficient estimats if the efors al &nd e1 are correlatd. The emrs ae liy to be correlated if men and woraen participate in t0e wage market from the same household. A bivariate probit is necessary to estimate the labor market partioipation 6qui.tions, I and 2, to obtain efficient estimks. However. !ar our sampie of 898 women and 2,255 men, olmy 6 pecent of men and women beoog to tho same household that participate in the labor market. We awsume, therfore, tw the correlatica bween errors is zeo. Probit Esimates for Female end Male Parlicipation Variables Females Mala Constant. -1.452 -0.753 (-15.190) (-9.504) Gen. Experin 0.062 0.058 (8.918) (10.231) Education Primary -0.013 0.023 (-0.195) (0.450) Secondary 0.399 0.187 (5.052) (2.993) Post-Secondary 0.743 0.355 (S.S9S) (3.283) Vocationa Training 0.363 0.261 (7.034) (5.808) Secondary technical diplom 0.300 0.213 (1.977) (1.520) Post-Sc.ndairy diploma 0.704 0.432 (5.376) (3.125) Univenity diploma 0.818 0.291 (S.559) (2.374) Atten-id public school 0.080 0.079 (1.495) (1.634) Unearned real mnome -0.0058 -0.0024 (-2.01!) (-1.259) Landholding -0.00003 -0.0016 (.0.053) (-1.685) Married or cohabiing -0.556 0.125 (-ICi.968) (2.555) OUA reaideac 40.320 -0.329 (-6.300) (-7.684) Runa residenc .0530 (-15.364) (-8.301) (-15.364) Salcted ample (aripl san) 6,942 6,429 Log-lelbiood -2145.15S -3686.749 Note: Numben in pu _am am! t4siatica c"14U16 WUULCLL V jJi y vi A U 6 MA UA* W45- bCi ii WULiI~LL teu 4Ve MU Lue rie way as men in responding to market incentives to participate in the wage market. With other sample characteristics remaining the same, we predict the probability of women's labor market participation for different educational levels (including general and technical education), public- private school attendance, marital status and regions where they live. The information is given in Table 17.5. The mean predicted probability of women's being in the wage sector is 9 percent against a 13 percent actual participation. This suggests that the participation model works well in explaining variations in women's labor market participation decision. However, the mez,tn predicted probability of women's market participation almost quadruples if women behave the same way as men in responding to market incentives (column B). ibis is an interesting finding that suggests that the labor market response pattern is different for women than men in Peru. In particular, there may be structural differences in men's and women's job specialization which may produce these large variations in their response behavior. The predicted probability for changing an individual job characteristic is given by the predicted individual probability for each characteristic. As expected, an increase in educational attainment leads to an increase in women's labor market participation. Note that women's participation does not change substantially if women have primary instead of less than primary education. However, women's participation increases more as women attain higher education and the gain is the highest if they attain post-secondary level of education. It is interesting to l ote that the increases in labor market participation are even much higher for the same level of education if women were to respond in the same way as men to changes in market incentives. For instance, a woman with secondary education increases her probability of participation by an additional 23 points if she follows men's response behavior rather than women's response pattern. Women with vocational training have a 6 percent higher probability of being in the wage market than women with no vocational training. However, women with vocational tr-ining can do even better if they follow men's response behavior. Thus, a woman with vocational training has a 26 percent higher probability if she follows a man's response pattern. Women gain substantially in labor force participation if they have a university rather than secondary or post-secondary diploma. However, the gain is only marginal if they attended public rather than private school. A woman's gain does not vary by whether or nct she follows women's or men's response patterns in this respect. Single women participate mon by about 9 percent compared to married women and their probability is 15 percent higber .r they follow single men's response pattern. In coitrast, a married woman's partcipation rate ;s 29 percent higher if she follows a married man's response behavior. The predicted participation ate for women is highest in Lima (15 percent) followed by other urban areas (9 percent) and rural areas (6 percent). The predicted probability of being in the wage market increases if women follow men's response pattern: 48 percent ir. Lima, 35 percent in other urban areas, and 21 percent in rural areas. 5. Wage Deir-nlnants and Returns to Education Following Becker (1964) and Mincer (1974), assume that variations in wages arise from differences in the stock of human capital such as schooling and experience. Thic assumption can be formally represented in an estimable equ-ion form 3 below: InW; = %cr + &S, + &Ki + ,1l0 + e4 (3" where InW1 is the natur.1 log of the hourly wage rate of the ith indMvidual (i=m for male, i=f for female wage worker); S is the individual's years of schooling, K is the individual's postschoxI Pmdictei 3 F Ptde Pmbabilities by acteriatic (%) aracteritics Prelicted Proba ty (A) (B) Educatin Laws dun prury 7.07 30.57 Pimauy 6.89 31.39 Soconday 14.21 37.40 Pods Seondawy 23.36 43.91 Vocational Training No 7.54 30.70 Yoe 14.14 40.40 Secondary Tech. Diploma No 8.79 32.84 Yes 25.48 49.50 Post Seconday Dipoma No 8.64 32.64 Yes 25.48 49.30 Univerdt Dipom No 8.53 32.67 Yes 29.03 43.72 Aade p c No 8.00 30.97 Yes 9.25 33.82 Maital Sab Single 14.90 30.45 Married 5.52 34.96 Reddeie Litma .4.87 48.14 Odher Ur 8.65 35.34 RnUrl uep 5.80 21.51 Pedited Mean PSrcipo 8.87 3W93 Now: (A) ca1ma lbs tpndcted priuLbiay using coedfcieo of minas rtiatiA n and (B) i basd on th coefficic of the nzmak puaticatio equoeti of Tabe 17.L experience (defined as age - S - school entry age, say, 6); K2 is the individual's experience squared; ei. and , (= 1,2,3) are, respectively, the intercept and slope coefficients to be estimated; and fi is the individual specific unobserved error. If the error is normally and independently distributed, an ordinary least squares (OLS) technique can be applied to estimate the wage equation. The estimated oDefficient P, measures the proportional increase in the wages associated with each additional year of educaton. As postschool experience increases, productivity and wages tend to rise. But firther increases in postschool experience may lead to a decline in wages and productivity because of diminishing marginal returns. The concavity of the wage -profile is thus captured by the quadratie experience terns.7 According to human capital theory, education and experience are likely to have mL;or effects on productivity. Two possible interpretations of wage equation 3 are found in the literature. The first is due to Rosen (1974) who interprets the equation as an hedonic index on cracateistics which affect the price of the individual's time. The more dominant interpretation is given by Mincer (1974) who v'.ews equation 3 as a generalization of the equilibrium relation baween wages and education, where the coefficient fl, is the estimate of the private rate of return to the time spent in school instead of in the labor market. Mincer's interpretation is widely applied in the empirical literature and is derived as follows. Assume that: (1) the only cost of schooling for an individual is his or her forgone earnings; (2) indivus enter the labor force immediately after completion of schooling; and (3) each individual's wking life of N years is independent of his years of education. Given the additional assumption of a steady state with no productivity growth, one can write the present value of the life earning of an individual with S years of schooling as: N 1I I V(S) = I W(S) - e = W(-(ep -en (4) S r r where r is the rate of diswount indicating people's time preference. If r is the same for everyone (and N is large), the equation becomes: I V(S) = W(S)-ten = VO for all S; (5) r and the present value of income streams are equalized among ind.viduals. The above can then be rewritten as: W(S) = W0e', where W. = V0r. (6) Taking log on both sides, we have: InW = W. + rS, (7) whero W. may be interpreted as thc permanent labor income of a worker. Individuals facing a given market interest rat, r, choose that level of schooling that maximizes the present value of 7 Although information on job-eci& experiace is avilable, we cannot includo it in the wae equation because it is an endogenous variable. In conast, post-lool experience is exogen to the extent that the individual's education is purentally ddet4ined and henc predetermined. their lifetime earnings. Thus r also represents the internal rate of return. Specification 7 then justifies using a semilogarithmric wage function, as in equation 3 to estimate the economic returns to education, (the estimated coefficient of S). Note that as Mincer's assumptions may not hold in the real world, the estimated schooling coefficient, , in equation 3, is only an approximation of the internal rate of return. Thus, if S takes the value of years of schooling in 3, then its coefficient ,B measures an average economic rate of return to an additional year of schooling. Equat;on 3 may be too simple to estimate an individual's productivity in the wage market when factors other than human capital influence the wages and hence the economic returns to education. Moreover, education quality may not be homogenous as assumed in equation 3. Thus, the basic wage mcdel needs to be adjusted to reflect reality. Tlree adjustments in the functional form of equation 3 are undertaken in this paper.' Furst, there is a possibility that regional labor markets may behave differently and hence yield quite different estimates. Three distinct markets (metropolitan Lima, other urban areas, and rural areas) have already been identified (Stelcner et. al., 1988). T he wage rate and labor market participation equations are thus estimated separately for men and women in these three regions. While this method is Dreferred where there is no interregional migration, such migration does occur as educated workers move to higher wage markets. But interregional migration may bias estimates of the returns to education as weli as to labor market participation. In Latin American countries, as much as half the life-cycle returns to schooling of rural residents result from migrating to urban centers (Schultz, 1988). This bias could not be reduced even if the migrants' original location were known, because migration is a self-selection process. Using regional 'shifters' in the wage equation fitted for the country as a whole, one can illustrate the potential severity of interregional migration on the estimated returns to schooling and labor market participation. In particular, because high-wage urban regions have more and better schooling, introducing regional shift variables in the wage or participation equations reduces the estimated remurms to schooling, or the influence of schooling on participation in the labor market. Second, since differert levels of schooling impart different skills and wages, an adjustnent is necessary to quantify the effect of the quality of different categories of education on wages. Tnmre are at least three ways one can quantify the effect of heterogenous quality of education. The first method is by including schooling squared, SI, as an additional variable. In this case the derivative of the dependent variable (log wage) with respect to S gives us an estirrate of (A + 2pS), where p is the estimated coefficient of sI. By inserting different values of schooling levels, say, 5 for primary level, 10 for secondary, and 14 for post-secondary education in this expression, we can estimate the private rate of return for each category of schooling. The drawback of this method is that it gives equal weight to each category of education in the seose that only the incremental return varies by the level of education, but not the basic reun to education. The secoad method requires an introduction of different education dummies for different levels of schooling where an education dummy is defined as a value of 1 if the individual belongs to a particular schooling level and 0 otherwise. In this second case, the educational-level-specific economic rate of return is calculated by deflating the estimated coefficient of a particular schooling dummy with the difference in years of schooling between this particular schooling level and the reference or control school group. The problem with t&is approach is that it understates the returns to primary education (Psacharopoulos, 1981). The third Nute ha tee adjuta aze al apped to Ot labor market p cipaton equationw I d 2. method is a more direct way of estimating the economic returns to various categories of education, and it involves using splines of schooling years in the wage equation 3. For example, if an individual has 9 years of school!ng, the value for his or her education takes 5 years in primary schooling, 4 years in secondary and O years in post-secondary education. This method is better in the sense that it estmates directly the ec).omic reurns to different quality education. This paper employs all three methods to estimate and compare the school returns for three categories of education - primary, secondary, and post-secondary. Third, an adjustment is necessay to control for the effect on wages of the quality of education across schools, particularly between private and public schools. Attendance or non-attendance in pub!ic school is included in the wage and participation fuictions to control for the influence of unobserved school quality. Parental characteristics also often contribute to children's unobserved ability by giving them a better aducation (Schultz, 1988). Thus, by including this school quality variable in the wage function, we may reduce the impact of parental charactei-ittics on an individual's productivity and hence returns to education.9 With these three adjustments in equation 3, the extended wage equation can be written as: 3 2 InWI % + E tSp + AA + AK2} + E,6flREGh + #IPUBSCL + q (8) j=1 h=1 where Si is the jth-leve1 education of the ith individual, REG represents regional dummies such as Lima, other urban areas, ai rural areas, and PUBSCL indicates wnether or not an individual attended a public school. Again, like equation 3, we may assume that the errors are independently and normally dimstibuted in which case an OLS can yieid unbiased estimat. An adjustment is, however, essential in our OLS srategy to esimate either model 3 or 8 free from sarnple selection bias. The sample selection bias arises for endogeneity if the decision to participate in the labor market is conditioned by the worker's labor-laisure choice. Thus the estimates of either equation will be biased if it is estimated by including only wage-earers - thus excluding persons not reporting a wage yet who are part of the potent-il labor force. The decision to join the labor market influences wages because the characei cs that affect labor market participation may aiso interact with wages. Thus the wage estimates need to be independent of the possible impact of these characteristics. Estimating model 3 or 8 in conjunction with labor market participation equation I or 2 may reduce sample selection bias from the wiage estimates. Heckman (1979) has suggested a two-step procedure to estimate the wage and labor market participation equatiens. In the first stage the expected values of the residuals of equation 3 or 8 that are truncated are obtained by estimating the labor market participation equatoo I or 2 by the probit method. By introducing the estimated values of r^siduals from the participation equation into wage equation 3 or 8, we can use ordinary D One may inslude paronts' chduaisi dircly in the wa greson. But parents may influence children's earnings only va children's shool atainment Tus, by including parevtat chamcteristics in the wage equation one would only rduce the rturns to individual's educsticn (se S!'lcner d.al., 1988.) Table 17.6 Eanipgs FPim for Wags Equation (3) Pemle Males Variables OLS Adj. OLS OLS Adj. OLS Constant -0.583 -0.779 -0.159 1.359 (-5.366) (-2.965) (-2.305) (5.597) Years of Schooling 0.124 0.131 0.11S 0.081 (16.674) (8.584) (11.287) (0.304) Gen. Expenrece 0.076 0.079 0.055 0.003 (9.291) (8.S84) ( 1.28n) (0.304) Gie. Exper. Squred (xlOO) -0.129 -0.136 -0.068 0.031 (16.746) (.6.448) (-6.392) (1.676) OUA Reidence -0.125 -0.144 -0.170 0.03S (-2.131) (-2.281) (-4.S13) (0.718) Rural Residence -0.339 -0.369 -0.358 0.164 (-4.12T) (-4.096) (-7.887) (1.786) Lambda 0.085 -1.019 (0.815) (-6.52) R-Squared 0.35S 0.35S 0.331 0.343 N 898 898 2,255 2,255 Note: Numbers in parenthe amr t-uta±si. least squares to estimate the wage fnctiDon in the seoond stage. Heckman's two-step procedure yields consistent estimates.'° An identification problem emerger, however. The variables that explain wages may also explain individual labor market participation. That is, the vector X and Z in equaion I or 2 contains the variables included in the wage equation 3 or 8. 