WPS3794 Disability, Poverty and Schooling in Developing Countries: Results from 11 Household Surveys Deon Filmer Development Research Group The World Bank dfilmer@worldbank.org This paper analyzes the relationship between whether a young person has a disability, the poverty status of their household, and their school participation using 11 household surveys from nine developing countries. Between 1 and 2 percent of the population is identified as having a disability. Youth with disabilities sometimes live in poorer households, but the extent of this concentration is typically neither large nor statistically significant. However, youth with disabilities are almost always substantially less likely to start school, and in some countries have lower transition rates resulting in lower schooling attainment. The order of magnitude of the school participation disability deficit is often larger than those associated with other characteristics such as gender, rural residence, or economic status differentials. World Bank Policy Research Working Paper 3794, December 2005 The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org. I thank Tomoki Fuji, Johannes Hoogeveen, Elizabeth King, Daniel Mont, Susan Peters, and Dan Rees for helpful comments on an earlier draft. Partial funding for this work was provided by the Disability Team of the Human Development Network of the World Bank. Excellent research assistance work was provided by Dilip Parajuli. 1. Introduction With over 100 million primary school age children not in school worldwide (UNESCO 2005) the target of universal education, endorsed by over 180 countries as a part of the Millennium Development Goals, remains elusive. Children with disabilities face particular hurdles to attend, and complete, school in developing countries. While there has been much policy discussion about interventions to increase access to schooling for children with disabilities (for example see Peters 2003, World Bank 2003), there has been little systematic empirical analysis on which to base this policy. A large part of this is due to the lack of appropriate and comparable data. Despite Elwan's (1999) description of the more general lack of empirical work on the association between disability and poverty in the developing world, such work is still missing.1 This study aims to start filling some knowledge gaps using existing data on the prevalence of disability and its association with poverty and schooling among youth in 8 developing and 1 transition country. Defining disability is complicated--and controversial. Purely medical definitions used in the past are giving way to definitions that incorporate continuous measures of the activities that people can undertake, the extent of participation in society and social and civic life, as well as the role of adaptive technologies. The World Health Organization's International Classification of Functioning, Disability and Health (ICF) describes disability as an umbrella term for impairments, activity limitations, and participation restrictions as a part of a broader classification 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. The main goal of this paper is descriptive. Many of the basic facts about disability, poverty and schooling in developing countries are unknown, or not systematically addressed. In order to contribute to the foundations of policy development, this paper analyzes available data to investigate the interactions between physical impairment and participation in schooling, and the intermediary relationship with poverty. The analysis finds that disability among youth is sometimes, but not always, associated with household poverty, but that it is systematically and significantly related to lower school participation. The paper proceeds as follows. Section 2 compares definitions and the prevalence of disability across the household surveys covered. Section 3 investigates the association 1An early exception to this 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 literature they refer to is typically not based on large-scale surveys. 2An online guide to the ICF is available at http://www3.who.int/icf/. 1 with poverty by examining the extent to which young people with disabilities live in households with lower economic status. Section 4 investigates the long run association with poverty by examining the association between disability and school participation among school-aged youth. 2. Data The data used for this analysis are from 11 nationally representative household surveys from 9 countries. Three of the surveys are associated with the Living Standards Measurement Study (LSMS) surveys: Jamaica 1998, Jamaica 2000, and Romania 1995. Three of the surveys are national socio-economic surveys (SES): Cambodia 1999, Indonesia 2000, and Mozambique 1996. One survey is a Demographic and Health Survey (DHS): Cambodia 2000. Four of the surveys are End of Millennium Multiple Indicator Cluster Surveys (MICS2) carried out under the guidance of UNICEF in 2000: Burundi, Myanmar, Mongolia, and Sierra Leone.3 These surveys are typically used to calculate poverty statistics, or derive basic health indicators such as child mortality or the use of health services, and underlie much empirical poverty and social analysis in developing countries. Most of the surveys have a sample size of between about 4,000 and 25,000 households (with Jamaica and Myanmar being outliers with 1,800 and over 65,000 households surveyed respectively). In order to select these datasets all LSMS, DHS, and MICS surveys were reviewed for any questions on disability and all those with a clear question on disability for a relevant age-range were included. In addition, the SES from Cambodia, Indonesia and Mozambique are accessible from national statistics offices and are some of the most recent in the world with information on disability. There is relatively little data of this kind in developing countries: the datasets, and therefore the countries, for this analysis were selected on the basis of data availability. The countries were not selected to be representative of developing countries in general. This is clearly a heterogeneous group of countries. Population living on less than a dollar a day ranges from 55 percent in Burundi to two percent in Jamaica and Romania; under-5 mortality--an indicator of basic health status--ranges from 206 per thousand live births in Mozambique to 15 in Romania (Table 1). There are three countries from Africa, four countries from Asia, one country from the Caribbean, and one country from Eastern Europe. While country variety is good since the results will reflect on a range of underlying conditions, little draws these countries together besides having the data available for this analysis. 