WPS5743 Policy Research Working Paper 5743 Does Parental Disability Matter to Child Education? Evidence from Vietnam Nguyen Viet Cuong Daniel Mont The World Bank East Asia and Pacific Region Poverty Reduction & Economic Management Sector Department August 2011 Policy Research Working Paper 5743 Abstract This paper examines the effect of parental disability performance is small and not statistically significant. The on school enrollment and educational performance conclusion of the paper is that to achieve the Millennium for children in the 2006 Vietnam Household Living Development Goal of universal primary school as Standards Survey. Results from instrumental-variables well as increased coverage of secondary education, the regressions indicate that children of parents with a government should have policies and programs that disability have a lower enrollment rate in primary either directly support the education of children with and secondary school of about 8 percentage points: disabled parents and/or have policies that support 73 percent compared with 81 percent. However, the disabled adults, thus lessening the incentive for their association of parental disability with educational children not to attend school. This paper is a product of the Poverty Reduction & Economic Management Sector Department, East Asia and Pacific Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http:// econ.worldbank.org. The author may be contacted at dmont@worldbank.org. 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 views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Does Parental Disability Matter to Child Education? Evidence from Vietnam Nguyen Viet Cuong Daniel Mont National Economics University, Hanoi The World Bank Keywords: Child education, disability, impact evaluation, household survey, Vietnam. JEL classification: I12; I21; O15 The findings, interpretations and conclusions expressed herein are those of the authors, and do not necessarily reflect the views of the International Bank for Reconstruction and Development/The World Bank and its affiliated organizations, or those of the Executive Directors of The World Bank or the governments they represent. 1. Introduction Child education is of fundamental importance to all governments, as shown by the inclusion of universal primary school enrollment as one of the Millennium Development Goals. Thus educational economics has long investigated the determinants of child education (e.g. see Becker, 1965; Becker and Tomes, 1976; Leibowitz, 1974; Tansel, 1997, Black et al., 2005). Sometimes investment in childhood education loses out to the use of child labor. In developing countries, when families are faced with poverty they often turn to their children to provide additional income (Basu and Van, 1998, Swinnerton and Rogers, 1999, Grootaert and Kanbur, 1995) and so make the decision to forego their children‟s education explicitly to generate more current household income (Edmonds, 2003; Edmonds and Turk, 2004; Beegle, Dehejia, and Gatti, 2005). Other parental characteristics, apart from income, also influence children‟s enrollment. For example, there is a clear correlation between high levels of education in parents and the enrollment and achievements of their children (e.g., Behrman et al., 1999; Glick and Sahn, 2000). Childhood characteristics also matter. For example, historically girls have been less likely to attend school than boys (World Bank, 2001), although this gap has been closing (United Nations 2008). Another factor that can limit children‟s education is the presence of a disability. In fact, Filmer (2008) found that childhood disability was a better predictor of school enrollment than gender and other socio-economic characteristics. Evidence in Vietnam also points towards lower enrollment in school for people with disabilities (Mont and Cuong, 2011). Yet, little is known about the effect of parental disability on child education. Parental disability could potentially be associated with poor education outcomes among children for a few reasons. Disability can result in lower labor productivity and less income. Some evidence suggests a strong association with poverty (Braithwaite and Mont 2009, Hoogeveen 2005, Yeo and Moore 2003), primarily through the effect on employment (Mitra and Sambamoorthi 2008, Contreras et al. 2006, Eide and Loeb 2006, 2 Eide and Kamaleri 2009, Mete 2008, Trani and Loeb 2010). Therefore, children who have disabled parents might have to work for income, undertake household production typically done by parents, and even possibly care for their parent. These extra demands on their time could lead to less time spent on education. Moreover, disability can decrease parents‟ involvement in their children‟s education because of the increased costs and demands on their time (Zaidi and Burchardt 2005). Child education depends crucially on the extent parents can get involved with their children in learning activities at home (e.g. Williams and Chavkin, 1989; Greenwood and Hickman, 1991; Trusty, 1996; Desforges, 2003). Disabled parents may be less able to supervise their children‟s educational activities because of various barriers in society - for example lack of accessible transportation to visit schools and teachers, or lack of training in sign language or Braille which limits their communication skills. If a parent has significant cognitive disabilities, that may also limit their ability to assist their children in learning. However, once education and income are controlled for, it is not clear whether parental disability can have negative impacts on child education, since parents with disabilities who have managed to obtain high levels of education and earnings have shown evidence of overcoming the barriers disabled people may face in their societies. In this paper, we will examine the extent to which the disability status of parents can affect the school enrollment and educational performance of children aged from 6 to 17 years old (age of primary and secondary school) in Vietnam. Around the world many more children are attending primary school, but as countries approach universal primary education, the remaining out of school students often have particular challenges that are not easily