7hus we need some identif'ing variables in equation I or 2 not included in the wagg equation to belp dLstisguish a participant from a non- participant. Three variables are considered herm s potential candidates for identiffing the labor market participation equation from the wage equotio. The first two variables are included in vector Z: landholding and uneuned income. Both are .cted to influecae the likelfhood tbat a peson will work for wages by affecting the perso. i resumntion wage. If an individual has a considerable amount of land or unearned inowme, he or she wil be less likely to work for wages 1o Note that aalo selection ce cdtim dbe ian pedict a priod the afiretion in hich this would altor th. waSg 'eAtiAZ because the returns in other ctivities are .gier. These two variables are expected to influen^e only labor market participation - not wages. The third identiiying variable is marital status which is included in the X vector of the participation equation 1 or 2. Married couples can specialize more easily than unmarried indiViduals, which usuaily encourages mafried men to work for wages and married women to work in the home (Schultz, 1988). Marital sus thus can influence labor market participatien, but not an individual's market productivity." The wage models - equations 3 and 8- are esfimated by the OLS method with and without sample selectivity bias. The results of these wage regressions for men and women are sown in Tables 17.6 and 17.7, respectively. Tle basic wage model explains about 36 percent and 33 percent of the wage variations among womn aid men. Te extended model, on the other hand, explains about 37 percent of women's and 34 percent of men's rvage variations. Thib cuggesta that the human capitul model explains more than one-third of wage variations among men and women in Peru. Either model's explanatory power does not change very much if we use Heckman's approach to correct for the endogeneity of labor market participation decisions. Nevertheless, sample selection bias correction has an important influeace on women's as well as men's productivity in the wage sector. The sign of the coefficienu of the correlatidon w wage earner and wage rate errors (i.e., Lambda) detmines d2e tye of selection that generaes the group of men and womren workers. Tables 17.6 and 17.7 suggest that the mast able men select non-wage employment, since men who work for wa earn less than an average man in Peru. Among women, on the other hand, the most aDe indivldwals seem to seect wage employment. Tle results thus indicate tha unobserved charateristics that influence labor market participation lso influence an individua's productivity. Among he important detemnaimts of productivity according to vage equation 8, education and experience are crucial; runs to experiece howver, are bigb-x fbr women than for meL Education on averdge has an imporant influerce on both mm's and women's productivity. Furthennore, education at all levels influences both men's and women's producivity in leau. Technical education increases labor market productivity among men and women in Peru. Women', wages increas by 15 percent and men's wages by 19 peret if they have had vocational training. But when sample selection correction Is int-oduced, women's wage gain increse t 28 percent, while men's gains drop to 4 percent. In corast, the wage change of both men and women with secondary diplomas are not significsc X a result of changes in secondary diploma holdings among men and womei.. Conversely, the wages of male university graduates are about 32 percent higher for men than for women before selectvity correcion. Tne adjusted OLS increases women's retns to university diploma by 22 pet but reduces men's re -ns by 12 percent. In comparison wi6 private schooling, the -turns to public school attendance are lower for both nale and female productivity. Wages are 2 percent lower for women and about 4 percent lower for men who atended public school than for those who attended private school. When sample electon correction is made, the wage differences fal to 17 percent for women but increase to 10 percent for men. The differmec in the productivity of public versus prvate school graduates indicates that the public scbool system should be improved. This finding s conistent with other studies (Stelcner etal., 1988; Ing, 1988). 11 Martal status, howover, influee an individual', pndctivity if we a& tat mrioed pc.pl am balthier than noai-mm people and heath aife productivity. Table 17.7 Earnings Picbioas for Wag- Eqpuaion 8 Females Males Variables OLS Aej. JLS OL Adj. OIS Constant -0.030 -0.861 -0.337 1.606 (.0.245) (-3.079) (4.766) (3.770) Gen. Experience 0.072 0.083 0.048 0.007 (8.601) (9.248) (9.63S) (0.473) Gen. Expor. Squared (xi00) -0.131 -0.162 -0.064 0.016 (-6.815) (-7.607) (-5.91S) (3.556) EducAtion Prfimary 0.311 0.300 0.270 0.256 (3.254) (3.163) (4.904) (4.637) Secondary 0.a74 0.990 0.696 o.S77 (8.434) (9.094) (11.246) (7.832) Post-Seco Jary 1.224 1.432 0.9?? 0.763 (8.415) (9.076) (!0.389) (6.417) Ser.onday technic diploma -0.133 -O.C38 -0.030 -0.087 (-0.873) (-0.244) (0.276) (-0.757) Post-seconali diploma 02.24 0.528 0.365 0.134 (2.141) (3.977) (3.601) '1.043) University diplma 0.153 0.372 0.484 0.343 (1.169) (2.54i) (5.184) (3.262) Attended public shool -0.195 -G.171 -0.043 -0.098 (-3.073) (-2.7CO) (-0.927) (-1.964) OUA Residex -0.089 -0.185 -0.176 0.093 (-1.492) (-2.79S) (-4.462) (0.106) Rural Residenco -0.38i -O.%67 -0.420 O.OSI (4.447) (-5.551) (-9.0141 (0.305) Lambda o.456 -1.909 (3.302) (-2.935) R-Squared 0.372 0.380 0.339 0.34: N mb 898 2,255 2,255 Note: Numbers in pantheaca re tsadziaks Table 17.8 Estimates of Privdte Rat r of Retum to Schooling by Gender Using Alternaave Estimation Procedures Private Rates of Return by School Level Method of Fsiimatioa Primar Secondary Post-secondar Mlale Female Male Female Male Female 1. Schooling Squared' OLS 0.09 0.11 0.09 0.11 0.08 0.11 OLS with selectivity 0.08 0.13 0.07 0.14 0.06 0.14 2. Schooling Dummy' OLS 0.05 0.05 0.12 0.15 0.16 0.20 OLS with slectivity 0.04 0.05 0.10 0.17 0.13 0.24 3. Schooling Splines OLS 0.09 0.09 0.09 0.13 0.09 0.09 OLS with setecivity 0.09 0.09 0.09 0.15 0.09 0.10 a. Schooling Squared neans schooling squared is to the rcgrewsion equation 8. b. S rbooling Dummy uJa wue school dummies in the regression. Workers in Lima are paid more than their counterparts with he same education in other urban and rural areas. Accoiding to the extended model, female workers in Lima earn 9 percent more than workers in other urban areas, and 38 percent more than workers in rural areas. When selectivity bias is corrected, the wage differences increase to 19 percent in ather urban areas and 57 percent in rural areas. For mea the wage differences are, respectively, 18 percent and 42 percent without selectivity correction. When sample selection bias is corrected, the differences seem to disappear. .gs1Jxana of refur to scheilng. Table 17.8 presents three sets ^' esfimaes of returns to '.ucation based on the three methods outlined earlier. Each set contains two types cf results, one istheOLSandtheotheristbeOLScorrectedforsampleselectivitybias. Theyarereported for both men and women. A comparison of regular and adjusted OLS results suggests that the estimates are sensitive to sample selection correction. Moreover, they are sensitive to the method used for estimating the return to education of various categories. Men lose from education because of sample selection correction. This is true for eacs method except for the splines method. Thus for males the returns dsase from 9 to 7 p,.cent at the secondary level, and from 8 to 6 percent at the posecondary level if we look at the schooling squared method. The decease Is from 12 to 10 and 16 to 13, -.espectively, at the secondary and postsecondary levels with the schooling dunmmy approach. Women, on the other hand, gain in almost all approaches used for ca!culting the school returns. For women the retums increase from 11 to 14 both at secondary and postsecondary levels when the schooling squared technique is used. For women the gain is the largest at the postsecondary level under the dummy schooling method, i.e., the increase is from 20 to 24 =ercent wIhen sample selection correction is made. The differences among three alternative methods for calculating the returns to education of different categories are substantial. As expected, the return to schooling is biased downward at the primary level with the dummy variable method. The OLS estimate of the returns to primary education is 5 percent for both men and women under the dummy schooling method, while it is 9 percent for both men and women with the splines method. In contrast, the returns are 9 percent for men and 11 percent for women using the schooling squared method. As the schooling squared method gives equal weight to both primary and post-primary education it seems to overestimate the returns to primary education. The schooling dummy method registers a much higher return for both secondary and postsecondary education tian any other method for both men Pnd women. Thus the return to postsecondz ry education for women is 24 percent compared to 14 and 10 percent under squared and splines methods, respectively. The results, therefore, show that the esimates of the returns to education vary remarkably with the kind of method used for calculating these estimates. The differences in returns to schooling for mer and women are also worth noting. The returns to schooling are higher for women than for men, especially at the secondary and postsecondary level. This is true for all three methods used for calculating the returns to schooling. ThL! finding contrasts with studies from other countries that suggest that the returns to schooling are similar for men and women (Schultz, 1989). The reurn to schooling is higher for women at the secondary level than at any other level with the splines method, a result which is consistent with other Latin American and Asian countries (Schultz, 1988, Mohan, 1986). However, with the dummy and squared methods, the return to schooling is higher for women at the postsecondary level than at any other level. 6. Male-Female Wage Differences A large number of sudies based on United States' data and a few studies from developing countries attempt to identify the extent of male-female wage differences that is explained by differences in hunman capital and other observed job-related characteristics (Becker, 1985; Birdsall and Fox, 1985; Gronan, 1988; Mncer and Polachek, 1974; Oaxaca, 1973; Gannicott, 1986). One standard procedure to measure the male-female wage gap is to fit equation 3 or 8 by ordinary least squares separately to a sample of male (m) and female (f) worke:s as follows: InW = X, B. + (9) and lnWf = X, 8 + 4 (10) where: B5 and Bk are the vectors of unknown coefficients, including the intercepts; X., and X, which are, respectively, the vector of males' and females' observed characteristics; and e. and (, are, respectively, the males' and females' individual specific error. A property of ordinary least squares is that the regrcssion lines pass through the mean values of the variables so that: InW =X,.,8 (11) lnW, =X,fi (12) ,, . ,,A~'U~44 gMftdCI~JWMwn WI- ma-remaie wage VVerences in Peru 791 The hats denote the estimated values of the coefficients. By simple manipulation of equations 11 and 12 the male-female wage gap function can be written as: InW@;- InW= (X=,- XD B + 14 (B,U- B, = (X X -xi + (8-) (13) where the first part of the right-hand side of equatlon 13 measures the wage gap due to male- female differences in wage-related charaisdcs and the sewond part measures the gap explained by the differences in male-female wage structures for the same observed job-related characteristics. Thus, one can measure the wage gap in two ways: Using the male wage structure or, alternatively, using the female wage structure.'2 Both the ba;ic and extended wage models are used here to measure and compare the wage gap that is explained by the job-related characteristics. Detemtinants of mal-female wage dtfference. As rable 17.3 showed, men ear more than wcmen in Peru. In fact, women earn about nalf of men's wages, when the log wage differences are adjusted for male-fem?le sample size differences. What explains the male-fernale wage differences? Table 17.9 shows the wage variations between males and females that are explained by the wage equations 3 and 8 under the OLS estimation method, with and without sample selection correction. Table 17.9 Male-Female Wage Gap Deoompoition Estwmes for AlU Peu By Alternave Sample Selection Mtdb a Earnings Function Saple Size Peretagc Explaied Type By Hunxa Capital Vuiables Usig OLS Method Widtou Sample Writh Samle Mm Wom Selection Correction Selection Correcion (A) (B) (A) (B) Basic 2,255 898 -24 -40 218 -62 Exteuded 2,255 898 -71 -89 167 -232 Notc: Two wage sttructura am used (A) malk wage ucusc, ad (B) fueak wg srucu The OLS results of the basic and extended equations explain nothing in terms of male-female differences in human capital variables. That means the wage differences are not explained by 12 Note that th od compont of equation 13 is often taken as relecting wage discrminao Because it is difficult to uov the effects of all possble wage-determining factors, including those that may reflect fedae ditrimin outside the labor market, it is indeed dit.cult to attibute the secoDd compooent u a measure of as-dicriminaion in the wag market (Gunderson, 1989). 392 Women's E5p1y me wd Pay in Lain America male-female dif^: rences in job-related characteristics, but by differences in the wage structures. In fact, wage structures for men and women are so different that women are not paid consistently according to their human capital endowment. Tlus, human capital differences produce even negative contributions in the zalol7 of male-female wage differences. However, when the sample selection correction is appli Al male wage structure is used to calculate the wage gap, the model explains more than 100 percent of the differences in wages in terms of the differences in job-related characteristics. This method includes, in addition with the standard variables, a correlation factor that measures the relationship between the errors of the wage equation and the labor market participation. This result suggests that the unobserved characteristics that influence both the labor market participation and productivity explain fully the male-female wage differences in Peru.' However, when female wage structure is used to calculate the wage gap, even sample selection correction does not help the human capital model to explain the wage gap that exists in Peru. If the unobserved characteristics explain the wage gap, it follows that we need to identify more obsefvable characteristics of a worker other than his or her human capital variables to explain the male-female wage differences that exist in Peru.'4 7. Discussion This paper addresses four critical questions. First, what influences men and women to participate in the labor market? Although education and training raise labor market participation of both men and women, vocational training and secondary school increase the labor market participation of women more than that of men. Thus, improving education for women can increase their participation faster than a similar increase in men's education would affect the participation of men. Unearned income and landholding reduce the participation of both men and women. The probability of being in the wage sector is high for married men and low for married women, indicating an expected job specialization after marriage. Second, what determines the productivity of men and women in the wage market? Expezience, education, and training are all effective. The quality of education is also significant: Those employees educated in private schools are more productive than those with a public schoul education. Moreover there are sharp regional differences in productivity. Men and women from other urban areas and rural areas are paid less than their counterparts in Lima. The extent of male-female differences in productivity depends on the impact of sample selectivity bias. Third, is there any systematic gender bias in the estimated retrs to schooling if we ignore the possib:e sample selection rule of who is a wage earner? The results suggest that sample selection correction decreases the reurns to schooling for men but increases them for women. Sample selection bias is substantial for both men and women, showing that the selecttd wage earners are not a random sample. The magnitude and direction of the bias, however, vary by method used 13 This is an inersting finding becaw it does not include any controversial control variable uch as occupational status in the wage regresion. The wage function includes an additional variable-the sample selection correction fictortha accounts for the unobsrved chanteristics influencing labor market participation. 