77561 Disability, Poverty, and Schooling in Developing Countries: Results from 14 Household Surveys Deon Filmer Analysis of 14 household surveys from 13 developing countries suggests that 1 –2 percent of the population have disabilities. Adults with disabilities typically live in poorer than average households: disability is associated with about a 10 percentage point increase in the probability of falling in the two poorest quintiles. Much of the association appears to reflect lower educational attainment among adults with disabilities. People of ages 6 –17 with disabilities do not live in systematically weal- thier or poorer households than other people of their age, although in all countries studied they are signi�cantly less likely to start school or to be enrolled at the time of the survey. The order of magnitude of the school participation de�cit associated with disability—which is as high as 50 percentage points in 3 of the 13 countries—is often larger than de�cits related to other characteristics, such as gender, rural resi- dence, or economic status differentials. The results suggest a worrisome vicious cycle of low schooling attainment and subsequent poverty among people with disabilities in developing countries. JEL codes: O15, J14, I32, I20, I10 With more than 100 million primary school–age children not attending school worldwide (UNESCO 2005), the target of universal education, endorsed by more than 180 countries as a part of the Millennium Development Goals, remains elusive. Children with disabilities face particular hurdles in attending and completing school in developing countries. While there has been much dis- cussion about policy interventions to increase access to schooling for children with disabilities (see, for example, Peters 2003; World Bank 2003), little sys- tematic empirical analysis has been conducted on which to base this policy. The lack of analysis largely reflects the lack of appropriate and comparable data. Almost a decade ago, Elwan (1999) described the lack of empirical work Deon Filmer is a senior economist in the Development Research Group at the World Bank; his email address is d�lmer@worldbank.org. Tomoki Fuji, Johannes Hoogeveen, Elizabeth King, Daniel Mont, Susan Peters, and Dan Rees provided helpful comments on an earlier draft. Dilip Parajuli provided excellent research assistance. A supplemental appendix to this article is available at http:// wber.oxfordjournals.org/. THE WORLD BANK ECONOMIC REVIEW, VOL. 22, NO. 1, pp. 141 –163 doi:10.1093/wber/lhm021 Advance Access Publication January 15, 2008 # The Author 2008. 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@oxfordjournals.org 141 142 THE WORLD BANK ECONOMIC REVIEW on the association between disability and poverty in the developing world; such work is still missing.1 This study aims to start �lling some knowledge gaps using existing data on the prevalence of disability and its association with poverty and schooling in 12 developing, and 1 transition, countries. De�ning disability is complicated—and controversial. The purely medical de�nitions used in the past are giving way to de�nitions that incorporate con- tinuous measures of the activities that people can undertake, the extent of their participation in society and social and civic life, and the role of adaptive tech- nologies (Mont 2007). The World Health Organization’s International Classi�cation of Functioning, Disability and Health (ICF) describes disability as an umbrella term for impairments, activity limitations, and participation restrictions as part of a broader classi�cation scheme covering three main domains: body functioning and structure, activities and participation, and environmental factors.2 The interaction of aspects of all three of these domains determines individual welfare and social policy choices facing governments.3 The main goal of this article is descriptive. Many of the basic facts about disability, poverty, and schooling in developing countries are unknown or have not been systematically addressed. To contribute to the foundations of policy development, this article analyzes the data to investigate the interactions between impairment and schooling and their relation with poverty. It �nds that disability among youth is not typically associated with household poverty but that it is systematically and signi�cantly related to lower school participation, which in turn increases poverty in adulthood. The article is organized as follows. Section I compares de�nitions and the prevalence of disability across the household surveys covered. Section II investi- gates the association with poverty by examining the extent to which young people with disabilities live in households with lower economic status and the extent to which disability and schooling are related to poverty in adulthood. Section III investigates the association between disability and school partici- pation among school-age youth. Section IV draws conclusions and makes the case for better data. I . D ATA The data come from 14 nationally representative household surveys in 13 countries. Five surveys—from Bolivia (1997), Cambodia (2000), Chad (2004), Colombia (1995), and India (1992)—are Demographic and Health Surveys 1. An early exception is Afzal (1992), who analyzes disability and its correlates in Pakistan. Yeo and Moore (2003) review some of the literature on poverty and disability, but the work they refer to is typically not based on large-scale surveys. 2. A guide to the ICF is available at http://www3.who.int/icf/. 3. Haveman and Wolfe (2000, p. 998) emphasize that an economic de�nition of disability refers to characteristics that “constrain normal daily activities or cause substantial reduction in productivity on the job.� Filmer 143 (DHS). Two surveys—from Jamaica (1998) and Romania (1995)—are associ- ated with the Living Standards Measurement Study (LSMS) surveys. Two other surveys—from Burundi (2000) and Mongolia (2000)—are End of Millennium Multiple Indicator Cluster Surveys (MICS2) carried out under the guidance of the United Nations Children’s Fund (UNICEF). Five surveys—from Cambodia (1999), Indonesia (2000), Mozambique (1996), South Africa (1995), and Zambia (2003)—are national socioeconomic surveys (SES).4 These types of surveys are typically used to calculate poverty statistics or to derive basic health indicators, such as child mortality, or the use of health services; they underlie much empirical poverty and social analysis in developing economies. Most of the surveys have a sample size of about 5,000–25,000 households, with India (88,512 households) and Indonesia (65,762 households) as outliers (table 1). All DHS, LSMS, and MICS2 surveys were reviewed for questions on disabi- lity, with all surveys with a clear question on disability for the relevant age range included.5 The SES from Cambodia, Indonesia, Mozambique, South Africa, and Zambia are some of the most recent in the world with information on disability. There are relatively few data of this kind in developing countries: the datasets, and therefore the countries, for this analysis were selected on the basis of data availability and are not necessarily representative of developing countries in general. This is clearly a heterogeneous group of countries. The population living on less than $1 a day ranges from 2 percent in Jamaica and Romania to 55 percent in Burundi; under-�ve mortality—an indicator of basic health status— ranges from 20 per 1,000 live births in Jamaica to 212 per in Mozambique (see table 1). The sample includes �ve countries in Sub-Saharan Africa, four in Asia, three in Latin America and the Caribbean, and one in Eastern Europe. While heterogeneity has the advantage that the results will reflect a range of underlying conditions, it has the disadvantage that little draws these countries together other than the availability of data for this analysis. The de�nitions of disability in these datasets are most closely consistent with a focus on impair- ment—and as such fall mostly under ICF’s “body functioning and structure� domain. This is arguably an advantage, because impairment such as blindness or the lack of a limb is typically easy to verify. Selective misreporting of morbidity has long been recognized as a potential problem in studies of the relation between health and other socioeconomic 4. DHS data are available at http://www.measuredhs.com; LSMS data are available at http://www. worldbank.org/lsms; MICS2 data are available at http://www.childinfo.org; national SES are available from the countries’ national statistics of�ces. Despite the general consistency of DHS surveys across countries, disability is not a part of the “core� DHS module; information on disability is therefore not typically collected as a part of DHS. Identifying questions relating to disability required reviewing the country-speci�c components of the datasets. 