3LSMS data are available online at http://www.worldbank.org/lsms; national socio-economic surveys are available from the countries' national statistics offices; DHS data are available online at http://www.measuredhs.com; MICS2 data are available online at http://www.childinfo.org. 2 Table 1. Basic statistics about the countries and surveys GDP per capita Population <$1 a Under-5 Number of households PPP day mortality surveyed Burundi 2000 590 55 190 3,979 Cambodia 1999 1710 34 135 6,001 Cambodia 2000 1804 34 135 12,236 Indonesia 2003 3213 8 48 65,762 Jamaica 1998 3366 2 20 7,375 Jamaica 2000 3395 2 20 1,800 Mongolia 2000 1620 27 75 6,000 Mozambique 1996 700 38 206 8,250 Myanmar 2000 - - 110 25,545 Romania 1995 5965 2 15 24,560 Sierra Leone 2000 464 - 186 3,916 Source: World Bank, World Development Indicators. Poverty rates are for the following years: Burundi 1998; Cambodia 1997; Indonesia 2002; Jamaica 1999 and 2000; Mongolia 1998; Mozambique 1996; Romania 1998. Under-5 mortality data are for 2000 except Romania which is for 1995. The datasets covered in this study are all most closely consistent with an impairment definition of disability--and as such fall under ICF's "body functioning and structure" domain. The definition does not include mental health, chronic illness or the inability to carry out specific activities. The latter approach is an alternative that is attractive since it is arguably easier to verify. Indeed, selective misreporting of morbidity has long been recognized as a potential problem in studies of the relationship between health and other socio-economic characteristics (Gertler, Rose and Glewwe 2000). To overcome this problem Gertler and Gruber (2002) use responses on questions regarding Activities of Daily Living (ADLs) when analyzing the impact of disability on household consumption in Indonesia, and Yount and Agree (2005) use ADLs when analyzing sex and gender differences in disability among older women and men in Egypt and Tunisia. The impairments reported in the surveys in this study are typically easily verified, for example blindness or missing a limb. Nevertheless it is possible that there is selective reporting in so far as some respondents and interviewers interpret blindness as partial sight whereas to others it means complete inability to see, for example. Or it is possible that mental disability is selectively recognized and reported by some respondents. Typically, however, selective reporting is assumed to operate such that higher socio-economic groups report higher morbidities. Under this assumption, the estimates reported below would be underestimates of the relationship between disability and poverty.4 4Interestingly, Benitez-Silva, Buchinsky, Chan, Cheidvasser, and Rust (2004) find no bias in self-reported disability as compared to bureaucratic assessment among adult US social security benefit applicants. 3 Table 2. Types of disabilities included in definition of "person with a disability" Type of Visual Hearing Speech Physical Mental survey disability disability disability disability disability Burundi 2000 MICS2 X Cambodia 1999 SES X X X X X Cambodia 2000 DHS X Indonesia 2003 SES X X X X X Jamaica 1998 LSMS X X Jamaica 2000 LSMS X X X X X Mongolia 2000 MICS2 X X Mozambique 1996 NHS X X X X X Myanmar 2000 MICS2 X X Romania 1995 LSMS X X X X X Sierra Leone 2000 MICS2 X X X X Note: See Annex Tables more precise wording and disaggregations. Despite the fact that all 11 surveys have an impairment definition of disability, non comparable definitions remain an issue in any effort to compile data across countries. Table 2 summarizes the items covered in each survey that define a person as having a disability. Clearly the definitions are non-comparable, even across surveys within the same country. Six of the surveys use an "extensive" definition that includes visual, hearing, speech and physical disability. But even in this group of six surveys, the definition of each type of impairment varies. For example, in Cambodia 1999 the physical disability category contains 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"--whereas in Jamaica 2000 there is simply one category described as "physical disability (legs and arms)". More generally, in some countries the definition is stricter than in others. In Mongolia and Myanmar sight and hearing are described as "problematic" whereas in other surveys they are characterized as "blind" and "deaf".5 The second main data constraint in carrying out this analysis is the fact that surveys do not identify large numbers of individuals as having a disability. Therefore, any subsequent analysis such as the correlation between disability and poverty, or disability and schooling, will suffer from imprecision. Table 3 highlights this point by showing the number of youth identified in each survey and the subset with a disability. For some surveys the small sample problem is especially acute, for example the Jamaica 2000 survey identifies only 14 youth as having a disability, the Sierra Leone survey identifies only 28. 5Note that another non-consistent aspect of the data is the coverage in terms of age: the upper age limit is sometimes 14 in Burundi and Myanmar. 4 In order to not give undue weight to these surveys, the results on poverty and schooling for datasets that identify fewer than 50 children with a disability are not reported.6 Table 3. Number of youth 6 to 17+ defined as having a disability in each survey Maximum Number Number of youth with Country and year Type of survey age of youth a disability Burundi 2000 MICS2 14 5,865 73 Cambodia 1999 SES 17 10,881 96 Cambodia 2000 DHS 17 23765 214 Indonesia 2000 SES 17 64,136 326 Jamaica 1998 LSMS 17 6,964 58 Jamaica 2000 LSMS 17 1,640 14 Mongolia 2000 MICS2 17 7,645 245 Mozambique 1996 NHS 17 14,520 156 Myanmar 2000 MICS2 14 26,329 41 Romania 1995 LSMS 17 13,777 82 Sierra Leone 2000 MICS2 17 7,534 28 Note: Data are unweighted in order to show the actual number of observations underlying the analysis. +Maximum ages are 18 in Mongolia, and 14 in Burundi and Myanmar. A last data constraint concerns the measurement of household poverty. All LSMS and SES surveys include household per capita consumption expenditures (PCE), the variable typically used in poverty analysis. DHS and MICS2 data, however, do not include those variables. In this study, quintiles based on per capita consumption expenditures are used when available. In other datasets, an index of household consumer assets and housing characteristics (an economic status index) was used to classify households into quintiles (following Filmer and Pritchett 2000). The exception is the SES from Cambodia 1999 in which there was a problem in the collection of expenditures data. An economic status index is therefore used in that survey to classify economic status.7 3. Prevalence of disability and its association with household economic status The first issue these data can be used to explore is the prevalence of disability and the association with household economic status. Prevalence estimates range between 0.13 (Myanmar) and 2.77 (Jamaica 2000) percent of the population as having a disability (Table 4). These numbers are consistent with those compiled by the United Nations statistical database on disability (DISTAT).8 In that source of over 65 surveys and censuses between 1970 and 1992 in developing countries, the mean prevalence rate for the entire population 6 Results for Jamaica 2000, Myanmar 2000, and Sierra Leone 2000 are available from the author on request. 