What Explains the Stagnation of Female Labor Force Participation in Urban India? Stephan Klasen and Janneke Pieters* Female labor force participation rates in urban India between 1987 and 2011 are sur- prisingly low and have stagnated since the late 1980s. Despite rising growth, fertility decline, and rising wage and education levels, married women’s labor force participa- tion hovered around 18 percent. Analysis of five large cross-sectional micro surveys shows that a combination of supply and demand effects have contributed to this stagna- tion. The main supply side factors are rising household incomes and husband’s educa- tion as well as the falling selectivity of highly educated women. On the demand side, the sectors that draw in female workers have expanded least, so that changes in the sectoral structure of employment alone would have actually led to declining participation rates. JEL codes: J20, J16, I25, O15 India’s economy has grown rapidly over the past two decades, with the services sector accounting for a large share of growth (Bosworth and Collins 2008). India has also experienced a sizable fertility decline, a rapid education expansion, and a decline in the education gender gap, while the labor market returns to educa- tion increased (Kijima 2006; Pieters 2010). Against this background, it is puz- zling to see that the reported female labor force participation rate in urban India has stagnated at around 18 percent since the 1980s. Considering these circum- stances, one would expect a rising share of women to enter the labor force, espe- cially in urban India where women have gotten much more educated and where white-collar jobs are concentrated. The aim of this paper is to investigate why female labor force participation (FLFP) in urban India stagnated despite rising education levels and rapid *Stephan Klasen is Professor at the University of Go ¨ ttingen and Research Fellow at IZA; his email address is sklasen@uni-goettingen.de. Janneke Pieters (corresponding author) is Assistant Professor at Wageningen University and Research Fellow at IZA; her email address is janneke.pieters@wur.nl. This paper has benefitted from the comments of Sonia Bhalotra, Francine Blau, Tanika Chakraborty, Lawrence Kahn, Stefan Klonner, David Lam, Ghazala Mansuri, Vijayendra Rao, Sher Verick, two anonymous referees, and the editor of this journal. We also had helpful comments from seminar participants at the Delhi School of Economics, the Harvard Center for Population and Development Studies, The New School, IZA, Passau, the World Bank, the IZA/World Bank Conference on Employment and Development, and the annual conference of the European Society for Population Economics. Two supplemental appendixes to this article are available at http://wber.oxfordjournals.org/. THE WORLD BANK ECONOMIC REVIEW, VOL. 29, NO. 3, pp. 449– 478 doi:10.1093/wber/lhv003 Advance Access Publication March 18, 2015 # The Author 2015. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. 449 450 THE WORLD BANK ECONOMIC REVIEW economic growth. Standard labor supply models and previous research on FLFP and economic development (Goldin 1994; Mammen and Paxson 2000; Blau and Kahn 2007) suggest that rising education and growth of white-collar services em- ployment should draw more women into the labor force by increasing their earn- ings capacity and reducing social stigma against women’s work. On the other hand, rising household incomes could lead to a withdrawal of women from the labor market due to well-known income effect. India is not the only country where female labor force participation has stag- nated despite overall growth, fertility decline, and rising female education. In the Middle East and North Africa, one can observe vast and persistent gender gaps in employment despite rising female education levels in many countries. This has been partly ascribed to conservative social attitudes towards women’s work in the region (see Gaddis and Klasen 2014), and similar factors might be at play in India as well. Understanding the causes of stagnation in FLFP matters for several reasons. India currently has an advantageous age structure of the population with a large and growing share of working age people and relatively few dependents. Optimistic predictions for India’s future growth often refer to this demographic dividend, which is alleged to have accounted for about a third of East Asia’s high per capita growth rates in the period between 1965 and 1990 (Bloom and Williamson 1998; Bloom 2011). However, the benefits of a country’s demographic dividend hinge on the productive employment of the male and female working-age population. In fact, high and rising female employment levels contribute to greater productivity growth (World Bank 2011) and have in fact been critical in sustaining East Asia’s high economic growth rates (Klasen and Lamanna 2009; Young 1995). Beyond women’s contribution to growth, stagnation in FLFP has implications for the degree to which women benefit from growth. Employment and earnings are robust determinants of bargaining power, with impacts on female and child- ren’s well-being (Qian 2008; Anderson and Eswaran 2009; Afridi et al. 2012). If there are structural economic or cultural barriers preventing women’s labor force participation, women are unable to capitalize on these opportunities. In this paper, we estimate a simple model of female labor force participation using individual level cross-section data spanning the period 1987 to 2011. The model is estimated separately for each survey year and also for women with low and high educational attainment. Our estimation results provide a detailed account of the impact of various factors on women’s labor force participation and their changes over time. We find a strong conditional relationship between educa- tion and labor force participation that is U-shaped, which points at the importance of social stigma for women with some education to work in low-skilled jobs. Accordingly, we show that women with low education appear to be boxed in by the necessity to work if household incomes are very low or insecure, and stigmas attached to working in low-skilled jobs if they are somewhat more educated and in more secure economic environments. Highly educated women appear generally less constrained by family circumstances in their labor force participation decision. Klasen and Pieters 451 The estimates are further used to decompose the stagnation of FLFP between 1987 and 2011 into contributions by different covariates and changes in behav- ior and unobservables (Fairlie 2006). On the supply side, we find that rising male incomes and education contributed to a withdrawal of women from the labor force, showing that the classic income effect is at work in urban India. On the demand side, changes in the sectoral structure of employment account for a further reduction in FLFP, in particular related to the declining shares of agricul- ture and manufacturing which tend to employ more women ( particularly in recent years). The effect of rising female education on female labor force participation is more complicated. Besides a U-shaped relationship between education and labor force participation, our estimates show a large decline in the positive participa- tion effect associated with secondary and graduate education. As a result, the substantial increase in educational attainment of women contributed only mod- erately to FLFP growth. We provide suggestive evidence that the declining posi- tive effect of higher education is partly accounted for by an erosion of positive selection into higher education, that is, a declining correlation between determi- nants of higher educational attainment and unobserved determinants of labor force participation. Reasons for this could be the rapid expansion of education supply, but also rising marriage market returns to education, leading women to pursue higher education regardless of their expected labor market attachment. The paper is organized as follows. Section I discusses the literature on female labor force participation determinants, focusing in particular on economic devel- opment and rising education levels. Section II describes patterns of FLFP, wages, education, and employment in urban India. Section III presents our empirical FLFP model and estimation results, followed by the decomposition analysis in Section IV. Section V further investigates the relationship between women’s edu- cation and labor force participation. Section VI concludes. I . D E V E LO P M E N T, E D U C AT I O N , A N D F E M A L E L A B O R F O R C E PART ICIPATION Labor force participation decisions can be the outcome of individual preferences of the woman, her family circumstances, as well as labor demand conditions for jobs that women are particularly suited for, or where employment in these jobs is seen as socially acceptable. Education can play a key role in shaping these supply and demand conditions. We will discuss these issues in turn. A common starting point for the analysis of female labor force participation is the basic static labor supply model (see Blundell and MaCurdy 1999), in which an increase in the wage rate reduces demand for leisure as its opportunity cost rises, increasing labor supply. If leisure is a normal good, an increase in a person’s income will increase the demand for leisure and thus reduce labor supply. These are the well-known substitution and income effects. For a person currently not working, an increase in the wage rate only has a substitution effect, 452 THE WORLD BANK ECONOMIC REVIEW increasing her incentive to work (i.e., one would always expect a positive own wage effect at the extensive margin). An increase in unearned income (nonlabor income or labor income earned by other household members, particularly the husband) constitutes a pure income effect and therefore reduces labor force participation. In initial stages of economic development, education levels typically increase much more for men than for women. Women’s wages and opportunities for work change relatively slowly while their husband’s income rises fast, so the neg- ative income effect is likely to dominate any positive substitution effect of rising female wages. This is what drives reductions in FLFP according to the so-called Feminization-U hypothesis (Boserup 1970; Goldin 1994; Mammen and Paxson 2000; Gaddis and Klasen 2014).1 Participation is further reduced because of social stigma against women working outside of the home, especially in factory work, and the difficulty of combining household production with market work in nonagricultural occupations; these effects are held to be particularly strong for married women. In later stages of development, women’s education starts to catch up to men’s, their earnings capacity increases and they gain access to socially acceptable types of work, especially if demand for white-collar workers increases with the expan- sion of the services sector. This will result in higher FLFP,2 but country-specific labor demand conditions clearly play a role in this process. The increase in FLFP could depend on growth in employment opportunities of the kind deemed appro- priate for educated women, relative to growth in the educated working age popu- lation. Boserup (1970) describes how the feminization of clerical jobs proceeds very slowly when the number of educated men is in excess of demand for clerical workers. In those cases, there is likely to be considerable resistance against women’s employment in white-collar jobs, as this would reduce the opportuni- ties for men (Boserup 1970: chapter 7). How the education-labor force participa- tion link evolves over time will thus depend on the structure of labor demand growth in the economy and the status associated with different types of work.3 If female labor mobility is limited, as is the case in India, the growth in desirable jobs relative to the educated population can generate local mismatches with impacts on female labor force participation rates. 