5. Surveys with fewer than 50 disabled people between the ages of 6 and 17 were dropped from the analysis, because they represented too few observations on which to draw inferences. DHS data from Mozambique and MICS2 data from Myanmar and Sierra Leone were excluded on this basis. T A B L E 1 . Basic Statistics on Countries and Surveys 144 GDP per Population living Under-�ve Number of Type of Size of Number of people Country/year of Type of capita (2000 on less than a $1 a mortality rate (per households disability population ages 6 – 17 with survey survey PPP dollars) day (percent) 1,000 live births) surveyed covered ages 6 – 17 disabilities Bolivia 1997 DHS 2,349 20 105 12,028 All 16,605 75 Burundi 2000 MICS2 584 55 190 3,979 Physical 5,865 73 Cambodia 1999 SES 1,741 34 135 6,001 All 10,881 96 Cambodia 2000 DHS 1,859 34 135 12,236 Physical 23,765 214 Chad 2004 DHS 1,241 — 208 5,366 All 9,952 57 Colombia 1995 DHS 6,207 3 31 10,107 All 11,951 130 India 1992 NFHS 1,692 42 123 88,512 Visual, 140,297 1,337 (DHS) physical Indonesia 2003 SES 3,204 8 48 65,762 All 64,136 326 Jamaica 1998 LSMS 3,611 2 20 7,375 Physical, 6,964 58 mental THE WORLD BANK ECONOMIC REVIEW Mongolia 2000 MICS2 1,610 27 65 6,000 Visual, 7,645 245 hearing Mozambique SES 700 38 212 8,250 All 14,520 156 1996 Romania 1995 LSMS 6,210 2 26 24,560 All 13,777 82 South Africa SES 11,044 11 68 28,192 All 30,151 454 2005 Zambia 2003 SES 823 76 182 9,713 All 19,075 223 — not available. Note: PPP is purchasing power parity; DHS is Demographic and Health Survey; MICS2 is End of Millennium Multiple Indicator Cluster Survey; SES is Socioeconomic Survey; NFHS is National Family Health Survey; LSMS is Living Standards Measurement Study survey. Data for all countries except Burundi cover 6- to 17-year-olds; age range in Burundi is 6- to 14-year-olds. Poverty rates are for the following years: Bolivia 1997; Burundi 1998; Cambodia 1997; India 1993; Indonesia 2002; Jamaica 1999; Mongolia 1998; Mozambique 1996; Romania 1998; South Africa 2000; and Zambia 2003. Under-�ve mortality data on Burundi, Cambodia, Colombia, Indonesia, Jamaica, and South Africa are for 2000. Data on India are for 1990. Data on Bolivia, Mozambique, and Romania are for 1995. Data on Chad, South Africa, and Zambia are for 2005. “All� types of disabilities include visual, hearing, speech, physical, and mental disabilities. See appendix table A-1 for the precise wording and types of disabilities covered by each survey. Source: World Bank (various years); author’s analysis based on data described in the text. Filmer 145 characteristics (Gertler, Rose, and Glewwe 2000). To overcome this problem Gertler and Gruber (2002) use responses to questions regarding activities of daily living when analyzing the impact of major illness on household consump- tion in Indonesia. Yount and Agree (2005) use activities of daily living in ana- lyzing gender differences in disability among the elderly in Egypt and Tunisia. Despite the relative ease of verifying the types of disabilities in the study datasets, it is nevertheless possible that there is selective reporting. Some respondents and interviewers, for example, might interpret blindness as partial sight, whereas others might interpret it to mean complete inability to see. It is also possible that mental disability is selectively recognized and reported by some respondents. Selective reporting is usually assumed to result in higher reporting of disabilities by wealthier socioeconomic groups. Under this assump- tion the estimates reported here would underestimate the relation between dis- ability and poverty.6 Despite the fact that all 14 surveys have an impairment de�nition of disabil- ity, substantive differences remain across datasets. Nine surveys use an “exten- sive� de�nition that includes visual, hearing, speech, physical, and mental disability. But even within this group, the de�nition of each type of impairment varies. In the Cambodia SES, for example, the physical disability category con- tains a detailed list of potential cases (“amputation of one limb; amputation of more than one limb; unable to use one limb; unable to use more than one limb; paralyzed lower limbs only; paralyzed all four limbs�). In contrast, in Jamaica a single category covers “physical or mental disability.� More gener- ally, in some countries the de�nition is stricter than in others. In Mongolia sight and hearing are described as “with dif�culty�; in other surveys they are characterized as “blind� and “deaf� (the wording of the questions on disability in these surveys is provided in appendix table A-1). The second main data constraint is the fact that the surveys do not typically identify large numbers of people with disabilities. Any subsequent analysis therefore suffers from imprecision. The smallest number of cases of disability among 6- to 17-year-olds are in Chad (57) Jamaica (58), Burundi (73), and Bolivia (75). Standard errors are often large for the results reported below, although the main �nding—the de�cit in school participation among people with disabilities—is consistently statistically signi�cant. 6. A large body of literature covers the selective reporting of disability in the context of social programs targeting disability. Higher bene�ts are typically hypothesized to result in higher rates of reported disability. For a recent empirical demonstration, see Duggan, Rosenheck, and Singleton (2006). Some program aspects, such as hurdles in accessing bene�ts, may reduce self-declared disability (see the discussion in Parsons 1991). At least one study (Benitez-Silva and others 2004) �nds no systematic bias in self-reported disability compared with bureaucratic assessment among adult applicants for Social Security bene�ts in the United States. Reported disability might also be an unintended consequence of a different set of programs. Figlio and Getzler (2002) argue that increases in the use of learning achievement tests for school accountability in the United States has led to an increase in reported disability among students, because schools can exclude these students from average scores. 