addressed by standard policies aimed to boost enrollment. Parental disability – at least in Vietnam – appears to be one of these factors. For several reasons, Vietnam is an interesting case. First, there have been no studies on the relation between parents‟ health, especially parental disability, and child education in Vietnam. Vietnam has implemented a large number of policies to promote education, especially primary and lower secondary education. Findings from this study can also provide helpful information for improving these policies on educational support. Currently, supports on education are provided mainly for the poor and ethnic minorities in Vietnam. If parental disability can have negative impacts on child education, then 3 maybe children whose parents are disabled should also receive educational supports or their families should receive supports and the creation of a more inclusive environment that eases their participation in society. Presumably, that would make education a more attractive choice for their children. Second, the Vietnam Household Living Standards Survey, which was conducted in 2006, has a special module on disability. Detailed data on disability allow for the analysis of the relation between disability and education. Third, Vietnam has a similar education system and economic development as many Asian countries, such as Thailand, Philippines, Indonesia, Laos, and Cambodia. Findings from this study might be relevant to not only Vietnam but also to other Asian developing countries. The remainder of this paper is organized as follows. Section 2 introduces the data sources used in this study. Section 3 provides some descriptive data on disability and child education in Vietnam. Section 4 presents estimation of the impact of parental disability on child education. Finally, Section 5 concludes. 2. Data source Data used in this study are from the Vietnam Living Standards Survey (VHLSS) in 2006. The 2006 VHLSS was conducted by the General Statistics Office of Vietnam (GSO) with technical support from World Bank. The survey contains detailed information on the characteristics of individuals, households and communes. Individual and household data include basic demographics, employment and labor force participation, education, health, income, expenditure, housing, fixed assets and durable goods, and the participation of households in socioeconomic programs. VHLSSs have been collected every two years since 2002, meaning there are four available VHLSSs from 2002 to 2010. However, unlike other VHLSSs, the 2006 VHLSS contains two special modules on education and disability. Information on education of individuals includes educational attainment and educational records in the schooling year 2005-2006 in primary, secondary and high schools. Information on disability includes types and reasons of different disabilities such as difficulty in hearing, seeing, walking, etc. 4 The commune questionnaires collect information on commune characteristics that affect the local living standards such as infrastructure, demography, and socioeconomic conditions. Commune data can be linked to household and individual data, but are only available in rural areas. The 2006 VHLSS covers 9,189 households. The number of individuals in the survey is 39,071. This sample is representative for rural and urban areas, and all eight geographical regions. 3. Parental disability and child education in Vietnam One of the difficulties in analyzing disability is in definition and measurement. In this paper, we follow the approach of the UN Statistical Commission‟s Washington Group on Disability Statistics (Washington Group, 2009), which relies on the functional model of disability underlying the International Classification of Functioning, Disability and Health (WHO, 2009). Incorporated in this approach is the social model of disability, which conceives of disability as emerging from the interaction of functional limitations with barriers in the environment. That is a person is disabled – that is, not capable of fully participating in various aspects of society – if they have difficulties in functioning that are not accommodated for in the environment – where environment is interpreted broadly as including the physical, cultural, and policy environments. Identifying disability in this way is quite complex. For the purposes of this paper – and following the Washington Group recommendations for assessing the impact of disability on equal opportunity – this paper uses the presence of difficulties in basic activities including seeing, hearing, walking, cognition, communication, and self-care as an operational proxy of functional limitation that puts him or her at risk of being disabled in the social model sense (Mont, 2007). Following Loeb, Eide, and Mont (2008), we will define a person to be disabled if she or he has a little difficulty in at least two of the 5 functional domains (seeing, hearing, walking, cognition, communication, and self-care), or a lot of difficulty in (or unable to do) one or more of the domains.1 The 2006 VHLSS shows that the fraction of pupils aged from 6 to 17 who have at least a mother or a father with disabilities is around 5.8 percent (Table 1). Older children are more likely to have parents with disabilities. This is because older children tend to have older parents and people are more likely to become disabled as they get older. Table 1 also shows there is a negative correlation between disability and household welfare. People who have higher education and consumption are less likely to have disabilities. Table 1 also presents the schooling rate of children with parents having a disability and children with parents not having disability. There is a large difference in the schooling rate between these two groups of children. Children with disabled parents have a lower school enrollment rate (73.1 percent) than other children (85.6 percent). Interestingly, girls have a higher proportion of education enrollment than boys. Among children whose parents have disabilities, the schooling proportion for girl was 77.5 percent, which is much higher than the proportion of 68.4 percent for boys. The gap in education between children with and without a disabled parent is