14 Even controlling the wag equation for occupational differences which may imply some var..tiona in uno4served characteristics does not solve the puzle of why m eam more tan women in Peru. Although occupational sau is a choice variable, we cntrol for it's effect in the wage equation by including a number of occua dummies. This method still does not alter the conclusion of this study. Wmn 's Labor Mark* Pardcpafion and Mak-Fmak Wage D;fferewcs in Peru 393 for the calculation of these returns. The retLrns to schooling are higher for women at the primary school level. The results confirm that sample selection bias is an iL-portant factor in labor market participation The most able men select non-wage employment, while the most able women select wage employment And finally, why do men earn more than women? Although there are some differences in hulman capital, the extent to which these differences explain the wage gap depends critically on sample selection correction factors and the wage structure used to calculate. the wage gap. Thus when sample selection correction is not included in the wage regression of a random sample of males and females, the human capital model does not explain any portion of the wage gap that exists in Peru. This is true no matter whether we use the male or female wage structure to calculate the wage gap. When the correction factor is included and the male wage structure is used, the model explains 100 percent of the wage gap. Thi suggests that the unobserved characterstics that influence labor market participation and productivity also affecs the productivity differences between males and females. Clearly it would be usefud to identify other observable characteristics that affect wage differences. Two policy implications that result from our answers to these questions should be mentioned. First, since public schools are less effective than private schools in raising productivity and reuacing the wage gap, policymakers should take steps to make the public school system more effective. Second, as the school returns are higher for women than for men, parents should invest equally, if not more, in female education. However, the PLSS survey data indicate that parents enroll more male than female children in schools, especially at the secondary level (Schafgans, 1990). This clearly supports the notion that school investment in children is not gender neutral, nor is it governed by the private rate of returns to schooling for men and women. Apart from equity reasons, there is a strong case for an efficiency-based argument for investing more equally in male and female children. The results of this paper indicate that investments in education and training for women raise their participation and productivity in the labor market more than a similar investment in men's education. In additicn, these investments reduce fertility, improving the education of children a'd the health and nutrifion of all family members. Thus human capital investment in women is a high reun activity and at least as good as an equivalent investment in men. The government, therefore, must identify ways to channel more resources toward women's education. References Becker, G.S. Hnuman Capital. New (ork: Columbia University Press, 1964. '. A Theory of the Allocation of Time." Economic Journal, Vol. 75 (1965). pp. 493-517. -. 'Human Capital, Effor and the Sexual Division of Labor." Journal of Labor Economics, Vol. 3 (1985). pp. 533-558. Behrman, J.R. and N. Birdsall. Tlhe Quality of Schooling." American Economic Review, Vol. 73, no. 5 (1983). pp. 928-946. Birdsall, N. and M.L. Fox. '"Why Males Earn-More.- Economic Development and Cultural Change, Vol. 33, no. 3 (1985). pp. 533-556. Dagsvik, J. and Aaberge, R. 1990. "Household Production, Time Allocation, and Welfare in Peru." PRE Working Paper No. 503. Washington, D.C.: World Bank, 1990. Gannicott, K. "Women, Wages and Discrimination: Some Evidence from Taiwan." Economic Development and Culwral Change, Vol. 39, no. 4 (1986). pp. 721-730. Grootaert, C. and A.M. Arriagada. 'lTe Peruvian Livings Standards Survey: An Annotated Questionnaire.- Washington, D.C.: World Bank, 1986. Griliches, Z. "Estimating Returns to Schooling: Some Econometric Problems." Economeiica, Vol. 45, no.1 (1977). pp. 1-22. Gronau, R. "Sex-related Wage Differentials and Women's Interrupted Labor Careers - The Chicken or the Egg." Journal of Labor Economics, Vol. 6 (1988). pp. 277-301. Gunderson, M. "Male-female Wage Differentials and Policy Responses." Journal of Economic Literature, Vol. 27 (1989). pp. 46-72. Heckman, J. "Sarnple Selection Bias as a Specification Error." Econometrica, Vol. 47, Jamnary - (1979). pp. 153-161. Khandker, S.R. 'Labor Market Participation of Married Women in Bangladesh." Review of Economics and Statsrics, Vol. 71 (1987). pp. 536-541. 394 Womm's Labor Marka Parkipraon ad Male-Female Wage Dfferences w Pens 395 King, E.M. ODoes Education Pay in the Labor Force.0 PHREE Working Paper. Washington, D.C.: World Bank, 1988. King, E.M. and R. Bellew. 'Gains in the Education of Peruvian Women, 1940 to 1980. PRE Working Paper No. 472. Washington, D.C.: World Bank, 1990. M&Wa, &.S. Iimited-d4umdw and Qualitative Variables in Econometrics. New York: Cambridge University Press, 1983. Meer, J. Schooling, Exwrience and Earnings. New York: Columbia University Press, 1974. Mincer, J. and S. Polachek. 0Family Investments in Human Capital: Earnings of Women.' Journal of Poli.cl Economy, Vol. 82 (1974). pp. S76-S108. Mohan, R. Work, Wages and Welfare In a Developing Metropolis. New York: Oxford University Press, 1986. Newman, J. 'Labor Market Activitiy in Cote d'Ivoire and Peru." LSMS Working Paper No. 36. Washington, D.C.: World Bank, 1987. Oaxaca, R. 'Male-female Wage Differentials in Urban Labor Markets. a Internaional Economic Review, Vol. 14, no. 1 (1973). pp. 693-709. Psacharopoulos, G. 'Rets to Education: An Updated International Comparison.' Conparadie Editcaxion, Vol. 11, no. 3 (1981). pp. 321-341. Robb, R. 'Earnings Differeals between Males and Females in Ontario, 1971.0 Canadian Journal ofJEconocs, Vol. 11, no. 2 (1978). pp. 350-359. Rosen, S. "Hedonic Function and Implicit Markets." Journal of Politcal Economy, Vol. 82 (1974). pp. 34-55. Schafgars, M.M.A. OA Comparison of Men nd Women in the Labor Force in Peru.' in B. Hen and S. Khander (eds.). Women's Wo k, Education and Welfare in Peru. Forthcoming. Schultz, T.P. 'Women and Development: Objectives, Framework, and Policy Interventions.' PHR Working Paper. Washington, D.C.: World Bank, 1989. . 'Educational Invesment and Returns.0 in H. Chenery and T.N. Srinivasan (eds.). Handbook of Dewoprne Economics, Vol. 1. Amsterdam: North Holland, 1988. Stelcner, M., A.M. Anriagada, and P. Mook. 'Wage Determinants and School Attainment Among Men in Peru.' LSMS Working Paper No. 41. Washington, D.C.: World Bank, 1988. Suarez-Berenguela, R. 'Peru lnformal Sector, Labor M arkets, and Returns to Education.' LSMS Working Paper No. 32. Washington, D.C.: World Bank, 1987. World Bank. 'Peru, Politices to Stop Hyperinflation and Initiate Economic Recovery." Mimeograph. Country Study Series. Washington, D.C.: World Bank, 1989. 18 ,~~~~~~~~~ Is there Sex Discrimination in Peru? Evidence from the 1990 Lima Living Standards Survey Indenmit A. Gll 1. Introduction Allegations of imperfecdy functioning labor markets have been commonplace in the literature for many years. Researchers have stuggled to provide reliable estimates of discrimination against women, racial and ethnic minorities, of the degree of segmentation in labor markets by occupation and locaion, and the effects of government intervention on these 'market failures." For example, it has been argued that while the government often creates jobs that are protected from market forces, it also - sometimes simultaneously - serves as an employer fur 'unfairly' disadvantaged groups swch as women. This chapter readdresses these issues using a somewhat novel approach: It combines analysis of one form of alleged labor market failure - gender discrimination - with the examinaion of another facet - segmentation of the labor market by type of employer. More precisely, I examine if the degree to which similar observed skills of men and women are differentially rewarded depends upon whether an incividual works in the wage sector or as a self-employed worker. This provides crude indizators of two forms of market imperfection: First, it throws up first-round estimates of the differenc in renuns to human capital (schooling, general work experience, and job-specific skills) of the self-employed and wage workers. Since in Peru these classes roughly correspond to the informal and formal sector, respectively, the restults can be used to determine whether the pecuniary rewards to human capital differ across sectors, i.e., whether the labor market is occupationally segmented. Second, it provides a preliminary measure of gender biises in remuneration under differing employment regimes, indicating whether the gender gap in earnings is driven by market structure or skill differentials. Uncovering differences by employment type and gender is just the first step, though. There are good reasons to believe that while self-employment is often harder to initiate than paid employment (because it may require high startup costs), it provides workers with relatively flexible work schedules. Married and cohabiting women (who, facts indicate, often balance two 397 398 Women '5 E wlotywnt an Pay in Lati Amrica careers - household and market work) are very likely to benefit from this flexibility.' The upshot of the discussion is that while married women are better suited for self-employment or informal market work, single women and men constitute relatively 'fungible' human capital. Standard measures of skills must be augmented by considerations of gender and marital status in studying the effectiveness of skill accumulation (investment in human capital) as a welfare enhancing device. Not realizing this will lead to inefficient policy design. A number of studies have been written on the Peruvian labor market in the years since the 1985-86 Peruvian Living Standards Survey results were made available. Labor market participation of Peruvian men and women, their schooling decisions and earnings determinants, and other forms of market segmentation have been examined by a battery of capable researchers. In other words, all the favorite 3reas of labor economists have been explored. Why another study on Peru? While repetition in scientific inquiry rarely needs to be justified, this study more than jutst duplicates past efforts: First, while human capital effects on work participation and earnings have been repeatedly explored for men (e.g., Stelcner, Arriagada, and Moock, 1988) and for women (e.g., King, 1990; Khandker, 1992), explicit gender comparisons of these phenomena are relatively rare. Tbis study does just that. Second, as discussed above, while allegations of market imperfection (referred to as 'labor market segmentation,' 'duality' or 'sex discrimination') are implicit in many studies of labor markets in Latin America, there has been no comprehensive examnination of these aspects of market failure in a unified analytical setting. Finally, since this study uses the 1990 Living Standards Survey, it is worthwhile examining whether the market structure revealed by these data differs substantially from that indicated by the 1985-86 survey. To sharpen the discussion, consider the following facts: 1. While about half of the women who worked in the martet were self-employed, only about a third of n.n were self-employed. 2. Both wage and self-employed males worked about 48 hours per week, while self- employed women worked about 7 hours less than wage and Ularied women (35 and 42 hours respectively). 3. The variance of hours worked in the salaried sector is less than half the value of the variance of hours worked per week by the self-employed for both men and women. 4. About 62 percent of self-employed women wcG° married or cohabiting, as compared with 37 percent, _= so 60 40- 20 0 2 4 S 8 10 12 14 1s 18 20 Highest Grade Completed - Mae Self-Employed | Female Self-Employed o Male Wage & SalarMed - Female Wage-SdJaried Suppose that the aim of policy is to improve the econmic position of women. This means that we want women to move vertically (along the Y-axis). According to the graph, there are two vehicles for this movement Mie first is an increase in the education of women, which moves women along either the self-employed or the wage sector schooling-earing profUe. The second is a movement of women from the wage sctor (where, aside from lower earing relative to the self-employed, the male-female earnings gap is large) to the self-employed sector (where mean hourly earings are roughly he same for men and women). Naturally, the relative effectiveness of these policies depends upon their costs and not just the benefits illustrated above. But there seems to be evidence that t policy of education subsidies wil encourage women to work in the w-e sector, where there are higher returns to schooling but where average earnings are lower than in the self-employed sector. Given the difference in levels, there are clearly obsales to being self-mployed. One likely obstacle is startup capital. Improving access to credit would help to remove this barrier. If women (especially mothers) obtain greater overall benefits from flexibility in work schedules, a policy of credit subsidization would be more compatible with the objective of improving the economic sus of women than educaton subsidies. There are probably still good arguments for schooling subsidies. However, this paper provides some evidence dtat increased schooling, by lowering fertility, can make women choose work patern similar to thooe of males only up to a point Women will always choose to have some chldren, and childcare will always remain a relatively female-intensive activity. Policy design must recognize that women need more flexibility in work schedules than men. If there are positive extenalities associted with the improved status of women, credit subsidization for women may be the more eftective instrument in attaining these benefits, at least in urban areas where scool-ig has a:ready reached reasonable levels. The discussion above seems to justify subsidization of work schedule flexibility primarily for women with children. This creates a targeing problem, because a policy of credit subsidization for mothers will lead to fertility that is higher than opdmal as women try to qualify for this subsidy. Subsidized credit for women without children (younger, umarried women) can be radonalized if returns to sector- specific experience among the self-employed are high, so that there are advantages to early entry. The evidence in this study indicates high returns to sector-specific experience for self-employed women. A simple policy implication emerges as a result of this: Education policy must be supplemented by a policy that facilitates the transition of women from the wage to the self- employed sector, for example, through credit subsidization. _ ., IJ &ui.4_UWd4_UjO 1 F( r.m encJr0m sh )990 Lb.a Uvb5g Siandard Swy 423 4ppenx Table 18A1 Reaso for Not Wog Females, by Marital Satus and Age Age Gro (Yeas) 14-20 21-35 31-40 41-50 S1-6S All Married, Col.bitin & Widowed Womn 1. Studying 3.8S 1.83 1.12 0.00 0.00 0.93 2. Household Work 80.77 90.37 86.59 86.62 68.S4 75.47 3. Retired, Raistir etc. 0.00 0.00 0.00 2.82 12.92 5.00 4. Unable to Work 0.00 0.46 0.00 0.70 7.87 8.9S S. Sick 3.85 3.67 5.03 6.34 7.30 5.58 6. Job Related Reams 0.00 0.00 3.35 0.70 0.00 0.81 7. Other Rea*os 11.54 3.67 3.91 2.82 3.37 3.26 Total Observations 26 218 179 .142 178 860 Sbgle & Separaed Women 1. Studying 87.80 46.90 15.00 5.56 0.00 83.22 2. Household Work 6.21 27.S9 35.00 S0.00 S6.25 8.92 3. Retred, Rentier etc. 0.00 0.00 0.00 0.00 6.25 0.38 4. Unable to Work 0.22 1.38 0.00 11.11 15.63 1.30 5. Sick 1.11 4.83 10.00 11.11 9.38 1.45 6. Job Relate Reasons 1.33 9.66 15.00 11.11 3.13 1.98 7. Other Reans 3.33 9.66 5.00 11.11 9.38 2.75 Total Obnseaons 4S1 145 20 18 32 1,311 Married, Codbtg & Widowed Men 1. Studying .. 38.46 0.00 833 0.00 3.41 2. Household Work .. 23.08 0.00 16.67 1.54 4.SS 3. Reftired, Rentier etc. .. 0.00 0.00 16.67 60.00 46.02 4. Unable to Work .. 7.69 0.00 0.00 3.08 1S.34 S. Sick .. 15.38 0.00 16.67 1846 12.50 6. Job Related Reasons .. 15.38 83.33 16.67 12.31 13.64 7. Odher Reasos .. 0.00 16.67 25.00 4.62 4.55 Tota Observaions * 13 13 12 65 176 Single & Men 1. Studig 90.30 64.36 .. .. .. 91.16 2. Household Work 0.81 2.97 .. .. .. 1.20 3. ReFtired, Rentier dc. 0.00 0.99 .. .. .. 0.69 4. Unable to Work 0.00 0.99 .. .. .. 0.69 S. Sick 1.62 8.91 .. .. .. 1.89 6. Job Reated Reass 4.04- 12.87 .. .. .. 2.49 7. Other Reasons 3.23 8.91 .. .. .. 1.89 Total Obsevatios 371 101 * * 486 N. x * lticae w than 7 observatiom. 424 Woma ', Eiyxw,u and Pay i Lai Au*ka App4udx Table 18A.Z Famle Modhl Earning RagrAnms Depahdm Vrib3o Log (Monthly Iom. from Mami Job) (2) (26) (27) (28) C29) (30) Schooling 0.0733 0.0628 0.0618 0.0500 0.0609 0.0383 (8.77) (6.62) (6.95) (5.00) (6.87) (3.71) Age-School-6 0.0589 0.0497 0.0465 0.0362 0.0384 0.0119 (7.04) (5.40) (5.15) (3.68) (3.99) (1.04) (ASe-School-V 4.000S -0.0007 ^0.0007 -0.0006 -0.0006 -0.0002 (-5.08) (-3.92) (-4.00) (-2,89) (-3.28) (-0.94) Tenure 0.0451 0.0465 0.0449 0.0471 (3.49) (3.60) (3.48) (3.68) Tanure2 -0.0012 -0.0013 -.0012 -0.0012 (-2.66) (-2.72) (-2.61) (-2.66) Married & 0.1618 0.3157 Cohabiting Dummy (2.33) (4.04) Log (HIoursWeek) 0.3932 0.3816 0.3707 0.3571 0.3875 0.3798 (8.38) (8.11) (7.87) (7.57) (8.16) (8.06) Work Participatiol -0.1080 -0.1162 -0.2145 Selectivity ( (-2.35) (-2.55) (4.18) Constant 5.5340 5.984S 5.7271 6.2208 5.6869 6.SS39 (24.68) (20.39) (24.99) (20.86) (24.81) (21.36) F-Statistic 39.81 33.00 29.17 26.02 25.90 25.21 Adjusted R2 0.1526 1567 0.1639 0.1690 0.1682 0.1837 Sample Siz 862 862 862 862 862 862 Mom of Variables LOg (Mobly hTcome) 8.2610 8.2610 8.2610 Schooling 9.4030 9.4030 9.4030 Aae-S ng4 18.48S0 18.4580 18.4580 Tenure 5.1286 S.1286 Mri & Cohabiting 0.5041 Avege Hours Per Week 34.3635 34.3635 34.3635 Lambda (Wodsku/Nom-wokui) 0.7074 0.7074 0.7074 Nowc t-.bZiatic in pahmis. 