146 THE WORLD BANK ECONOMIC REVIEW A last data constraint concerns the measurement of household poverty. All LSMS and SES surveys include household per capita consumption expendi- tures, the variable typically used in poverty analysis. For these datasets econ- omic status quintiles based on per capita consumption expenditures can therefore be used. DHS and MICS2 data do not include consumption expendi- tures. For these datasets, an index of household consumer assets and housing characteristics (an economic status index) are used to classify households into quintiles (following Filmer and Pritchett 2001). In the Cambodia SES, which encountered problems collecting consumption data, and the South Africa SES, which does not include a full consumption module, an index based on assets and housing quality variables is also used.7 I I . P R E VA L E N C E O F D I S A B I L I T Y A N D A S S O C I A T I O N W I T H H O U S E H O L D E CO N O M I C S TAT U S The �rst issue these data are used to explore is the prevalence of disability and its association with household economic status among youth 6- to 17-years-old. The next issue is the relation among disability, poverty, and education attain- ment among adults. Prevalence of Disability among 6- to 17-Year-Olds Estimates of the prevalence of disability range from 0.49 percent (Chad) to 3.2 percent (Mongolia) (table 2). These �gures are consistent with those that appear in the United Nations statistical database on disability (DISTAT).8 In that database, which compiles results from more than 65 surveys and censuses in developing economies between 1970 and 1992, the mean prevalence rate is 1.7 percent for the entire population and 0.7 for children under 14. Using a de�nition of disability consistent with the one adopted in the datasets analyzed here, LeRoy, Evans, and Deluca (2000) �nd a prevalence of disability among 5- to 15-year-olds of 2.07–2.62 percent in Ireland, Italy, Switzerland, and the United States.9 Perhaps surprisingly, of the 14 surveys analyzed here, those that list more types of impairments do not systematically identify a higher percentage of the population as disabled. In countries that include visual, hearing, speech, physi- cal, and mental disabilities, for example, prevalence ranges from 0.49 in Bolivia and Chad to 1.38 in South Africa—close to the entire range across all 7. Consistent with typical poverty analysis, quintiles are derived on the basis of the distribution of people across the economic status measure. 8. The database is available at http://unstats.un.org/unsd/demographic/sconcerns/disability/disab2. asp. A summary of the country-level DISTAT information is available in the supplemental appendix to this article, accessible at http://wber.oxfordjournals.org/. 9. The U.S. rate of 2.1 rises to 4.4 percent if “speech and language disability� (a separate category from “mute and deaf/mute�) is included. When people with dif�culty in learning, remembering, or concentrating are added, the rate increases to about 6 percent (Freedman, Martin, and Schoeni 2004). T A B L E 2 . Prevalence of Disability among 6- to 17-Year-Olds, by Household Economic Status Quintile p-values on economic status variablesa Quintile 1 Quintile 5 Dummy Country All (poorest) Quintile 2 Quintile 3 Quintile 4 (richest) variables Continuous Bolivia 0.49 (0.07) 0.45 (0.12) 0.43 (0.12) 0.54 (0.15) 0.64 (0.19) 0.39 (0.14) 0.82 0.78 Burundib 1.24 (0.21) 1.28 (0.50) 1.19 (0.69) 1.19 (0.42) 1.28 (0.36) 1.28 (0.30) 1.00 0.62 Cambodia (Socioeconomic 0.87 (0.12) 0.91 (0.24) 0.84 (0.24) 0.87 (0.34) 0.81 (0.28) 0.94 (0.18) 1.00 0.95 Survey) Cambodia 0.86 (0.07) 1.00 (0.17) 1.01 (0.16) 0.77 (0.15) 0.65 (0.14) 0.90 (0.16) 0.40 0.39 Chad 0.49 (0.09) 0.46 (0.19) 0.32 (0.14) 0.64 (0.22) 0.49 (0.20) 0.55 (0.12) 0.72 0.70 Colombia 1.08 (0.10) 1.24 (0.24) 1.34 (0.25) 1.03 (0.22) 0.72* (0.17) 1.05 (0.26) 0.24 0.15 India 1.03 (0.04) 1.20 (0.09) 1.13 (0.09) 0.92** (0.08) 1.03 (0.07) 0.84*** (0.08) 0.01 0.01 Indonesia 0.50 (0.03) 0.70 (0.08) 0.55 (0.08) 0.41** (0.07) 0.50* (0.08) 0.38*** (0.06) 0.02 0.08 Jamaica 0.82 (0.11) 1.01 (0.28) 1.06 (0.26) 0.48 (0.18) 0.68 (0.24) 0.88 (0.30) 0.29 0.63 Mongolia 3.20 (0.27) 3.40 (0.57) 3.01 (0.50) 2.88 (0.56) 2.81 (0.50) 3.92 (0.63) 0.62 0.14 Mozambique 1.19 (0.13) 0.87 (0.17) 0.81 (0.20) 1.58* (0.36) 1.39 (0.29) 1.29 (0.28) 0.14 0.60 Romania 0.60 (0.07) 0.91 (0.19) 0.47* (0.13) 0.54 (0.16) 0.47* (0.13) 0.58 (0.14) 0.38 0.13 South Africa 1.38 (0.09) 1.50 (0.20) 1.46 (0.19) 1.67 (0.21) 1.22 (0.17) 1.06 (0.24) 0.27 0.26 Zambia 1.32 (0.11) 1.46 (0.26) 1.32 (0.22) 1.40 (0.28) 1.24 (0.23) 1.16 (0.22) 0.88 0.36 Difference with poorest quintile: ***Statistically signi�cant at the 1 percent level; **statistically signi�cant at the 5 percent level; *statistically signi�- cant at the 10 percent level. Note: The Numbers in parentheses are robust standard errors. a The p-values report of the test of joint sigini�cance of the set of dummy variables for quinitiles 2– 5 and of the continuous measure of economic status and its square in a probit regression of disablity on economic status. b Age range Burundi is 6 – 14. Filmer Source: Author’s analysis based on data described in the text. 147 148 THE WORLD BANK ECONOMIC REVIEW the surveys. The highest prevalence rate in this collection of datasets is observed in Mongolia (3.20 percent), which includes only visual and hearing impairments, while the rates are lower in Burundi (1.24 percent) and the Cambodia DHS (0.86 percent), which cover only physical disabilities. Of course, this variability combines both actual differences in prevalence and differences in survey techniques. In Cambodia, which conducted two surveys separated by only one year, the survey with the more extensive de�- nition of disability does not yield the higher prevalence: the SES in 1999, with a broad de�nition of disability, identi�es 1.51 percent of the population as dis- abled, whereas the DHS in 2000, with a narrow de�nition (restricted to physi- cal disabilities), identi�es 1.57 percent of the population as disabled. Clearly, there is substantial variation across surveys in how people with disabilities are identi�ed; cross-country comparisons in prevalence can be made only with caution.10 Despite the lack of cross-country comparability in the de�nitions and measurement of disability, these surveys are still useful in describing the associ- ation of disability with other characteristics. That is, conditional on a particu- lar de�nition, the analysis is valid for a given survey because the de�nition is common to all individuals in that survey. Moreover, it is less likely that cross- country comparisons of the association between disability and other characte- ristics suffer from these problems. But even this comparison needs to be treated with some caution: if, for example, some types of disabilities are more closely associated with a correlate than others, then surveys that include that type of disability will show a higher association with the correlate than those that do not. If loss of a limb is more closely associated with poverty than are other types of impairments, for example, then (everything else being equal) a survey that includes loss of a limb in its de�nition of disability will yield a higher cor- relation between disability and poverty. Do Youth with Disabilities Live in Poorer Households? The prevalence of disability among 6- to 17-year-old tends to be slightly lower in richer quintiles, but the association is not always monotonic. Moreover, India and Indonesia are the only countries in which the prevalence of disability in the richest quintile is statistically signi�cantly different from that in the poorest quintile. In India, prevalence in the richest quintile is about a third of that in the poorest quintile; in Indonesia it is about half (see table 2). Two additional tests of the association between disability and poverty were carried out. The �rst entailed estimating a probit regression of disability on dummy variables for living in a household of each economic status quintile and then determining the joint statistical signi�cance of these dummy variables (that is, a joint version of the quintile-by-quintile tests). This approach allows 10. Developing good data on disability is dif�cult: United Nations (2001) contains a guide to doing so. See Mont (2007) for a recent review. Filmer 149 for a