7 Consistent with typical poverty analysis, quintiles are derived on the basis of the distribution of people across the socio-economic status measure. Specifically, quintiles are defined such that 20 percent of youth live in each quintile. 8 Available online at http://unstats.un.org/unsd/demographic/sconcerns/disability/disab2.asp. 5 is 1.7 percent, and for those countries with statistics for children under age 14 the prevalence rate is 0.7 (see Annex Tables for a summary of the data from DISTAT). Perhaps surprisingly, of the 11 surveys analyzed here, those that list more types of impairments do not systematically identify a higher percentage of the population as disabled. For example in the six countries that include visual, hearing, speech, and physical disabilities the percentages are 1.51 (Cambodia 1999), 0.64 (Indonesia); 2.77 (Jamaica 2000); 1.19 (Mozambique); 1.32 (Romania); and 0.55 (Sierra Leone) which span close to the entire range of prevalence across all the surveys. In Mongolia which inquires only about visual/hearing impairments the prevalence is the highest observed in this collection of datasets (3.2 percent), while in Burundi and the 2000 DHS in Cambodia which cover only physical disabilities the prevalence rates are 1.24 and 0.86 percent respectively. Table 4. The prevalence of disability among youth ages 6 to 17+ by household economic status quintile Concen Std. Poorest 2nd 3rd 4th Richest -tration Error All quintile quintile quintile quintile quintile Index of CI Burundi 2000 1.24 1.28 1.19 1.11 1.36 1.28 0.032 (0.064) Cambodia 1999# 0.88 0.91 0.84 0.87 0.81 0.94 -0.007 (0.687) Cambodia 2000 0.86 1.08 0.71 0.86 0.82 0.86 -0.044 (0.045) Indonesia 2003# 0.51 0.70 0.55 0.41 0.50 0.38 -0.084 (0.038)* * Jamaica 1998 0.82 1.01 1.05 0.48 0.68 0.89 -0.064 (0.082) Mongolia 2000 3.20 3.40 3.01 2.88 2.81 3.92 0.019 (0.037) Mozambique 1996# 1.19 0.86 0.81 1.57 1.40 1.29 0.111 (0.045)* * Romania 1995# 0.60 0.91 0.47 0.54 0.47 0.58 -0.110 (0.067)* Note: Maximum age is 14 in Burundi and Myanmar. Survey includes vision, hearing, speech, and physical + # disabilities. ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels respectively. Standard errors in parentheses. Of course this variability combines both actual prevalence and differences in survey techniques. However, in the countries with more than one survey (Cambodia and Jamaica) the survey with the more extensive definition of a person with a disability does not always result in the larger prevalence. The SES in Cambodia in 1999 characterizes 1.51 percent of the population as having a disability with an extensive definition, whereas the DHS in 2000 characterizes 1.57 percent of the population as having a disability with a definition restricted to physical disabilities. In Jamaica the more extensive definition in 2000 characterizes 2.77 percent as having a disability--more than the 2.09 percent identified in 1998 with a more limited definition. Clearly there is substantial variation 6 across surveys in how people with disabilities are identified and cross-country comparisons in prevalence can only be made with caution, if at all.9 Despite the lack of cross-country comparability in the definitions and measurement of disability, these surveys are still potentially useful in describing the association of disability with other characteristics. That is, conditional on a particular definition, the analysis is valid for a given survey (the definition is common to all individuals in the survey). Moreover, it is less likely that cross-country comparisons of the association between disability and other characteristics would suffer from these problems. Nevertheless, if some types of disabilities are more 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. For example, say loss of a limb was typically more associated with poverty than other types of impairments, then a survey that included loss of a limb in its definition of disability would yield a higher correlation between disability and poverty. Therefore even the cross-country comparisons of the relationship between disability and correlates needs to be treated with caution. The analysis of the relationship 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 as it lowers earning power and consumption expenditures (Gertler and Gruber 2002) and a consequence of poverty as the cumulative depravations of poverty can manifest themselves in disability (e.g. infant and child development, exposure to dangerous working conditions). Moreover, the presence of a person with a disability entails direct costs which result in lower standards of living (Jones and O'Donnell 1995, Zaidi and Burchardt 2005). Indeed, Hoogeveen (2005) estimates that in Uganda, households headed by a person with a disability have substantially lower consumption-- and are significantly more likely to be poor. Children in those households are also more likely to have lower education attainment for their age.10 Table 4 reports the percent of youth ages 6 to 17 characterized as having a disability in each economic status quintile: it is lower in the richest than in the poorest quintile in all surveys except Burundi, Cambodia 1999, Mongolia and Mozambique. But the relationship is not neatly ordered with lower prevalence in each higher quintile. A useful way of summarizing the entire distribution of a characteristic (such as disability) across the economic status distribution is through the use of concentration curves. These plot percentiles of a population ranked by economic status on the horizontal axis, against 9Developing good data on disability is complex, United Nations (2001) contains a guide to doing so. 10Disability among household heads is defined differently in the survey used in Hoogevenn (2005). A head of household is considered disabled if this "prevents him or her from being actively engaged in labour activities during the past week". 