15 Phre Sa Dbcria h ParN? Ev"aCefioe thu 1990 Lha lfiag Stard SaTdy 425 AppatEx Table 1.3 Male Monddy Earmings Regessions Depedent Varib1e Log (Monthly Incone from Main Job) (31) (32) (33) (34) (35) (36) Schooling o.osso 0.0815 0.0o6 0.0831 0.0838 0.0824 (15.53) (14.53) (15.41) (14.24) (14.83) (14.32) Age-School-6 O.OS13 0.0365 0.0524 0.0383 0.0442 0.0379 (8.85) (4.79) (8.38) (4.81) (5.68) (4.77) (Age-School-) k..07 -0.0004 -0.0007 -0.0004 -0.0005 -0.0004 (-S.77) (-2.82) (-5.36) (-2.75) (-4.43) (-2.93) Tenure -4.001S -0.0024 -0.0041 -0.0041 (-0.32) (-0.33) (-0.56) (-0.57) Tenure2 -0.0001 -0.00O0 -0.0000 -0.0o00 (-0.22) (-0.21) (-0.16) (-0.08) Married & 0.1907 0.1564 Cohabiting Duhmy (3.77) (2.80) ILog (Hoursi/WeeO 0.3171 0.2983 0.3197 0.2994 0.2955 0.2888 (6.93) (6.40) (6.98) (6.46) (6.41) (6.22) Work Participation -.1059 -0.1029 -0.0562 Selectivity (Xj) (-2.96) (-2.87) (-1.42) Constant 6.0386 6.3790 6.0071 6.3404 6.1226 6.2834 (32.14) (28.93) (31.73) (28.51) (n3.5) (28.34) F-Statistic 99.07 31.00 66.47 58.11 59.48 52.04 Adjusted R2 0.1948 0.1984 O.l951 0.1983 0.2016 0.2017 Sample Size 1,622 1,622 i,622 1,622 1,622 1,622 Meuns of Variables Log fonthly Icomne) 8.7233 8.7233 8.7233 Schooling 9.7477 9.7477 9.7477 Age-Schooling-6 19.8175 19.8175 19.8175 Tenure & 8.0601 8.0601 Married J& CoShabitinS 0.6486 Average Hours Per Week 45.6457 45.6457 45.64S7 Lambda (Workers/Non-wkvorks) 03505 0.3505 0.3505 No(c. t-ticaw in penttshs. 426 Wmn 's Emiopkym4o and Pay in La Amnwica Appendix Table 18A.4 FPenal Eunings Regresdons: By Employment Staus Dependent Varable: Log (Monddy Incom fiom Main Job) Self-Employed Wag & Salaried Worke (37) (38) (39) (40) (41) (42) Schooling 0.0439 0.0430 0.416 0.0998 0.0916 0.0872 (2.64) (2.09) (2.03) (9.88) (6.80) (6.35) Age&Scbool-6 0.0526 0.0416 0.0172 0.0471 0.0495 0.0378 (3.00) (1.85) (0.74) (S.00) (4.31) (2.70) (Age,School-V -0.0008 -0.0006 -0.0004 .0.0008 -0.0008 -0.0007 (-2.64) (-1.63) (-0.90) (-3.54) (-3.29) (-2.33) Tenure 0.0712 0.0232 (2.75) (1.43) Tenure2 -0.0018 .0.0004 (-1.61) (-0.75) Log (Average Hours 0.4168 0.40S4 0.3702 O.S380 0.5361 0.5212 Worked) (6.29) (6.02) (5.53) (6.3i) (6.25) (6.06) Firm Size 0.1313 0.1286 0.1239 (2.64) (.57) (2.489) Union 0.0931 0.0882 0.0530 (1.72) (1.77) (1.78) Work Participation -0.0986 .0.1413 -0.0125 -0.0291 Selectivity (\) (-1.08) (-1.55) (-0.23) (-0.52) Sector Choice -0.0353 -0. 13S8 0.041S 0.0130 Selectivity (,) (-0.34) (-1.28) (0.86) (0.26) C;onstant 5.Tf73 6.2218 6.7557 4.7280 4.7079 4.9707 (15.56) (10.62) (11.30) (12.81) (10.89) (10.97) F-Statistic 13.08 9.38 8.98 30.41 21.80 17.46 Adjusted R2 0.1516 0.1482 0.1757 0.2383 0.2365 0.2396 Sample Size 338 338 338 471 471 471 Mens of Variables Log (Moothy Income from Main Job) 8.3596 8.2750 Schooling 7.4378 11.1423 Age-Schoeling-6 24.0976 14.5053 Tenure 5.0099 5.3786 Average Weekly Hours 27.8853 39.2125 Firm Size 1.6213 Unionizaton 0.3737 Note: t-taistics in paralbesiL 1i Then Sex Diaa*nlaatin fhi Peru? EvW--i~from die 195(JLhua Llvb.g Siwd 427 Appedix Table 18A.5 Male Eanins Re remiona By Employmat Sbfts Depadt Vaiablr. Log (Monthly Inoome from M iin Job) Self-Employed Wag & Slare Workers (43) (44) (45) (46) (47) (48) Schooling 0.0970 0.0930 0.0945 0.0808 0.0794 0.0792 (8.98) (7.94) (8.00) (13.10) (11.48) (11.31) ASe,School-6 0.0453 0.0287 0.0318 0.0458 0.0293 O.029O (4.06) (1.86) (2.02) (7.03) (3.36) (3.12) (Age-School-6)2 -O.O.OS -0.0002 4.0003 -0.0006 -0.0003 -0.0003 (-2.44) (40.81) (-0.91) (-4.56) (-1.92) (-1.74) Tenure 0.0142 0.0037 (-1.03) (0.45) Tanure' 0.0003 -0.0003 (0.77) (-0.69) Log (Avemag Hours 0.2842 0.2577 0.2687 0.3743 0.3565 0.3565 Worked) (4.22) (3.74) (3.85) (5.70) (5.39) (5.39) Firm Size 0.0102 0.0102 0.0043 (2.41) (2.41) (2.49) Union 0.0421 0.0410 0.0435 (0.89) (0.87) (0.91) Work Participation -0.1178 -0.1201 -0.1144 -0.1104 Selectivity Q) (-1.67) (-1.68) (-2.84) (-2.72' Sector Choice 0.0163 0.0205 -0.9416 -0.035S Selectivity ) (0.22) (0.25) (-087) (-0.73) Constat 6.2339 6.58S6 6.5413 5.8188 6.2711 6.2420 (22.61) (16.26) (15.67) (21.58) (19.39) (19.04) F-Statistic 29.56 21.58 16.91 54.82 40.31 31.38 Adjused R2 02088 0.2103 0.2093 0.2034 0.2077 0.2066 Suaple Size 542 512 542 1,055 l,OSS 1,055 Mean of Variables Log (Monthly Iwcome from Main Job) 8.8725 8.6422 Schooling 9.0738 10.1045 Ago-Schooling-6 21.9686 18.7878 Tenure 8.1274 8.0542 Aveago Weekly Hours 44.7459 46.3467 firm Siz 2.5134 Unioniation 0.3810 Notec: t-stats in parenthesis. References Cornwell, Christepher anl Peter RuperL 'Unobservable Individual Effects, Marriage and the Earnings of Young Men,' working paper. &cate University of New York at Buffalo, 1990. Gill, Indermit "Is there Sex Disaimination in Chile? Evidence from the CASEN Survey,' in Female Employment andt Pay In Lain America, A Regional Study, edited by George Psacharopoulos, Human Resources Division, Technical Departmet. Lan America and the Caribbean Region, The World Bank, 1991a. Gili, Indermit. 'Gender, Occupational Choice, and Earnings in Lain America: TMe Cases of Lima and Santiago,0 Human Resources Division, Techical Department, Latin America and the Caribbean Region, The World Bank, 1991b. Glewwe, Paul. 'The Distrbtionof Welfare in Peru in 1985-86,' Living Standards Measurement Study Working Paper No. 42, The World Bank, 1988. Heckman, James J. 'Sample Selection Bias as a Specification Error,' Econometrca, Volume 47: pages 153-161, 1979. Hecklan, James J. an,- Richard Robb. 'Alterative Methods for Evaluating the Impact of Interventions,' in Longitudinal Analysis 'f Labor Market Data, edited by J. Heckman and B. Singer. Cambridge Universiy Press: pages 156- 245, 1985. Killingsworth, Mark R. and James J. Heckman. 'Female Labor Supply: A Survey,' in Handbook of Labor Economics, Volume 1, edited by 0. Ashenfelter and R. Layard. Elsevier Science Publishers: pages 103-204, 1986. Khandker, Shahidur. 'Women's Labor Force Participation and Male-Female Wage Differences in Peru." This volume, 1992. King, Elizabeth M. 'Does Educaion Pay in the Labor Market? The Labor Force [articipation, Occupation, and Eanings of Peruvian Women," Living Standards Measurement Study Working Paper No. 67, Tle World Bank, 1990. Mincer, Jacob. Stoo1ing, Exrienc ana Earnings. Columbia University Press, New York, 1974. 428 l Thsre 3c Diaimbknatin. Perx? Evw&isc u ae I9V L&xa Lhmg Sdards hrwyey 429 IM>ock, Peter, Philip Musgrove and Morton Stdcner. Education and Ernings in Peru's Informal Nonfarm Family Enterprises," Living Standards Masurement Study Working Paper No. 64, lhe World Bank, 1990. Oaxaca, Ronald.'Male-female Wage Differeals in Urban Labor Markets,' Inwaiowa l Economic Review: pages 693-709, 1973. Steicner, Morton, Ana-Maria Arriagada, and Peter Moock. Wage Determinants and School Attainment among Men in Peru,' Living Standards Measurement Study Working Paper No. 38, lhe World Bank, 1988. Unite Nations Development Programme. Hwnw Dewlopment Rporr, New York: Oxford University Press, 1990. World Bank. World Dewlopment Repor. Washington, D.C. Oxford University Press, 1990. 19 Women's Labor Force Participation and Earnings: The C_ase of Uruguay Mary Arends 1. Introduction Women's wages are about 75 percent of men's wages in Uruguay. This study investigates this differential using econownc-ic analysis. Uruguay is an interesting country to study because of its high female labor force p-nticipation rate (about 50 percent of females betweer. the ages of 14 and 65 participate) and beciuse of a long-standing commitment to publk; educatior. There is a wide scope for investigation of human capital characteristics and their effect on ft aie earnings. A description tr the Uruguayan labor market is given in Lse next section. Section 3 discusses the sample used in the analysis. Section 4 looks .t the characteristics that influence a "man's decision to participate in the labor force. It examines the selectivity problem, wfich arises because working women are a self-selected group out of the entire ferJle sample. Section 5 uses a Mincerian earn-Lgs function to estimate retuias to hunan capital etadowments in the labor market and considers how returns differ between men and womec, alking into account the selectivity problem. Section 6 breaks down the earnings differential to determine how much can be explained by differences irn endowments and what is the upper bound of possible labor market discrimination. 2. The Uruguayan Economy and Labor Market Uruguay's demographics are characterized by low growth rates, a high emigration rate, an aging population, and a high degree of urbanization. The average annual population growth rate from 1965 to 1980 was .4 percent, and was .6 percent from 1980 to 1988. (f the total population in 1988, 26.2 percent was aged 0 to 14, 62.7 percent r:as aged IS to 64, and 11.1 percent was over 65. Eighty-five percent of the population lived in an uban area, and about 52 percent lived in Montevideo. Due to higher emigration rates among men, in 1989, 53 percent of the population of 2,747,800 people were women. Considering the population aged 14 to 65 the disparity was even greater; 55 percent of a total of 2,116,200 people in this age group were female. Economically, although Uruguay has one of the highest GNP per capita in Latin America at US$2,620 in 1989, the country has experienced stagnation since the mid-1950s due to import- substitution policies in the fiftier and sixties, and stabiization policies in the late seventies and eighties. The annual growth rate was 1.3 percent from 1965 to 1988, and -.4 percent from 1980 to 1988. The share of industry in the economy lhs declined from 32 percent in 1965 to 29 percent in 1988. Sixty percent of the country's GNP in i988 came from the service sector. In addition, the country has suffered pclitical upheaval. In 1973, there was a military coup, and a 431 432 WOne w's E op ymc2 and Pay in Latin AnicJ dictatorship ruled the country until March 1, 1985. This period coir,cided with a fall in real wages, which had fallen to 62 percent of their 1968 level by 1984.' The stagnation, coupled with the recent militai.y regime, has led to high emigration rate as Uruguayans move to Brazil and Argentina. Emigration was concentrated in the years 1973 to 1977 and reached its peak in 1974, when 62,400 people left the country. There were approximately 300,000 emigrants from 1963 to 1981, whicb represents about 10 percent of the current Uruguayan population. Emigrants tended to be males, in their early or late twenties, and married. They also tended to be workers in the private industrial sector. Becase of the ag: and educational attainment of emigrants, the impact of the emigration on the labor market was a very well developed preschool program and primary school begins at age 6. Preschool education covers about 40 percent of children aged 3-5; 75 percen of 5 year olds attend school. Sixty-nine percent of the chidren in the sample aged 4 to 6 atend school. lbis would help women w-vth their childcare duties, and therefore enable them to participate in the labor force when the children reach a younger age than in other Latin American countries, where schooling might begin at age six. However, about 60 percent of children without access to preschool educvton are from the poorest households. Women that have more of a need for income may not have the opportnity to send their young children to school. The other variables used in the probit are total household income and the number of working people in the household.'4 Table 19.4 presents the results of a simulation testin for each characteristic whie holding all other charaaeristics at the value of their sample mean. It is apparent that education plays a role in predicting whether a female works. For example, the probability ranged from .28 for women with some primary education to .54 for women with university level education. Also, at lower levels of education the effect is not as significant as for higher levels; looling at the t-statistics, they are insignificant at the 5 percent level for all education levels except the second cycle of secondary school, the university, and teacher school. Apparently, the opportnity cost of staying out of the labor market is higher for women with more education, and the opportunity cost effect outweighs the positive effect that education has on the reservation wage. The net effect is that higher education is associated with higher participation rates in the labor force. " Standing, (1982), p. 55. '4 As shoin the oter chapters in this book, being maied ha a significant negative effect on the probability of working in uV- ealier equaton, but was left out in the final probit equation because of coelaion effects with tha number of chiL. Wwm 'a Labor Force Paricoaiou and Eansgs.: The Case of Uruguay 441 Ta 19.4 Predicted Waiting Probabiltis by Cbhzcestic Chacteristic Predicte Probability odcain Lvs No Edocafim .35 Some Pnimzy .28 Completed Priury .34 Fiast Cycle Secoodary .37 Secod Cycle Scoonday .46 Technical .38 Teacbe .52 Univenuiy .54 Odher Edation Level .45 # of Child= 0 to 3 None .39 One .32 ITWO .25 Thre .20 # of Childri 4 to 6 None .38 One .34 TWO .30 Tree .26 # of Childre 7 to 12 None .39 one .35 Tvo .31 Three .28 Ago 14 to 19 .12 Age 20 to 24 .38 Age 25 to 29 .54 Age 30 to 34 .62 Ag 3S to 39 .55 Ag 40 to 44 .55 Ag 4S to 49 .47 Age SO to S4 .35 Age 55 to 59 .28 Age 60 to 65 .20 HTead of Houb4 .65 No .34 ontiud- 442 Women ' Enp1oymen w zad ay in Latin America Table 19.4 (continued) Predicted Wozing Probabilities by C hareceristic Charscteaiaic Predicted Probabiity Live in Montevideo Yes .5 No .49 Number of Ocuied Perso in HousehQld No= .08 one .22 .44 Three .67 Four .8S As for age effecs, participation is high at all ages compared to other Latin American countries, and peaks at the age of 30 to 34. This indicates that Uruguayan women have a high dedication to the work force thrughout their life cycle. Age is a highly significant determinant of participation at all levels, and the age splines have the highest-valued partial derivaives of all the explanatory variables. Participation is lowest at ages 14 to 19, which is expected given Uruguay's high enrollment rates in secondary and tertiary education. Tle number of children is also a negative and significant determinant of labor force pzticipaton With no children aged 0 to 3 and other things being equal, the participation rate would be .39. With one child aged 0 to 3, the probability drops to .32, with two to .25 and with 3 to .20. There is also a significant difference in the effect of 0 to 3 year old children compared vith 4 to 6 year old children. The mber of chidren aged 7 to 12 has less of an impact than the numiber of younger children. Being the head of a household also significantly increases the probabiity that a woman will woot from .34 to .65. This ma sense becse female headed households are likely to be poorer than male headed households, increasing the woman's necessity to work. The mmiber of employed persons in the household has a significant, positive effect on the probability that a female will be working. The explanation fo. the sign is not immediately obvious. This may show that wealthier households tend to divide up labor, with women working at home and n working outside the home, while poorer households have to send more members, incuding children, into the wage earning market It may also show a kind of family 'work ethic' with members preferring to work outside the home. The coefficient on household income is small and negative, which is as expected. Lastly, living in Montevideo had a small, positive, but not very strong effect on the decision to work. S. Earnings Functions Regression results for men and for women, both corrected for selecivity and uncorrected for selectivity, are presented in Table 19.5. Obviously, the sample for the earnings regression includes only working men and working women who reported positive income, positive hours, Woe 's L1bO7 Force PW=W& and Emiags: T Case qf Uraguy 443 Tabau 19.5 ,,_____,_ Earnings aFuctions Females Females (Coffecm _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ~~~~~for Selectivity) basic alternate basic alternate basic alterna Constant 1.1:2 1.545 .421 1.092 .351 .990 (14.362) (20.492) (5.467) (13.766) (4.139) (11.610) Schooling .098 .OS5 .110 .076 .112 .078 (Years) (41.524) (36.974) (36.244) (23.347) (35.532) (23.468) Experiene .057 .042 .042 .036 .044 .039 (29.110) (19.712) (16.152) (13.758) (15.86 (14.028) Expeiene Squared -.001 -.001 -.001 -.001 -.001 -.001 (-20.582) (-14.737) (-11.576) (-10.110) (-11.S18) (-10.584) Los Hours .586 .516 .685 .628 .684 .626 (31.298) (28.620) (39.653) (37.709) (39.656) (37.674) Married .274 .040 .037 (14.310) (1.987) (1.80i) lnfornnl -.277 -.427 -.432 (-14.2U4) (-48.007) (-18.216) Public Sector -.075 .152 .151 (-4.270) (5.704) (5.688) EmploWy .421 .50 .5 (13.891) (7.269) (7.258) l hambda .060 .093 (2.001) (3.243) Adjusted R-squared .352 .397 .401 .464 .402 .465 N 6,646 4,484 4,484 Not= Bae gSup fDr regrea including scorn and Mri SUM am mmarnied ardeum, eir s- nplbyed or vmgeam*ra in the priva* seWr, who am not employe. and for whom the yeas of schooling and experience could be esdmated. The model estiat is the stndard Mincer wagernings equation, where the log of wage is regressed on years of schooling, eperience, and experience squared. In these regressions, the dependent variable is the log of the prinmy monthly earnings, and the independent variables include the log of weekly hours. The log of the hourly wage is not used as the dependent variable because then the elasdcity of the wage with respect to hours would be constrained to be one. In the final results, this eaicity is always significanty differt from 1. 