great degree of nonlinearity in the association, because the coef�cient can be different for each quintile and the approach does not impose monotonicity. However, it is possible that the small number of young people with disabilities means that there will be very few cases within each quintile and that therefore even a joint test of the coef�cients on the quintile dummy variables may not have enough power to identify a signi�cant association. To address this potential problem, the second approach entailed a probit regression of disability on the continuous variable measuring economic status ( per capita household expenditures or an index of assets and housing charac- teristics) and its square. This approach allows less flexibility, but it does not rely on quintile-speci�c estimates of prevalence, which may be imprecise. In both tests, India and Indonesia are the only countries in which either the joint test on the dummy variables or on economic status and its square yield a statistically signi�cant association. These are the countries with the largest sample sizes, which gives rise to the concern that it is simply the power of the test that is low in the other countries. However, as discussed below, the same datasets yield large and statistically signi�cant gaps in schooling in all countries, suggesting that it is not simply an issue of power.11 In sum, these results do not suggest that, as a general rule, youth with disabilities are more likely to live in poorer households, although this is the case in two of the 14 datasets. Disability, Poverty, and Schooling in Adulthood Disability is a strong correlate of poverty in adulthood. Haveman and others (1999) show that income in the United States in 1991 of households headed by a person with a disability was roughly half the mean for the population as whole (even after accounting for transfers from social programs) and the rate of poverty about twice as high as the overall population’s. Hoogeveen (2005) esti- mates that 42 percent of households headed by a person with a disability were poor in Uganda in 1991 but that just 25 percent of other households were.12 The analysis of the relation between disability and economic status should be interpreted as an association and not necessarily a cause or consequence. Disability is both a determinant of poverty, because it lowers earning power and consumption expenditures (Haveman and Wolfe 2000; Gertler and Gruber 2002), and a consequence of poverty, because the cumulative deprivations of 11. The results are consistent if (as suggested by a referee) one estimates the reverse regression of the probability of being poor as a function of whether a school-age child has a disability controlling for other covariates (that is, a speci�cation analogous to the schooling model estimated below). In this approach, disability is statistically insigni�cantly related to poverty in all the countries when poverty is de�ned as being in the poorest two quintiles. The association is statistically signi�cant in India and Indonesia if poverty is de�ned as being in the poorest quintile (results are available in the supplemental tables of this article, available at http://wber.oxfordjournals.org/). 12. Hoogeveen (2005, p. 606) de�nes disability among household heads differently than the datasets used here. In that survey, a head of household is considered disabled if the disability “prevents him or her from being actively engaged in labor activities during the past week.� 150 THE WORLD BANK ECONOMIC REVIEW poverty such as inadequate infant or child development, or exposure to danger- ous working conditions, can manifest themselves in disability. Moreover, the presence of a person with a disability entails direct costs, which lower stan- dards of living (Jones and O’Donnell 1995; Haveman and Wolfe 2000; Zaidi and Burchardt 2005). Twelve of the datasets include information on the disability status of adults. A probit regression model was estimated in which the dependent variable is a dummy variable equal to 1 if a person of 20 –50 lives in a household in the two poorest quintiles.13 The �rst model estimates the association with disability after controlling for a set of basic characteristics: age, age-squared, a dummy variable for being male, urban residence, and dummy variables for region of residence. Disability is statistically signi�cantly related to an increase in the probability of being poor in eight of the datasets (table 3). Among these, having a disability is associated with a 5.0–14.5 percentage point increase in the probability of being in the two poorest quintiles. In the remaining four datasets the association is positive but not statistically signi�cant. The second model adds years of schooling completed to the set of correlates of poverty, thereby controlling for the effect of schooling on poverty. The coef- �cient on disability becomes statistically insigni�cant in several countries—and turns from being signi�cantly positive to signi�cantly negative in several others.14 In all cases, the coef�cient on years of schooling is statistically signi- �cantly negative: each additional year of schooling is associated with about a 2–5 percentage point reduction in the probability of being in the two poorest quintiles.15 These results suggest that disability and poverty are related in adulthood and that much of this association is mediated by education: after accounting for the lower educational attainment of adults with a disability there is no longer a sys- tematic positive relation between disability and poverty. This �nding suggests that ensuring that youth with disabilities do not have lower educational attain- ment could be a powerful way to reduce the likelihood that they live in poverty as adults. The next section shows just how great a challenge this is. III. DISABILITY AND SCHOOLING Consider now the relation between disability and schooling among youth (table 4). Six- to seventeen-year-olds with disabilities are almost always substantially less likely to be in school than their peers without disabilities. 13. Qualitatively similar results are obtained if just the poorest quintile is used. 14. This effect is caused by years of schooling alone. The results are qualitatively similar if only years of schooling enter the model and all other correlates are dropped. 15. The results presented in the supplemental appendix, available at http://wber.oxfordjournals.org/, do not support the notion that there is an interactive effect between disability and schooling: when included, an interaction term is always insigni�cantly different from zero and small in magnitude in all countries. T A B L E 3 . Association between Disability, Schooling, and the Probability of Being in the Bottom Two Economic Status Quintiles among Adults 20 –50 Model 2, including schooling and other control Model 1, excluding schooling but variables including other variables Coef�cient on: Coef�cient on: years of Number of Number of people Country Coef�cient on: disability (1) disability (2) schooling (3) observations (4) with disabilities (5) Bolivia 0.145*** (0.052) 0.013 (0.057) 2 0.037*** (0.002) 18,632 179 Cambodia 0.015 (0.040) 0.013 (0.046) 2 0.027*** (0.003) 9,195 251 (Socioeconomic Survey) Cambodia 0.129*** (0.028) 0.092*** (0.029) 2 0.047*** (0.002) 23,191 504 Chad 0.056 (0.077) 0.021 (0.074) 2 0.037*** (0.005) 7,128 138 Colombia 0.131*** (0.033) 2 0.052* (0.031) 2 0.050*** (0.002) 18,807 309 India 0.068*** (0.012) 0.052*** (0.012) 2 0.049*** (0.001) 199,140 309 Indonesia 0.083*** (0.024) 2 0.030 (0.023) 2 0.042*** (0.001) 121,964 774 Jamaica 0.133*** (0.048) 0.084* (0.048) 2 0.024*** (0.002) 10,197 188 Mozambique 0.015 (0.032) 0.006 (0.032) 2 0.019*** (0.003) 13,909 366 Romania 0.050* (0.028) 2 0.068** (0.028) 2 0.046*** (0.002) 30,584 356 