largest for children aged from 15 to 17, which is the age for upper-secondary school. This gap in education tends to be larger for disadvantaged groups such as rural children, ethnic minority children, children living with a household head of low education degree, and poor children. For example, in the lowest expenditure quintiles, children whose parents have disabilities have an enrollment rate 14 percentage points lower than children whose parents do not have disabilities. Among children in school, those with disabled parents also have worse educational performance. However, this difference is not very large. Table 2 shows that the fraction of pupils achieving excellent academic performance in the last school year is 14 percent for children without disabled parents, while this figure is around 11 percent for 1 In the 2006 VHLSS, people are asked about their difficulties in different functions including seeing, hearing, walking, cognition, communication, and self-care, and their responses are classified into 4 categories: a. No - no difficulty b. Yes – some difficulty c. Yes – a lot of difficulty d. Cannot do at all. 6 children whose parents have a disability. The table also shows a strong relation between children‟s educational performance with other household characteristics. As expected, children in better-off households such as urban, Kinh (the ethnic majority whom all are native Vietnamese speakers), and households with high income and education are more likely to have better academic performance. Table 1. Proportion of children with parents with disability and proportion of schooling children Proportion Schooling rate of children by disability of of children their parents (%) Groups with parents Parents Parents with without with Total disability (%) disability disability Total 5.64 85.59 73.11 84.88 Gender Girls 5.84 86.93 77.49 86.38 Boys 5.46 84.31 68.42 83.44 Age Age 6-10 3.21 95.72 91.94 95.60 Age 11-14 6.00 91.82 82.85 91.28 Age 15-17 7.37 69.10 56.66 68.19 Ethnic minorities Kinh 5.17 86.69 75.12 86.09 Ethnic minorities 8.00 79.98 66.26 78.88 Rural/Urban Rural 5.67 84.44 70.61 83.65 Urban 5.55 89.95 82.29 89.52 Regions Red River Delta 4.42 90.23 83.70 89.94 North East 6.98 90.71 70.50 89.30 North West 3.37 81.48 81.21 81.47 North Central Coast 4.21 86.79 79.50 86.48 South Central Coast 4.44 88.26 90.59 88.36 Central Highlands 8.18 86.79 69.55 85.38 South East 7.18 83.66 73.97 82.96 Mekong River Delta 5.97 76.06 55.29 74.82 Education of household head No degree 9.65 72.57 58.85 71.24 Primary school degree 5.23 83.73 68.10 82.91 Lower-secondary school degree 4.19 89.14 85.33 88.98 Upper-secondary school degree 4.43 93.63 96.99 93.78 Post secondary degree 4.19 96.62 94.52 96.53 7 Proportion Schooling rate of children by disability of of children their parents (%) Groups with parents Parents Parents with without with Total disability (%) disability disability Consumption quintiles Poorest 7.12 78.15 63.94 77.13 Near poorest 5.66 84.45 74.25 83.88 Middle 4.65 87.58 71.48 86.83 Near richest 4.91 89.35 78.88 88.84 Richest 5.34 93.23 88.18 92.96 Number of observations 8,790 507 8,283 8,790 Source: Estimation from the 2006 VHLSS Table 2. Proportion of schooled children by their educational performance Children with parents without disability Parents with disability Groups Excellent Good Middle, Total Excellent Good Middle, Total Bad Bad Total 14.01 37.29 48.7 100 11.02 39.82 49.16 100 Gender Girls 17.03 40.51 42.46 100 13.27 49.22 37.51 100 Boys 11.04 34.11 54.85 100 8.38 28.86 62.76 100 Age Age 6-10 20.85 35.71 43.44 100 12.05 41.47 46.48 100 Age 11-14 12.27 39.31 48.42 100 11.13 38.61 50.26 100 Age 15-17 8.53 36.03 55.44 100 10.27 40.56 49.17 100 Ethnic minorities Kinh 15.89 40.15 43.96 100 13.31 46.62 40.07 100 Ethnic minorities 3.5 21.27 75.23 100 2.77 15.45 81.78 100 Rural/Urban Rural 10.88 35.75 53.37 100 6.09 38.63 55.28 100 Urban 25.15 42.76 32.09 100 27.53 43.81 28.66 100 Regions Red River Delta 15.41 46.35 38.24 100 6.76 50.3 42.94 100 North East 8.98 37.85 53.17 100 7.8 18.01 74.18 100 North West 5.92 19.53 74.55 100 0 21.27 78.73 100 North Central Coast 7.83 37.42 54.75 100 6.64 51.06 42.3 100 South Central Coast 19.98 30.85 49.17 100 6.01 31.86 62.13 100 Central Highlands 10.62 28.29 61.09 100 7.91 43.68 48.41 100 South East 22.21 38.59 39.2 100 24.03 37.17 38.8 100 Mekong River Delta 15.41 36.69 47.9 100 14.74 47.32 37.94 100 Education of household head No degree 4.95 27.52 67.53 100 4.29 30.78 64.93 100 Primary school 11.58 34.02 54.41 100 10.86 41.11 48.02 100 degree Lower-secondary 11.59 40.39 48.03 100 9.26 43.75 46.99 100 school degree 8 Upper-secondary 18.03 44.39 37.58 100 27.37 53.71 18.92 100 school degree Post secondary 33.71 43.56 22.73 100 18.37 38.19 43.44 100 degree Consumption quintiles Poorest 4.87 26.51 68.62 100 4.11 34.12 61.77 100 Near poorest 8.26 37.48 54.27 100 6.81 40.01 53.17 100 Middle 13.2 39.79 47.01 100 9.19 32.97 57.84 100 Near richest 18.32 43.51 38.17 100 12.13 50.35 37.52 100 Richest 32.85 41.88 25.27 100 30.96 45.95 23.09 100 Number of 34 141 205 380 924 2,471 3,650 7,045 observations Source: Estimation from the 2006 VHLSS 4. Impacts of parental disability on child education 4.1. Estimation method To measure the impact of disability of parents on child education, we assume the follow function: PY  j | X , D  G  X  D , (1) where Y is an indicator of educational performance of children, D is the dummy indicating the disability of parents of the children, and X is a vector of control variables including individual and household characteristics which can affect educational performance. In this paper we use two education indicators. The first is school enrollment, which is equal 1 for children attending school, and 0 otherwise. The second is their academic record, which is equal to 1, 2 and 3 for „excellent‟ record, „good‟ record and „normal and bad‟ record, respectively.2 When the dependent variable is school enrollment, we will use a probit model to estimate equation (1). A problem in non-linear functions is that the meaning of the coefficients is not clear. Thus to estimate the impact of parental disability on school 2 As mentioned, we combine pupils with the „bad‟ record and pupils with the „normal‟ record, since there are only around 1 percent of pupils having the „bad‟ record. 