7 the cumulative percentage of a characteristic on the vertical axis. When the concentration curve lies above the 45 degree line this means that the characteristic is concentrated among the poor--with larger deviations indicating higher concentration among the poor. The left panel of Figure 1 shows the concentration curves for disability among youth ages 6 to 17 for the 4 surveys with an extensive definition of disability and more than 50 children identified as having a disability. The right panel of Figure 1 shows the deviation of the concentration curve from the 45 degree line--a transformation that sharpens the distinction between the lines. In this set of countries, disability is concentrated among the poor in Indonesia and Romania. It is concentrated among the wealthy in Mozambique. In Cambodia 1999 it is evenly spread across the economic distribution. Figure 1. The distribution of disability across economic quintiles among youth 6 to 17 Concentration curves Deviation of concentration curves from the 45 degree line 100 20 15 80 10 45 degree line Cambodia 1999# Cambodia 1999# 60 5 Indonesia 2003# Indonesia 2003# Mozambique 1996# Mozambique 1996# 0 Romania 1995# Romania 1995# 40 -5 -10 20 -15 0 -20 0 20 40 60 80 100 0 20 40 60 80 100 The Concentration Index (CI) is a summary measure of the entire distribution of an indicator by the welfare ranking--in this context it is therefore a summary statistic for inequality in disability by economic status. Intuitively, the CI is defined as twice the area between the concentration curve and the 45 degree line, with area below the 45 degree line counted as positive and area above the 45 degree line as negative. Note that while Figure 1 is drawn in terms of quintiles, the CI is derived on the basis of the full (continuous) distribution of the welfare ranking variable. In all but 3 of the surveys, the CI is negative indicating a concentration of people with a disability among the poor (Table 4). In Indonesia and Romania this negative value is statistically significantly different from zero; in Mozambique the positive value is statistically significantly different from zero.11 It is hard to determine whether these numbers are "high" or "low": there is no "expected" degree of concentration of disability among the poor. A comparison to a different outcome--child mortality--provides a sense of the order of magnitudes. 11Standard errors are obtained through bootstrapping the calculation of the concentration index 1,000 times for each survey. The standard deviation of the estimate of the CI across those replications is reported here as the standard error of the CI. 8 Wagstaff (2000) calculates the CI of child mortality for 9 developing countries using a similar approach to that applied here.12 He finds that the index ranges from -0.322 in Brazil to -0.016 in Vietnam with a median value of -.132 in Nepal. In all but two of the countries he studies he finds the CI to be negative and highly significantly different from zero. The order of magnitude of the CI of disability among youth is somewhat lower than that of mortality. The median CI of child mortality across the nine developing countries in Wagstaff (2000) was -.132, while it is ­0.02 for disability among youth 6 to 17 in the 8 surveys reported in Table 4. In the two surveys where the CI for disability is negative and significantly different from zero it is -0.084 (Indonesia) and ­0.110 (Romania) suggesting that in these two countries the order of magnitude is similar to that for child mortality. 4. Disability and schooling We turn now to the relationship between disability and schooling among the school-age population (defined for the purpose of this analysis as ages 6 to 17). Table 5 shows the percent of youth that are currently in school disaggregated between those who are generally of primary (6 to 11) and secondary (12 to 17) school age. Youth with a disability are almost always substantially less likely to be in school than those without. The deficit among children 6 to 11 years old ranges from a shortfall of 15 percentage points in Mozambique to 59 percentage points in Indonesia. In the latter country, whereas 89 percent of children 6 to 11 without a disability are in school, only 29 percent of those with a disability are in school. Among older children and youth the gap covers a similar range (from 15 percentage points in Cambodia to 58 percentage points in Indonesia), with the exception of Burundi where the gap is zero. On average the gaps are larger among the older group: the median is a 26 percentage point shortfall among 6 to 11 year olds, and a 31 percentage point shortfall among 12 to 17 year olds. Table 5. Percent reported to be in school Ages 6-11 Ages 12-17+ With Without With Without Difference Difference disability disability disability disability Burundi 2000 14.6 37.2 -22.6 48.0 47.8 0.2 Cambodia 1999# 18.1 58.2 -40.1 30.6 68.0 -37.4 Cambodia 2000 37.8 66.8 -29.0 46.5 61.7 -15.2 Indonesia 2003# 29.2 88.5 -59.3 18.3 75.9 -57.6 Jamaica 1998 70.5 99.4 -28.9 50.2 85.9 -35.7 Mongolia 2000 41.0 58.0 -17.0 47.1 72.6 -25.5 Mozambique 1996# 34.2 49.2 -15.0 29.3 48.4 -19.1 Romania 1995# 57.7 79.2 -21.5 35.7 83.7 -48.0 Note: Maximum age is 14 in Burundi. Survey includes vision, hearing, speech, and physical disabilities. + # 12The countries included in the Wagstaff (2000) study are Brazil, Cote d'Ivoire, Ghana, Nepal, Nicaragua, Pakistan, Philippines, South Africa, Vietnam. 9 To the extent that disability in a given country is correlated with other factors that affect schooling, such as poverty, age, or urban/rural residence, the unadjusted difference in school participation between youth with and without a disability might give a misleading picture of the deficit. Column (i) of Table 6 reports the unadjusted percentage point deficit in current school participation among school-aged children with a disability, and column (ii) reports the deficit after adjusting for potential confounding factors (standard errors are reported in parentheses). The adjustment is carried out, for each survey, using a multivariate Probit model with school participation as the dependent variable. The independent variables include a dummy variable for whether a child has a disability as well as a set of variables capturing potentially confounding variables: age and age squared; a dummy variable for a child's gender; a dummy variable for urban residence; and dummy variables for each economic status quintile. The effect of the change in the dummy variable for disability--evaluated at the means of all the other variables--is the number reported in column (ii). Table 6. Schooling deficits among children ages 6 to 17+ with a disability: "raw" differential, and differential after controlling for age, gender, urban residence, and economic status quintile (percentage points). Current school participation Ever attended school Deficit adjusted for Deficit adjusted for Unadjusted other factors Unadjusted other factors (i) (ii) (iii) (iv) Burundi 2000 -12.2 (5.3) ** -15.8 (4.8) *** -13.5 (5.5) ** -18.7 (4.9) *** Cambodia 1999# -38.8 (5.0) *** -45.2 (5.5) *** -45.7 (5.1) *** -56.4 (5.7) *** Cambodia 2000 -22.0 (3.8) *** -26.6 (4.5) *** -20.3 (3.9) *** -31.6 (4.7) *** Indonesia 2003# -58.8 (2.7) *** -67.4 (3.1) *** -45.8 (3.3) *** -52.9 (4.6) *** Jamaica 1998 -32.7 (6.4) *** -27.5 (8.0) *** -24.6 (5.7) *** -18.5 (5.8) *** Mongolia 2000 -20.3 (3.2) *** -27.9 (3.6) *** -16.9 (3.1) *** -36.7 (4.2) *** Mozambique 1996# -17.7 (4.5) *** -17.5 (5.0) *** -12.2 (5.2) ** -14.3 (5.4) *** Romania 1995# -38.9 (5.5) *** -53.2 (6.4) *** -30.0 (5.4) *** -50.4 (7.1) *** Notes: Maximum age is 14 in Burundi. Survey includes vision, hearing, speech, and physical + # disabilities. 