1. Though the feminization-U is sometimes considered a stylized fact, the empirical evidence in support of it is mostly based on cross-country analysis, while panel analyses have produced mixed results (Tam 2011; Gaddis and Klasen 2014). 2. Over the course of development, changes in circumstances may also be accompanied by changes in women’s behavior, i.e., the degree to which wages, income, and social restrictions affect FLFP. As Goldin (1990) describes the history of women’s work in the United States, economic development is reflected in an increasing own wage effect while women’s responsiveness to other family income declines. Blau and Kahn (2007) and Heim (2007) find similar evidence for women in the United States. 3. A gender assessment for Pakistan (World Bank 2005) shows that the increase in urban female labor force participation in the 1990s was driven by an increased demand for teachers. But overall, still a much higher proportion of urban men than women are engaged in white-collar jobs. Klasen and Pieters 453 One might further hypothesize that similar factors produce a U-shaped rela- tionship between economic or educational status and women’s labor force par- ticipation at a given point in time within a country—as is indeed observed in India. Among the poorest with no or very little education, women are forced to work to survive, while among the very highly educated, high wages induce women to work and stigmas militating against female employment in white- collar jobs that are open to highly educated women are low. Between these two groups, women face barriers to labor force participation related to both the absence of an urgent need to work (the income effect), and the presence of social stigmas associated with female employment in menial jobs. A correlation between education levels and labor force participation can also appear when both are outcomes of unobserved preferences for work (e.g., related to family background), such that women with a greater taste for work are more likely to attain higher education. Recent research has shown that primary and secondary school enrollment in India respond to the perceived returns to schooling, in particular, the availability and awareness of job opportunities in business and IT-services (Jensen 2012; Shastry 2012; Oster and Millett Steinberg 2013). These studies also show, however, that responses are limited to very local opportunities and that girls’ schooling is affected by active recruit- ment rather than the mere availability of jobs (Jensen 2012). It apparently takes more than growth and rising wages to raise awareness of labor market opportu- nities, and despite its fast growth, the business services sector still accounts for only a small share of total employment in India. Nonetheless, when analyzing the effect of education on labor force participation it is important to keep in mind the potential endogeneity of education through nonrandom selection into education. Education could be endogenous to labor force participation in the exact opposite direction as well. In India, social restrictions on the lifestyles of women tend to become more rigid as households move up in the caste hierarchy (Chen and Dre ` ze 1992). If education of women and restrictions on women’s mobility and work both increase with families’ social status, one would observe a negative correlation between education and labor force participation, at least for some levels of education. Eswaran et al. (2013) find supporting evidence for this nega- tive endogeneity channel in rural India (based on data for 1998–99), but Das and Desai (2003) find no support in a sample of rural and urban women in India in 1993–94. In all, major determinants of women’s labor force participation over the course of economic development are income, wages, and access to jobs deemed appropriate for women. Education levels shape female labor force participation partly through these channels. But women’s education is likely to reflect other, unobserved determinants as well. We analyze the role of education and other determinants in detail in our econometric analysis in Section III. First, Section II gives a descriptive analysis of the most important trends in the data. 454 THE WORLD BANK ECONOMIC REVIEW I I . F E M A L E L A B O R FO R C E PA R T I C I PAT I O N IN URBAN INDIA The descriptions and empirical analysis in this paper are based on the so-called thick waves of the NSS Employment and Unemployment Survey, in 1987 – 88, 1993 – 94, 1999 – 2000, 2004 – 05, 2009 – 10, and 2011 – 12 (henceforth 1987, 1993, 1999, 2004, 2009, and 2011). This cross-sectional survey is the official source of nationally representative employment and earnings data used by the Government of India. Before discussing labor market developments in urban India in detail, Table S1.1 in the online appendix S1 (supplemental material is available at http:// wber.oxfordjournals.org/) summarizes the trends in the female population and labor force for urban and rural India, for different age groups. Female population growth in the age group 15–64, and especially 25–54, exceeds total female popu- lation growth. Due to this demographic change, the female labor force has almost doubled in urban India and stayed constant in rural India, despite stagnating female labor force participation rates in urban India, and declining rates in rural India. In that sense, India is already benefitting from its demographic dividend related to a favorable age structure (although to a much lower extent than it would if more women were working). As still more than two-thirds of the Indian popu- lation lives in rural areas, the rural trend of declining participation rates dwarfs the urban trend, so that overall female labor force participation declined from 22 percent in 1987 to 17 percent in 2011. Note, however, that rural numbers are more likely to suffer from underreporting of women’s work in agricultural activities. In our analysis we focus on married women in urban India in the age group 25–54. Their labor force participation rate fell slightly from 18.5 percent in 1987 to 17.9 percent in 2011. Breaking down the labor force into different com- ponents, one can see in Figure 1 there has been little change in the different types of work and unemployment rates of married women, except for a peak in self- employment and unpaid family work in 2004.4 Throughout the period, the labor force participation rate of urban married men in the same age group was stable at around 97 percent and hardly differs by education levels. Marriage is almost universal in India, with average age at marriage around 19 in urban India in 2004–05 (Desai et al. 2010). According to the NSS survey data, the share of ever-married women among urban women age 25 –54 was 95.3 percent in 2011 and 96.5 percent in 1987. Because in the age group 20–24 the marriage rate declined over time as more women pursue higher education and postpone marriage, we focus our analysis on women age 25 and older. 4. We always use the sampling weights provided. There is some question whether differences in sampling strategy between rounds might have a (slight) impact on the comparability of levels in the female labor force participation shown here, particularly also the peak reported in 2004–05. As discussed above, however, they will not affect the observation of low levels, and the stability over time. For a discussion of these issues, see Klonner and Oldiges (2014). Klasen and Pieters 455 F I G U R E 1 . Urban Female Labor Force Participation Rate Notes: Married women age 25– 54. Self-employment includes employers and own account workers. Unpaid refers to unpaid family workers. Regular employees receive salary or wages on a regular basis. Casual workers receive a wage according to the terms of the daily or periodic work contract. Source: NSS Employment and Unemployment Survey. We also exclude the 2 percent of all married women in this age group who report being head of their household.5 Female participation rates are calculated using women’s reported usual status, which refers to a reference period of 1 year in which the principal activity is the activity in which the respondent spent the majority of time. Subsidiary activity status is recorded as well but is not taken into account in our analysis, as it affects less than 5 percent of the adult urban female population and its definition is not consistent over time.6 Nonetheless, the pattern of female labor force partic- ipation is similar across different age groups, when including unmarried women, and when including labor force participation in both principal and subsidiary activities (see Table S1.1 in the supplemental appendix). One might worry that even though unpaid family workers and own account workers are considered part of the labor force, women’s work is underreported. Survey respondents may be reluctant to report women’s contributions to family businesses or may not consider a woman’s work to be different from her general domestic duties. This type of underreporting will mainly affect participation rates in rural areas, where women spent much more time on farm activities that are less likely to be considered as work, and will affect subsidiary status activities more than principal status activities because the former includes work done for only a few hours per day or during peak season only, etc. Principal status 5. We exclude them because we are interested in the role of women’s own versus household head education, among others. Participation rates of female household heads are higher, and declined from 39 percent in 1987 to 30 percent in 2011. 