9 enrollment, we use the Average Treatment Effect on the Treated, which is the most popular parameter in the impact evaluation literature (Heckman et al., 1999). This parameter is expressed as follows: ATT  E Y( D 1) | D  1  E Y( D  0) | D  1, (2) where E Y( D0) | D  1 is the expected value of educational variables of children whose parents have a disability had there not been a disability. This is not observed and has to be estimated. Once equation (1) is estimated, ATT can be computed as follows:     nD ˆT  1     X   ˆ     X  , AT ˆ ˆ ˆ ˆ (3) i i nD i 1 where nD is the number of children whose parents have a disability, and  is the cumulative distribution function of the standard normal distribution (since we use a probit ˆ model).3 Since the probit function is non-linear, and ATT estimated by (3) can vary slightly across X. We can allow the effect of parental disability to vary more remarkably across X by including interactions between D and X. The equation (1) becomes: PY  j | X , D  G  X  D  XD  . (4) The ATT is estimated by: ˆ ATT  1 nD  G   X i ˆ  ˆ  X iˆ  G   X i ˆ . nD i 1 ˆ  ˆ   (5) ATT can be estimated for children who have the value of the X variables equal to x:  G  x ˆ  ˆ  x ˆ  G   x ˆ  nD,X x ˆ 1 ATTX  x  ˆ i i ˆ i (6) nD , X  x i 1 Where nD , X  x is the number of children who have parents with disabilities and the value of the X variables equal to x. 3 We estimate the standard error of the ATT estimates by using a non-parametric bootstrap technique. More specifically, we repeatedly draw random samples from the original VHLSS 2006. Since the VHLSSs sample selection follows stratified random cluster sampling, communes instead of households are bootstrapped in each stratum (Deaton, 1997). The number of replications is 1000. 10 An important issue in estimating the effect of disability of parents on child education is the endogeneity of parental disability. There can be unobserved variables which affect both disability of parents and child education. When randomization is not possible, instrumental variables regression is the traditional way to deal with endogeneity. Finding a convincing instrument when both outcome and intervention (disability and education in this case) are indicators of individual welfare is challenging. In this study, we use the age of father as an instrument for parental disability. For children who do not live with a father, mother‟s age is used instead. We expect that age of parent is strongly correlated with their disability (Mont and Cuong 2011), but not correlated with child education given that the model controls for important factors affecting child education such as demography of children, education and income of parents. The second instrument is the proportion of people with disabilities within a district.4 This proportion reflects the risky level in the district. It can strongly affect the probability of being disabled of an individual within district but would not affect the school enrollment of children. This so- called internal instrumental variable is also used in empirical studies such studies on the effect of migration (e.g., Mansuri, 2006; Acosta, 2006). The migration network is often used as the instrument for the probability of migration of a household or individual. When the dependent variable is the academic record of children in school with the three mutually exclusive choices of „excellent‟, „good‟ and „normal and bad‟, we will use an ordered probit regression. A problem is that there are no available models of ordered probit (as well as other multiple responses models such as ordered logit, multinomial logit or probit) with instrumental variables. In this study, we follow a control function approach. As noted in Wooldridge (2007), we can estimate an ordered probit with instrumental variables regression using a two-step approach. In the first step, we regress disability on explanatory variables and instruments, and predict residuals for all the observations. In the second step, we estimate the effect of parental disability consistently using an ordered probit model with the dependent variable of children‟s academic record and independent variables including parental disability, explanatory variables and predicted residuals from the first step. 4 For each household, we calculate the proportion of disabled people within the district of residence, excluding the household itself. The average number of sampled households per district in the VHLSSs is 15. 11 4.2. Estimation results Impacts on school enrollment The regressions of children‟s school enrollment on parental disability are reported in Table A.3. Following studies on child education, the control variables (variables X) are education of parents, household income, age and sex of children, and geographic variables (e.g., Black et al., 2005; Lee, 2008). It should be noted that parents are labeled as household head and head‟s spouse to indicate the role of household head. The summary statistics of control variables are reported in Table A.1 in the Appendix. To examine the sensitivity of the estimates to model specifications, we compare three different models, which mostly vary in the number of explanatory variables. Model 1 includes only exogenous variables, and Model 2 adds household heads‟ education. Model 3 also includes per capita income. It should be noted that parents‟ education and household income can be affected by disability, thus they should not be included in the model (Heckman et al., 1999). However, since these variables are important factors for education of children, we examine whether the estimates of the impact of parental disability are sensitive to these variables. For each model, we estimate using both probit and probit with instrumental variables regressions. Table A.2 presents the first-stage regression of parental disability. It shows that both age of father and proportion of disability strongly increase the probability of parental disability. We perform a Cragg-Donald weak identification test of the instruments, and the statistic is extremely high, indicating that the instruments are strong.5 Very strong instruments are expected to reduce the bias caused by invalid instruments. We can also perform the test of valid instruments by an overidentification test which is reported in Table A.3. It shows that for the three models, we are not able to reject the null hypothesis of valid instruments at the 5% significant level. However, the test on endogeneity of 5 As a rule of thumb, if a test is under 10, the instruments might be weak (Staiger and Stock 1997). 