South Africa 0.097*** (0.018) 2 0.020 (0.020) 2 0.061*** (0.002) 44,539 1,766 Zambia 0.015 (0.034) 2 0.039 (0.035) 2 0.032*** (0.002) 18,596 377 ***Statistically signi�cant at the 1 percent level; **statistically signi�cant at the 5 percent level; *statistically signi�cant at the 10 percent level. Note: The Numbers in parentheses are robust standard errors. Explanatory varible differ across models. Model 1 includes disability, age, age-squared, and dummy varibles for gender, urban residence, and region. Model 2 includes years of schooling and all the variable from model 1. Source: Author’s analysis based on data described in the text. Filmer 151 152 THE WORLD BANK ECONOMIC REVIEW The shortfall among children aged 6 –11 ranges from 10 percentage points in India to almost 60 percentage points in Indonesia. The gap is also large among older children, ranging from 15 percentage points in Cambodia to 58 percen- tage points in Indonesia (exceptions are India, where the gap is just 4 percen- tage points, and Burundi, where there is no gap). The gaps are typically larger among the older group: the median shortfall is 21 percentage points among 6- to 11-year-olds and 25 percentage point among 12- to 17-year olds. Schooling De�cits Controlling for Individual, Household, and Community Characteristics To the extent that disability is correlated with other factors that affect school- ing, such as poverty, age, and urban or rural residence, the raw difference in school participation between children with and without disabilities may give a misleading picture. For each survey, an adjustment was carried out by estimat- ing a multivariate probit model with school participation as the dependent variable and an indicator of disability as the explanatory variable (table 5). The estimates also include, as explanatory variables, the potentially confound- ing variables—age and age-squared, a dummy variable for a child’s gender, a dummy variable for urban residence, dummy variables for each economic status quintile, and dummy variables for region of residence.16 The adjusted de�cit in school participation is more than 50 percentage points in Bolivia, Indonesia, and Romania; 24 –45 percentage points in Cambodia, Colombia, Jamaica, Mongolia, South Africa, and Zambia; 14 –18 percentage points in Burundi, Chad, and Mozambique; and 8 percentage points in India. In all countries, the difference is large and statistically signi�- cantly different from zero. In most countries, the unadjusted de�cits are of comparable orders of magnitude: the estimated de�cits are usually smaller after adjusting for confounding variables, but the effect of the adjustment is not typically large. The results for the probability that a person has ever attended school are similar to those for current school participation. As the de�cit is of a similar order of magnitude, the results imply that a substantial part of the de�cit in schooling attainment among people with disabilities comes from the fact that they never attended school at all. Analysis of the Kaplan-Meier grade survival curves (table S.2 in the supplemental appendix, available at http://wber.oxford- journals.org/) suggests that most of the difference in attainment can be attribu- ted to the decision (or the ability) to enter school. Nevertheless, in seven countries, the de�cit at grade 1 widens as children progress through the school system: in Bolivia, Colombia, Indonesia, Jamaica, Romania, South Africa, and Zambia the de�cit associated with disability increases by about 7 –10 16. Similar results ( presented in the supplemental appendix, available at http://wber.oxfordjournals. org/) are found if nearest neighbor matching (using the same set of explanatory variables for matching) is used. T A B L E 4 . Percentage of 6- to 17-Year-Olds Reported to Be in School, by Country Age 6 – 11 Age 12– 17 Country Without disability With disability Difference Without disability With disability Difference Bolivia 0.95 (0.00) 0.38 (0.08) 0.57*** (0.08) 0.83 (0.01) 0.39 (0.11) 0.44*** (0.11) Burundia 0.38 (0.01) 0.19 (0.06) 0.19*** (0.06) 0.48 (0.02) 0.48 (0.10) 0.00 (0.10) Cambodia (Socioeconomic Survey) 0.58 (0.01) 0.18 (0.06) 0.40*** (0.06) 0.68 (0.01) 0.31 (0.08) 0.37*** (0.08) Cambodia 0.67 (0.01) 0.38 (0.06) 0.29*** (0.06) 0.62 (0.01) 0.47 (0.06) 0.15*** (0.06) Chad 0.36 (0.02) 0.24 (0.12) 0.12 (0.12) 0.43 (0.03) 0.25 (0.09) 0.18* (0.10) Colombia 0.92 (0.01) 0.56 (0.08) 0.36*** (0.08) 0.74 (0.01) 0.29 (0.06) 0.45*** (0.06) India 0.70 (0.01) 0.60 (0.02) 0.10*** (0.02) 0.35 (0.00) 0.32 (0.02) 0.04* (0.02) Indonesia 0.89 (0.00) 0.29 (0.04) 0.59*** (0.04) 0.76 (0.00) 0.18 (0.04) 0.58*** (0.04) Jamaica 0.99 (0.00) 0.71 (0.09) 0.29*** (0.09) 0.86 (0.01) 0.50 (0.11) 0.36*** (0.11) Mongolia 0.58 (0.01) 0.41 (0.05) 0.17*** (0.04) 0.73 (0.02) 0.47 (0.05) 0.26*** (0.05) Mozambique 0.49 (0.01) 0.34 (0.08) 0.15** (0.08) 0.48 (0.01) 0.29 (0.06) 0.19*** (0.06) Romania 0.79 (0.01) 0.58 (0.11) 0.22** (0.11) 0.84 (0.01) 0.36 (0.07) 0.48*** (0.07) South Africa 0.96 (0.00) 0.76 (0.04) 0.21*** (0.04) 0.95 (0.00) 0.70 (0.04) 0.25*** (0.04) Zambia 0.62 (0.01) 0.42 (0.06) 0.20*** (0.06) 0.75 (0.01) 0.58 (0.06) 0.17*** (0.06) ***Statistically signi�cant at the 1 percent level; **statistically signi�cant at the 5 percent level; *statistically signi�cant at the 10 percent level. Note: Numbers in parentheses are robust standard errors. a Age range in Burundi is 6 – 14. Source: Author’s analysis based on data described in the text. Filmer 153 154 T A B L E 5 . Schooling De�cits among 6- to 17-Year-Olds with Disabilities Current school participation Ever-attended school Average among Average among 6- to 17-year olds 6- to 17-year olds without Unadjusted De�cit adjusted for without Unadjusted De�cit adjusted for Country disabilities (1) de�cit (2) other factors (3) disabilities (4) de�cit (5) other factors (6) Bolivia 0.90 –0.51*** (0.07) – 0.61*** (0.08) 0.98 – 0.47*** (0.07) – 0.46*** (0.08) Burundia 0.41 –0.12** (0.06) – 0.16*** (0.05) 0.45 – 0.14** (0.06) – 0.19*** (0.05) Cambodia 0.63 –0.39*** (0.05) – 0.45*** (0.05) 0.72 – 0.46*** (0.05) – 0.56*** (0.06) (Socioeconomic Survey) Cambodia 0.65 –0.22*** (0.04) – 0.26*** (0.05) 0.74 – 0.20*** (0.04) – 0.31*** (0.05) THE WORLD BANK ECONOMIC REVIEW Chad 0.39 –0.14* (0.08) – 0.14* (0.08) 0.43 – 0.12* (0.07) – 0.13 (0.09) Colombia 0.83 –0.42*** (0.05) – 0.43*** (0.06) 0.95 – 0.47*** (0.05) – 0.48*** (0.06) India 0.54 –0.08*** (0.02) – 0.08*** (0.02) 0.74 – 0.05*** (0.02) – 0.07*** (0.02) Indonesia 0.82 –0.59*** (0.03) – 0.67*** (0.03) 0.94 – 0.46*** (0.03) – 0.53*** (0.05) Jamaica 0.93 –0.33*** (0.08) – 0.28*** (0.09) 1.00 – 0.25*** (0.07) – 0.19*** (0.06) Mongolia 0.65 –0.20*** (0.04) – 0.28*** (0.04) 0.80 – 0.17*** (0.04) – 0.37*** (0.05) Mozambique 0.49 –0.18*** (0.05) – 0.18*** (0.05) 0.61 – 0.12** (0.05) – 0.14*** (0.05) Romania 0.82 –0.39*** (0.06) – 0.53*** (0.07) 0.90 – 0.30*** (0.06) – 0.50*** (0.07) South Africa 0.96 –0.23*** (0.03) – 0.25*** (0.03) 0.85 – 0.20*** (0.03) – 0.30*** (0.05) Zambia 0.68 –0.18*** (0.04) – 0.24*** (0.05) 0.77 – 0.16*** (0.04) – 0.23*** (0.05) ***Statistically signi�cant at the 1 percent level; **statistically signi�cant at the 5 percent level; *statistically signi�cant at the 10 percent level. Note: Adjusted differentials correspond to the marginal effect of disability in a probit regression of school participation that includes age, age-squared, and dummy variables for gender, urban residence, economic quintile, and region. Numbers in parentheses are robust standard errors. a Age range in Burundi is 6 – 14. Source: Author’s analysis based on data described in the text. Filmer 155 percentage points between grades 1 and 8.17 These countries have relatively high grade 1 completion rates, suggesting that even in countries that are able