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 sex, urban residence, and economic quintile. ***, **, * indicate statistical significance at the 1%, 5%, and 10% levels respectively. Standard errors in parentheses. In most countries controlling for confounding factors leads to an increase in the school enrollment deficit that can be attributed to disability. This deficit is over 50 percentage points in Indonesia and Romania; between 25 and 45 percentage points in Cambodia, Jamaica, and Mongolia; and slightly less than 20 percentage points in Burundi and Mozambique. In all countries the difference is large and statistically significantly different from zero. 10 There is substantial heterogeneity across countries in the schooling deficit associated with disability. Part of this variation might be due to differences in the definition of disability. That is, in a survey with a more "stringent" definition of disability one would likely observe a larger deficit since this survey would identify individuals who would have to overcome bigger obstacles in order to access education. The fact that the two surveys from Cambodia yield schooling deficits among youth with disabilities that are 15 to 20 percentage points apart suggests that this is likely a part of the story. Another part of this variation likely relates to overall enrollment. It would not be surprising to observe larger deficits in countries where enrollment among children without a disability is high: in these countries there would be more scope to observe a bigger deficit. The schooling deficit does tend to be smaller in the countries with the lowest overall enrollment (Burundi and Mozambique) and is larger in countries with higher enrollment (Romania and Indonesia). The relationship is not perfect, however: Jamaica has the highest overall enrollment, but the deficit associated with disability is about average for the surveys reviewed here. Last, a part of the variation is likely related to differences in the social and policy environment. Countries where there is greater stigma towards a person with a disability, or where less effort has been made to ensure equal access to schooling, will undoubtedly have a larger deficit associated with schooling. But this is only a part of the cause for cross- country variation. It would therefore be beyond the reach of these data to attribute differences across countries in Table 6 entirely to differences in policies towards people with disabilities. Patterns of school participation The last two columns of Table 6 show analogous results for the percentage of children who have ever attended school. The pattern of results is similar to the current school participation results, and the deficit is of a similar order of magnitude suggesting that a large part of the schooling deficit among children with disabilities comes from the fact that they never attended school at all. Figure 2 illustrates this issue by showing the grade "survival" profile of the cohort of 10 to 17 year olds. Each line shows the percentage of children of this cohort who have completed each grade--allowing for the fact that some of the cohort are still in school--as derived by the Kaplan-Meier survivor function.13 There are clearly large differences in the patterns of attainment between youth with and without disabilities. In all countries the 13A similar approach which doesn't adjust for right-censoring was used in Filmer and Pritchett (1999). This model allows all children to be "at risk," i.e. even those who have never attended school. Because of the computations of the survival estimation, children who have never attended school enter into the calculation of the probability at grade 1, and implicitly assumes that they will not attend school in the future. The lower age bound of 10 allows for late starting. Results for Jamaica 2000 and Myanmar are not available because sample sizes are too small. 11 difference exists in the probability of ever attending school. In some counties, these differences are exacerbated as children progress through the school system. In particular, in Indonesia, Jamaica, and Romania, where the gap at the start of schooling is on the order of 30 to 45 percentage points, the shortfall in grade completion increases to about 60 or 70 percentage points by grade 8. Figure 2. Grade "survival" profiles for ages 10 to 17: Kaplan-Meier estimates of the probability of completing a grade. Benin 2000 Cambodia 1999 Cambodia 2000 100 100 100 80 80 80 60 60 60 40 40 40 20 20 20 0 0 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Indonesia 2003 Jamaica 1998 Mongolia 2000 100 100 100 80 80 80 60 60 60 40 40 40 20 20 20 0 0 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Mozambique 1996 Romania 1995 100 100 80 80 60 60 40 40 Without disability With disability 20 20 0 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Note: Estimates account for right hand censoring. Jamaica 2000 and Myanmar are not reported because there are too few observations to estimate survival profiles. 12 Relative magnitude of school participation deficits How large is the deficit in school participation relative to other sources of inequality? The multivariate models can be used to compare school participation gaps associated with disability, gender, urban/rural residence, as well as economic status. Figure 3 shows, for each survey, the school participation deficit among children with disabilities (relative to those without disability); girls (relative to boys), children in rural areas (relative to those in urban areas), and children in households in the poorest quintile (relative to the richest quintile).14 Figure 3. Magnitudes of school participation deficits associated with: having a disability, being a girl, living in a rural area, and being in the poorest relative to the richest quintile. Percentage point differences for children ages 6 to 17 Burundi 2000 Cambodia 1999 Cambodia 2000 Indonesia 2003 Jamaica 1998 Mongolia 2000 Mozambique 1996 Romania 1995 -80 -60 -40 -20 0 20 With disability Female Rural Poorest-Richest quintile Note: Deficits shown are the marginal effects of dummy variables for each characteristic in multivariate probit models. 