444 Women 's Fmpkymat and Pol in Latin AnXrca The earnings functions are corrected for selectivity using Heckman's two-stage procedure (1979). By computing the probit equation, it is possible to compute the ir /erse Mill's ratio (Lambda) for each working woman in the sample. The Lambda is then inclu Jed in the explanatory variables for the female regression, and the results are compared with the uncorrected regression. In the uncorrected regression, the return on education for women is seen to be about 11.1 percent. A one percent increase in weekly hours worked is associated with a .7 percent increase in monthly earnings. ITe remrn on experience is 4 percent, and the sign on experience squared is negative, implying a concave earnings function. When Lambda is added to the regrcssion determining women's earnings, the return to schooling increases very slightly to 11.2 percent and a one percent increase in hours worked is still associated with a .7 percent increase in wages. There is very little difference in the results when Lambda is included. Lambda is positive and barely significant at the 5 percent level. This indicates that those characteristics that are associated with high earnings also increase the probability that the woman will be in the labor force. Another way of saying this is that working women have a comparative advantage in work outside the home. Those women who earn higher wages are also likely to have lower reservation wages. The return to schooling is higher for women than men (11.1 percent versus 9.9 percent), but men have a higher return to experience at about 5.8 percent. The elasticity of earnings to hours worked is about the same for both groups. There is no sel-ctivity correction foe mnen because their parLicipation rate is high at 84 percect. Interesting results occur when dummy variables representing sectors and marital statLs are added to the wage equation. The rate of return for education to women declines to 7.7 percent from 11.1 percent in the uncorrerted regression, and the reurn to experience also declines to 3.7 percent. The increase in adjusted R-squared implies that these variables were omitted variables in the first regression and perhaps some of the retrn to ed-,cation should be attributed to sectoral choice. It is apparent that werking in the informal sector is associated with a 43 percent decline in earnings when compared to the reference group of those employed in the private formal sector, while working in the public sector or being an employer implies earnings premiums of 15 percent and 51 percent, respectively, compared to the reference group. The sector effects are significant. When Lambda is added to the equation, its coefficient is larger and more significant than when the sectors are not included in the regression. There must be interactions between the decision to work and the choice nf sector. Again, the other coefficients are not affected much wnen Lambda is added to the right hand side of the equation, although the returns to schooling and experience increase slightly. Being married, ceterisparabis increases wages by about 4 percent, but the effect is barely significant at the 5 percent level when Lambda is not in the equation, and is insignificant when correcting fur selectivity. For men, adding the marital status and sectoral dunmmies decreases the return to schooiing by about 1.3 percent and reduces the return to experience by about 1.5 percent. The reurn to schooling becomes higher fur men than women. For the men, working in the informal sector or the public sector is associated with lower wages than the reference group by 28 percent and 8 percent respectively, while employers earn a 42 percent premium. This implies that the sectoral choice is more important in determining female earnings than male eirnings. However, married males earn approximately 27 percent more thts- single men, and thbe effect is significant. The rationale fur including marital status in the regression is that it is often observed that married people earn more than single people across countries. One explanation could be that skills valued in the household are also beneficial in the work place. Another explanation is that marriage Wowu j Labw Force Paricoao aed E6ndags: 7he Ca of Uragwy 44S allows male workers to increase titeir dficieny dLrough specialization of labor, whth the female concentrating on household tasks and the mmn cow e on outside work. The experience profile for adl specifications pab lata for men than women. For the women's regressions with selectivity, it peak at about 35 years of age; for women's regressions without selectivity, it peaks at about 36 years o' age; and for men, the profile peaks at about 38 years of age. 6. Discrimiation Having estimated the acofficients for males and females, the Oaxaca decomposition can be determined. Oaxaca (1973) devised a medthd to break down the earnings differential into two parts; differences explained by differentis in human capital endowments (endowments) and differences caused by variations in retrmn to human capital in the labor market (the wage structure). The latter represents he upper bound to discrimin on. In symbols, the difference between males and females is the following: In (Earninpsj) - In (Earnins) = X*bb,, - X,b, Xi represents the mcans of the sample parameers, and b, their corresponding coefficents. There are two equations that can Fe used to do the decomposkien, and they will give different results. One equation measares the differeztiial using tie female means and the other measures it using the male means. Xhb, - X1b, = Xj(b. - b) + b,.(X, - X) (1) Xbb. - X,b, = X.(bD - b + bA(X - X Q The first term in both equations on the right side refer to the differences in earings due to difference in wage structure, while the second tem refes to the differences due to the differences in endowments. The two equations present a base mimber problem, and there is no economic reason to use one of the two equations over the other. Table 19.6 includes the reslts using both equations. Using the eamings coefficients from the selectivity correctd ordinary least squares (OLS) estimates for the regression hinuding only experence, schooling and log hours, 23 Percent of the difference in earnings can be attributed to differences in endowments ard 77 percent to the differer^e in wage structure betwee men and wom The percenez are coincidentally the same whether evaluated at the male or female mea=s. The OLS esthztes uncoffected for seliwtivity imply that a higher percentAge of the difference can be explaine-a by the differences in endowments. For the regressions with the dmnmy variYies for sectors and marital satus included, the percentage explained by endowments is higher, whether the estimates are corrected for selectivity or not. A higher percentage is explained by endowments with the uncoffected OLS equations than the corrected OLS equations. The uppcr bound on discrimination Is estimated at about 55 to 60 percent for the equations icluding the sectrD and marial staus dummy variables. 446 Womn 's Eipoymem and PeW i Lan Aaia Trbl 19.6 DwoCDi& of So Mdflemle E=Bs Dff aid Fescg of MaePy dwe Due to Diffei In Specification (%) Wage Srucur () Corrected for Selctvity Evaluated at FePale Mews (Equati ot) Simple Regreson 23 77 Regression with Sect= 35 65 Evaluated at Male Means (Euan 2) Simple Regressiklt 23 77 Regressiun with Sectom 40 60 Unodm fo S *mivt Evaluated at Female Mews (Equation 1) Simple Regresion 24 76 Regression with Sect= 39 61 Evaluae at Male Means (Equaio 2) Simple Regression 26 74 Regression with Sector 44 56 7. Discussion What conclusions can be made about discrimination against women in Uruguayan labor markets? Overali, Uruguayan women esrn about 75 percent of what men earn, and in some sectors that ratio is higher than in others. Specifically, women in the public sector edrn about 90 percent of what their male colleagues make, whie women in the informal sector eam between 65 to 75 percent of male earnings. The relatively strong position of females in Uruguayan labor markets can be atributed to high educational attaiments and to the recent emigrations. The exodus of educated, prime working age men in the 1970s provided labor market opportunities to womn who were prepared to take advantage of them, specifically dose with higher educational levels. However, there are market fores working against woenw. especially those with lowar educational levels. Declining standards of living have pushed more women out intr, the labor market, and this has tended to decrase women's real wages in the Fectors that wome. with hILle education and little work eerience are liky to enter, and especially in the informal sector. The greatest expansion in absorbing the female labor force has come in social and personal services, traditionally female occupatio. They also happen to be among the lowest paying occupations. Looking at the e-irngs regresions, d4 seems that the choice of sector has a larger impact on women's thau men's earning It coud h that the pay in informal sector activities is lowe in wnica case tuet, wudO no De tenor maricet aiscrinaon, but lower wages as the result of trade-offs made by working women. From the decomposition, it is true that in every case differences in wage structure are more important ir exrlaining the dixterential than differences in eaiowments. However, the difference in coefficieuts could be biased upwards. For example, the proxy for experience in the female regression is likely to be. ovdmated, becse women typically have intenuptod c:aer. ir order to riaise children. This will bias the return to experince downward, since the wage exper. - - profile is concave. Ihis will alo increase the percentage c-hbe diffrental attributed to wage stucture, and therefore, discrimination. Also, it would be heip:; >o have measrements of job tnre, or uninterrupted time in the labor force. This is a proxy for dedicadom to labor market activities, and could be a missing variable that is higher for men than women and which is desirable to employer. It is imporutt to note also that addir, the secteral variables increases thc percenutge attrbutable to endowmnts; the choice of secto. s obviously an important factor in examining discrimination. Uruguayan wornen coud benefit from po!icies that would make it easier to combine household work and work ir ;?e formal sector, such as expandeod provision of daycare. roverage of the already-exing presehool program could be expanded to poorer families. Pay is low in dte informal sector, and the ratio o fwale to male wage is also low. lTe sarme is tuue of thc self- etnplr; rector. The sWte has already d-,ne much in the area of preschool edt'caion. Also, fiunh. . v shoulid be done to determine why the wage sturure is different ber veen men and wome., ind whethe;' women are constrained by custom or habit from more highly paid occupAioMs. References Agu far, C. El ImpaLlo de las Migracdones Internado.ales en el Mercado de Empleo del Pals de Orlgen: El Caso Uruguayo. Montevideo: Centro Interdisciplinario de Estudios Sobre el Desarrollo Uruguay, 1984. Alderch-Langen, C. 7he Admlssion and Academc Placement of Students ifm Seleced Souwh Amecar Counries. Wshington, DC: Nationai Association for Foreign Student Affairs, 1978. Fortuna, J. C. and S. Prates, 'Informal Sector versus Intormalized Labor Relations in Uruguay.' In A. Portes, M. Castells, and L. Benton (eds.) The Informal Economy: Studies In Advanced and Less Developcd Countres. Baltimore: Ihe Johns Hopsins University Press, 1989. Heclnman, J. 'Sarmple Selection Bias as a Specification Error.' Economesrica, Vol. 47 (' J79). pp. 153-161. Lubell, H. The Informal Sector in the 19 and 199's. Paris: Development Centre of the Organisation for Economic Co-operation and Development, 1991. Mann, A. and C. Sanchez, 'Labor Marke )onses to Southern Cone Stabilization Policies: Tle Cases of Argentina, Cbile, U. q.f Inter-American Economic Affars, Vol. 38 (Spring 1985) pp. 19-39. Mincer, I. Schooling, Erperience and Earnings. New York Columbia University Press, 1974. Oaxaca, R. 'Male-female Wage Differentials in Urban Labor Markets.' InternasloncdEconomic Review. Vol. 14, no. 1 (1973) pp. 693-701. Portes, A., S. Blitzer, and J. Curtis, pThe Urban Informal Sector in Uruguay: Its Intenal Stucture, Chacterisdcs, and Effects.' World Dewlopment, Vol. 14, no. 6 (1986) pp 727-741. Solari, A. "Analysis of Educational Finaicing and Adminisraion in a Contet of Austerity: the Case of Uruguay.' Major ProJeat in the Reld of Education in LarN America and dte Caribbean (a.'Oe). No. 21, April 1990, pp. 34-56. 448 - ---~o -- --- - - -- *. : "MV.416 AU4"UU1MA L^IW Organization, 1982. Taglioretti, G. Women and Work In Uuguay. Paris: UNESCO, 1983. Weal, T. Area Handboo&for Uruguay. Washington, DC: American University, 1971, pp. 117- 144. Weinstein, M. Uruguay: Dowarcy at dte Crossroads. Bolder, CO: Westview Press, 1988. World Bank, *Uruguay: Employment and Wages. Country Operations, Division 4. Report No. 9608-UR, May 1991. 20 Female Participation and Earnings, Venezuela 1987 Doal Cox md George Pachampoulos 1. Introduction In 1987, the average eamings of Venezuelan working women were 70 percent of the aveage earnings of Venezuelan working men. What accounts for the pay gap? Are huffn capital indicators lower for women? Or is the gap due in part to labor market dicimination? This chapter sheds some light on these qucstions by analyzing Venezuelan household survey micodata Simple comparisons of differences in average earnings can be misleading when making inferences about possible discrimination becs skill indicators can differ between men and women. So we seek to estimate earnings differences while controlling for earnings deerminan. 2. The Venezuelan Labor Market Modenization, improved access to educadon, the growth of the public sector, and long term declines in ferdlity rates have all contributed tO significant increases in female labor force participation rates in Venezuela and by 1989, women accounted for 30 percent of the labor force. Women's participation rates have increased in both public and private sectors, although the increase has been most pronounced in the public sector. A number of studies (World Bank, 1990) show that women earn lower wages than men in VenezzOc!s. While the wage differentials are pardy explained by women's concentradons in lower-paying kndustries and in -he informal sector, there is some evidence suggesting dtat women earn less than men even when they have similar levels of education. years oi experience in the labor force, and when they hold similar jobs. This study uses 1987 Houslhold Survey datm to deternine what proportion of the niale/female earnings differential is due to differences in huma capi.l endowments and how much of the differential can be attbuted to diffrfences la the way employ'ers value male and female labor. 3. Data Characteristics We use data from the Encuesta de Hogares, a survey covering a representative -rossecton of households in Venezuela. The survey was conducted t-y the Oficina Central de Estadistica e Infonnatica (OCEI) in the second semester of 1987. nTe data set contains observations from 131,032 households covering 681,Z28 persons and contains information about labor market earnings and individual characteristics such as age, schooling, gender, ad place of residence. 4,S From the large data set we selected a random sample of 10 percent of the individuals. Since we are interested in the behavior of prime-age individuals, we restricted our sub-sample to individuals aged between 20 and 55 years. Further, we dropped persons with inconsistent labor- market infonnation: Those wlbo reported having earing but no hours worked or vice versa. These sample-selection criteria rewilt in an aggrate sample of 17,725 individuals: 8,375 working males, 4,131 working females and 5,219 non-working females. Clbe reasons for including non-working females in our analysis is explained in a later section.) Table 20.1 displays rmeans and standard deviations (S.D.) of selected variables from the sample. The average age of working men and women is about the same, but average schooling among working women exceeds that of working men by almost a year. Eleven percent of working women ittended university, compared with 8 percent of men. And half of the working women attended secondary school compared with 41 percent of men. A higher proportion of men were self-employed-28 percent compared to 22 percent for working women. But female workers were better represented in the public sector than men-31 verss 17 percent. The earnings of women are 70 percent those of men C2,700/3,827). Non-working women are slightly older than their working counterparts, and haoe approximately 2 years less schooling. The variable 'years of experience' is computea in the standard way by subtracting years of schooling plus 6 from age. Note that we do not have measures: of actual labor market experience in our data set So we must use potential years of labor market experience as a proxy. WVe reoDgnize, of course, that women frequently experience interruptions in their careers, so that our experience measure is an imperfect proxy for actual years spent working. 