to get most children into school, special effort may be needed to get children with disabilities into school and to keep them there. There is substantial heterogeneity across countries in the schooling de�cit associated with disability. Part of this variation might be due to differences in the de�nition of disability. In a survey with a more stringent de�nition of disability, one might observe a larger de�cit, because such a survey would identify individ- uals who would have to overcome greater obstacles in order to access education. The fact that the two surveys from Cambodia yield schooling de�cits that are about 20 percentage points apart suggests that this is a plausible explanation. Another part of this variation likely relates to overall enrollment. It would not be surprising to observe larger de�cits in countries in which enrollment among children without disabilities is high: in these countries there would be more scope to observe a bigger de�cit. The schooling de�cit does tend to be smaller in countries with the lowest overall enrollment (Burundi, Chad, India, and Mozambique) and larger in countries with higher overall enrollment (Bolivia, Indonesia, and Romania): the correlation between the school partici- pation de�cit and the level of participation among youth without a disability is about –0.4 across the 14 datasets (the correlation is similar for the probability of ever having attended school). But the relation is not perfect: in Jamaica and South Africa, for example, where overall school participation is high, the de�cit associated with disability is about average for the surveys analyzed here. Finally, part of the variation in the schooling de�cit associated with disabil- ity is likely related to differences in the social and policy environment. Countries in which there is greater stigma toward people with disabilities or less effort has been made to ensure their access to schooling will undoubtedly have larger de�cits associated with schooling. However, this is but one of many potential explanations of why results might differ across countries. “Endogeneity� of Disability and Schooling Disability among 6- to 17-year olds could be partly the result of poverty, which may have a direct effect on the probability of school attendance. Adjusting for economic status in estimating the association between schooling and disability mitigates the potential for such bias. More generally, however, it is possible that other—unobserved—factors affect both the probability of being disabled and the probability of attending school. Indeed, households that disfa- vor investing in both children’s health and education in favor of other types of expenditures are more likely to have infants and children with poor health— who might develop a disability as a result—and low schooling. In this case dis- ability and schooling would be related, but the association would merely reflect 17. The de�cit in grade 1 ranges from 15 percentage points in Zambia to 48 percentage points in Bolivia. 156 THE WORLD BANK ECONOMIC REVIEW parental neglect (see Strauss and Thomas 1995; Haddad, Hoddinott, and Alderman 1997). One way to address this potential problem is to use a household �xed-effects approach. Such an approach controls for all—observed and unobserved— household-level characteristics common to all children in a household. In such a model the source of identi�cation of the difference in school participation is between children with and children without a disability in the same household. Any generalized household-speci�c above- or below-average investment in chil- dren will have been netted out. Implementing such an approach involves revis- ing the set of control variables used in the “adjusted� models reported in table 5 and replacing all household-, community-, and regional variables with a set of dummy variables each equal to one for each household.18 A household �xed-effects speci�cation can be estimated only on the sub- sample of households that include at least one youth with disabilities and one without (table 6). The results of re-estimating the basic multivariate results without household �xed-effects on the subsample are consistent with those obtained using the full sample, despite the potentially selected nature of this subsample. The household �xed-effects results for current school participation and for ever-attended school are likewise similar to those that exclude �xed-effects for the subsample. In one country (Burundi) the magnitude of the effect increases, in another (Mozambique) it decreases. But in most countries the estimated impact is virtually indistinguishable, suggesting that the associ- ation between disability and schooling among 6- to 17-year-olds is not simply a reflection of �xed household attributes, such as parental neglect, but rather a more direct effect of disability on schooling. Relative Magnitude of School Participation De�cits How large is the de�cit in school participation relative to other sources of inequality? The multivariate models can be used to compare school partici- pation gaps associated with disability, gender, urban or rural residence, and economic status (�gure 1).19 The de�cit associated with disability is clearly large compared with other sources of inequality. In almost all countries it is larger than the de�cit associ- ated with being female (which is a “surplus� in some countries). In most countries it is substantially larger than the de�cit associated with rural residence, and it is usually larger even than the gap between the poorest and richest quin- tiles, typically one of the strongest predictors of enrollment.20 The exceptions are Burundi, Chad, and particularly India, where wealth gaps are larger than all other gaps; Burundi and Mozambique, where rural-urban gaps are larger than 18. In this section, the probit model is replaced by a linear probability model. 19. In each case the de�cit is estimated at the means of the other variables. 20. See Filmer (2005) for a comparison of wealth and gender gaps. See Ainsworth and Filmer (2006) for a comparison of the gaps associated with wealth and with orphan status. T A B L E 6 . Effect of Disability on Schooling of 6- to 17-Year-Olds in Households with at Least One Child with and One Child without a Disability Current school participation Ever attended school Basic Household Basic Household Total number Number ages multivariate �xed-effects multivariate �xed-effects of observations 6 –17 with Country speci�cation (1) speci�cation (2) speci�cation (3) speci�cation (4) (5) disabilities (6) Bolivia – 0.50*** (0.08) – 0.49*** (0.09) – 0.43*** (0.08) – 0.44*** (0.10) 187 61 Burundia – 0.24*** (0.06) – 0.31*** (0.08) – 0.22*** (0.06) – 0.30*** (0.08) 136 61 Cambodia (Socioeconomic Survey) – 0.36*** (0.06) – 0.39*** (0.07) – 0.40*** (0.06) – 0.42*** (0.07) 265 82 Cambodia – 0.21*** (0.04) – 0.22*** (0.05) – 0.23*** (0.04) – 0.23*** (0.04) 649 189 Chad – 0.10 (0.10) – 0.07 (0.11) – 0.07 (0.10) – 0.03 (0.11) 218 52 Colombia – 0.39*** (0.05) – 0.39*** (0.07) – 0.48*** (0.06) – 0.48*** (0.07) 276 98 India – 0.05*** (0.02) – 0.06** (0.02) – 0.04** (0.02) – 0.05** (0.02) 3,574 1,138 Indonesia – 0.51*** (0.04) – 0.51*** (0.06) – 0.43*** (0.04) – 0.44*** (0.06) 545 208 Jamaica – 0.34*** (0.12) – 0.40*** (0.12) – 0.28*** (0.09) – 0.34*** (0.11) 110 42 Mongolia – 0.25*** (0.04) – 0.24*** (0.05) – 0.28*** (0.04) – 0.28*** (0.04) 557 201 Mozambique – 0.23*** (0.06) – 0.13** (0.06) – 0.16** (0.05) – 0.11* (0.06) 370 121 Romania – 0.41*** (0.09) – 0.44*** (0.10) – 0.34*** (0.07) – 0.36*** (0.10) 136 52 South Africa – 0.18*** (0.03) – 0.21*** (0.04) – 0.19*** (0.03) – 0.21*** (0.04) 1,039 361 Zambia – 0.17*** (0.05) – 0.20*** (0.05) – 0.18*** (0.04) – 0.19*** (0.05) 700 253 ***Statistically signi�cant at the 1 percent