12 parental disability does not reject the exogeneity of parental disability in the child education equation. Table 3 presents the estimate of ATT of parental disability. The estimates are quite similar in the three models. The effect of parental disability on children‟s education enrollment is negative and statistically significant at the 5% significant level in all the models. The point estimates are also quite similar between different models. According to Model 1 in Probit with IV regressions, parental disability reduces child schooling rate by around 8.1 percentage points from 81.3 percent to 73.2 percent. This means children whose parents have a disability have a school enrollment rate of 73.2 percent, but if their parents had not been disabled, their schooling rate would have been around 81.3 percent. It should be noted that adding education and income of households (Model 2 and 3, both with and without instruments) leads to a smaller estimate of the effect of parental disability. This implies that education of parents and income can be channels through which disability can affect children‟s education. Put differently, disability can lead to lower education and lower income, thereby lower education for their children. Table 3: The impact of parental disability on child schooling rate Probit regressions Probit with IV regressions Y1 Y0 ATT=Y1–Y0 Y1 Y0 ATT=Y1–Y0 Model 1 0.732*** 0.811*** -0.081*** 0.732*** 0.813*** -0.081*** (0.025) (0.010) (0.024) (0.024) (0.011) (0.024) Model 2 0.732*** 0.788*** -0.056*** 0.732*** 0.793*** -0.061*** (0.024) (0.012) (0.022) (0.024) (0.013) (0.023) Model 3 0.732*** 0.786*** -0.054** 0.732*** 0.793*** -0.061*** (0.024) (0.012) (0.021) (0.024) (0.013) (0.022) Note: Y1 is the observed schooling rate of children whose parents have disabilities. Y0 is the counterfactual schooling rate that children would have if their parent not been disabled. ATT=Y1–Y0 is the effect of parental disability on children’s schooling rate. * significant at 10%; ** significant at 5%; *** significant at 1% Figures in parentheses are standard errors. Standard errors are corrected for sampling weights and estimated using bootstrap (non-parametric) with 500 replications. Source: Estimation from the 2006 VHLSS. We estimated a probit with instrumental variables regression including interactions between parental disability and different characteristics of children including gender, urbanity, ethnicity and age. The instrumental variables for the interactions are the interactions between these interacted characteristics and the instruments for parental 13 disability (i.e., age of father and the district proportion of disabled people). The probit regressions are reported in Table A.4 in the Appendix. Table 4 presents the estimate of ATT for different child groups. We are not able to reject the equality of the impact estimates between different groups. However, Table 4 shows some interesting findings. The point estimate shows a stronger (negative) effect of parental disability on boys than on girls, given that culturally females are more likely to care for people in need, this may suggest that the main driver for lower enrollment may be resulting more from the need to undertake livelihood production to account for the extra costs associated with disability. This is also consistent with the finding that the biggest impact on enrollment is for children aged 15 to 17. These results are also consistent with the descriptive data presented in Table 1, in which there is a large gap in the schooling rate between girls and boys in households with parental disability. The adverse effect of parental disability on child education is found larger in urban areas and Kinh children. Table 4: The impact of parental disability on child schooling rate for different groups Model specification Groups Y1 Y0 ATT All 0.731*** 0.816*** -0.086*** (0.024) (0.011) (0.023) Model including interaction between Boy 0.684*** 0.810*** -0.126*** child gender and (0.032) (0.014) (0.030) disability of parents Girl 0.776*** 0.822*** -0.047 (0.030) (0.014) (0.030) All 0.731*** 0.818*** -0.087*** (0.024) (0.011) (0.023) Model including interaction between Urban 0.823*** 0.870*** -0.046 urbanity and (0.048) (0.021) (0.051) disability of parents Rural 0.708*** 0.805*** -0.096*** (0.027) (0.013) (0.026) All 0.731*** 0.817*** -0.087*** Model including (0.024) (0.011) (0.024) interaction between Ethnic minorities 0.662*** 0.713*** -0.051 ethnic minority of children and (0.049) (0.028) (0.051) disability of parents Kinh 0.750*** 0.850*** -0.101*** (0.026) (0.011) (0.026) Model including All 0.731*** 0.822*** -0.091*** interaction between (0.024) (0.011) (0.023) children’s age and disability of parents Aged 6-10 0.920*** 0.945*** -0.024 14 (0.032) (0.010) (0.032) Aged 11-14 0.829*** 0.856*** -0.028 (0.031) (0.011) (0.030) Aged 15-17 0.569*** 0.743*** -0.174*** (0.039) (0.016) (0.037) Note: Y1 is the observed schooling rate of children whose parents have disabilities. Y0 is the counterfactual schooling rate, that is their enrollment if their parent had not been disabled. ATT=Y1–Y0 is the effect of parental disability on children’s schooling rate. * significant at 10%; ** significant at 5%; *** significant at 1% Figures in parentheses are standard errors. Standard errors are corrected for sampling weights and estimated using bootstrap (non-parametric) with 500 replications. Source: Estimation from the 2006 VHLSS. Table A.5 in the Appendix presents the ordered probit regressions of academic records. We tried both ordered logit with and without instruments for parental disability. Estimates of the parental variables are very small and not statistically significant in all the regressions. This means that once attending school, children with disabled parents do not have worse educational performance than children with non-disabled parents given that the estimation procedure controls for other observed characteristics. Finally, the regressions of school enrollment and academic records show some interesting findings on other factors associated with education. Girls have higher school enrollment as well as better educational performance than boys even after other observed variables are controlled. Gender equality in children‟s education is very important, since it can contribute to gender equality in wage and other social economic activities when the children become adults. Other control variables have expected signs. Older children tend to drop out of school. Urban children, Kinh children, and children whose parents have higher education and income are much more likely to attend school and have better educational records once attending school than rural and ethnic minority children, and children in lower education and income households. 