6. Before the 2004 wave, there was no lower bound on the number of hours spent on a particular activity to be considered as subsidiary activity, but in 2004 the minimum was set at 30 days of the reference year. 456 THE WORLD BANK ECONOMIC REVIEW participation rates in urban India are arguably least affected by underreporting of women’s work but have the disadvantage that women working part time are not considered as active in the labor force if they spend the majority of their time in domestic duties. This is important to keep in mind, but as noted above this affects at most 5 percent of the sample and does not affect time trends. To provide some verification of participation rates from the NSS Employment and Unemployment Survey using other data, we compare the 2004-05 numbers to the 2004-05 wave of the India Human Development Survey (Desai et al. 2009). There is no major time criterion to be considered as a worker in the IHDS survey, and there is considerably more probing as respondents are asked to specify each household member’s contribution to each family business as well as any other ac- tivities earning an income or a wage.7 For married women in urban India in the age group 25–54, the IHDS data show an employment rate (unemployment is not recorded) of 19.8 percent, which is very close to 19.6 percent based on the NSS data. The participation rate for married women in urban Delhi is also very similar between the NSS 2004–05 wave (19.4 percent) and a survey done for a study on women’s work in 2006 (19 percent; see Sudarshan and Bhattacharya 2009). Figure 2 shows that the urban FLFP rate has a U-shaped relationship with educa- tion and that the stagnation in FLFP hides a combination of rising participation among women with low education and a decline in participation rates of highly edu- cated women. As income, wages, and access to different types of jobs are important candidates for explaining these patterns, we now turn to a brief discussion of those. In line with India’s high growth rates, earnings data form the NSS surveys show that real wages roughly doubled between 1987 and 2011 (see Fig. S1.1 in the supplemental appendix). In absolute terms, real wages increased almost equally for men and women, but the ratio of male to female average weekly earn- ings declined from 1.6 in 1987 to 1.3 in 2011. Given the very high participation rates of married men in this age group and their higher average wages, one can safely assume that most women in urban India are secondary earners. Rising earnings of men most likely had a strong negative impact on female labor force participation. But with women’s wages increasing more than men’s, positive sub- stitution effects could at least partly offset the negative impact of men’s income. However, patterns in the data suggest there is no close link between women’s wages and their labor force participation. Figure 3 shows women’s real wages by education level. Returns to secondary and graduate education are high and rising, especially returns to graduate education. Real wages at lower levels of schooling, on the other hand, show a strong convergence due to relatively fast wage growth of illiterate women. If anything, own wage effects would thus have led to increases in labor force participation rates at both the ends of the edu- cational distribution—for illiterate women on the one side and highly educated women on the other side. Going back to Figure 2, however, we see that these are exactly the groups of women for whom participation rates declined. 7. See http://ihds.umd.edu/questionnaires.html for the IHDS household questionnaires. Klasen and Pieters 457 F I G U R E 2. Urban Female Labor Force Participation Rate by Education Level Notes: Married women age 25 – 54. Education is the highest level completed, see Table S1.2 (supplemental appendix) for definitions. Source: NSS Employment and Unemployment Survey. F I G U R E 3 . Log Real Wages for Women Age 25 –54 by Education Level, Urban India Notes: Wages are average total weekly earnings for casual and regular employees. Earnings are spatially deflated and in 1987– 88 Rupees, based on the Labour Bureau Consumer Price Index for Industrial Workers and Deaton (2003). Source: NSS Employment and Unemployment Survey. Wages may have little impact on FLFP, despite rising returns to higher educa- tion, if employment growth in the activities appropriate for educated women is limited. As shown in Figure 4, the distribution of female workers across indus- tries varies substantially with education. The distribution is shown separately for women below secondary education and those with secondary or higher educa- tion. Women with lower education work mainly in agriculture, manufacturing, and domestic services (included in “other services”). Access to white-collar ser- vices sector jobs is mostly confined to women with at least secondary education, but declined steeply over time. By and large, this pattern is consistent with the 458 THE WORLD BANK ECONOMIC REVIEW F I G U R E 4 . Distribution of Urban Female Workforce across Industries, by Education Notes: Distribution of female workers across industries, including employees, self-employed, and unpaid family workers. Shares are in percentage of all female workers in the respective education group. Low education is below secondary schooling; high education is secondary or higher. Source: NSS Employment and Unemployment Survey. U-shape in education suggesting that only with at least secondary education, women gain access to jobs that are not subject to social stigma. It could also account for the declining participation rates of highly educated women whose supply increased much faster than available jobs, thereby reducing employment opportunities for them in these sectors. In 1987 the vast majority of highly educated women worked in public admin- istration, education, and health. This share declined to 50 percent in 2011, mainly driven by a declining share of public administration. Although financial and business services have become more important, they still account for only a small fraction of female employment. Consequently, highly educated women are increasingly working in industries such as manufacturing (mostly in textiles), wholesale and retail trade, and domestic services; within these sectors, many highly educated women work in professional and administrative occupations so that there are still white-collar occupations.8 For low-educated women, 8. Due to changes in the occupational classification used in the NSS data, we cannot construct a consistent definition of white-collar occupations over time. We do find that in all sectors, highly educated women are more likely to work in professional and administrative occupations than women with less than secondary education. Klasen and Pieters 459 employment has shifted from agriculture into food and tobacco, textiles, con- struction, and domestic services.9 The changing industrial distribution of workers is consistent with Boserup’s (1970) description of white-collar jobs becoming increasingly scarce when edu- cation levels grow rapidly. The distribution of male workers (not shown) con- firms that the share of white-collar services employment has declined not just for women but for the entire labor force, while employment growth has been con- centrated largely in construction and retail and wholesale trade. At the same time, educational attainment has indeed grown rapidly. The share of women with at least secondary education grew from 21 percent in 1987 to 45 percent in 2011. Among men, this share increased from 38 to 56 percent over the same period (for more detail, see Fig. S1.2 in the supplemental appendix). Besides rising incomes, this growing supply of highly educated workers combined with employment shifting toward less skill-intensive sectors could be an important reason why participation rates among highly educated women have declined. I I I . E S T I M AT I N G THE DETERMINANTS OF WOMEN’S LABOR FORCE PART ICIPATION Using the NSS survey data, we test how the different factors discussed above have contributed to the stagnation of FLFP in urban India. We first estimate the effect of education, income, and other variables on women’s labor force participation in a reduced form labor supply model. In the next section, a decomposition analysis is used to show how changes in the explanatory variables and changes in coefficients and unobservables contributed to the stagnation of FLFP between 1987 and 2011. The probability of woman i in year t (1987, 1999, 2004, 2009, and 2011)10 being in the labor force (including self-employment, unpaid family work, regular and casual wage employment, and unemployment) is modeled as ! X Pit ¼ F ast þ bE E t Dit þ bXt Xit þ bZt Zit ; ð1Þ E where F is the standard normal cumulative distribution function. The model is estimated separately for each year to allow for changes in behavior over time. The first right-hand side term is a state fixed effect. Education is measured through dummies for the highest education level completed, D E (E ¼ 2, . . ., 6), with illiterate (E ¼1) as the reference level. Xit is a vector of explanatory variables at 9. Domestic services could be considered appropriate for women, but they are typically not covered by existing legislation and are relatively easy victims of exploitation due to their invisibility, lack of education and, often, migration background (NCEUS 2007). 10. Data for 1993 are not used in the econometric analysis because the 1993 data do not contain district identifiers, which are needed to construct district-level explanatory variables. Districts are administrative units at the substate level in India. Our sample covers 362 districts across India’s 18 main states. 