40 Detes-minants of Female Labor Fmer Partipatlon We seek to estimate earnings functions for men and women, but our analysis for women poses a special problem because of the intermittency of female labor force participation. Hence we are estimating earnings functions for self-selected samples of women. That is, we estimate earnings functions for women whose market wage exceeds the value of time spent at home. Our analysis is based on the occupational choice model of Roy (1951) that has been explored by Hec.inan (1979), Lee (1978), Willis and Rosen (1979), Borias (1987) and others. Suppose the market w4c of woman 'iP is given by the equation: w(m) = bX + e(m), (1) where X denotes a vector of wage determinants, b measures the returns to those determinants, and e(m) is the error term for woman *i. CThe subscript '"i applies to the terms w(m), X, and e(m) and is suppressed for convenience.) The value of time spent at home for woman a; is expressed as: w() aZ + e(). (2) The vector Z denotes the detrminants of the woman's productivity at home. We assume the vector Z contains all of the elements of X, plus other determinants. This assumption makes sense since variables such as being a household head would affect productivity at home but not in the Memo (d S D of S V Variab1e Woz1dns Mea Workng Woma Nca-Working Womza APg 33.6 33.50 34.15 (9-78) (9-00) (10.36) Yam of Schooling 6.97 7.86 5.50 (3.79) (4.02) (3-64) Yer of Ezpeimc 20.65 19.6 22.65 (11.00) (10.03) (240) Educatio Level No Educadcm 0.05 0.05 0.16 (0.24) (0.24) (0.36) PriUY 0.47 0.34 0.51 Swonyw 0.41 0.50 0.34 Univouity 0.08 0.11 0.04 Self.Emp, Sey 0.28 0.22 (0.45) Publi-Secor Wortw 0.17 0.31- (0-37 We&ely eming (ba&) 3,826.5 2,700.2 (3523.6) (2,060.6) E&ningn of Otw 41,98.8 6,734.7 - (4,496.3) (5,763.7) N 8,375 4,131 5,219 Moto: Mean Fmnl Peatciptina .4418. Fre i back. am gdai dcvno1 Sour= Eacecat de Hogpuc, 1987. marke. On the other han, it is hard to dtink of personal attrbus tha affea the m agm esa but nt the value of tim spew at home. The vector a denotes the retrmn to attributes Z and e(h) is the ew te asso with the value of tm at hom. (Ch subscript i applies to the tem w(h), Z, and e(b) and is upresand for cvenie nce.) The efror terms e(.) are assm to be normal with expecteatin 0 and covaian matrix of fill rnnL A woman chooses to work if, and only if: I = w(m) - w(h) = bX - aZ + t(m) - e(h) > 0, (3) where I denotes an index of labor force participation. The variable 1, which differs for each individual in the cross-section, is a continuous variable that indexes the propensity for a woman to enter the labor force. If the variable I crosses a cenain threshold, the woman enters the labor force (otherwise she does not). We do not lose anything essential by normalizing this threshold to 0, as in expresskon (3). Consider the expected value of women's wages, conditional on working in the market and on X. E(w(m)) = bX + E(e(m)II > 0). (4) 'E' denotes the expectations operator. This is the standard sample-selection problem considered by Heckman (1974, 1979). The expected value of the wage is conditional on the sample-selection rule, which in this case is that the women work (I> 0). Focus on the last term in 4, the expectation of the error term. That expression can be written as: E(e(m) II > 0) = c(f(l)/FQ)) = cL, (5) vihere f(l) denotes the ordinate of the standard normal density evaluated at the index Y, F(l) the standard normal distribution function evaluated at (I), and the ratio of the two (rewriven as L) is the inverse Mill's ratio term. It can be show. that the variable c can be written as: c = s(h)(s(m)/s(h) - r(m,h)), (6) where s(.) denotes the standard deviation of e(.), (e.g., s(m) is the standard deviation of the error term associated with market wage offers) and r(m,h) is the correlation between e(m) and e(h) (-1 <= r <= 1). The derivation in equation 6 is useful for determining the sign of the coefficient of the Mill's ratio, or selectivity term. The sign of c can be positive, negative, or zero, depending on the dispersion and covariance of the error terms e(.). For example, if e(m) and e(f) are inversely correlated (r < 0), so that unobserved characteistics that raise market productivity (e.g., aggressiveness) lower productivity in the home, and the dispersion of marlet wages (captured by s(m)) is low relative to that of bome production (s(h)), then the coefficient of the selection term wil be positive. In this case positive self-selection will occur; women sort themselves into the sectors in which they are most productive. On the other hand, it 3(m) and s(h) are roughly equal and r is approximately 1, the coefficient c will be close to zero. In this case, there is a strong positive correlation between unobserved market and home traits and the dispersion in the market and home error terms is the same. It nukes sense that self-selection effects are minimal in this case because unobservables in each secVor are strongly correlated and their dispersion is the same. Now consider the case in which r- 1 but s(h) > s(m). Unobservables that help boost the payoff to home production boost the payoff to market work as well. And the dispersion of rewards in the howe ,txotr is higher than that of the market. The most productive women will be attracted to the se=-.c with the greater dispersion. The reason is that if they are going to be at the top, they might as well be at the top of a wide distribution-this strategy maximizes the reward from sector choice. On the other hand, women whose e(.) terms are lowest will be attracted to the sector with the lesser dispersion (in this case the market sector). The reason is that if they are going to be at the bottom of the distribution, they might as well choose the sector with the smaller dispersion so that their reward, which will be smallest, will at least not be too small. In this instance, then, the market sector attracts the women who are least productive in terms of their unobservables, so that the selection effects in the market wage equation are negative. The first step in estimating selectivity-adjusted earnings functions is to estimate the index functior' 3 using probit. The vector Z contains age dummies, schooling dummies, dummies for region of residence, dummies for whether the woman was a wife or partner of the household head, and earnings of other household members, and a rural residence dummy. The dependent variable in the probit analysis is labor force participation, which is defined as earning at least 200 bolivares per month in 1987. This definition is not strictly comparable with the official one, which counts both the unemployed and employed as members of the labor force. But the concept of unemployment is complicated; experts disagree on exactly who is unemployed. The concept of employment is unambiguous; it is easy to identify those who have earned over a threshold amount. The estimation results are presented in Tables 20.2 and 20.3. The probability of participating in the labor force steadily rises with age until women reach their late forties, then it declines. The age effects are large. For example, the probit coefficients indicate that, controlling for other factors, the probability of participation is about 30 percentage points higher for women in their early 40's than for those in their early 20's. Education has powerful effects on participation too. All else being e-jual, secondary school graduates have an estimated participation probabUity that is 30 percentage points higher than primary school graduates. University grauuates have a participation probability 70 percentage points higher than primary school graduates. Rut graduating from a technical secondary school results in a lower participation probability than graduating from an academic secondary school. This result is puzzling, but also likely to be imprecise-less than 2 percent of the sample attended technical school. Tnose with some university education are less likely to participate in the labor force than secondary school graduates. Part of the reason might be b.at attending a university raises reservation wages leading to longer spells of unemployment. The marital status and headship variables are very large, precisely estimated and have the anticipated sign in the participation probit. Being a wife or partner reduces the probability of participation by 22 percentage points, so family responsibilities compete for time spent in the market. Being a household head raises the participation prohability by 23 percentage points. Income of other family members reduces the probability of working. Ihis is most likely due to income effects which raise the demand for time spent at home (Mincer, 1962). A 15,000 bolivare increase in other income reduces the probability of working by 4 percentage points. Finally, participation probabilities follow distinct regional patterns. Women from Caracas are more likely to work than those from 5iuayana. A 9 percentage point difference in participation probabilities exists between the two regions. And living in a rural area reduces the probability of participating by 13 percentage points. 456 Women's T kTylna a: tray L A0rIca Irb220.2 Pmobit EstntA for Female Prtcipton Variable Coefficient t-valuw Mean Part derivative Constant -.861 -10.80 1.00 Age < 25 -.153 -2.33 .211 -.06 Aged 25-29 346 5.49 .187 .136 Aged 30-34 .473 7.51 .170 .186 Aged 35-39 .526 8.36 .143 .2I Aged 40-44 .60 9.17 .103 .239 Aged 45-49 .452 6.64 .086 .179 Sone Prinam .191 3.52 .172 .075 Primary grad. .265 5.08 .258 .104 Somn seondary .77S 14.34 .250 .306 Secondssy gd. .991 L5.56 .115 .390 Technical .531 4.36 .015 .209 Some university .735 8.74 .039 .289 Univessity grad 1.809 16.24 .031 .712 Wife or parinr -.561 -16.18 .563 -.221 Household head b587 9.42 .085 .231 Other earnings -.674E-oS -2.59 6492.1 -.000 CaGacs .443 6.10 .055 .174 Cental .303 6.16 .282 .119 W Cental 325 6.75 .318 .129 Guayan .226 4.38 .222 .0897 Rural -.338 -7.25 .142 -.133 Nem: Samplew Women agod 20 to 55 yenr. Observations 93 Mean Parficipn .4418 Log-Likelihood -5457.3 Chi-Squued Statistic 19M3 5. Eamings Functions The next equation to explore is the selectivity-adjusted earings function for women. lt'ber dt deflate monthly log earninp by hos wored, we include the lo, of hours worked as a sepaate regressor. This f onal form is more flexible than using log (eau.ln hurs), which restricts the elasticity of earnings with respect wo hours to be unity. We estimate the sbndard fincerian eaming function, which includes years of education, experience and experience swared as regreso, in addition to the log of hours worked. The regression results are given ii Table 20.4. This table displays the earing function adjusted for selection bias. The estimated rae of r urn to schooling for women is about 12 percnt, whih is high by United Stat' sumdards but loe than that found in other Latin American countries (see other chapters in this volume). The log earnin increase with experience at a decreasing rate, which is a famiiar ret for eanings equa s of this sort. Recall that it is potential experience that Is measured hae, since most women have interruptdons in their careers. At Femal ParI4Paon and EaniRgs, YVawaI 1987 457 Table 2@.3 Predicted paticipahan probabilitte by characteristic Charaberic Predicted Probability 20-24 .27 25-29 .45 30-34 .50 35-39 .53 40-44 .56 45-49 .50 50-55 .32 No education .25 some pfimay .31 Primar grad .34 Some secondary .54 Secondary grad .62 Technical .44 Some univeeaBty .52 University grd .87 No .56 Yes .34 Hmt,ld kg No .41 Yes .64 Region and l1catigg .50 Central .45 W Central .46 Gusyana .42 Other region .33 Urba .45 Runa .32 sample means, the rate of return to experience is 1.8 percent The peak of the eamings- experienceprofile implied by the esdmates is 50, which means that the estimated peak in earnings occurs at about age 64. So eaminp do not turn down until women are well into their potPntial retirement years. 'he estimated dastity of earing with respect tX hours worked is significantly different from unitv. The oefficient In Table 20.4 indicates that a one percent increase in weekly hours workd is a5sociated with about a half a percent rise in monthly earnings. Finally, the coefficient of the seletvity variable (inverse Mill's rado) is negative and significan at the .05 level. Multiplying the coefficient of the sdectivity variable with its sample mean gives the average error term conditional on being in the labor force, which is about 5 percent. 4) W,men a r'"WWymens ana ray in LaLn Amana Table 20.4 Ernings Functions Variable Men Wome Womm (Correced for (Uncorrcted for Selectivity) Selwtivity) Constant 3.986 4.425 4.302 (38.369) (38.878) (43.623) Schooling (years) .106 .117 .121 (63.587) (35.725) (47.657) Experience .052 .030 .031 (25.568) (10.051) (10.841) Experience squared -.0006 -.0003 -.003 (-14.806) (-4.924) (-5.329) Ln (hours) .682 .535 .541 (25.327) (21.863) (22.210) Lambdg -.064 (SelecivityVariable) (-2.164) R2 .379 .426 .426 N 8,375 4,131 4,131 Notes: Figues in s are t-ratios. Depwdnt variable = log (hourly earnings) Analysts are sometimes puzzled by negative selection effects, but they are consistent with one of the sctnarios discussed in the theoretical section-namely, (1) a strong positive correladon between unobservables in market and home productivity and (2) a greater dispersion in rewards to hometime compared to market work. To see whether adjustment for sample-selection bias makes a difference for rates of reun to schooling and experience we re-esi d the earnings fiucsion by simple ordinary least sqites (OLS) (rable 20.4). The esimated rate of return to schooling is a fraction of a percenage point higher for the corrected esdmate. Both the slope and concavity of the earnings profile increase a bit in absolute value. The net effect is a 1.9 percent rate of return to experience at sample means, compared with a 1.8 figure. The effect of omitting the se;ection terms from the earnings- function estimates is to bias upward the marginal rate of return to human-capital indicators by about 5 percent (wnt percentage points). The earnings function for men is also given in Table 20.4. Note that we do not correct for selection bias in the male earning functions. The reason is that labor force participation for prime-aged males should be exogemo. If we were to include males that are close to school age or retirement age the decision would be endogenous, but recall that our samples are for people aged 20 to 55. Some males might have earnings below the threshold of 200 bolivares, but for Fwmak Pafp n and EaniLngs, Veneza 1987 459 reasons that are likely to be exogenous to the model; illness, unemployment caused by deficient demand, or search unemploymenL The estimated rate of return for schooling is slightly lower for males than females. But the returns from labor market experience are a lot higher for men than women. At sample means the rate of retn to experience is 2.4 percent for men, a third higher than the comparable figure for women. nis result is consistent with human-capital-invesument theory, which predicts that workers with long horizons will invest in skills morc than those with short ones. And men are likely to have much longer horizons than women who drop out of the labor force to raise children. How is the investment effect reflected in the experience-earnings profile? Investir, a lot early in the areer entails foregone earnings, which lowers starting wages. But as skills accumulate with experience investment declines. The latter occurs because It pays to invest the most when young. The two effects combine to steepen the earnings profile. 6. Disa imination Now that we have estimated earnings functions for men and women, we can addiess the question posed at the beginning: how much of the male-female earnings differental can be explained by observed factors? How much might be caused by discrimination? The technique used to answer this question is the wide-y-used Oaxaca (1973) decomposition. The idea is to split the difference in log wages into that accounted for by differences in observed variables, and that accounted for by differences in the way those variables are rewarded. We can write the difference in log earnings of men and women as: BmX - B;X, = (B - B)X. + B D- X) (7a) = (B. - BJX, + BD(X. - X, (7b) whae Bi i = m,f are the estimated coefficient of the earmings functions and Xi i = m,f are the averages of the explanatory variables in the earnings functions. Focus on expression 7a. The first term is the difference in rewards, for those having mean attributes of men. The second term is the differential due lo differences in atibutes, weighted by the vector of female coefficients. Expression 7b does the same job as 7a, but the weights are different. (We discuss this below.) Before we proceed fiurher in calculating the 'explained' component of the wage gap, we need to address three further Lssues. Fur, what sample mean should we use for women-workers only or the entire sample? The answer is that we should use the entire sample, because the selectivity corrected equations gives us an estimat, of the population regression function when we base our predictions o'i the estimate of b in equation 1. Second, we do not include the Mill's ratio terms or their coefficients in making our predictions because we seek to measure the conditional mean f,r the population, not just the sample of working women. Third, though we use entire-sample means for female schooling and experience, we use the working-sample mean for log hours, sincet hours for non-working women equal zero because they do not participate in the labor force. Second, note that the Oaxaca-decomposition can be done two ways, hence expressions 7a and 7b. Which way is best? Economic theory gives little guidance; this is an example of the index- 460 Womeni 's Emp&ymem and Pay in Latn Amaica Table 20.5 DwouTosifim of the MalFemale Eatnings Diffenl Perntage of Male Advantage Due to Differes in Male Pay Advantage Eadowiit Wape StrUture Estimated at Male Means (7a) 12.87 29.33 42.20 (30.5) (69.5) (1X.0) Esfimated at Female M ran (7b) 14.96 27.24 42.20 (35.4) (64.6) (100.0) Nota: Figures in parfndw= an percatages Male pay advantage = 42.2% number problem which arises in many problems in applied economics. So we vill d the decomposition both ways. lhe male pay advantage is 42.2 percent. lhis is the empirica1inalogue of cxpression 7. How much of the advantage is explained by observable nactors? The answer is 12.9 percentage points. This is the empirical analogue of L e expression B1(XQ - XJ. The rest of the advantage is due to the way attributes are rewarded. So obswvbles explain a little less than a third of the pay advantage for men. If we do the calculations according to expression 7b instead (to explore the index number problem) the anount of the pay advantage explained is 15 percentage points, or a little over a thid of the acala pay advantage. A couple of caveats about the Oaxaca decomposition technique should be zoted. First, the right- hand-side variables do not captire every skill component that affecs eaaiings. So if we attrbute all of the nmexplained pay gap tn discrim tion, we must recognize that it is an upper bouid. After all, some of the unexplained advantage could be due to male skill advantages that we did not measure. Left-out variables bias measures of discrimination upward. On the other hand, the right-hand-side variables themselves could be affected by discrimination. Suppose discriminarion led women to go to school for fewer years than they would have liked. If discrimination affects right-hand-side variables, this cculd bias discrimination measures downward. Referonces Borjas, G.J. "Self-selection and the Earnings of Immigkants." American Economkc Review, Voi. 77 (1987). pp. 531-555. Heckman, J. "Sample Selection as a Specification Error.' Econometira, Vol. 47, no. 1 (1979). pp. 153-161. . Shadow Prices, Market Wages, and Labor Supply." Economet#ca, Vol. 42, no. 4 (1974). pp. 679-694. Lee, L. "Unionism and W_ge Rates: A Similtaneous Equations Model with Qualtative and Limited Dependent Variables.' Internauonal Economic Review, Vol. in (1978). pp. 415- 433. Mincer, J. 