14In each case the deficit is estimated at the means of the other variables. 13 Clearly the deficits associated with disability are large compared to other sources of inequality. In all these countries the gender gap in enrollment is small relative to that associated with disability. Perhaps more surprising, the gaps associated with rural residence, and the large gap between rich and poor, are usually substantially smaller than that associated with disability. The exceptions are Burundi where wealth gaps dominate the gaps associated with disability, and Mozambique where rural/urban differences dominate. Typically, however, the gap in school participation between children with and without a disability is on the order of twice as large as those associated with rural residence or wealth. The interaction of disability with other characteristics in the association with schooling Comparing the magnitudes of the schooling gaps associated with disability and other characteristics helps to get a sense for orders of magnitude. But an interesting additional question is 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. In Romania, the disability deficit in school participation among boys is about 9 percentage points smaller than that among girls. In Mongolia the school participation deficit associated with disability is about 17 percentage points larger in rural areas than in urban areas. In Cambodia, the school participation deficit associated with disability is smaller in the poorer quintiles (largely because overall school participation is lower in those quintiles). But other than these specific cases, there are no statistically significant interactions. There is an important caveat to this finding, however: small sample sizes make it hard to estimate these effects with much precision. Not only does one need enough observations to estimate average differences, one needs enough cases of the various combinations of characteristics in order to identify their association with enrollment. Given the small numbers overall (see Table 3) it is perhaps unsurprising that the models do not yield significant estimates.15 5. Conclusions This analysis of data from 11 nationally representative surveys has confirmed the many data problems that earlier discussions have identified as hampering the establishment of a broad empirical base for developing policies targeted to people with disabilities in poor countries. The fundamental variation across surveys in the definition of "disability" makes cross-country comparisons difficult. The small number of people identified as 15In order to increase the power of these tests, the models were also run separately for each interaction. The pattern of statistically significant results is not affected under this alternative approach. 14 having a disability makes it hard to precisely estimate patterns in the data beyond simple correlations. Despite these limitations, but keeping them in mind, the data are nevertheless revealing. Consistent with other similar surveys, the 11 surveys analyzed here identify on the order of 1 to 2 percent of the population as having a disability. Countries with two surveys and varying definitions suggest that the percentage is not always sensitive to the exact definition (e.g. different definitions can give similar prevalence rates, and vice versa). In addition, other aspects of the surveys, such as the training of enumerators or the use that interviewees expect the survey to be put, might affect the overall estimated rates. Youth with disabilities sometimes live in poorer households--but the extent of this concentration is typically neither large nor statistically significant. On the other hand, youth with disabilities are almost always substantially less likely to participate in schooling--and significantly so. Children with disabilities are less likely to start school, and in some countries have lower transition rates resulting in lower schooling attainment. The order of magnitude of the school participation disability deficit is often larger than those associated with other characteristics such as gender, rural residence, or economic status differentials. The data do not suggest that there are typically interactive effects-- although the small number of disabled youth in these surveys makes hard to identify those effects. 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. However, the results should be treated as tentative at best. Establishing clear and consistent measures of disability for use in household surveys, and implementing these in the context of samples that are large enough to identify sufficient observations to allow detailed analysis (perhaps in the context of a census), will be a pre-requisite for further work on the relationship between disability and poverty. This should be a high priority for building empirically grounded policies to address the issue of disability, poverty and schooling. References Afzal, Mohammad. 1992. "Disability Prevalence and Correlates in Pakistan: A Demographic Analysis." Pakistan Development Review 31(3):217-257. Benitez-Silva, Hugo, Moshe Buchinsky, Hiu Man Chan, Sofia Cheidvasser, and John Rust. 2004. "How Large is the Bias in Self-Reported Disability?" Journal of Applied Econometrics 19(6):649-670. Elwan, Ann. 1999. "Poverty and Disability: A Survey of the Literature." Social Protection Discussion Paper No. 9932. World Bank, Washington, D.C. 15 Filmer, Deon, and Lant Pritchett. 1999a. "The Effect of Household Wealth on Educational Attainment: Evidence from 35 Countries." Population and Development Review 25(1). 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):115-132. 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 Margaret Grosh and Paul Glewwe, eds., Designing Household Survey Questionnaires for Developing Countries: Lessons from 15 Years of the Living Standards Measurement Study. Washington, D.C.: World Bank. Hoogeveen, Johannes G. Forthcoming. "Measuring Welfare for Small but Vulnerable Groups Poverty and Disability in Uganda." Journal of African Economies. Advance Access available at http://jae.oxfordjournals.org/ Jones, Andrew, and Owen O'Donnell. 