level; **statistically signi�cant at the 5 percent level; *statistically signi�cant at the 10 percent level. Note: Coef�cients are from linear probability models. Basic speci�cation includes age, age-squared, and dummy variables for gender, urban residence, economic quintile, and region; household �xed-effects speci�cation includes age, age-squared, and a dummy variable for gender. Numbers in parentheses are robust standard errors. a Age range in Burundi is 6 – 14. Filmer Source: Author’s analysis based on data described in the text. 157 158 THE WORLD BANK ECONOMIC REVIEW F I G U R E 1. De�cits in School Participation Associated with Various Characteristics Note: De�cits shown are the marginal effects of dummy variables for each characteristic in multivariate probit models for 6- to 17-year-olds, except in Burundi, where the sample covers children ages 6 – 14. Source: Author’s analysis based on data described in the text. those for disability; and Mongolia and Zambia, where wealth gaps are only slightly smaller than those for disability. In most countries, however, the gap in school participation between children with and without disabilities is about twice as large as that associated with rural residence or wealth.21 21. An interesting additional question would be whether disability interacts with other characteristics in a way that reduces or exacerbates inequalities. A straightforward way to investigate this hypothesis is to estimate the multivariate model of school participation and include interaction terms between disability and each of the other covariates. Given the relatively small number of sample observations, however, the data do not typically yield statistically signi�cant interaction terms—even when the magnitude of the interaction is relatively large—suggesting an inability to estimate these interactions with much precision. Because it is hard to assess whether this is caused solely by statistical power, these results are not reported here. They are reported in the supplemental appendix, available at http://wber.oxfordjournals.org/. Filmer 159 I V. C O N C L U S I O N S This analysis of data from 14 nationally representative household surveys con- �rms the many data problems that earlier research has identi�ed as hampering the establishment of a broad empirical base for developing policies targeted to people with disabilities in poor countries. The variation across surveys in the de�nition of disability makes cross-country comparisons dif�cult. The small number of people identi�ed as disabled in surveys makes it hard to precisely estimate patterns in the data beyond simple correlations. Despite these limitations, but keeping them in mind, the data are neverthe- less revealing. Consistent with similar surveys the 14 surveys analyzed here identify about 1–2 percent of the population as having a disability. One country with two surveys and varying de�nitions suggests that the percentage is not always sensitive to the exact de�nition: different de�nitions can yield similar prevalence rates, and similar de�nitions can yield different prevalence rates. In addition, other aspects of the surveys, such as the training of enumer- ators or the use to which interviewees expect the survey to be put, might affect the overall estimated prevalence rates. Analysis of these datasets provides little evidence to suggest that children with disabilities are generally more or less likely to live in richer or poorer households. Adults with disabilities do typically live in poorer households, but much of this association appears to come from the fact that they have lower educational attainment. Given this �nding, it is particularly worrisome that children with disabilities are almost always much less likely to participate in schooling than are other children. They are also less likely to start school, and in some countries they have lower transition rates. The school participation dis- ability de�cit is typically larger than de�cits associated with characteristics such as gender, rural residence, or economic status. This analysis suggests that in developing countries disability is associated with long-run poverty, in the sense that children with disabilities are less likely to acquire the human capital that will allow them to earn higher incomes. In all countries, the schooling gap between children with and without disabilities starts at grade 1, suggesting that efforts are needed to boost enrollments of children with disabilities at the earliest grades in order to increase education attainment for this population. The result that the disability de�cit widens from grade to grade in countries that have achieved relatively high enrollment among children without disabilities suggests that special effort may be needed to keep children with disabilities in school. The results of this analysis should be treated as tentative. Better data are needed. Establishing clear and consistent measures of disability for use in household surveys and national censuses would be a start. A recent review (Mont 2007) suggests that questions that focus on functionality, concentrate on a core set of activities, and allow for variation in the degree of functional limitation (as opposed to the mere presence or absence of a limitation) should 160 THE WORLD BANK ECONOMIC REVIEW be preferred. To build the quantitative evidence base for empirically grounded policies, these questions must be implemented within samples that are large enough to allow detailed analysis. An important complement to the infor- mation that would emerge from the analysis of such data would be evaluations of the impact of alternative interventions that attempt to increase enrollments among children and youth with disabilities. FUNDING The Disability Team of the Human Development Network of the World Bank provided partial funding for this work. APPENDIX T A B L E A - 1 . De�nition of Disabilities in Covered Surveys Country/year Name of Question from survey of survey survey De�nition used in survey instrument Bolivia 1997 DHS Mentally retarded, deaf, mute, Does [X] have any of the blind, paralyzed, crippled following extreme physical impediments? Burundi 2000 MICS2 Presence of a physical handicap Speci�c wording not available. (missing upper or lower limbs, or other body part) Cambodia SES Amputation of one or more Does X have a disability? If 1999 limbs, inability to use one or “yes,� what type of more limbs, blind, deaf, disability does X have? mute, mentally disturbed, [respondent chooses from permanent dis�gurement, coded answers]; What was other the cause of the disability? [respondent chooses from coded answers]. Cambodia DHS Physical impairment Is there a person who usually 2000 lives in your household who has any type of physical impairment? Please give the name of each individual who has a physical impairment. For each individual, then ask: “Has X been physically impaired since birth, or was X’s impairment due to an accident?� [respondent chooses from coded answers]. (Continued ) Filmer 161 TABLE A-1. Continued Country/year Name of Question from survey of survey survey De�nition used in survey instrument Chad 2004 DHS Missing limb, deformed limb, Is there a person in your blind, deaf, mute, missing household who . . .is missing body part, behavioral a part of the body, for example a hand, an arm, or a leg, has dif�culty seeing or is almost or completely blind; has dif�culty hearing or is deaf; who has dif�culty speaking or is mute; who is missing an extremity such as �nger tips, toes, nose or ear; who has behavioral problems. Colombia DHS Blind, deaf, mute, paralysis or Does [X] have one of the 1995 missing limb, mental following health problems retardation, behavioral and how did they acquire the problem. problem? [respondent chooses from coded answers]. India 1992 NFHS (DHS) Blind, limb impairment of Does anyone [in household] suffer from: with separate answers for: “Blindness?