'Labor Force Participation of Married Wcmen: A Study of Labor Supply' in National Bureau of Economic Research. Asperu of Labor Econondcs. Princeton, New !ersey: Princeton University Press, 1962. Oaxaca, R.L. 'Male-female Wage Didferentia in Urban Labor Markets.' Inernadonal Economic Review, Vol. 14. no. 1 (1' 73). pp. 693-709. Roy, A.D. 'Some Thoughts on the Distribution of Earnings." Oford Economic Papers, Vol. 3 (1951). pp. 135-146. WU-ilis, R. and S. Rosen. "Education and Self-selection." Journal ofPolitical Economy, Vol. 87, no. 5, part 2 (19s9). pp. S7-S36. Woild Bank. 'Venezuela: A Cou'try Assessment on the Role of Women in Development. Mimeograph. Washington, ;. C.: Latin American and Canrbbean Regioa, World Bank. 1990. 461 21 Female Earnings, Labor Force Participat'lon and Discrimination in Venezuela, 1989 Caromlysi W er 1. Introduction In this chapter we try to determine (1) what factors are most ikely to influence female labor force participation snd (2) what factors accoiunt for existiiw male-femnale wage differentials in Venezuela. TMe 1989 house*old survey data .htw wor ing female montl:ly earnings to be approximately 78 percent of rWale earnings. Although this differential is not as large as that reported in many other Latin American co,intries, it is still substantial.' It is important to determine whether this di feremtial is the result of different endowmnents in productivity-related chracteristics between 'n'e sexes, sucb as education and work experience, or whether t is a consequcnce of labor market disaimination. If men and women in Venezuela are paid according to the satne wage structure differences in endowments shoula account for all the obstrved earnings differentials. If, however, we adjust for difflrences in endowments between the sexes and we continue to find a wage g-p, this can be int.i.;reted as evidence of wage disrimination between the sexes. Fellowing Osiaca's (1973) approach, we decompose sex-specific earnings into an 'ei-Iowment' componxt and a 'discrimination' component and attempt to estimdte the exten zo which wage differentials result from discrimination. The fbl!owing section i rovies z bief description of the Veraezueln labor market, its fluctuating fortunes since the end of flme 'oil boom,' and general factors affecting women's labor force participation. Section 3 describes the data base used in the analysis and sore basic features of the data. In Stction 4 we presen piobit estimates showing the determnats of women's labor force participation and inr Section 5 we consider earnings functions estimates for working males and fetnales and inciude corroctions for possibie selectiv.ty bias among women. Section 6 presents the estimate of the exten tJ which earnings differentials can be explained' ty discriutination. 2. The Setiung: .e Venezuean Economy and labor Market Tle discovery and widespread exploitation of oil meant that Venezuela changed rapidly from an 3griculture-based econoLny to one of the largest oil exporLers by mid-centry. Although tb_ oil industry itself has never been a large empluyer of iabor (in 1989 it employed oa:y 0.7 percent I Many of the other btudies reported in this volume (Preil and Peru, for exnmple), report om'sa eamings to be abou: two-hirds of mWL- Dirdsai & Fox (1985) repoul that female teaches in Brazil earn less thin 55 pemt of male teaher's eamuwzs. 463 -- -- ---- --~~ -- c,-- - t-e -.- r - - - ,Y AO "V W predominantly urban-based. In 1989 approximately 82 percent of the population lived in urban areas, principally in the northern industri?Ji; states. The rapidly growing population and the economic windfalls of the oil boom' during the sixties and seventies prompted the government to give priority to the expansion of education, particularly tertiary education where enrollments increased by 9.1 percent per anium between 1975 and 194 (Psacharopoulos and Steier, 1988). Improved access to schooling has especially benefitted female5 who now have, on average, 1.6 years more schooling than males. increased access to education has meant that women's labor force participation has increased significantly, from 22 percent in the 1970s, to 29 percent in 1982, and to 38 percent in 1989 (de Plai:chart, 1988; Psacharopoulos and Planm 1991). The proportion of women holding professional and technical jobs grew from 15.2 percent in 1961, to 22.1 perceat In 1987, and 21.1 percent in 1989 (de Planchart, 1988). In terms of tarnings, however, women continue to be .oncentrated in lower paying occupations. The proportion of women in the highest paying category, managerial occupations, has changed little over the part two decades (see Appendix table) and in professional occupations, women are predominantly found in the lower paying areas, such as nursing and teachiny Economic growth halted abruptly in 1979 with the end of the 'oil boom' and labor shortages were replaced by rising unemployment which peaked at 14 percent in early 1985, stabilized around 6.9 percent in 1988 and began to rise sharply again in 19Z). Workers in low-paid, low sklills jobs have been most immediately affected and womrn are often beavily represented among these groups. Women's labor force participation has also been affected by 'protective' labor legislation laws introduced in the seventies which inadvertently work to exclude women from certain sectors of the labor market. These laws prohibit employers from hiring women for 'physically and morally" dangerous work, for night werk, or in industries with numerous dailv shifts. It is also illegal for women to work in most occupations in the mining sector. In addition, legisla:-an stipulating generous materal leave privileges at full pay makes female labor potentially more costly to employers than male labor. Recorded incidents of discrimination against female workers are few, but there is evidence that emphlyers seek not to hire women and actively discriminate against hiring married women (Rakowski, 1985). Clauses supporting equal pay for equal work have really only been enforced in the public sector which possibly accxunts for the high proportion of women (more than twice as many women as men) employed in this sector. Venezuela's rapid population growth rate, averaging 3.5 percent in the previous two decades and 2.?, percent in the 1980s, means that 40 percent of the population is now under i5 years of age. To keep unemployment at its current levels, a real annual growth rate in the GDP of 5 percent would have to be achieved and maintained. This !s not expected (Economist Intelligence Unit, 1989). Changes in women's labor force participation and the extent to which discrimination affects their earDings will thus be a real concen in any poverty alleviation efforts; women are more heavily represented among lower inconme groups than men and the proportion of female headed households, dlready accounting for over 20 percent of all households, is continuing to increaF-. J. Jata LJrIdIcteISc The analysis is based on data from the 1989 Venezuela Household Survey conducted by the Oficina Central de Estadisticas e Informatica (OCEI). Such Household Surveys were conducted t-vice yearly between 1968 and 1983, and qvsrterly from 1984. The survey covers nine political idrninistrative regions, Caracas Metropolitr a area, Capital, Central, West-central, Zuliana, Los Andes, Southerr., Trn-easten, and Gv.,ana. Data from the Guyana region were collected independently in the 1989 survey and technical difficulties with the data prevented its inclusion in this analysis. Th available survey data included 159,818 individuJ observations from which a 10 percent random sample was drawn for use in this analysis. The survey provides detailed information on labor issues including employment status, weekly hours worked, occupation and industry category, and monthly income. Data on socio-economic characteristics such as age, educational attainment, marita; stat is, number of children and household size is also available. Labor par6cipation, as commonly defined, includes those 'mployed and those seeking employment. However, because a large informal sector exists ii Venezuela, it was difficult to identify individuals as unemployed or as being employed in the .brmal sector in the data base. Consequently, only actively employed individuals, identified by their positive responses to questions concerning employment status, weekly hours worked and monthly income, were defined as participating in the labor force. Individuals with incomplete or inconsistent labor market information were excluded from the sample. This included unpaid family workers and woikers who did not report hours worked. Tncome was reported erraticaliy by younger and older respondents. Consequently the sample was restricted to prime-age working males (20-460 years) and prime-age females (20-55 years). Within the samples of working males and females, individials who reported earning less than 10 percent of the mean hourly wage for their sex or more thin 5 times the mean hourly wage were excluded. Fifteen cases, reporting either extremely high or low earnings, were excluded in this way. This resulted in a sample of 2,408 working males and 3,143 females, of whom 1,181 were working in either the public or private sector. The proportion of employed men and women were 76 and 38 percent respectively. A proxy for labor force experience was cnstructed as age minus years of schoolinig mirus six years. This proxy measure will almost certainly overstate experience since no adjustments can be made for periodic abseaces from the labor force. Overestimates will be most severe for women since they are more likely to withdraw during childrearing. Table 21.1 gives means and standard deviations of the sample variables by gcnder. Working women earn approximately 78 percent of men's weekly earnings but, on average, work fewer hours per week (38.48 compared to 43.71 hours). After adjusting for differences in weekly hours worked, women's hourly earnings are 12 percent less than men's. The proxy measure for labor forc.e experience is lower for women. As in most Latin American countries, female wo;:kers in Venezuela have, on average, approximately one and one half years more schooling than male workers. This educational advantage holds true at all education levels beyond primary school, even at tertiary levels. Working women are also more likely than men to be studying whilo they are working. Mprried!cohabiting women are less likely to participate than married/cohabiting men. Table 2L1 Venenela - Means (and Standard Deviations) of Sample Variables Wonidng Woreing Non-working Variable descriptions Men Women Womea Age (year) 35.97 34.01 33.6 '10.68) (8.90) (10.25) Years of schooling 6.93 8.52 6.31 (4.14) (4.23) (3.85) Experience 23.05 19.55 (12.17) (10.8) Eauninge (weekly) 1518.23 1179.91 (1292.81) (773.07) Hours worked (per week) 43.71 38.48 (8.37) (9.73) Distribufion by Education (pescen*- No oduattion 0.08 0.05 0.11 (0.27) (0.21) (0.31) Incomplete primuy 0.19 0.11 0.19 (0.39) (0.31) (-.39) Primary 0.27 0.22 0.27 (D.44) (0.41) (0.44) lncomplete seondary 0.23 0.27 0.22 (0.42) (0.45) (0.41) Secondiry 0.11 0.16 0.09 (0.32) (0.37) (0.29) Seondary Whnical (.02 0.02 0.01 (0.13) (0.14) (0.10) lncomplee uriversty 0.04 0.09 0.07 (0.21) (0.289) (0.26) Universty 0.06 0.1 0.01 (0.23) (0.30) (0.08) CmuTatly a student 0.03 0.09 0.09 (0.18) (0.29) (0.29) Marit Sts(percent): Marriad (or cohabiting) 0.73 0.55 0.71 (0.44) (0.50) (0.45) Distnibution by Enmlovmnt- SW=u poeaent): Public Sctor 0.15 0.36 (D.36) (0.48) Private Sector 0.74 0.61 (D.44) (0.49) Number of Observatios 2408 1181 1962 a. Bolivar Notes: - Stundud deviations am gwen in puahadm - Sample nludes working maks agod 20 to 60 yen and working mad nonworking emnaes aged 20 to 5S years. - Female labor force p ripn rut- 38% - Male kbor force paip oo rade - 76% Source: Vcnezuela Household Survey. 1M. 4. Determinants of Female Libor Fore Participation Numerou3 factors influence a woman's decision to participate in the labor market - her investments& in human capital, personal characteristics such as her marital status and whether she has young children, and other factr, such as the availability of suitable childcare options. Ultimately, her decision to partic;na;e wfll rest upon the comparison of her market wage with the value of her tixreI in the howe (i.e., her reservation wage). This means that if we estimate earnings functions using data from our sample of working women, the sample will include only women whose market wage exceeds their reservation wage. Consequently, we will be estimating earnings functions for a self-selected sample of women. To r'oi., t for this we follow Heckman's (1979) widely adopted procedure and estimate a probit a, . .a for the fill sample ox women (working and non-working) in which the probability that woi_an will participate is estimated given various conditions, in this case whether she has young ependent children, her age, r?gion of residence and educational attainment. The dependent tariable in this model is a dummy variable for labor force participation (I if a participant and 0 if not). The inverse Mill's ratio (Lambda) is estimated in this equation and entered in the earnings equations to adjust for !he possible selectivity biai inherent in our sample of working women. The probit estimates are shown in Table 21.2. Table 21.3 estimates predicted participation rates for each cnaracteristic while the values of other characteristics is held at their sample mean. In line with the general human capital literaure, education is found to have a powerful effect on participation. The probit coefficients in Table 21.2 show that the probability of participaiing rises steadily with each successive level of education. The predicted probabilities in Table 21.3 makes this very evident A woman with mean values of all other characteristics and completed university education has a predicted probability of labor force participation 37 percentage points higher than a woman with completed secondary education (probability = .87 versus .50). Similarly, a woman with completed secondary education has a predicted probability of participation 21 percentage points higher dhan a woman who has only completed primary education (probability = .50 versus .29). The effects of age on participation are as expected, with women's probability of working increas;ng steadily from their mid-twenties and peaking between ages 41 and 45. Low participation rates among women in their early twenties are consistent with the high enrollment (44 percent) of women in this age group in higher education. It is widely posited that being the mother of young children (under 6 years of age) significantly increases the opportunity costs of women's labor force participation and increases the probability that they will withdraw from the labor force.2 Our estimates support this finding. Table 21.3 shows that a woman has a predicted probability of participation of .32 if she has young children and .41 if she does not. 2 See Behduan id Wolfe (1984) sad GO a (1988). Table 21.2 Probit Estimates for Female Participation Variable Patiad Variable Coefflicst t-rao Mean Derivaive Constant -1.054 -8.78 1.000 Age 20 to 25 -.052 -.47 .253 -.019 Age 26 to 30 .320 .81 .185 .120 Age 31 to 35 .434 3.78 .153 .164 Age 36 to 40 .448 3.93 .153 .168 Age 41 to 45 .495 4.12 .098 .186 Age 41 to 50 .208 1.67 .0&4 .078 Education Incomplete pimary .103 1.01 .160 .038 Primary .270 2.81 .254 .101 Incomplete seoondary .607 6.11 .240 .228 Secondary .814 7.31 .119 .306 Secondary Technical .2:9 1.31 .013 .097 Incomplee university .782 5.57 .078 .294 University 1.9q3 11.15 .041 .724 Children -.254 -8.01 .461 -.