15 5. Conclusions In this paper, we used data from the 2006 VHLSSs and probit regressions to estimate the effect of parental disability on education of children age from 6 to 17. Our estimates indicate that disability among parents reduces the school attendance rate of their children around 8.1 percentage points from 81.3 percent to 73.2 percent. The effect tends to be larger for boys, Kinh children, and children in urban areas. We found an especially strong effect of parental disability on children aged from 15 to 17, i.e., upper secondary school age children. However, disability of parents does not have a significant effect on the academic performance of their children in school. There can be several reasons why children whose parents are disabled are less likely to attend school. First, parental disability can reduce the education as well as income of parents, which then decreases child education. Second, children might have increased responsibilities at home if their disabled parents cannot generate a sufficient livelihood or if they require personal assistance that they cannot access outside the family. Third, children might be reliant on their parents to physically get to school, which disabled parents – e.g., facing transportation barriers – cannot do. Fourth, the increased costs and demands on disabled parents‟ time might leave them less time and energy to look after their children‟s education. This fourth reason, however, is not supported by the fact that once they attend school, children of disabled parents have the same level of achievement as their peers. These findings suggest clear policy implications. To achieve the Millennium Development Goal of universal primary school as well as increased coverage of secondary education, the government should have policies and programs supporting the education of children whose parents have a disability. One approach would be to implement education supports, such as a reduction or exemption of tuition fees and contributions or possibly assistance with transportation. But an approach which gets more directly at the costs of disability would be policies that provide supports to parents (such as rehabilitation) or help build a more inclusive environment so that a parent having a functional limitation imposes less of an imperative for their children to forego their education to address their family‟s immediate needs. 16 References Acosta, Pablo (2006), “Labor Supply, School Attendance, and Remittances from International Migration: The Case of El Salvador”, World Bank Policy Research Working Paper No. 3903. Basu K. and Van P.K. (1998), “The Economics of Child Labour”, American Economic Review, Vol. 88 Becker, G., and N. Tomes (1976), “Child Endowments and the Quantity and Quality of Children”, Journal of Political Economy, 84, 398-419. Becker, G.S. (1965), “A Theory of the Allocation of Time”, Economic Journal, 75: 493– 517. Beegle, K., R. Dehejia, and R. Gatti (2005). “Why Should We Care about Child Labor? The Education, Labor Market, and Health Consequences of Child Labor.” World Bank Policy Research Working Paper 3479. CEPR Discussion Paper 4443. NBER Working Paper No.10980. Behrman, J.R., A.D. Foster, M.R. Rosenzweig and P. Vashishtha (1999), “Women‟s Schooling, Home Teaching, and Economic Growth”, The Journal of Political Economy, 107(4, August): 682–714. Black, Sandra, Paul J. Devereux, Kjell G. Salvanes (2005), “The More the Merrier? The Effect of Family Size and Birth Order on Children's Education”, Quarterly Journal of Economics, Vol. 120(2), pp. 669-700. Braithwaite, J. and d. Mont (2009), “Disability and Poverty: A Survey of World Bank Poverty Assessments and Implications”, forthcoming in Alter: The European Journal of Disability Research, 2009 Contreras, D., J. Ruiz-Tagle, P. Garces‡ I. Azocar (2006), “Socio-Economic Impact of Disability in Latin America: Chile and Uruguay,” Working Paper, Department of Economics, University of Chile Deaton Angus (1997), “The Analysis of Household Surveys”, the Johns Hopkins University Press, Baltimore, Maryland, U.S.A. 17 Desforges, C (2003), “The Impact of Parental Involvement, Parental Support and Family Education on Pupil Achievement and Adjustment”, DfES Research Report 433. Eide, A. and Y. Kamaleri (Eds.) (2009), “Living Conditions among People with Disabilities in Lesotho: A National Representative Study,” Report A17239, SINTEF Health Research, Oslo. Eide, A.H., & Loeb, M.E. (Eds.) (2006), “Living Conditions among people with activity limitations in Zambia: a National Representative Study,” Report No. A262, SINTEF Health Research, Oslo. Edmonds, E. V.(2003), "Does Child Labor Decline with Improving Economic Status?," NBER Working Papers 10134, National Bureau of Economic Research, Inc Edmonds, E, and C. Turk. (2004), "Child Labor in Transition." In P. Glewwe, N. Agrawal and D. Dollar, eds., Economic Growth, Poverty and Household Welfare: Policy Lessons from Viet Nam. Washington D.C.: World Bank Glick, P. and D.E. Sahn (2000), “Schooling of girls and boys in a West African country: The effects of parental education, income and household structure”, Economics of Education Review, 19: 63–87. Greenwood, G. E., & Hickman, C. W. (1991), “Research and practice in parent involvement: Implications for teacher education”, Elementary School Journal, 91(3), 279-288. Grootaert, C. and Kanbur, R. (1995), “Child Labour: A Review” Policy Research Working Paper No. 1454, Washington, D.C: The World Bank. Heckman, J., R. Lalonde and J. Smith (1999), "The Economics and Econometrics of Active Labor Market Programs", Handbook of Labor Economics 3, Ashenfelter, A. and D. Card, eds., Amsterdam: Elsevier Science. Hoogeveen, J. (2005), “Measuring Welfare for Small but Vulnerable Groups: Poverty and Disability in Uganda”, Journal of African Economies, Vol. 14, No. 4, pp.603-631, August 2005. 