MPRA Munich Personal RePEc Archive Spatial Variation in the Disability-Poverty Correlation: Evidence from Vietnam Mont Daniel and Cuong Nguyen 15. June 2013 Online at http://mpra.ub.uni-muenchen.de/48659/ MPRA Paper No. 48659, posted 28. July 2013 09:10 UTC 1 Spatial Variation in the Disability-Poverty Correlation: Evidence from Vietnam Daniel Mont Leonard Cheshire Disability and Inclusive Development Centre University College London Cuong Nguyen National Economics University in Hanoi Abstract Poverty and disability are interrelated, but data that can disentangle to what extent one causes the other and vice versa is not available. However, data from Vietnam allows us to examine this interrelationship in a way not done previously. Using small area estimation techniques, we uncover three findings not yet found in the literature. First, disability prevalence rates vary significantly within a county even at the district level. Second, the correlation between disability and poverty also varies at the district level. And most importantly, the strength of that correlation lessens based on district characteristics that can be affected by policy. Districts with better health care and infrastructure, such as road and health services, show less of a link between disability and poverty, supporting the hypothesis that improvements in infrastructure and rehabilitation service can lessen the impact of disability on families with disabled members. Keywords: Poverty, disability, small area estimation, household survey, population census, Vietnam. JEL codes: I12, I31, O15 Acknowledgments: The authors would like to thank Sophie Mitra for her comments on an earlier draft of this paper. 2 1. Introduction Growing evidence documents a link between disability and poverty across the globe, (WHO/World Bank 2011, Mitra, et al., 2013, Trani and Loeb 2010, Rischewski et al. 2008, Hoogeveen 2005, Yeo and Moore 2003,Elwan 1999), and in Vietnam, (in particular see:Mont and Nguyen 2011, Palmer et al. 2010, Braithwaite and Mont 2009,). However, the relationship between disability and poverty is complex. Often it is characterized as a vicious circle, with poverty as both a cause and consequence of disability (Yeo and Moore, 2003). Poverty creates the conditions that increase disability – for example, malnutrition, poor sanitation, dangerous working conditions, and lack of access to good health care. Disability can create poverty – or prevent its escape – because of barriers to education and employment. However, when one looks at the empirical relationship between consumption measures of poverty and disability, the link is not always strong. In the broadest available look at the relationship of disability and poverty, Mitra et al. (2013) found that only four countries showed a significant relationship. In fact, while growing incomes can lessen the rate of poverty by ameliorating many of the factors mentioned above, growing incomes can also increase disability rates. First, primarily by leading to longer life expectancies. Disability rates are much higher for older people (WHO/World Bank 2011). And not only do richer societies have longer life expectancies, but among people with later onset disabilities the link to poverty is weaker (Mont and Nguyen 2011, Demographic Institute, 2013). Not being disabled when of working age, people who become disabled as older adults have not had their education, training, employment, and years of asset building affected by disability. And the richer they are, the more they have been able to afford health care, rehabilitative services, or assistive devices that can help them survive disabling conditions that might have otherwise proved fatal. Nevertheless, Mitra et al. (2013) found a significant correlation between disability and multidimensional poverty in most of the developing countries under study when looking at various measures of exclusion, such as deficits in education, employment, life expectancy, etc The World Report on Disability (WHO/World Bank 3 2011) reports a wide literature showing this to be the case. It also points out that disability is not a rare event. Globally, the prevalence rate for disability is about 15 percent, and about 4 percent for those with severe disabilities. The percentage of people living in households with a disabled member is much higher. And it should be remembered that disability impacts family members, as well, by affecting their schooling and work decisions. In Vietnam, for instance, children of parents with disabilities are significantly less likely to attend school (Mont and Nguyen 2013). Moreover, having a disability imposes extra costs (Tibble 2005, Zaidi and Burchardt 2005) such as extra medical costs, assistive devices, and special transportation needs. In fact, studies estimate that in Vietnam disability increases the cost of living by about 10% (Braithwaite and Mont 2009, Mont and Nguyen 2011). Thus, the relationship between disability and poverty – adjusting for those costs – is even stronger. Disentangling the effects of disability on poverty and vice versa, though, is difficult. To our knowledge, a panel data set that could be used to examine the transitions in and out of both states is not available.Moreover, as Mitra et al. 2013 state “whether disability and poverty are causally related is an empirical question and the answer will be environment specific.”Indeed, we hypothesize that various factors may lessen the link between disability and poverty. For example, improved roads and transportation systems could lessen the barriers that disabled people face in obtaining education and employment, or even participating in community events. To the extent those systems are more inclusive, the barriers to participating in things such as work would become even less. Also, improved access to health and rehabilitation services could increase functional capabilities of individuals. And, the more people with disabilities move about their communities, the more they can break down existing stereotypes and misconceptions that might be serve as attitudinal barriers to their increased participation in society. This paper uses a unique source of data to explore how local characteristics – within a single country – could influence the link between disability and poverty. While data directly related to inclusion – for example, accessibility audits of infrastructure and the availability of assistive devices – are not available, the hypothesis is that improved 4 infrastructure related to those concepts – better roads, more doctors, and a more