460 THE WORLD BANK ECONOMIC REVIEW the individual and household level, including household income excluding her own earnings, and education of the household head. Income is measured as total house- hold earnings in the reference week excluding the woman’s own earnings.11 We use income per capita to control for differences in the number of people depending on that income.12 Instead of an asset index or similar measure of wealth, the education level of the household head is included to capture household wealth or permanent income beyond total earnings. If higher socioeconomic status leads to more restric- tions on women and if greater wealth reduces the need for women to work, the edu- cation level of the household head should have a strong negative effect on participation. We also control for the security of household income through an indi- cator for having at least one male household member with salaried employment.13 Further individual controls are age, age squared, number of children, and social group (indicators for scheduled caste and tribe–SCTS–, non-SCST Hindu, Muslim, and other). These caste and religion dummies are included to capture direct impacts of culturally or religiously determined restrictions on women, which are expected to be strongest among Muslim and high-caste Hindu households (Chen and Dre ` ze 1992; Das and Desai 2003). Zit is a vector of local labor demand and supply variables, included to capture the effect of the availability of suitable or attractive jobs. We include the district share of male workers in agriculture14, industry, construction, white-collar ser- vices (financial and business services, public administration, education, health, and social work), and other services; and the share of the district working age population with a graduate degree, to control for the local supply of high-skilled labor. We expect that female participation is higher in districts that are relatively specialized in white-collar services. The relative supply of graduates in the district is expected to depress participation rates through a crowding-out effect. These factors are expected to be particularly important for highly educated women. Standard labor supply models would also include the woman’s own wage, but identification of own-wage effects is challenging. To estimate the effect of wages on labor force participation it is necessary to use predicted wages for workers and non-workers, corrected for selection into employment and predicted based on at least one exogenous variable (for a discussion see Heim 2007). Following 11. For total household earnings, the earnings of self-employed household members are imputed based on the earnings of employees. Although this is a fairly rough approximation, it appears this imputation serves the purpose of measuring household income well: results are very similar when households with at least one self-employed adult are excluded from the sample. 12. Data on other income sources are not collected in the survey, nor are household assets. 13. Households, especially in developing countries, can use women’s labor supply to deal with negative income shocks or uncertainty (Attanasio et al. 2005; Bhalotra and Uman ˜ a-Aponte 2010). Implicit in the empirical model is the assumption that women’s participation decision is made conditional on men’s: we do not consider joint utility maximization or bargaining within the household. Given the very high and unresponsive labor force participation rates of men, we believe that this assumption is warranted. 14. Employment shares are measured within urban areas of districts, but even in urban areas, households engage in agricultural and livestock production. Klasen and Pieters 461 leading studies in the literature (Blau and Kahn 2007; Heim 2007), we estimate two different specifications to identify the own-wage effect. One exploits wage var- iation across districts, the other exploits wage variation across state-age-education groups. The two sources of variation give very different estimates, which do not allow us to draw firm conclusions about the impact of wages on women’s labor force participation. However, the estimated effects of other explanatory variables, including education, are robust to the different specifications and to excluding the own wage from the model.15 Given the difficulties in identifying the own wage effect and the inability to include the self-employed in models that include own wages, we focus here on the results without including own-wage effects. Results for own-wage estimates and more details of the estimation method are discussed in supplemental appendix S2 (available at http://wber.oxfordjournals.org/). Table 1 contains sample means for all variables. The largest changes over time are increasing education (women’s own education levels and those of their male partners and household heads, as well as the district population share with grad- uate level education), fertility decline, and increasing household income. Further note that employment has shifted over time from agriculture, manufacturing, and white-collar services to construction and other services. To explore in more detail to what extent the determinants of labor force participation differ between low- and highly educated women, equation (1) is also estimated separately for women with less than secondary education and women with secondary or higher education. Estimation Results Estimation results are reported in Table 2 as average marginal effects, showing the change in the probability of being in the labor force associated with a unit change in the explanatory variable (for categorical variables this is the difference with the reference category). First of all, the U-shape in education appears even stronger in the marginal effects than in the average unconditional participation rates in Figure 2. Even conditional on variables that could have explained the un- conditional U, such as partner’s income and education, caste, or religion, partici- pation declines as education increases up to intermediate levels. A remaining channel—one not captured in control variables—is through social stigma. It is possible that the stigma associated with low-skilled jobs (including in agriculture, menial jobs in manufacturing and construction, or in domestic services) increases as women attain low and intermediate levels of education. This means education is associated with an increased (fixed) utility cost of engaging in low-skilled work. This channel can also explain the upward sloping part of the ed- ucation U-curve: once women attain secondary or post-secondary education, the utility cost disappears as they gain access to white-collar jobs that are not subject to social stigma. In all, the education U-curve conditional on income, social status, 15. The level of the own education effects does change but in all specifications we find a strong U-shape and a large decline in the effect of secondary and graduate education over time. 462 THE WORLD BANK ECONOMIC REVIEW T A B L E 1 . Sample Means 1987 1999 2004 2009 2011 Labor force 0.18 0.18 0.20 0.17 0.18 Illiterate 0.42 0.32 0.28 0.23 0.22 Literate 0.11 0.08 0.07 0.07 0.07 Primary 0.15 0.12 0.13 0.12 0.11 Middle 0.11 0.14 0.16 0.15 0.15 Secondary 0.14 0.20 0.22 0.26 0.27 Graduate 0.07 0.13 0.14 0.17 0.18 Log income 3.63 3.99 4.08 4.36 4.47 Salaried employment 0.52 0.48 0.46 0.46 0.46 Hh head Illiterate 0.21 0.18 0.16 0.15 0.14 Hh head Literate 0.14 0.10 0.10 0.08 0.08 Hh head Primary 0.17 0.12 0.13 0.11 0.11 Hh head Middle 0.14 0.15 0.16 0.15 0.16 Hh head Secondary 0.22 0.26 0.28 0.30 0.29 Hh head Graduate 0.13 0.18 0.18 0.21 0.22 Hindu non-SCST 0.66 0.65 0.65 0.66 0.65 SCST 0.14 0.17 0.17 0.16 0.16 Muslim 0.14 0.14 0.14 0.14 0.15 Other social group 0.05 0.05 0.05 0.04 0.04 Age 35.6 36.1 36.6 36.7 36.8 Children 0 – 4 0.69 0.53 0.50 0.42 0.39 Children 5 – 14 1.59 1.37 1.20 1.05 1.05 Agriculture 0.07 0.06 0.05 0.05 0.05 Construction 0.07 0.10 0.10 0.12 0.12 Manufacturing 0.27 0.23 0.24 0.22 0.23 White-collar services 0.20 0.18 0.18 0.19 0.19 Other services 0.38 0.43 0.43 0.41 0.41 Graduate share 0.11 0.16 0.17 0.21 0.22 N 29031 32541 29513 27198 27306 Notes: Averages for married women age 25– 54, calculated using sampling weights. Education levels (Illiterate-Graduate) refer to the highest level completed, where Literate means the person at- tained literacy without formal schooling or below the primary level; Secondary includes secondary and higher secondary schooling; and Graduate includes graduate and post-graduate degrees. Source: NSS Employment and Unemployment Survey. and the sectoral structure of employment (which we further discuss below) can be explained by social stigma associated with different types of jobs. In addition, the positive effects of higher education could reflect higher wages or positive selection effects. As we find very similar estimates when controlling for the own wage, we think the own wage channel does not play an important role, despite high returns to education in the labor market. On the other hand, it is quite plausible that the positive higher education effect partly reflects an upward bias due to endogenous selection into higher education of girls with stronger labor market orientation. This would not only bias upwards the effects of higher education, but is also consistent with the large decline in the positive effects of secondary and graduate education over time. If increasing educational attainment has been driven by the growing supply of education, for example, Klasen and Pieters 463 T A B L E 2 . Estimation Results (Average Marginal Effects) Pr (Labor force) 1987 1999 2004 2009 2011 Own education (Ref. ¼ Illiterate): Literate 2 0.050*** 2 0.063*** 2 0.051*** 2 0.030** 2 0.026* (0.009) (0.011) (0.013) (0.014) (0.014) Primary 2 0.072*** 2 0.074*** 2 0.062*** 2 0.025** 2 0.027* (0.008) (0.011) (0.015) (0.012) (0.014) Middle 2 0.067*** 2 0.084*** 2 0.100*** 2 0.060*** 2 0.064*** (0.011) (0.009) (0.013) (0.013) (0.011) Secondary 0.086*** 2 0.020* 2 0.048*** 2 0.056*** 2 0.041*** (0.013) (0.010) (0.014) (0.014) (0.014) Graduate 0.324*** 0.217*** 0.144*** 0.131*** 0.151*** (0.020) (0.018) (0.023) (0.019) (0.022) Log income 2 0.034*** 2 0.017*** 2 0.032*** 2 0.028*** 2 0.022*** (0.003) (0.003) (0.003) (0.003) (0.003) Male salaried emp. 2 0.022*** 2 0.032*** 2 0.019** 0.002 2 0.005 (0.007) (0.008) (0.009) (0.009) (0.010) Household head education (Ref. ¼ Illiterate): Literate 2 0.067*** 2 0.045** 2 0.040** 2 0.025 0.017 (0.012) (0.020) (0.017) (0.018) (0.019) Primary 2 0.107*** 2 0.078*** 2 0.053*** 2 0.035** 2 0.057*** (0.012) (0.012) (0.015) (0.014) (0.018) Middle 2 0.126*** 2 0.115*** 2 0.078*** 2 0.072*** 2 0.083*** (0.013) (0.011) (0.014) (0.014) (0.015) Secondary 2 0.170*** 2 0.150*** 2 0.121*** 2 0.095*** 2 0.103*** (0.013) (0.013) (0.014) (0.015) (0.015) Graduate 2 0.166*** 2 0.154*** 2 0.110*** 2 0.084*** 2 0.126*** (0.016) (0.014) (0.018) (0.017) (0.019) Social group (Ref. ¼ Hindu non-SCST): SCST 0.089*** 0.054*** 0.036*** 0.059*** 0.033*** (0.010) (0.011) (0.013) (0.014) (0.009) Muslim 2 0.058*** 2 0.069*** 2 0.086*** 2 0.087*** 2 0.082*** (0.010) (0.010) (0.014) (0.012) (0.011) Other 0.021 0.013 0.004 0.022 0.039** (0.015) (0.017) (0.013) (0.015) (0.018) Age 0.020*** 0.026*** 0.027*** 0.016*** 0.014*** (0.004) (0.004) (0.005) (0.005) (0.004) Age2 2 0.0003*** 2 0.0003*** 2 0.0004*** 2 0.0002*** 2 0.0002*** (0.000) (0.000) (0.000) (0.000) (0.000) Children 0– 4 2 0.022*** 2 0.018*** 2 0.024*** 2 0.036*** 2 0.039*** (0.004) (0.004) (0.005) (0.007) (0.008) Children 5– 14 2 0.005** 2 0.001 0.008** 2 0.003 0.007* (0.003) (0.002) (0.004) (0.004) (0.004) District male employment shares (Ref. ¼ construction): Agriculture 0.255* 0.171* 0.165 2 0.008 0.220** (0.132) (0.097) (0.116) (0.108) (0.096) Manufacturing 0.073 2 0.097 2 0.039 0.155* 0.268*** (0.118) (0.087) (0.099) (0.091) (0.077) Services 2 0.048 2 0.090 2 0.015 0.120 0.014 (0.121) (0.084) (0.093) (0.088) (0.073) (Continued ) 464 THE WORLD BANK ECONOMIC REVIEW TABLE 2. Continued Pr (Labor force) 1987 1999 2004 2009 2011 White-collar serv. 2 0.036 2 0.022 2 0.086 0.170 0.052 (0.124) (0.099) (0.120) (0.106) (0.100) District grad. share 0.004 2 0.087 0.017 2 0.402*** 2 0.051 (0.111) (0.083) (0.105) (0.091) (0.082) N 29031 32541 29513 27198 27306 Pseudo R2 0.145 0.126 0.125 0.114 0.097 FLFP rate 0.185 0.179 0.203 0.174 0.179 Notes: Married women age 25– 54. All estimations include state fixed effects. District-clustered standard errors in parentheses, *** p , .01, ** p , .05, * p , .10. highly educated women in 1987 were more positively selected than those in 2011. We cannot test this directly without modeling educational attainment itself, which is beyond the scope of this paper, but we provide suggestive evidence for a decline in positive selection in Section V. Moving down in Table 2, we find a negative income effect as expected. Security of income (male salaried employment) also reduces women’s labor force participation, and we find a particularly strong negative effect of education of the household head. The size of the effects declines over time, suggesting that women have become less responsive to income insecurity and to overall socioeconomic status of the household. Household head education effects remain, however, quan- titatively large. Looking at the social group variables, we find that women in SCST house- holds are most likely to work, but the gap between SCST and non-SCST Hindus declined.16 The impact of religion appears stronger, with Muslim women 6 to almost 9 percentage points less likely to work than non-SCST Hindus; a differ- ence close to half the female labor force participation rate. Women’s age and the presence of children in the household have the expected effects. Participation first increases and then declines with age. We see that the age profile shifted somewhat after 2004, reducing the peak age from around 40 in 1987–2004 to around 35 years in 2011. At the same time, the negative effect of young children almost doubled in magnitude. Though the estimates cannot be interpreted as causal due to joint fertility and participation decisions, we clearly see an increasing negative association between women working and the presence of young children in the household. 16. From 1999 onwards, the NSS data also distinguish Other Backward Class (OBC) households. Including a separate category for OBC—rather than including them in the reference non-SCST group—we find that OBC women have higher FLFP than high caste Hindus in 1999 and 2004, but there is no difference in 2009 and 2011. The estimated gap between SCST and high caste Hindus is somewhat larger but still declines over time. Klasen and Pieters 465 The bottom of the table shows the estimated effects of district demand and supply variables.17 We do not find that local supply of high-skilled workers affects FLFP much, except for a strong negative effect in 2009. This could reflect the impact of the global financial crisis, which led to substantial declines in employment (see for example Kucera et al., 2012). Regarding the sectoral struc- ture of employment, in 1987 and 1999 FLFP was highest in districts where agri- culture makes up a higher share of urban employment suggesting that agricultural activities in and at the fringes of cities generate higher employment opportunities for women. Manufacturing has started drawing in more women since 2009. Surprisingly, the employment shares of white-collar services and other services are not significantly positively related to FLFP. Subsample estimates (reported in Tables S1.3 and S1.4 in the supplemental appendix) shed more light on this. In the low-education sample, we see the strong association of FLFP with agricultural employment in the early years and again in 2011. This same pattern is not present for highly educated women. The negative effect of high-skilled labor supply is much stronger in the high- education sample, where it remains significantly negative (though smaller in magnitude) in 2011. We also find a positive effect of white-collar services in the high-education sample, but it is significant only in 2009. It appears that 2009 was characterized by fiercer competition for skilled jobs, and only districts with a large share of employment in white-collar services could keep highly educated women in the labor force. By 2011, the crowding out effect is weaker and at the same time, highly educated women are shifting more toward the manufacturing sector, similar to the low-education sample. There are other important differences between the two education groups. First of all, the negative income effect in the high-education sample halves between 1987 and 2009 and disappears in 2011, as opposed to a slight increase in the low-education sample. Second, education of the household head has a strong negative effect in the low-education sample, but no effect on highly educated women in any of the years. Third, male salaried employment—our proxy for income security—is associated with lower labor force participation of women with low education, while the opposite is found for highly educated women (though for both subsamples, the effect declines in magnitude and is no longer statistically significant in 2011). These results suggest that highly educated women are less constrained by family circumstances in their labor force partici- pation decision, while women with less than secondary education appear to be boxed in by necessity to work if household incomes are very low or insecure, and stigmas attached to working if they have some education and are in more secure economic environments. The finding that manufacturing employment is nowa- days associated with higher FLFP is an indication that manufacturing jobs are becoming less stigmatized. 17. If we replace the district-level variables and state fixed effects by district fixed effects, results are virtually unchanged. These results are available on request. 466 THE WORLD BANK ECONOMIC REVIEW Regarding changes in participation rates over time, it follows from the probit estimates that the increase in women’s educational attainment and fertility decline should have translated into higher labor force participation. Working in the opposite direction, however, is the increase in educational attainment of men and the increase in male incomes. Changes in the structure of employment also appear to have contributed to lower participation rates. The employment shares of agriculture, manufacturing, and white-collar services declined, while these are the sectors positively associated with FLFP. The decomposition analysis in Section IV quantifies the contributions of changes in the different covariates in- cluded in the model. Before turning to those results, we first discuss the potential influence of migration on our estimation results. The Potential Role of Rural-Urban Migration Rural to urban migration is an important part of India’s development process. The population numbers in Table S1.1 show that the total female population grew by 50 percent in the period 1987-2011, comprising a 91 percent growth of females in urban India and 38 percent growth in rural India. Rural-urban mi- grants are typically not a random selection of the rural population, so one might worry that the stagnation of urban FLFP and the changes in labor force partici- pation determinants are driven by the inflow of migrants. The National Sample Survey Organisation (NSSO) integrated the collection of migration data with the employment surveys in 1987 and 1999 but not with those of 2004 and after. In 1987 and 1999, almost 70 percent of women in our sample have a migration background, of which almost 60 percent (i.e., around 40% of the total female urban sample) migrated from a rural area. The most im- portant reason for migration is marriage (about 65 percent of all female mi- grants). On average, rural-urban migrants are more likely to be at either end of the education distribution (illiterate or with at least secondary education) than nonmigrants, and less likely to have low or intermediate educational attainment (see Fig. S1.3 in the supplemental appendix). Given the U-shaped relationship between education and labor force participation, the compositional impact of migration should have been to increase urban FLFP, if anything. But despite the large share of migrants and differences in educational attainment, controlling for rural-urban migration background in the 1987 and 1999 estimations has no impact on the results. The migrant variable itself has no significant effect on labor force participation. Results are also unchanged if we add dummies for both the woman’s and her spouse’s migration background and their interaction, which captures the difference between women migrating to join their husbands’ family versus families migrating jointly. Even if we drop all rural-urban migrants from the sample, the results do not change.18 Unfortunately, no migration data was collected in the employment surveys after 1999, but the NSSO does provide descriptive statistics on migrants from a 18. Detailed results for this section are available on request. Klasen and Pieters 467 special migration survey conducted in 2007–08 (NSSO 2010). Based on these statistics we find that the educational attainment differences between urban female migrants and the total urban female population have been quite stable between 1987–88 and 2007–08. In other words, the educational composition of female migrants has not changed differently from the educational composition of the entire urban female population. We are therefore fairly confident that changes in female migration patterns are not driving the stagnation of FLFP. I V. D E C O M P O S I T I O N A N A L Y S I S In order to quantify the contribution of different explanatory variables to the observed change—or lack thereof—in the female labor force participation rate, we now turn to a decomposition analysis, using the probit estimates for 1987 and 2011. Following Fairlie’s (2006) extension of the Oaxaca-Blinder decom- position (Blinder 1973; Oaxaca 1973), the labor force participation equation t (Eqn. 1) is expressed as Yit ¼ FðXt i b Þ, resulting in the following decomposition: " # " 11 # N11 X N 87 X X N N11 X  11  87 FðX11 b87 Þ i FðX87 b87 Þ i FðX11 i b Þ 11 FðX11 87 i b Þ Y ÀY % À þ À : i N11 i N 87 i N 11 i N11 ð2Þ The Nt are sample sizes and Y  t the sample average probability of being in the 19 labor force. The first right-hand side term in (2) measures the contribution of changes in covariates, evaluated at 1987 coefficients. It measures how changes in covariates would have translated into changes in FLFP if coefficients and unob- servables had remained at their start-of-period (1987) levels. The second term is the remaining change, which measures the change in FLFP that would have re- sulted from changes in coefficients and unobservables if covariates had been constant at their end-of-period (2011) levels. The contribution of a single explanatory variable is computed using a counterfac- tual predicted participation rate, by replacing only this particular variable, say X87 1i , by its 2011 counterpart X111i , while keeping all other variables at their 1987 values. This is done by drawing a 2011 subsample of size equal to the 1987 sample, match- ing women on their predicted probability of working (based on a pooled probit esti- mation), and assigning women in the 2011 subsample the value of X1 observed for their 1987 match. The results reported below are based on 1000 random subsam- ples, in which furthermore the order of variables is randomized to account for the fact the contribution of one variable depends on the value of other variables.20 19. The equality in equation (2) does not hold exactly for the probit model, because the average value of predicted probabilities does not equal the average of the dependent variable. It is a close approximation, however, and we find very similar results when we use a logit model instead (for which the equality does hold exactly, see Fairlie [2006]). 