1995. "Equivalence Scales and the Costs of Disability." Journal of Public Economics 56(2):273-289. Peters, Susan J. 2003. "Inclusive Education: Achieving Education for All by Including those with Disabilities and Special Needs." Prepared for the Disability Group, Human Development Network, The World Bank. http://siteresources.worldbank.org/DISABILITY/Resources/Education/Inclusive_Education_En.pdf United Nations. 2001. "Guidelines and Principles for the Development of Disability Statistics." Statistics on Special Population Groups, Series Y, No. 10, United Nations Statistics Division. UNESCO. 2005. Education for All Monitoring Report 2005: The Quality Imperative. Paris: UNESCO. Wagstaff, Adam. 2000. "Socioeconomic Inequalities in Child Mortality: Comparisons Across Nine Developing Countries." Bulletin of the World Health Organization 78(1): 19-29. 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, The World Bank. http://siteresources.worldbank.org/DISABILITY/Resources/Education/Education- Notes/Education_Notes_En.Aug.pdf Yeo, Rebecca, and Karen Moore. 2003 "Including Disabled People in Poverty Reduction Work: `Nothing About Us, Without Us'". World Development 31(3):571-590. Yount, Kathryn M., and Emily Agree. 2005. "Differences in Disability among Older Women and Men in Egypt and Tunisia." Demography 42(1):169-187. 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. 16 Annex Table 1. Defining a person as having a disability in the covered surveys Type of Country and year survey Definition used in survey Question from survey instrument Burundi 2000 MICS2 Presence of a physical handicap (missing upper Specific wording not available. or lower limbs, or other body part). Cambodia 1999 SES Amputation of one or more limbs; inability to "Does X have a disability?"; "If use one or more limbs; blind; deaf; mute; `yes', what type of disability does mentally disturbed; permanent disfigurement; X have?" [with pre-coded other. answers]; "What was the cause of the disability?" [with pre-coded answers]. Cambodia 2000 DHS Physical impairment. "Is there a person who usually 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?" [with pre-coded answers]. Indonesia 2003 SES Blind; deaf; mute; physical disability; mental "Have a disability?". If yes: "Type disability. of disability" [with pre-coded answers]; "Cause of disability" [with pre-coded answers]. Jamaica 1998 LSMS Physical or mental disability. "Is X physically or mentally disabled?" Jamaica 2000 LSMS Sight; hearing; speech; physical (legs and "Is X physically or mentally arms); multiple disability; mental retardation. disabled?" [with pre-coded answers that include types of disabilities]; "If yes, when did this disability occur?" [with pre-coded answers such as "since birth," "In child under 5 years," ...]. Mongolia 2000 MICS2 Difficulty seeing; difficulty hearing. Specific wording not available. Myanmar 2000 MICS2 Visual problem; hearing problem. Specific wording not available. Romania 1995 LSMS Amputation of limb(s); Paralysis of limb(s); "Do you suffer from a handicap?"; ankylosis of limb(s) or column; physical If yes: "Type of handicap" [with deformation(s); unilateral or bilateral cecity; pre-coded answers]. deaf; mute; epilepsy; mental retardation; mental disorder. Sierra Leone 2000 MICS2 Blindness; crippled; lost limbs; deafness; mute. Specific wording not available. 17 Annex Table 2. Number of people defined as having a disability in each survey Age range Number of people Country and year Type of survey considered Number of people with a disability Burundi 2000 MICS2 0 to 14 9,925 103 Cambodia 1999 SES all 32,348 504 Cambodia 2000 DHS all 66105 1017 Indonesia 2000 SES all 259,237 1,720 Jamaica 1998 LSMS all 26,458 558 Jamaica 2000 LSMS all 6,304 175 Mongolia 2000 MICS2 0 to 18 15,025 330 Myanmar 2000 MICS2 0 to 14 43,363 66 Romania 1995 LSMS all 72,726 962 Sierra Leone 2000 MICS2 all 24,254 131 Note: Data are unweighted in order to show the actual number of observations underlying the analysis. 18 Annex Table 3. Summary estimates of disability prevalence from the UN's DISTAT database Population Population 0-14 Country / Year Source disabled (%) disabled (%) Definition Algeria 1992 survey 1.2 anyone in hh: impairment types Bangladesh 1982 survey 0.8 impairment types Benin 1991 survey 1.3 "handicap" Bermuda 1991 census 7.6 2.0 chronic condition that affects activities of daily life Botswana 1991 census 2.2 0.9 impairment types Brazil 1981 survey 1.7 0.9 Brazil 1991 census 0.9 0.4 impairment types CAR 1988 census 1.5 0.7 impairment types Chile 1992 census 2.2 impairment types China 1983 survey 1.4 1.4 China 1987 survey 5.0 2.8 Colombia 1991 survey 5.6 2.9 Colombia 1993 census 1.8 0.7 impairment types Comoros 1980 census 1.7 0.9 impairment types Congo 1974 census 1.1 impairment types Egypt 1976 census 0.3 0.1 Egypt 1981 census 1.6 0.7 impairment types + activities Egypt 1996 survey 4.4 impairment types + activities El Salvador 1992 census 1.6 0.6 "disability" types India 1981 census 0.2 impairment types Iraq 1977 census 0.9 0.3 impairment types Jamaica 1991 census 4.8 1.3 impairment types Jordan 1991 survey 2.6 impairment types Jordan 1994 census 1.2 0.9 Kenya 1989 census 0.7 0.6 impairment types Kuwait 1980 census 0.4 0.4 Lebanon 1994 survey 1.0 0.5 Malawi 1983 survey 2.9 impairment types Mali 1987 census 2.7 0.8 "handicap" Mauritania 1988 census 1.5 impairment types Morocco 1982 census 1.1 0.1 Namibia 1991 census 3.1 1.0 impairment types Niger 1988 census 1.3 0.6 impairment types Nigeria 1991 census 0.5 0.3 impairment types Oman 1993 census 1.9 0.6 impairment types Pakistan 1981 census 0.5 0.1 impairment types Panama 1980 census 0.7 impairment types Panama 1990 census 1.3 0.7 impairment types Peru 1981 census 0.2 impairment types Peru 1993 census 1.3 0.7 impairment types Philippines 1980 survey 4.3 2.2 impairment types + activities Philippines 1990 census 1.1 0.7 impairment types Philippines 1995 census 1.3 0.4 Poland 1988 census 9.9 0.5 Senegal 1988 census 1.1 0.4 impairment types South Africa 1980 survey 0.5 0.2 Sri Lanka 1981 census 0.5 0.3 impairment types Sri Lanka 1986 survey 2.0 Sudan 1992 survey 1.1 0.7 long-term condition that affects normal activities 19 Sudan 1993 census 1.6 0.7 impairment types Swaziland 1986 census 2.2 0.9 impairment types Syria 1970 census 1.0 Syria 1981 census 1.0 0.7 Syria 1993 survey 0.8 0.6 long-term condition that affects normal activities Thailand 1986 survey 0.7 0.5 Thailand 1990 census 0.3 0.7 impairment types Togo 1970 census 0.6 0.1 Tunisia 1984 census 0.9 0.3 Tunisia 1989 survey 0.9 Tunisia 1994 census 1.2 0.6 Turkey 1985 census 1.4 0.6 Uganda 1991 census 1.2 0.7 impairment types Yemen 1994 census 0.5 0.1 Zambia 1980 census 1.6 0.7 Zambia 1990 census 0.9 0.6 impairment types Note: The UN warns against simple comparisons across countries since the data and definitions are not strictly comparable Annex Table 