� with options “yes: partial; yes: complete; “no� and for “Any physical impairment of limbs?� with options “yes: hands�; “yes: legs�; “yes: both�. Indonesia 2003 SUSENAS Blind, deaf, mute, physical Have a disability? If yes: “Type disability, mental disability. of disability� [respondent chooses from coded answers]; “Cause of disability� [respondent chooses from coded answers]. Jamaica 1998 LSMS Physical or mental disability. Is X physically or mentally disabled? Mongolia 2000 MICS2 Dif�culty seeing, dif�culty Speci�c wording not available. hearing Mozambique SES Blind, deaf, mute, mental Have a disability? If yes: “Type 1996 disability paralysis, of disability� [respondent amputated arm(s), amputated chooses from coded leg(s), other. answers]; “Cause of disability� [respondent chooses from coded answers]. (Continued ) 162 THE WORLD BANK ECONOMIC REVIEW TABLE A-1. Continued Country/year Name of Question from survey of survey survey De�nition used in survey instrument Romania 1995 LSMS Amputation of limb, paralysis Do you suffer from a of limbs, ankylosis of limbs handicap? If yes: “Type of or column, physical handicap� [respondent deformation, unilateral or chooses from coded bilateral blindness, deafness, answers]. muteness, epilepsy, mental retardation, mental disorder South Africa General Sight (blind/severe visual I am now going to ask about 2005 household limitation), hearing (deaf, disabilities experienced by survey profoundly hard of hearing), any persons within the communicating (speech household. Is X limited in impairment), physical (for his/her daily activities, at example needs wheel chair, home, at work or at school, crutches or prosthesis, limb because of a long-term or hand usage limitation), physical, sensory, hearing, intellectual (serious intellectual, or psychological dif�culties in learning, condition, lasting six months mental retardation), or more? If yes, “what emotional (behavioral, dif�culty or dif�culties does psychological problems), X have?� [respondent other chooses from coded answers]. Zambia 2003 LCMS Blind, partially sighted, deaf, Is X blind, partially sighted, dumb, crippled, mentally deaf, dumb, crippled, retarded, mentally ill, former mentally retarded, mentally mental patient ill, ex-mental [sic], or has multiple disabilities? Note: DHS is Demographic and Health Survey; MICS2 is End of Millennium Multiple Indicator Cluster Survey; SES is Socioeconomic Survey; NFHS is National Family Health Survey; SUSENAS is National Socioeconomic Survey; LSMS is Living Standards Measurement Study survey; LCMS is Living Conditions Monitoring Survey. Source: Author compilation. REFERENCES Afzal, Mohammad. 1992. “Disability Prevalence and Correlates in Pakistan: A Demographic Analysis.� Pakistan Development Review 31(3):217– 57. Ainsworth, Martha, and Deon Filmer. 2006. “Children’s Schooling: AIDS, Orphanhood, Poverty, and Gender.� World Development 34(6):1099– 128. Benitez-Silva, Hugo, Moshe Buchinsky, Hiu Man Chan, So�a Cheidvasser, and John Rust. 2004. “How Large Is the Bias in Self-Reported Disability?� Journal of Applied Econometrics 19(6):649– 70. Duggan, Mark, Robert Rosenheck, and Perry Singleton. 2006. Federal Policy and the Rise in Disability Enrollment: Evidence for the VA’s Disability Compensation Program. NBER Working Paper 12323. Cambridge, Mass.: National Bureau of Economic Research. Elwan, Ann. 1999. Poverty and Disability: A Survey of the Literature. Social Protection Discussion Paper 9932. Washington, D.C.: World Bank. Figlio, David, N., and Lawrence S Getzler. 2002. Accountability, Ability and Disability: Gaming the System. NBER Working Paper 9307. Cambridge, Mass.: National Bureau of Economic Research. Filmer 163 Filmer, Deon. 2005. “Gender and Wealth Disparities in Schooling: Evidence from 44 Countries.� International Journal of Education Research 43(6):351– 69. Filmer, Deon, and Lant Pritchett. 2001. “Estimating Wealth Effects without Expenditure Data—or Tears: With an Application to Educational Enrollments in States of India.� Demography 38(1):115– 32. Freedman, Vicki, Linda G. Martin, and Robert Schoeni. 2004. “Disability in America.� Population Bulletin 59(3):1–32. Gertler, Paul J., and Jonathan Gruber. 2002. “Insuring Consumption against Illness.� American Economic Review 92(1):51–70. Gertler, Paul J., Elaina Rose, and Paul Glewwe. 2000. “Health.� In M. Grosh, and P. Glewwe eds., Designing Household Survey Questionnaires for Developing Countries: Lessons from 15 Years of the Living Standards Measurement Study. Washington, D.C.: World Bank. Haddad, Lawrence, John Hoddinott, and Harold Alderman. 1997. Intrahousehold Resource Allocation in Developing Countries: Models, Methods, and Policy. Baltimore, Md.: Johns Hopkins University Press for the International Food Policy Research Institute. Haveman, Robert, Karen Holden, Barbara Wolfe, Paul Smith, and Kathryn Wilson. 1999. “The Changing Economic Status of U.S. Disabled Men: Trends and Their Determinants, 1982–1991.� Empirical Economics 24:571– 598. Haveman, Robert, and Barbara Wolfe. 2000. “The Economics of Disability and Disability Policy.� In A. J. Culyer, and J. P. Newhouse, eds., Handbook of Health Economics. Vol. 1. Amsterdam, New York: Elsevier Science B.V. Hoogeveen, Johannes G. 2005. “Measuring Welfare for Small but Vulnerable Groups Poverty and Disability in Uganda.� Journal of African Economies 14(4):603– 31. Jones, Andrew, and Owen O’Donnell. 1995. “Equivalence Scales and the Costs of Disability.� Journal of Public Economics 56(2):273– 89. LeRoy, Barbara W., Peter Evans, and Marcella Deluca. 2000. “United States and European School-Aged Disability Prevalence: An Investigative Study to Elaborate Differences.� Organisation for Economic Co-operation and Development, Paris. (www.oecd.org/dataoecd/9/59/27133749.pdf ) Mont, Daniel. 2007. Measuring Disability Prevalence. Social Protection Discussion Paper 706. Washington D.C.: World Bank. Parsons, Donald O. 1991. “Self-Screening in Targeted Public Transfer Programs.� Journal of Political Economy 99(4):859 –76. Peters, Susan J. 2003. “Inclusive Education: Achieving Education for All by Including those with Disabilities and Special Needs.� Disability Group, Human Development Network, World Bank, Washington, D.C. (http://siteresources.worldbank.org/DISABILITY/Resources/Education/ Inclusive_Education_En.pdf ) Strauss, John, and Duncan Thomas. 1995. “Human Resources: Empirical Modeling of Household and Family Decisions.� Handbook of Development Economics. Vol. 3. New York: North-Holland Press. United Nations. 2001. Guidelines and Principles for the Development of Disability Statistics. Statistics on Special Population Groups, Series Y, No. 10. New York: Statistics Division. UNESCO (United Nations Educational, Scienti�c and Cultural Organization). 2005. Education for All Monitoring Report 2005: The Quality Imperative. Paris: UNESCO. World Bank. 2003. “Education for All: Including Children with Disabilities: Summarized Lessons Learned and Key Policy Findings on the World Bank’s Work in Education.� Education Notes. Human Development Network, World Bank, Washington, D.C. ———. Various years. Word Development Indicators. Washington, D.C.: World Bank. Yeo, Rebecca, and Karen Moore. 2003 “Including Disabled People in Poverty Reduction Work: ‘Nothing about Us, without Us.’�World Development 31(3):571– 90. Yount, Kathryn M., and Emily Agree. 2005. “Differences in Disability among Older Women and Men in Egypt and Tunisia.� Demography 42(1):169 – 87. Zaidi, Asghar, and Tania Burchardt. 2005. “Comparing Incomes When Needs Differ: Equivalization for the Extra Costs of Disability in the U.K.� Review of Income and Wealth 51(1):89–114.