20. See Fairlie (2006) for a more detailed discussion of the method. 468 THE WORLD BANK ECONOMIC REVIEW The decomposition analysis can also be carried out with covariate contribu- tions measured at 2011 coefficients, and coefficient contributions measured at 1987 covariate levels, using the following expression: " # " 87 # N 11 X 11 N 87 X 87 11 X N 87 11 N87 X 87 87  11  87 FðX011 i b Þ F ð Xi b Þ F ðX i b Þ F ðXi b Þ Y ÀY % 11 À 87 þ 87 À 87 : i N i N i N i N ð3Þ If coefficients and covariates changed substantially from 1987 to 2011, which is the case for a number of variables, measured contributions will differ between (2) and (3). We therefore report results for both choices of reference year. The “true” contribution of a variable will lie somewhere in between, and of course this type of decomposition is an accounting exercise, ignoring the fact that changes in covariates and changes in coefficients are interdependent. Table 3 shows the decomposition results. In our sample of married women 25–59, the female labor force participation rate declined by 0.6 percentage points between 1987 and 2011. At 1987 coefficients (Eqn. 2), covariate changes account for a 0.3 percentage point decline. The remaining 0.2 percentage point is accounted for by changes in coefficients and unobservables. At 2011 coefficients (Eqn. 3), we find a much larger negative contribution from covariate changes, which is almost entirely offset by a large positive contribution from changing coefficients and unobservables. The variable-specific covariate contributions show that the increase in women’s education levels and declining numbers of children account for an increase in par- ticipation rates.21 These effects are offset, however, by negative contributions from rising household incomes, education of household heads, and the change in the sectoral structure of employment. The probit estimation results indicated that only agricultural and manufacturing employment are associated with significantly higher female labor force participation rates, but both sectors’ shares in employ- ment declined over time. This sectoral change alone can account for a 1 to 2 per- centage point reduction in FLFP. The negative contributions of household income and household head education are larger, but because women have become less re- sponsive to income and household head education, they account for a smaller decline in FLFP at 2011 coefficients than at 1987 coefficients. The total covariate contribution is negative and much larger at 2011 coeffi- cients. Sensitivity with respect to the choice of reference year is mainly due to the contribution of women’s education: at 1987 coefficients, increasing education accounts for an increase in FLFP of 4.7 percentage points, while at 2011 coeffi- cients the contribution is only 1.1 percentage points. This implies that FLFP would have increased by 3 percentage points (rather than decline by 0.6 21. The positive contribution of state dummies suggests that growth of the urban female population was higher in states with higher FLFP rates. This could be due to migration: either migration from low-FLFP states to high-FLFP states, or stronger urbanization within high-FLFP states. Klasen and Pieters 469 T A B L E 3 . Decomposition of FLFP, 1987–2011 Pr (labor force) 1987 0.185 (N ¼ 29031) Pr (labor force) 2011 0.179 (N ¼ 27306) Difference 2 0.006 At 1987 coefficients At 2011 coefficients Contribution SE Contribution SE Own education 0.047*** 0.003 0.011*** 0.003 Log income 2 0.030*** 0.002 2 0.020*** 0.003 Male salaried emp. 0.001** 0.000 0.000 0.000 Education household head 2 0.025*** 0.002 2 0.021*** 0.002 Caste and religion 0.001*** 0.000 0.000 0.001 Age 0.001*** 0.000 0.000 0.001 Children 0.009*** 0.001 0.007** 0.003 District employment shares 2 0.011*** 0.003 2 0.019*** 0.004 District graduate share 0.001 0.008 2 0.006 0.007 State dummies 0.003*** 0.001 0.003*** 0.001 Total covariates 2 0.003 2 0.045 Coefficients & unobservables 2 0.002 0.039 Note: SE ¼ standard error. *** p , .01, ** p , .05. percentage points) if educational attainment had increased without the effect of education declining. The remainder of this paper is devoted to a further analysis of precisely this declining effect of higher education. V. T H E D E C L I N I N G E F F E C T O F H I G H E R E D U C A T I O N ON LABOR FO R C E PA R T I C I PAT I O N As discussed in Section III, the effect of higher education on women’s labor force participation could partly reflect an upward selection bias, where completing education and joining the labor force are both outcomes of some unobservable determinants. Examples of such unobserved determinants could be the education or labor force participation of mothers, influences of peers, or a more general preference for work that motivates both higher education and labor force partici- pation. This selection effect might change over time. If the average woman completing secondary or higher education in 2011 is less positively selected than her 1987 counterpart, this could potentially account for the declining effects of secondary and graduate education between 1987 and 2011.22 A detailed analysis 22. Another explanation would be a decline in the quality of education. Azam and Kingdon (2013) study gender bias in education expenditure and show that in the period 1993–2005, when girls’ education caught up to boys’, households still spent less on girls, primarily by sending girls to public schools and boys to private schools. Girls are thus likely to receive lower quality schooling than boys, even if they attain the same level of education. Unfortunately we do not observe the type of institution a woman attended or any other indicator of the quality of education in the NSS data. 470 THE WORLD BANK ECONOMIC REVIEW of the determinants of women’s educational attainment would require data on women’s parental background and on schooling supply and other characteristics of the location where they grew up. Unfortunately, those data are not available, as we only observe married women in their husband’s household—which is often in a different area than where they grew up—and have no information on their location of origin or on their parents’ education and employment. The following section provides supporting evidence, however, for the empirical relevance of the selection channel, which is also consistent with an analysis by cohorts we discuss next. Among the factors that could explain a decrease in positive selection are the increasing supply of education and rising marriage market returns to education: both would be reasons for women to pursue more education even in absence of expected labor market returns and unrelated to the unobservable propensity to join the labor force. Given the high pace of expansion of schooling supply, it is likely that highly educated women in 2011 are less positively selected than highly educated women in 1987.23 Besides an expansion in the supply of education, rising marriage market returns to women’s education could be driving women to pursue higher educa- tion. Marriage is of great importance in India, especially to women and their parents (Anderson 2003), as women typically leave their parental household and migrate to live with their husband’s family (see Rosenzweig and Stark 1989; Banerjee et al. 2013). Despite a growing literature on women’s earnings capacity and their bargaining power within marriage in developing countries (e.g. Luke and Munshi 2011), there is little evidence on the marriage market returns to women’s education. According to Anderson (2003), the most important quality of women on the marriage market in India is a good appearance, while for men their ability to earn a living is most important. Data for India in Banerjee et al. (2013) show both men and women have a preference for marrying a highly edu- cated spouse, though their sample includes only educated persons from the urban middle-class in West Bengal. Rising labor market returns to education would be an obvious reason why men prefer more educated women. However, even in absence of labor market returns, women’s education can contribute to husbands’ social status directly and through higher productivity in status produc- tion (Eswaran et al. 2013). Another important channel could be increased pro- ductivity of maternal time in the production of child human capital (Lam and Duryea 1999), which raises the demand for educated wives if labor market returns to men’s education increase (Behrman et al. 1999). 