18 Lee, J. (2008), “Sibling size and investment in children's education: An Asian instrument”, Journal of Population Economics, Vol. 21(4), pp. 855-875. Leibowitz, A. (1974), “Home Investments in Children”, Journal of Political Economy, 82(2): s111–s131. Loeb, Mitchell E., A. H. Eide, D. Mont (2008), “Approaching the measurement of disability prevalence: The case of Zambia”, Alter: European Journal of Disability Research . Mansuri, Ghazala (2006) “Migration, Sex Bias, and Child Growth in Rural Pakistan,” World Bank Policy Research Working Paper No. 3946. Mete, C. (Ed.) (2008), Economic Implications of Chronic Illness and Disease in Eastern Europe and the Former Soviet Union, The World Bank. Mitra, S. and U. Sambamoorthi (2008), “Disability and the Rural Labor Market in India: Evidence for Males in Tamil Nadu”, World Development, Vol. 36(5), pp. 934-952. Mont, D. (2007), “Measuring Disability Prevalence”, SP Discussion Paper No. 0706. The World Bank, March 2007. Mont, D. and N.V. Cuong (2011), “Disability and Poverty in Vietnam,” World Bank Economic Review, forthcoming Staiger, Douglas, and Jame H. Stock (1997), “Instrumental Variables Regression with Weak Instruments.” Econometrica 65, no. 3: 557–86. Swinnerton, K. A. and Carol Ann Rogers (1999), "The Economics of Child Labor: Comment," American Economic Review, American Economic Association, vol. 89(5), pp 1382-1385, December Tansel, A. (1997), “Schooling Attainment, Parental Education, and Gender in Côte d‟Ivoire and Ghana”, Economic Development and Cultural Change, 45(4, July): 825–56. Trani, J. and Loeb M. (2010).” Poverty and disability: A vicious circle? Evidence from Afghanistan and Zambia,” Journal of International Development 19 Trusty, J. (1996), “Relationship of Parental Involvement in Teens‟ Career Development to Teens‟ Attitudes, Perceptions and Behavior”, Journal of Research and Development in Education, 30(1), 63-69. United Nations, (2008), “The Millennium Development Goals Report 2008” Washington Group (2009), http://www.cdc.gov/nchs/citygroup.htm WHO, ICF website (2009), http://www.who.int/classifications/icf/en/ Williams, D.L. & Chavkin, N.F. (1989), “Essential elements of strong parent involvement programs”, Educational Leadership, 47, 18-20. Wooldridge, J. (2007), “Control Function and Related Methods”, Lecture notes of course “What‟s New in Econometrics?” in 2007 in National Bureau of Economic Research. Yeo, R. and K. Moore (2003), “Including Disabled People in Poverty Reduction Work: Nothing About Us, Without Us”, World Development, Vol. 31, No. 3 pp.571-590, 2003. Zaidi, A., and T. 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. 20 Appendix Table A.1: Variable description Explanatory variables Type Children having parents Children having parents with disability without disability Mean Std. Dev. Mean Std. Dev. Gender (male = 1, female = 0) Binary 0.494 0.500 0.512 0.500 Age Discrete 13.436 2.910 12.420 3.253 Per capita income (million VND) Continuous 6.074 5.105 7.163 8.251 Head no degree Binary 0.359 0.480 0.201 0.401 Head with primary school degree Binary 0.256 0.437 0.277 0.448 Head with lower-secondary school degree Binary 0.230 0.421 0.314 0.464 Head with upper-secondary school degree Binary 0.069 0.254 0.089 0.285 Head with post secondary degree Binary 0.086 0.281 0.118 0.322 Red River Delta Binary 0.148 0.356 0.192 0.394 North East Binary 0.147 0.354 0.117 0.321 North West Binary 0.021 0.145 0.037 0.188 North Central Coast Binary 0.120 0.325 0.163 0.369 South Central Coast Binary 0.074 0.262 0.096 0.294 Central Highlands Binary 0.127 0.334 0.085 0.280 South East Binary 0.182 0.386 0.141 0.348 Mekong River Delta Binary 0.181 0.385 0.170 0.376 Urban (yes = 1) Binary 0.206 0.405 0.209 0.407 Number of observations 507 8283 Source: Estimation from the 2006 VHLSS. 21 Table A.2: First-stage regressions of parental disability Explanatory variables Model 1 Model 2 Model 3 Age of father 0.0070*** 0.0070*** 0.0070*** (0.001) (0.001) (0.001) Proportion of people with disability at the district 0.4960*** 0.5163*** 0.5143*** (0.065) (0.065) (0.066) Gender (male = 1, female = 0) -0.0046 -0.0052 -0.0050 (0.005) (0.005) (0.005) Age -0.0013* -0.0011 -0.0011 (0.001) (0.001) (0.001) Ethnic minorities (yeas = 1) 0.0339*** 0.0156* 0.0148 (0.008) (0.009) (0.009) Head no degree Omitted Head with primary school degree -0.0315*** -0.0310*** (0.009) (0.009) Head with lower-secondary school degree -0.0501*** -0.0487*** (0.009) (0.009) Head with upper-secondary school degree -0.0428*** -0.0401*** (0.012) (0.012) Head with post secondary degree -0.0637*** -0.0602*** (0.011) (0.011) Income per capita (million VND) -0.0005** (0.000) Red River Delta Omitted North East 0.0234** 0.0224** 0.0225** (0.009) (0.009) (0.009) North West -0.0079 -0.0121 -0.0121 (0.012) (0.012) (0.012) North Central Coast -0.0092 -0.0105 -0.0110 (0.008) (0.008) (0.008) South Central Coast -0.0006 -0.0104 -0.0101 (0.009) (0.009) (0.009) Central Highlands 0.0239** 0.0154 0.0155 (0.011) (0.011) (0.011) South East 0.0211** 0.0066 0.0086 (0.010) (0.010) (0.010) Mekong River Delta 0.0110 -0.0081 -0.0069 (0.008) (0.009) (0.009) Urban (urban = 1; rural = 0) -0.0092 -0.0002 0.0011 (0.006) (0.007) (0.007) Constant -0.2679*** -0.2269*** -0.2249*** (0.021) (0.022) (0.022) Observations 8,788 8,788 8,788 R-squared 0.060 0.066 0.067 Cragg-Donald weak identification test 242.7 243.7 243.2 Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Source: Estimation from the 2006 VHLSS. 