4. Percent distribution of types of disabilities among those with a disability Visual Hearing Physical Mental Multiple/ Total and/or Other Speech Burundi 2000 100 100 Cambodia 1999 12 7 55 7 19 100 Cambodia 2000 100 100 Indonesia 2003 16 22 40 23 100 Jamaica 1998 100 100 Jamaica 2000 27 21 14 23 15 100 Mongolia 2000 63 37 100 Myanmar 2000 28 72 100 Romania 1995 15 7 47 27 4 100 Sierra Leone 2000 31 33 36 100 20 Annex Table 5. Marginal Effects from probit regressions. Dependent variable: Participating in school, ages 6 to 17 Burundi 2000 Cambodia Cambodia Indonesia Jamaica 1998 Jamaica Mongolia Myanmar Romania Sierra Leone 1999 2000 2003 2000 2000 2000 1995 2000 Disabled (0,1) -0.179 -0.452 -0.263 -0.675 -0.279 -0.294 -0.279 0.014 -0.533 -0.108 (3.19)*** (6.27)*** (5.91)*** (18.73)*** (6.79)*** (3.13)*** (7.68)*** (0.17) (9.56)*** (0.96) Age (years) 0.509 0.457 0.507 0.269 0.036 0.003 0.548 0.237 0.340 0.104 (20.98)*** (32.72)*** (59.36)*** (75.51)*** (11.89)*** (1.64)* (38.25)*** (21.20)*** (51.76)*** (7.32)*** Age (years) squared -0.023 -0.019 -0.023 -0.012 -0.002 -0.000 -0.023 -0.012 -0.014 -0.005 (19.22)*** (31.98)*** (60.33)*** (80.30)*** (15.55)*** (3.26)*** (36.61)*** (21.56)*** (51.43)*** (8.31)*** Male (0,1) 0.092 0.085 0.120 -0.007 -0.012 -0.004 -0.071 -0.001 -0.007 0.099 (6.70)*** (6.91)*** (15.76)*** (2.30)** (4.93)*** (3.56)*** (5.94)*** (0.18) (1.31) (7.84)*** Urban (0,1) 0.191 0.098 0.044 0.059 -0.001 0.001 -0.087 0.123 0.095 0.073 (4.40)*** (7.59)*** (3.19)*** (16.25)*** (0.27) (0.97) (5.29)*** (13.74)*** (15.68)*** (3.80)*** Quintile 2 0.006 -0.044 0.034 0.044 0.014 0.003 0.056 -0.000 0.068 0.022 (0.24) (2.23)** (3.09)*** (11.52)*** (4.90)*** (2.51)** (3.09)*** (0.01) (10.73)*** (1.08) Quintile 3 0.126 -0.009 0.093 0.077 0.018 0.003 0.145 0.019 0.091 0.093 (5.54)*** (0.44) (8.37)*** (19.89)*** (6.52)*** (2.61)*** (8.19)*** (1.90)* (14.65)*** (4.49)*** Quintile 4 0.164 0.065 0.196 0.103 0.021 0.003 0.198 -0.010 0.091 0.172 (7.27)*** (3.36)*** (17.83)*** (25.83)*** (7.66)*** (3.43)*** (10.73)*** (0.98) (14.14)*** (8.10)*** Quintile 5 0.266 0.232 0.321 0.134 0.023 0.004 0.219 0.002 0.101 0.271 (10.91)*** (10.61)*** (25.50)*** (30.42)*** (9.13)*** (3.60)*** (10.33)*** (0.13) (15.21)*** (11.19)*** Observations 5834 10880 23719 64136 6952 1547 7593 25796 13777 7339 Absolute value of t statistics in parentheses. *, **, *** indicates significant at 10%, 5%; and 1% levels. Models include region dummies 21 Annex Table 6. Marginal Effects from probit regressions. Dependent variable: Ever attended school, ages 6 to 17 Burundi 2000 Cambodia 1999 Cambodia 2000 Indonesia 2003 Jamaica 1998 Mongolia 2000 Myanmar 2000 Romania 1995 Sierra Leone 2000 Disabled (0,1) -0.189 -0.565 -0.314 -0.529 -0.185 -0.367 -0.002 -0.503 -0.035 (3.36)*** (7.84)*** (7.50)*** (22.39)*** (9.26)*** (11.01)*** (0.03) (11.26)*** (0.34) Age (years) 0.495 0.281 0.308 0.024 0.004 0.317 0.135 0.103 0.094 (19.89)*** (23.27)*** (45.00)*** (41.25)*** (6.04)*** (34.48)*** (14.92)*** (31.88)*** (6.69)*** Age-Squared -0.021 -0.010 -0.012 -0.001 -0.000 -0.012 -0.006 -0.004 -0.004 (17.58)*** (19.44)*** (38.98)*** (38.75)*** (6.50)*** (30.38)*** (12.35)*** (29.38)*** (7.21)*** Male (0,1) 0.107 0.022 0.038 -0.001 -0.001 -0.015 -0.000 0.000 0.103 (7.58)*** (1.99)** (5.92)*** (2.55)** (1.32) (1.82)* (0.09) (0.03) (8.13)*** Urban (0,1) 0.198 0.078 0.057 0.002 0.001 0.019 0.083 0.004 0.094 (4.40)*** (6.67)*** (5.11)*** (5.66)*** (1.06) (1.66)* (11.13)*** (1.70)* (4.93)*** Quintile 2 0.010 -0.053 0.035 0.002 0.002 0.026 0.002 0.017 0.010 (0.42) (3.02)*** (3.94)*** (6.05)*** (2.50)** (2.28)** (0.23) (7.08)*** (0.50) Quintile 3 0.127 -0.018 0.075 0.003 0.001 0.061 0.015 0.021 0.088 (5.49)*** (1.01) (8.42)*** (8.19)*** (0.96) (5.85)*** (1.88)* (9.58)*** (4.35)*** Quintile 4 0.151 0.052 0.154 0.004 0.002 0.066 0.002 0.022 0.162 (6.67)*** (3.12)*** (17.73)*** (11.37)*** (1.95)* (5.74)*** (0.18) (10.60)*** (7.81)*** Quintile 5 0.261 0.183 0.228 0.005 0.001 0.107 0.005 0.019 0.243 (10.65)*** (9.41)*** (22.74)*** (13.32)*** (1.62) (9.18)*** (0.47) (7.64)*** (10.22)*** Observations 5838 10880 23737 64136 6952 7619 26317 13776 7448 Absolute value of t statistics in parentheses. *, **, *** indicates significant at 10%, 5%; and 1% levels. Models include region dummies 22 Annex Table 7. Marginal Effects from probit regressions. Dependent variable: Participating in school, ages 6 to 17 Burundi Cambodia 1999 Cambodia 2000 Indonesia 2003 Jamaica Jamaica Mongolia 2000 Myanmar 2000 Romania 1995 Sierra Leone 2000 1998 2000 2000 Disabled (0,1) -0.163 -0.581 -0.414 -0.620 -0.363 -0.632 -0.449 0.069 -0.682 0.131 (1.49) (5.85)*** (5.20)*** (6.89)*** (4.32)*** (3.65)*** (5.57)*** (0.50) (5.17)*** (0.61) Age (years) 0.510 0.457 0.507 0.269 0.036 0.002 0.549 0.237 0.340 0.104 (21.00)*** (32.69)*** (59.38)*** (75.49)*** (11.94)*** (0.85) (38.25)*** (21.20)*** (51.76)*** (7.33)*** Age - squared -0.023 -0.019 -0.023 -0.012 -0.002 -0.000 -0.023 -0.012 -0.014 -0.005 (19.25)*** (31.95)*** (60.35)*** (80.28)*** (15.60)*** (2.10)** (36.61)*** (21.55)*** (51.43)*** (8.32)*** Male (0,1) 0.091 0.083 0.119 -0.007 -0.012 -0.003 -0.074 -0.001 -0.008 0.101 (6.65)*** (6.77)*** (15.56)*** (2.28)** (4.87)*** (3.42)*** (6.08)*** (0.17) (1.55) (7.91)*** Urban (0,1) 0.189 0.097 0.043 0.059 -0.001 0.001 -0.094 0.123 0.095 0.074 (4.33)*** (7.48)*** (3.06)*** (16.34)*** (0.38) (1.21) (5.62)*** (13.77)*** (15.66)*** (3.83)*** Quintile 2 0.005 -0.044 0.034 0.044 0.013 0.002 0.055 -0.000 0.068 0.022 (0.24) (2.24)** (3.11)*** (11.52)*** (4.92)*** (2.30)** (3.07)*** (0.01) (10.75)*** (1.06) Quintile 3 0.125 -0.007 0.094 0.077 0.018 0.002 0.146 0.019 0.091 0.093 (5.46)*** (0.35) (8.50)*** (19.87)*** (6.58)*** (2.70)*** (8.20)*** (1.90)* (14.61)*** (4.46)*** Quintile 4 0.162 0.067 0.198 0.103 0.021 0.002 0.201 -0.010 0.092 0.172 (7.18)*** (3.44)*** (17.95)*** (25.81)*** (7.76)*** (3.54)*** (10.79)*** (0.98) (14.12)*** (8.10)*** Quintile 5 0.263 0.234 0.323 0.134 0.024 0.002 0.221 0.002 0.101 0.271 (10.80)*** (10.67)*** (25.60)*** (30.40)*** (9.22)*** (3.60)*** (10.37)*** (0.16) (15.15)*** (11.17)*** Disability*Male 0.025 0.179 0.096 -0.008 -0.011 0.001 0.094 0.031 0.086 -0.270 (0.17) (1.47) (1.26) (0.18) (0.45) (0.46) (1.47) (0.20) (2.35)** (1.47) Disability*Urban 0.291 0.056 0.131 -0.092 0.009 -0.001 0.171 -0.651 -0.034 -0.301 (1.03) (0.45) (1.29) (1.53) (0.58) (0.26) (2.54)** (2.13)** (0.50) (1.39) Disability*(Quint. 1 or 2) -0.167 0.196 0.153 0.005 0.012 0.091 -0.018 0.008 0.001 (1.17) (1.59) (2.18)** (0.10) (1.07) (1.23) (0.10) (0.13) (0.00) Observations 5834 10880 23719 64136 6952 1543 7593 25796 13777 7339 Absolute value of t statistics in parentheses. *, **, *** indicates significant at 10%, 5%; and 1% levels. Models include region dummies 23