23. As discussed in Section 2 some recent studies show that (girls’) enrollment responds to the growth of jobs in IT and IT enabled services (Jensen 2012; Shastry 2012; Oster and Millett Steinberg 2013), but this is a very recent development and this sector still accounts for a very small share of employment. We are not aware of any research on the causal effect of education on women’s labor force participation in India or on the impact of rapidly increasing schooling supply on women’s educational attainment, other than evaluations of recent policies and programs for primary education and a growing literature on higher education quota for low-caste men and women (e.g., Bertrand et al. 2010; Kingdon 2007). Klasen and Pieters 471 In Klasen and Pieters (2013) we show that marriage market returns to educa- tion became significantly more convex between 1987 and 2009. In 2009 women needed at least secondary education to have a reasonable chance of marrying a highly educated and high-earning spouse. In contrast, primary school and cer- tainly middle school generated reasonable odds to attract a high-quality spouse in 1987. This is consistent with our claim that at least some of the female educa- tion expansion is driven by expected marriage market returns rather than labor market returns. Selection Bias in the Effects of Education Without instruments for educational attainment we cannot control for selection into education, but the extreme case—assuming that the education effects capture endogenous selection only—can be used to estimate an upper bound on the contri- bution of changes in selection to changes in the estimated effects of education. For simplicity, suppose there is one unobservable characteristic that determines labor force participation, and let us call this ability. Assume there is perfect sorting on ability into education, the ability distribution in the population is fixed, and there are k different ability levels in the population with k ¼ 1, . . ., 6 (one could also think of a uniform distribution of ability, but the idea remains the same). We thus have six ability types with corresponding ability level ak, increasing in k. Taking 1987 as our benchmark year, the 1987 distribution of education (E ¼ 1, . . ., 6) corresponds one-to-one with the distribution of ability types in the female population, such that aE k 87 ¼ a for E ¼ k. As the supply of education expands, women end up with increasingly higher educational attainment, but the average ability of women at education levels E . 1 declines over time. In other words, between 1987 and 2011, low-ability types move into higher education levels. The shift is illustrated in Figure 5: by definition, illiterate women in 1987 are of ability level a 1, which is the ability level of the bottom 42 percent of the distri- bution. Literate women below primary school in 1987 are of ability level a 2, women with primary schooling in 1987 are of ability level a 3, and so on. By 2011, all literate women without or with primary schooling are of ability level F I G U R E 5 . Women’s Educational Attainment in Urban India Note: Distribution of married women age 25– 54 across levels of educational attainment in 1987 and 2011. Source: NSS Employment and Unemployment Survey. 472 THE WORLD BANK ECONOMIC REVIEW F I G U R E 6 . Estimated and Reweighted Marginal Effect of Education, 1987 and 2011 Notes: Estimated marginal effects from equation (1) in the main text. Reweighting based on the education distribution as reported in Table 1. Source: Authors’ analysis based on NSS Employment and Unemployment Survey. a 1, as they all fall within the bottom 42 percent of the distribution. Similarly, all women with middle school completed in 1987 are of ability level a 4, but their 2011 counterparts are of ability levels a 1, a 2 and a 3; and so on for higher educa- tion levels. Thus average ability declines at all education levels above illiteracy. Now assume that the estimated marginal effects of education on women’s labor force participation measure the pure ability-selection effect. The 1987 edu- cation effects bE 1987 quantify the effect of ability on labor force participation for each ability type k. The education effects in other years should differ from those in 1987 only by the change in the ability composition of education groups. That is, they are a weighted average of the 1987 estimates. Table S1.5 in the supple- mental appendix shows the original estimated marginal effects of education in columns 1 –5, and the reweighted effects for 1999, 2004, 2009, and 2011 in columns 6 –9. Results for secondary and graduate education are summarized in Figure 6. As Figure 6 shows, the reweighting exercise predicts the change in the margin- al effects of education from 1987 to 2011 quite well. While this is by no means conclusive evidence that the education effects are largely driven by selection,24 it does show that declining selection could potentially play a large role in the declining effect of higher education on female labor force participation, where de- clining selection means that women’s selection into education is increasingly based on characteristics that are not positively related to labor force participation. 24. If purely selection-driven, one would not find negative estimates at intermediate education levels. Klasen and Pieters 473 Cohort Analysis In a sample of women aged 25 to 54, educational attainment changes over time primarily because younger cohorts enter the sample with higher educational at- tainment while older cohorts with lower educational attainment age out of the sample. Consequently, the educational attainment by age group increases contin- uously, while the age composition of the sample is stable over time. To further probe the selection channel discussed above, we estimate age-group-specific mar- ginal effects of education on labor force participation. If the declining effect of higher education indeed reflects a weaker positive selection into higher education among younger cohorts, we should observe that the effect of education is lower for younger women than for older women in a given year. Furthermore, we should find that the effect declines over time for a given age group (e.g., as 30-year old women in 2011 went to school more recently than 30-year old women in 1987) but not for a given birth-year cohort (e.g., those born in 1970). For this analysis we use five 6-year age groups, which allows us to trace birth cohorts between the surveys in 1987, 1999, and 2011. The age group specific effects of education are estimated by including interaction effects of age group and education level in the probit specification (Eqn. 1) and calculating marginal effects of education for each age group. Figure 7 summarizes the results graphi- cally, plotting the marginal effects of graduate education across survey years, by age group in panel (a) and by birth-year cohort in panel (b).25 Full results are available in Table S1.6 in the supplemental appendix. These results are in line with the selection channel driving the effect of gradu- ate education: the marginal effect is lower for younger age groups than for older age groups in a given year. Furthermore, for each age group, the effect declines by about one third over time in each of the two 12-year periods. Looking at the same results presented by birth-year cohort in panel (b), we see some decline in the effect of graduate education over time in those cohorts we can trace, but the changes are much smaller.26 All in all, this is further supporting evidence that declining positive selection is a plausible explanation why graduate education has become less positively asso- ciated with women’s labor force participation in urban India. 25. Marginal effects in nonlinear models vary with the value of other explanatory variables. For interaction terms, this variation is typically large. Effects can range from significantly negative to significantly positive, so the average marginal effect does not necessarily convey anything meaningful (see Ai and Norton 2004). We find that the marginal effects of graduate education are always positive. Effects are statistically significant only for the oldest two age groups (i.e., the effect of education is significantly larger for the two oldest groups than for the youngest group). For reasons of space we report only the averages, but detailed results are available from the authors. 26. For secondary education, the pattern is similar but somewhat less clear than for graduate education (see Table S1.5). 474 THE WORLD BANK ECONOMIC REVIEW F I G U R E 7 . Marginal Effect of Graduate Education by Age Group (a) and Birth- Year Cohort (b) Note: Full results available in supplemental appendix table S1.6. Source: Authors’ analysis based on NSS Employment and Unemployment Survey. VI. CONCLUSIONS In this paper, we investigate the very low and stagnating female labor force par- ticipation rates in urban India over the past 25 years. This stagnation is surprising given that it took place at a time of high GDP and earnings growth, a sizable fer- tility decline, a rapid expansion of female education, and rising returns to educa- tion. A combination of demand and supply side effects appear to have played a role in accounting for this stagnation. On the supply side, rising male incomes and education have reduced female labor force participation, showing that the classic income effect is at work in urban India. The effect of rising female educa- tion on female labor force participation is more complicated. While the pure shift toward increasing the proportion of women with graduate education has in- creased labor force participation, this effect is moderated and counteracted by range of opposing effects. First, the strong conditional U-shape pattern of the effect of education on labor force participation suggests that, particularly in the middle of the education distribution, other factors depress female labor force participation. This is most likely a result of stigmas for these women to be working in low-skilled sectors. But it appears that the role of stigma, at least towards working in the manufacturing sector, has started to decline in 2009, when the manufacturing sector has started to draw women into the labor force. Second, the positive effect of secondary and graduate education on female labor force participation has fallen considerably. We show supporting evidence that this is related to a declining positive selection into higher education. So even Klasen and Pieters 475 though more women have attained higher education, positive selection effects have been diluted, contributing to falling labor force participation rates among the highly educated. Demand side changes also play a role. Changes in the sectoral structure of em- ployment contributed to a reduction in FLFP rates. Most employment growth oc- curred in construction and low-skilled services, while expansion of employment in manufacturing and white-collar services has not been sufficient for absorbing a growing female working-age population. These results suggest that, if current trends and preferences persist, there is little likelihood that women will drastically increase their labor force participa- tion rates in coming years. Although employment growth in manufacturing has the potential to contribute to rising FLFP, India is, on current trends, unlikely to fully reap the benefits of the demographic dividend