22 Table A.3: Logit regressions of school enrollment Probit regression Probit with IV regression Explanatory variables Model 1 Model 2 Model 3 Model 1 Model 2 Model 3 Parental disability (yes = 1, no = 0) -0.3134*** -0.2253*** -0.2161*** -0.6362* -0.6583** -0.7031** (0.072) (0.073) (0.072) (0.330) (0.335) (0.337) Gender (male = 1, female = 0) -0.1470*** -0.1422*** -0.1539*** -0.1321*** -0.1261*** -0.1366*** (0.038) (0.039) (0.039) (0.036) (0.037) (0.037) Age -0.1780*** -0.1889*** -0.1931*** -0.1719*** -0.1826*** -0.1868*** (0.009) (0.010) (0.010) (0.007) (0.007) (0.007) Ethnic minorities (yeas = 1) -0.5279*** -0.2642*** -0.2102*** -0.4716*** -0.1944*** -0.1343** (0.059) (0.063) (0.064) (0.056) (0.059) (0.059) Head no degree Omitted Head with primary school degree 0.3620*** 0.3335*** 0.3747*** 0.3420*** (0.049) (0.049) (0.049) (0.050) Head with lower-secondary school 0.6810*** 0.6187*** 0.7083*** 0.6443*** degree (0.060) (0.060) (0.057) (0.058) Head with upper-secondary school 0.9792*** 0.8708*** 1.0009*** 0.9044*** degree (0.107) (0.109) (0.095) (0.096) Head with post secondary degree 1.2942*** 1.1477*** 1.3249*** 1.1717*** (0.106) (0.111) (0.097) (0.100) Income per capita (million VND) 0.0343*** 0.0350*** (0.007) (0.005) Red River Delta Omitted North East 0.1957** 0.2431*** 0.2261*** 0.2118*** 0.2565*** 0.2409*** (0.083) (0.085) (0.086) (0.078) (0.081) (0.081) North West -0.0942 0.0238 0.0200 -0.2234** -0.0666 -0.0692 (0.104) (0.108) (0.108) (0.095) (0.100) (0.100) North Central Coast -0.1597** -0.1408* -0.1139 -0.1461** -0.1212* -0.0935 (0.072) (0.073) (0.073) (0.070) (0.072) (0.072) South Central Coast -0.1356* 0.0913 0.0924 -0.1556** 0.0729 0.0736 (0.078) (0.083) (0.084) (0.077) (0.082) (0.082) Central Highlands -0.1934** -0.0332 -0.0563 -0.1648** -0.0000 -0.0291 (0.081) (0.084) (0.085) (0.079) (0.082) (0.083) South East -0.4465*** -0.1840** -0.2746*** -0.4816*** -0.1942*** -0.2667*** (0.072) (0.078) (0.081) (0.070) (0.074) (0.075) Mekong River Delta -0.6554*** -0.3184*** -0.3741*** -0.6643*** -0.3236*** -0.3763*** (0.063) (0.069) (0.070) (0.062) (0.068) (0.069) Urban (urban = 1; rural = 0) 0.3121*** 0.1241** 0.0699 0.3183*** 0.1323** 0.0907 (0.055) (0.058) (0.058) (0.051) (0.055) (0.056) Constant 3.7922*** 3.2912*** 3.2095*** 3.7158*** 3.2029*** 3.1189*** (0.138) (0.148) (0.149) (0.107) (0.114) (0.115) Observations 8,788 8,788 8,788 8,788 8,788 8,788 R-squared 0.166 0.209 0.217 Wald test of exogeneity 1.14 1.82 2.27 (P-value in parentheses) (0.28) (0.17) (0.13) Test of overidentifying restrictions 2.96 0.12 0.28 (P-value in parentheses) (0.09) (0.73) (0.59) Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Source: Estimation from the 2006 VHLSS. 23 Table A.4. Probit with instrumental variables regressions including interactions between Parental disability and children‟s charactersitics Explanatory variables Dependent variable is school enrollment (yes = 1) Parental disability (yes = 1, no = 0) 8.9437*** 0.1884 -1.2452*** -0.6143* (2.482) (0.505) (0.467) (0.351) Parental disability * age -0.6616*** (0.167) Parental disability * gender -1.4416** (0.659) Parental disability * ethnic minorities 1.4787** (0.609) Parental disability * urban -0.1886 (0.942) Gender (male = 1, female = 0) -0.1406*** -0.0334 -0.1374*** -0.1320*** (0.037) (0.057) (0.036) (0.036) Age -0.1418*** -0.1726*** -0.1731*** -0.1717*** (0.010) (0.007) (0.007) (0.007) Ethnic minorities (yeas = 1) -0.4935*** -0.4661*** -0.6026*** -0.4725*** (0.058) (0.056) (0.075) (0.056) Red River Delta Omitted North East 0.2270*** 0.2109*** 0.2065*** 0.2124*** (0.080) (0.078) (0.078) (0.078) North West -0.1762* -0.2181** -0.1696* -0.2218** (0.100) (0.096) (0.097) (0.096) North Central Coast -0.1449** -0.1415** -0.1490** -0.1450** (0.073) (0.071) (0.071) (0.071) South Central Coast -0.1540* -0.1458* -0.1617** -0.1529* (0.080) (0.078) (0.078) (0.078) Central Highlands -0.1895** -0.1564** -0.1547* -0.1622** (0.082) (0.079) (0.079) (0.079) South East -0.4860*** -0.4806*** -0.4750*** -0.4802*** (0.072) (0.070) (0.070) (0.070) Mekong River Delta -0.6610*** -0.6563*** -0.6669*** -0.6636*** (0.065) (0.063) (0.063) (0.062) Urban (urban = 1; rural = 0) 0.3452*** 0.3218*** 0.3235*** 0.3255*** (0.053) (0.051) (0.052) (0.079) Constant 3.3009*** 3.6625*** 3.7702*** 3.7122*** (0.145) (0.110) (0.110) (0.108) Observations 8,788 8,788 8,788 8,788 Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Source: Estimation from the 2006 VHLSS 24 Table A.5. Ordered probit model of educational performance: marginal effect Ordered probit regression Ordered probit with instrumental variables Explanatory variables Prob Prob Prob Prob Prob Prob (Excellent) (Good) (Normal, (Excellent) (Good) (Normal, Bad) Bad) Parental disability (yes=1,no=0) 0.0019 0.0020 -0.0039 0.0024 0.0026 -0.0050 (0.0128) (0.0131) (0.0259) (0.0665) (0.0698) (0.1363) Gender (male = 1, female = 0) -0.0662*** -0.0685*** 0.1347*** -0.0662*** -0.0685*** 0.1347*** (0.0058) (0.0062) (0.0115) (0.0058) (0.0062) (0.0116) Age -0.0118*** -0.0124*** 0.0241*** -0.0118*** -0.0124*** 0.0241*** (0.0010) (0.0011) (0.0020) (0.0010) (0.0011) (0.0021) Ethnic minorities (yeas = 1) -0.1044*** -0.1750*** 0.2795*** -0.1044*** -0.1751*** 0.2795*** (0.0056) (0.0137) (0.0181) (0.0057) (0.0140) (0.0186) Red River Delta Omitted North East -0.0168* -0.0191* 0.0359* -0.0168* -0.0191* 0.0359* (0.0093) (0.0113) (0.0206) (0.0093) (0.0113) (0.0206) North West -0.0492*** -0.0699*** 0.1191*** -0.0492*** -0.0699*** 0.1191*** (0.0125) (0.0236) (0.0360) (0.0126) (0.0237) (0.0363) North Central Coast -0.0514*** -0.0671*** 0.1185*** -0.0514*** -0.0671*** 0.1185*** (0.0074) (0.0120) (0.0192) (0.0074) (0.0120) (0.0192) South Central Coast -0.0259*** -0.0310** 0.0568** -0.0259*** -0.0310** 0.0569** (0.0094) (0.0128) (0.0221) (0.0094) (0.0128) (0.0222) Central Highlands -0.0596*** -0.0864*** 0.1460*** -0.0596*** -0.0864*** 0.1460*** (0.0078) (0.0151) (0.0226) (0.0078) (0.0151) (0.0227) South East -0.0080 -0.0087 0.0168 -0.0081 -0.0088 0.0168 (0.0099) (0.0112) (0.0212) (0.0100) (0.0113) (0.0213) Mekong River Delta -0.0206** -0.0236** 0.0442** -0.0206** -0.0236** 0.0442** (0.0086) (0.0108) (0.0194) (0.0086) (0.0108) (0.0194) Urban (yes = 1) 0.1163*** 0.0815*** -0.1978*** 0.1163*** 0.0815*** -0.1978*** (0.0097) (0.0049) (0.0135) (0.0097) (0.0049) (0.0135) Observations 7,257 7,257 Note: the coefficients present the probability of different categories of academic record. Robust standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. Source: Estimation from the 2006 VHLSS 25