%5 Student -.284 -2.66 .091 -.107 Urban Residence .164 2.37 .818 .062 Notes: DpncVnix Var ble: Labor Fore Prticipaon Sample :Womcn aged 20 to 55 N: 3143 Log-Likihbood -1867.7 Mean Pazticipaion Rate 38% Many studies have shown that participadon rates are strongly affected by the woman's area of residence.' In Venezuela, urban residents have an estmated probability of participating 6 percentage points higher dtm rural residents. 5. Earnings Functions In estimating the earnings funcdt (Table 21.4) we ufilize a conventional human capital specification and specify the logrithm of the wage as a funcdon of years of schooling, years of experience and expeience squard. ITe experience proxy is entered as a squared term to test if the earnings function is parabolic in the experience tem. Eanings functions are estiat for males and the 1,181 working women.' Ihe inverse MfiUl's ratio, derived from the probit estmane, X See Bidsal and Fox (1985M) 2rn and WoUc (1964) and Khutam r in this volume. ' No correction is mad for sion bis in the mume wmple i= v tra labor fix= particiation as on exogenous variable. It is anumed da prime-e nuls do not hav the sae opioos regding labor force participation as do feales. Maie am trditionally viewed as provide for the family while females may have the opton of leaving de labor nurt to umdertake childreaing and homwereo acivities Ta"e 213 Predicted Pancicipation Probabilities by Chaderistic Characteristics Predicted Probability Eduction No education .21 Incomplete primary .24 Primary .29 Incomplete secondary .42 Secondary .50 SeczAdary technical .29 Incomplete university .49 University .87 Presence of Young C.iildren No .41 Yes .32 Stuient No .38 Yes .27 AMe Rural .32 Urban .38 Overall Mean Participation Rate .38 a. P noablB ofpuztion is reported for each coh io, holding other condioa ooeaaz at dow mcan vahus. is entered as a regressor to correct for simple selecdon bias. Is coefficient will pro-Ode an estimate of the covariance between the disurbances in the work/no work and wage equations. The rates of return to schooling are 9 percent for men which is comparable with the earlier findings of Psacharopoulos and Alam (1991). The rate of return for the 'corrected' and 'uncorrected estimates for womeare lOpercent and 11 percent, respectively. This shows that, had we omitted the selection term from the earnings eqation, the mariznal rate of return would have been biased upward. 'no log earmnigs Increase with expeience at a decreasing rate in accordance with the expectd ageearnings profies. It is impormat to remember that it is potential experience that is measured here, and that nost women have ierruptons in their careers. Hence, the experience variable is li}dy to be an overestima With respect to hours worked, the coefficient indicates that a one Table 21.4 Earnings Functions Men Women Women Variable Uncorfected (Corrected (Uncorrected for for Selectivity) Selectivity) Constant 3.91812 3.81234 3.51945 (20.692) (14.456) (17.993) Schooling .090507 .101094 .111062 (30.701) (13.237) (24.406) Ln Hours .541290 .545487 .554462 (11.127) (11.103) (11.307) Experience .034726 .023371 .028033 (9.705) (3.725) (5.078) Experience squared -.000389 -.000205 -.000283 (-5.925) (-1.575) (-2.376) Lambda -.137366 (-1.640) R1 .321 .397 .395 N 2,408 1,IS: 1,181 Notca: T-ratios arc in p sreee Dcpendan variable = log (wceky arings) percent increase in weekly hours worked is associated with just over a half percent rise in monthly earnings. Finally, the coefficient of the selectivity variable (the inverse Mill's ratio) is negative and significant only at the 10 percent level. Multiplying the coefficient of Lambda with its sample mean gives the average error term conditional on being in the labor force, which is about 12 percent. The negative and significant Lambda (at the 10 percent level) indicates that there is some correlation, although weak, between the unobserved characteristics that make women highly productive in the market and at home. Co -.paring the estimates for men and women, the coefficient on education is higher for females than men, indicating that additional schooling adds more to female than male earnirgs. lee returns to experience rise faster for males in their earlier working years than for women. Multiplying out the coefficients and sample means for eAperience and experience squared we find that male earnings peak at 43.7 years, while for females (using corrected data) they peak at 50 years. This difference may be partly explained by the fact that most women's labor force experience is interrupted by absences during childbearing. 6. Discrimination The standard Oaxaca (1973) decomposition pamits us to estimate what proportion of the male- female ear.iings differential is attributable to differences in observed characteristics (i.e., different human capital endowments) and that which is atributable to 'unexplained' factors, including discrimination. We write the difference in log weekly earninp of males and females as: B.Xr, - B,Xq = X+(b.W+bff-. (la) = X.(b,bi)+bAX.-Xg (lb) In both equations, the firt term is the part of the log earnings differential atributable to differences in the wage stuctures between the sexes and the second term is that part of the log emings differential aL-ibutable to differences in human capital endowments. An index number problem means that we can estimate discrimination in two ways. There is no reason to choose one method over the other, so we present resuwts for both. The fit expression la estmates disriminaton based on the suppwosition that women are paid on the same wage scale as men. In this case (see Table 21.5), differences in endowments account for 14 percent of wage differences and up to 86 percent of earnings differentials may be due to discrimination'. Table 215 Decompwition of the Wage DiffaeriaP Specification Differao due Diffece due Male PAy AdantaW to Eadowmca to unXplined fia 'Using expression la 14 (1.1) r6 (18.9) 10( (22) Usingexpression lb 5 (1.1) 9S (21.0) 20) (22) a. Meo of Working Wom Ouy - U Noses: Figurs in parcatbsm ame pcoeage showing the mac Fay ^adVa . (Wzn/W4-128.6%). Choosing the second expression lb, we find that only 5 percent of wage differences can be explained by differences In eadowments if all workers are paid as if they are females. As much as 95 percent of eaminp differenals are explained by discrimination if females have the same endowments as males. o Discussion The wage differential between men and women in Venezue!a is surpisingly low with working women earning, on average, 78 percent of men's wages. Tnis dhirendal is low even for induMriallzed nations (in Britain and Greece women eamr 74 and 73 percent of men's wages, respectively) and is among the lowest in La'in America. g It should be noed dat tis rXeamde 'upp bound' to discri i, i.e., that vaou fators odhs than discimination can acout for the wap diffemtiaL For instance, if we hav oautted variables fim tho oening equato Xt will bis the ad- of diseimination upwards. , Khanr pOls wom w 'es wagas beng about twc4hirda of mn's in Peu while Ng found wome's wages in Arguttun to be 65 percePt of mm's (both in ths volu). See also Oundao (1989). 472 Womn 's E)pIoyme and Pay In Loa Ai:fca 1°ere are two fictors that may pardy explain this low differential: First, women average more years of schooling than men and have significantly higher attendance rates at tertiary education - in 1989, 10 percent of women had tertiary education compared to 6 percent of men. Given -xising acute shortages of managerial personnel, skilled workers, and technicians in Venezuela, this must have provided women with some advantage in the labor market. Indeed, women in three industry grups (mining, construction and transport) earn more than men on average, being employed mosdy in higher skill occupations. Second, equal pay legislon enacted in the 1970s, although only enforced in the public sector, seems to have played a role in increasing women's wages. Cenainly, there is evidence that adherence to equal pay legislation by the public sector has attracted women employees; more than one third of all working womer were employed in this sector in 1989. Our estimates show only a small proportion of the earnings differential to be the result of differences in humarn capital endowments. Much of the earngs differential can thus be ascrbed to employer discrimination between the sexes.7 lbis discrimination may take various forms - women may be required to have higher levels of education and more experience than men to qualify for the same job, or they mav be paid lower wages for tie same work. Further studies are necessary to determine what forms discriminadon takes in Venezuela. This study suggestl dtat factors influencing women's participation in the labor force are also deserving of fiuther investigation. The probability that a woman will participate in the labor force is shown to decrease significantly if -she has children under six years of age. To date, litde consideration has been given to the provision of childcare facilities in Venezuela. Access to these services is likely to be important in enabling women to participate, particularly women in poorer areas who need to supplement household incomes. 7 We cannot, howv, dc= tdohea tha there may be ms ki advantae which hv not bee included in our ceimat. If this is the c our etimto of discrimination will bo biased upwad. Femae EaniAp, Labor Force ParzicpeAion and Disoimination in Venezuela, 1989 473 Appendi Table 21.i O puuional Charcte"ia of Employed Wom Percent in Percent of Avage Avesag Av. HKm Av. Ed. Occuqsioa1 occpation en*loyed mnthl y Age wo*od in yas Group who am woe in eaing per week womn tion (olivar) 1989 1989 1989 1989 82 87 89 82 87 89 hokiona 55 54 67.2 19.8 22.1 22.1 7161.81 34.5 35.7 12.2 Mnageral 10 14.2 14.7 1.6 2 1.4 9379.41 36.4 46.1 10.6 Ofric= 55 61 64.7 23.1 20.7 19.5 5015.16 30.8 39.8 9.8 Sake 29 30 34 13.7 14.8 13.5 4624.2 35.1 38.3 7.1 PFamcr 3 4 4.2 1.6 1.8 1.8 3454.52 39.8 37.6 2.2 Transport 2 .7 1.6 .6 .4 .4 5978 28.4 40.2 7.8 Crab 13 14 14.7 11.6 11.9 9.7 4009.99 343 36.8 6.2 Scrvice 57 54 41.1 27 26.1 5.7 3226.35 35.1 40.1 5.4 Miner 0 n/a n/a 0 n/an/a n/a n/a noa n/a Sources: 1982 and 1987 from Perez de Plcbast, Unted Na&OCJ IwrrinaWl Seminar, ScpL 1988. 1989 fom 02oC Houabold Survey, 1989. References Birdsal1, N. and N. Fox. OWby Males Earn More; Locadon and Training of Brazilian Scioolteachers." Economc Developmem and Culturc2 Qange, Vol. 33, no. 3 (198). pp. 533-556. Behrman, J. R. and B. L. Wolfe. 'Jabor Force Participation and Earnings D in for Women in the Special Conditions of Developing Countries.' Journal of DewJopmem Economics, Vol. 15 (1934). pp. 259-288. Ganniott, K. 9Women, Wages and Discrimination: Some Evidence from Taiwan.= Economic Dswlopment and Cultural Ozange, Vol. 39, no.4 (19' 6) pp. 721-730. Gronau, R. 'Sex-Related Wage Diffirentials and Women's Interrupted Labor Careers: The Chicken and Egg Question.' Journal of Labor Economics, Vol.6, no. 3 (1988). pp. 277- 301. Gunderson, M. 'Male-female Wage Differentials and Policy Responses.9 Journal of Economnic Literatre, Vol. 21, no. 1 (1989). pp. 46-72. Heckman, JJ. 'Sample Selection Bias as a Specification Error.0 Econowica, Vol. 47, no. 1 (1979).pp. 53-161. Xhandker, S.R. 'Labor Market Participation, Retrns to Education, and Male-Female Wage Diffe.ences in Peru.' In this volume. ight, J.B. and R.H. SabotL -LU%or Market Discrimination in a Poor Urban Economy.' 7he Journal of Dewlopment Studies, Vol. 19. no. 1 (1982). pp. 67-87. McCoy, I.L. ole Politics of Adjustment: Labor and the Debt Crisis.' Joumnal of IntAmerim Studies and World Affalr, Vol. 28 (196/87). pp. 103-38. Oaxaca, R.L 9Sex Earning Differentials9 in G. Psacharopoulos (ed.). Economics of Educarlon Research and Sndis. New York: Pergamon Press, 1987. Oaxaca, R.L. 'Male-female Wage Differendals in Urban Labor Markets. International Ecnomic Review, Vol. 14, no.1 (1973). pp. 693-709. Perez do Planchart, M.C. Case Study of Vcneela., Women and the Economic Crsi. Vienn: United National Interreglonal Seminar on Women and the Economic Crisis, 1988. 474 rami±, rAmump, LAOr rorDM rarucqpwn and L 1icnmmaz m VcCA, 1YU9 475 Psacharopoulos, G. and A. Alam. 'Earnings and Education in Venezuela: .4n Update from the 1987 Household Survey.' Economics of Education Review, Vol. 10, No. 1 (1991). Psachxiropoulos, G. and F. Steier. 'Education and the Labor Market in Venezuela, 1975-1984.' Economics of Educaton Review, iol. 7, no. 3 (1988). pp. 321-332. Psacharopoulos, 0. ari E. Velez. 'Does Training Pay Independent of Education? Some Evidence from Colombia.' Inmernadonal Journal of Educ4tional Research, forthcoming, 1991. Rakowski, C.A. Women in NontradJional Industry: 7he Case of Steel in Cudad Guayana, Venezuela. Working Paper No. 104. Michigan: Michigan State University, 1985. Rakowskli, C.A. Production and Reproduction In a Planned Industrl Cary: The Working- and Lower-aass Households of a d Guayana, Veneula. Working Paper No. 61. Michigan: Michigan State University, 1984. Ihe Economist Intelligence UniL Venzla to 1993. A O&znge In Direcion? Special Report No. 2003. London: The Economist Intelligence Unit, 1)A9. United Nations. F7w Studies on the Siuaion of Women in Latin America. Santiago: United Nations,1983. Urdaneta, L. Partlcipadon Economka de la MuJer YLa Dttribudon del Ingreso. 1986. Caracas: Banco Central de Venezuela, 1986. Appendix A Contents of Companion Volume Ackrowledgments Foreword 1 Introduction and Summary 2 Trends and Patterns in Female L3bor Force Participation 1950-1985 37 3 The Indus;rial and Occupational Distribution of Female Empioyment 71 4 Potential Gains from the Elimin3tion of Labor Mark-et Differentials 135 5 Gender Differences in the Labor M. rket: Analytical Issues 151 6 Sumnmary of Empirical Findings and LDrplicaions 183 Appendix A: Contcnts of Companion Volume Appendix B: Authors of Country Case Studies 217 References 219 477 ______ _____ Appendix B The Autbora Mhry Aramih is g Consultant for the World BknE's Tn America wd C an Techn'cal Departinant, Human Reources Division. Jon A. B.rlaw Is AssocUte Profe5sor in the Departme of Econoidcs, Concordla Univsity, Monteal. ornald Cov is Associa Profesor of Economics, Ecinomics Deparmwt, Boston College at Ciesnue H'11, Massachusem. .ndamhit Gil is Assistznt - w 'or the Schnol of Mag mnen e Sitte Univhy of NeV York at Bu:tilo. og h yo TA. Gludling is Assismt Professor in the Department of Eccnmics, Unnvershy of Maryland, laltimre Couty. Gewwg Jakubs is Associae wrofessr in die Schol of Indusrial Lsbor Rdatbns at Coru% UnIvershy. Shahidtw Khandziv is a flsexrch Economist in the Women in Development Dision, Population -n Human Reources D , e World Bank. re4rry Magrs te is afocied with IN2A, ESR Paris, Frace and dhe DWlrtmewt of Economics, Uaivers.y College of LoaCwn, LTnite Kingdow. Gem.es Monette is Associate P' fessor in die Depatme of Miath.acs, York Univesity, Tomwno. Yin Chu Ng is Assist Pcfesswr a Hong Ymg Baptist College. George PsacharopouIos is Senior Humsa Resooe Advisor, Technical Depwrwant, Latin America and the Canbbean Region. Ts World Bankl te;.! Sctt ha a Ccnsultant for the World Bank's Latin America and Carmen TectIcal - mmet HRn= Rscurces Diviion. J. ram Smith is Associate Proessor in ho Deparnt of E0comics, '(ork Univmkl, Toronto. MortGn Stalner is a Professor in the Department of Economics at Concordia Uniersity, Montreal. lailme Tenifo is Assistant Professor in the Department of Management and Eccaomics at the University of Toronto, Scarborough Campus. 1ill ITefenthaler is Assistant Professor in the Department of Economics at Colgate University. Zafirls Tzannstos is a Labor Economist in the Education and Employment Diviclon, Population and Human Resources Department of the World Bank. Eduardo VeIez is an Education Specialist in the Human Reso urces Division of the Latin Americ3n and Caribbean Region of the World Bank. Carolyn Winter is a Human Resources Specialist in the Women and Developmwt Division, Population and Human Resources Department of the World Bank. Horngyu Yang !s a Consultant in the Human Resources Division of zle L atin American and Caribbean Region of the World Bank. World Bank Regional and Sectoral Studies Non,governincn itl Organizations and the WVorld Bank: Cooperation for Development, cdited bv Samuel Paul and Arturo Israel Unnfair Advantage: Labor Markct Discrim,;ination in Dcveloping Countrie!, editcd by Nancy Birdsall and Richard Sabot Education in Asia: A Comparativc Stidy of Cost and Finan icinq, Jec-Peng Tan and Alain Mlingat Healtlh Cain in Asia: A Comparative Study of Cost and Financing, Charles C. Griffin Bolivia's A nsiver to Povcrty, Economic Crisis, and Adjtustncnt: 77Tc Emnernc)vcn Social Ftn^d, edited by Steen Jorgenscin, Margaret Grosh, and Mark Schacter Crop-Li,ctock Interaction in Stub-Sal,aran Africa, Johlin Mclintire, Daniel Bourzat, and Prabhu Pingali Commodity Price Strbilizationi and Policy Rcformn: An Approach to thjc Ealuahtion of the Brazilian Price Ban d Proposals, Avishay Braverman and othcrs Thle Transition fromn. Socialism in Easterni Eusropc: Domcstic Rcstruicturit,ni and Foreign Trade, edited by Aryc L. Hillman and Branko Milaaovic Wo,nen 's Employencti and Pay in Latin Amncrica: O;. iviecn, anld Methodology, Gccrge Psacharopoulos and Zafiris Tzannatos