associated with its high share of the working-age population. And importantly, rising education of women will not contribute much to their economic empowerment, which is typically associ- ated with employment and earnings. As such, the sustainability of India’s high growth is very much in question if it fails to substantially improve the integration of educated women into the labor force. It is difficult to make any definitive judgments on welfare effects. To the extent women’s labor force participation is decided by their families and does not reflect women’s own preferences, or is constrained by their inability to migrate for employment, policy action to promote female employment would be war- ranted. But even if the main constraint is women’s own preferences, the degree to which this impedes their labor force participation should be a concern to policy makers. Our findings point at the importance of mismatches between the sectoral structure of employment and women’s preferences. Employment growth in urban India has been concentrated in construction and low-skilled services, but from the perspective of female labor force participation a different growth strat- egy would be warranted; a more female-intensive export-oriented growth strat- egy (as has been pursued in many East Asian economies as well as in neighboring Bangladesh) would substantially increase female employment opportunities for those in the middle and even the top of the education distribution. On the supply side, policies explicitly promoting the acceptability of female employment outside the public sector, policies to allow a greater compatibility of female employment with domestic responsibilities and policies to improve the safety of female workers in the private sector could also draw more women into the work- force. Looking beyond India, the results could also hold considerable relevance for other parts of the developing world. In particular, in neighboring Pakistan as well as many countries of the Middle East and North Africa, substantial growth, fertili- ty decline, and an expansion of female education also coexist with very low female labor force participation rates (e.g., World Bank 2003; Gaddis and Klasen 2014). The insights from this paper might also inform research and policy-makers in those settings. 476 THE WORLD BANK ECONOMIC REVIEW REFERENCES Afridi, F., A. Mukhopadhyay, and S. Sahoo. 2012. “Female Labour Force Participation and Child Education in India: The Effect of the National Rural Employment Guarantee Scheme.” IZA Discussion Papers Series, No. 6593. Ai, Chunrong, and Edward C. Norton. 2003. “Interaction Terms in Logit and Probit Models.” Economics Letters 80 (1): 123– 29. Anderson, S. 2003. “Why Dowry Payments Declined with Modernization in Europe but Are Rising in India.” Journal of Political Economy 111 (2): 269–310. Anderson, S., and M. Eswaran. 2009. “What Determines Female Autonomy? Evidence from Bangladesh.” Journal of Development Economics 90 (2): 179 –91. Attanasio, O., H. Low, and V. Sa ´ nchez-Marcos. 2005. “Female Labor Supply as Insurance against Idiosyncratic Risk.” Journal of the European Economic Association 3 (2–3): 755–64. Azam, M., and G. G. Kingdon. 2013. “Are Girls the Fairer Sex in India? Revisiting Intra-Household Allocation of Education Expenditure.” World Development 42: 143– 64. Banerjee, A., E. Duflo, M. Ghatak, and J. Lafortune. 2013. “Marry for what? Caste and Mate Selection in Modern India.” American Economic Journal: Microeconomics 5 (2): 33 –72. Behrman, J. R., A. D. Foster, M. R. Rosenzweig, and P. Vashishtha 1999. “Women’s Scholling, Home Teaching, and Economic Growth.” Journal of Political Economy 107 (4): 682– 714. Bertrand, M., R. Hanna, and S. Mullainathan. 2010. “Affirmative Action in Education: Evidence from Engineering College Admissions in India.” Journal of Public Economics 94: 16– 29. Bhalotra, S., and M. Uman˜ a-Aponte. 2010. “The Dynamics of Women’s Labor Supply in Developing Countries.” IZA Discussion Paper Series No. 4879. Blau, F. D., and L. M. Kahn. 2007. “Change in the Labor Supply Behavior of Married Women: 1980– 2000.” Journal of Labor Economics 25 (3): 393– 438. Blinder, A. S. 1973. “Wage Discrimination: Reduced Form and Structural Estimates.” Journal of Human Resources 8 (4): 436–55. Bloom, D. E. 2011. “Population Dynamics in India and Implications for Economic Growth.” PGDA Working Paper No. 65, Harvard School of Public Health. Bloom, D. E., and J. G. Williamson. 1998. “Demographic Transitions and Economic Miracles in Emerging Asia.” World Bank Economic Review 12 (3): 419–55. Blundell, R., and T. MaCurdy. 1999. “Labor Supply: A Review of Alternative Approaches.” In: O. Ashenfelter, and D. Card, eds., Handbook of Labor Economics, 1st ed., vol. 3, chap. 27, 1559– 1695, New York: Elsevier. Boserup, E. 1970. Women’s Role in Economic Development. London: Earthscan. Bosworth, B., and S. M. Collins. 2008. “Accounting for Growth: Comparing China and India.” Journal of Economic Perspectives 22 (1): 45– 66. Chen, M., and J. Dre` ze. 1992. “Widows and Health in Rural North India.” Economic and Political Weekly 27 (43– 44): 24– 31. Das, M. B., and S. Desai. 2003. “Why Are Educated Women Less Likely to be Employed in India? Testing Competing Hypotheses.” Social Protection Discussion Paper Series, No. 313. Washington, DC: World Bank. Deaton, A. 2003. “Prices and Poverty in India, 1987– 2000.” Economic and Political Weekly 38: 362 –68. Desai, S., A. Dubey, B.L. Joshi, M. Sen, A. Shariff, and R. Vanneman. 2009. “India Human Development Survey (IHDS) [Computer file].” ICPSR22626-v2. Ann Arbor, MI: Inter-university Consortium for Political and Social Research. ———. Human Development in India: Challenges for a Society in Transition. New Delhi: Oxford University Press. Eswaran, M., B. Ramaswami, and W. Wadhwa. 2013. “Status, Caste, and the Time Allocation of Women in Rural India.” Economic Development and Cultural Change 61 (2): 311 –33. Klasen and Pieters 477 Fairlie, R. W. 2006. “An Extension of the Blinder-Oaxaca Decomposition Technique to Logit and Probit Models.” IZA Discussion Paper Series, No. 1917. Gaddis, I., and S. Klasen. 2014. “Economic Development, Structural Change, and Women’s Labor Force Participation: A Re-Examination of the Feminization U Hypothesis.” Journal of Population Economics 27 (3): 639– 81. Goldin, C. 1990. Understanding the Gender Gap: and Economic History of American Women. New York: Oxford University Press. Goldin, C. 1994. “The U-shaped Female Labor Force Function in Economic Development and Economic History.” NBER Working Paper Series, No. 4707. Heim, B. T. 2007. “The Incredible Shrinking Elasticities: Married Female Labor Supply, 1978–2002.” Journal of Human Resources 42 (4): 881– 918. Jensen, R. 2012. “Do Labor Market Opportunities Affect Young Women’s Work and Family Decisions: Experimental Evidence from India.” Quarterly Journal of Economics 127 (2): 753– 92. Kijima, Y. 2006. Why Did Wage Inequality Increase? Evidence from Urban India 1983– 99. Journal of Development Economics 81 (1): 97– 117. Kingdon, G. G. 2007. “The Progress of School Education in India.” Oxford Review of Economic Policy 23 (2): 168– 95. Klasen, S., and F. Lamanna. 2009. “The Impact of Gender Inequality in Education and Employment on Economic Growth: New Evidence for a Panel of Countries.” Feminist Economics 15 (3): 91– 132. Klasen, S., and J. Pieters. 2013. “What Explains the Stagnation of Female Labor Force Participation in Urban India?” IZA Discussion Paper No. 7597. Klasen, S., and C. Wink. 2003. “‘Missing Women’: Revisiting The Debate.” Feminist Economics (2– 3): 263–99. Klonner, S., and C. Oldiges. 2014. “Safety Net for India’s Poor or Waste of Public Funds? Poverty and Welfare in the Wake of the World’s Largest Job Guarantee Program.” University of Heidelberg Department of Economics Discussion Paper No. 564. Kucera, D., L. Roncolato, and E. Von Uexkull. 2012. “Trade Contraction and Employment in India and South Africa during the Global Crisis.” World Development 40 (6): 1122– 34. Lam, D., and S. Duryea. 1999. “Effects of Schooling on Fertility, Labor Supply, and Investments in Children, with Evidence from Brazil.” Journal of Human Resources 34 (1): 160–92. Luke, N., and K. Munshi. 2011. “Women as Agents of Change: Female Income and Mobility in India.” Journal of Development Economics 94 (1): 1 –17. Mammen, K., and C. Paxson. 2000. “Women’s Work and Economic Development.” Journal of Economic Perspectives 14 (4): 141– 64. National Sample Survey Organisation. 2010. “Migration in India, 2007– 2008.” NSS Report No. 533. NCEUS. 2007. “Report on Conditions of Work and Promotion of Livelihoods in the Unorganised Sector.” New Delhi, National Commission for Enterprises in the Unorganised Sector. Oaxaca, R. 1973. “Male-Female Wage Differentials in Urban Labor Markets.” International Economic Review 14 (3): 693–709. Oster, E., and B. Millet Steinberg. 2013. “Do IT Service Centers Promote School Enrollment? Evidence from India.” Journal of Development Economics 104: 123–35. Pieters, J. 2010. “Growth and Inequality in India: Analysis of an Extended Social Accounting Matrix. World Development 38 (3): 270– 81. Qian, N. 2008. “Missing Women and the Price of Tea in China: The Effect of Sex-Specific Earnings on Sex Imbalance.” Quarterly Journal of Economics 123 (3): 1251–85. Rosenzweig, M. R., and O. Stark. 1989. “Consumption Smoothing, Migration, and Marriage: Evidence from Rural India.” Journal of Political Economy 97 (4): 905 –26. Shastry, G. K. 2012. “Human Capital Response to Globalization: Education and Information Technology in India.” Journal of Human Resources 47 (2): 288–330. 478 THE WORLD BANK ECONOMIC REVIEW Singh, N., and M. K. Sapra. 2007. “Liberalization in Trade and Finance: India’s Garment Sector.” In: B. Harriss-White, and A. Sinha, eds., Trade Liberalization and India’s Informal Economy, 42 –127. New Delhi: Oxford University Press. Sudarshan, R. M., and S. Bhattacharya. 2009. “Through the Magnifying Glass: Women’s Work and Labor Force Participation in Urban Delhi.” Economic and Political Weekly 44 (48): 59 –66. Tam, H. 2011. “U-shaped Female Labor Participation with Economic Development: Some Panel Data Evidence.” Economic Letters 110 (2): 140– 42. World Bank. 2003. “Gender and Development in the Middle East and North Africa: Women in the Public Sphere.” World Bank, Washington, DC. ———. 2005. “Bridging the Gender Gap: Opportunities and Challenges.” Pakistan Country Gender Assessment, Report No. 32244-PAK. World Bank, Washington, DC. ———. 2011. World Development Report 2012: Gender Equality and Development, World Bank, Washington, DC. Young, A. 1995. The Tyranny of Numbers: Confronting the Statistical Realities of the East Asian Growth Experience.” Quarterly Journal of Economics 110 (3): 641–80.