developed infrastructure (e.g., communication and transportation systems, electrification, etc) – can make people with disabilities and their families less likely to experience poverty. As such, this is the first empirical paper the authors are aware of that explores not only the relation between disability and poverty, but also what specific factors influence that relationship. The findings in this paper can potentially be useful for policymakers in two regards. First, because these techniques can be used to identify potential policy levers for lessening the link between disability and poverty, but secondly because they can identify regional differences in disability rates and the disability-poverty connection that can be useful in targeting programs. The remainder of this paper is organised as follows. Section 2 briefly presents the data sets used in this study. Section 3 presents the methodology to investigate the association between poverty and disability. Next, section 4 presents the empirical findings. Finally, section 5 concludes. 2. Data This study relies on two main data sets. The first is the 15-percent sample of the Vietnam Population and Housing Census (referred as the 2009 VPHC). The 2009 VPHC was conducted in April 2009 by the General Statistics Office of Vietnam (GSO)with technical assistance from the United Nations Population Fund (UNFPA). The 2009 VPHC is designed to be representative at the district level. 1 It covered 3,692,042 households with 14,177,590 individuals. The 2009 VPHC contains data on individuals and households. Individual data include demographics, education, 1 Administratively (?) Vietnam is divided into 63 provinces. Each province is divided into districts, and each district is further divided into communes (communes are called wards in urban areas). Communes are the smallest administrative areas. In 2009, there were 690 districts and 10,896 communes. 5 employment, disability and migration. Household data include durable assets and housing conditions. The 2009 VPHC also contains data on disability of people aged 5 and above. Respondentswere asked about their difficulties in four basic functional domains including seeing, hearing, walking, and remembering. There are four multiple exclusive responses which are as follows: (i) no difficulty, (ii) some difficulty, (iii) a lot of difficulty and (iv) cannot do at all. 2 These were the minimum four census questions recommended by the United Nation Statistical Commission’s Washington Group on Disability Statistics (hereafter referred to as the Washington Group). 3 The second dataset is the 2010 Vietnam Household Living Standard Survey (VHLSS). The 2010 VHLSS was carried out by GSO with technical support from the World Bank in Vietnam. The 2010 VHLSS covers 9,402 households with 37,012 individuals, who are sampled from the population frame of the 2009 Population Census. The 2010 VHLSS is representative for rural/urban areas and 6 geographic regions. The 2010 VHLSScontains very detailed data on demographic and living standards of individuals, households and communes. Individual data includeinformation on demographics, education, employment, health and migration, while household data include information on durables, assets, production, income and expenditure, and participation in government programs. However, there are no data on disability in the 2010 VHLSS. In this study, we define a household as poor if their real per capita expenditure is below the GSO-World Bank expenditure poverty line of 653 thousand VND/month/person (7836 thousand VND/year/person). Under this line the poverty rate of Vietnam in 2010 is 20.7 percent. 3. Methodology 2 There is a full population census which was conducted in April 2009. However, this census contains onlylimited data on basic demographic and housing data. There are no data on disability in the full census. Thus we do not use the full census in this study. 3 See http://www.cdc.gov/nchs/washington_group.htm 6 Poverty gaps between households with and households without disabled members The main objective of this study is to examine the spatial correlation between poverty and disability and subsequently investigate several factors associated with this disability- povertycorrelation in Vietnam. We will estimate the poverty measures for households with and without disabled members at the provincial and district level. Although the 2010 VHLSS contains expenditure data for households, it is not representative at the provincial as well as district level. On the contrary, the 2009 VPHC is representative at the district level, but it does not contain expenditure data to estimate the poverty measures. To overcome this data limitation, we will use a small area estimation method that essentially links the information in both data sets (Elbers et al. 2002, 2003). In Vietnam, this method has been widely applied to construct the poverty and inequality maps. (e.g., Minot et al., 2003; Nguyen et al., 2010; Nguyen, 2011; Lanjouw, 2012). The Elbers et al. (2002, 2003) method is used to combine a population census and a household survey to predict welfare measures such as poverty and inequality indicators for small areas. It can be described in three steps as follows. First, we select common variables of the census and the households. The common variables can include household- level variables, commune-level and district-level variables. Secondly, we regress the log of per capita expenditures on the common variables using the household survey. More specifically, we use the following model: ,)ln( iccicic Xy εηβ ++= (1) where )ln( ic y is log of per capita expenditure of household i in cluster c, ic X the vector of the common variables, β the vector of regression coefficients, c η the cluster-specific random effect and ic ε the household-specific random effect. The subscript ic refers household i living in cluster c. In the third step, we use the estimated model to predictper capita expenditure of households in the census: ( ) , ˆˆ ˆ exp icc Census ic Census ic Xy εηβ ++= (2) 7 where β ˆ , c η ˆ and ic ε ˆ denote the estimates for β , c η and ic ε . The predicted per capita expendituresof households are then used to estimate the mean expenditure and poverty indexes of provinces and districts. The poverty indexes include the poverty rate, the poverty gap index, and the squared poverty gap index. 4 It should be noted that the point estimates as well as the standard errors of the poverty estimates are calculated by Monte-Carlo simulations. In each simulation, a set of values β ˆ , c η ˆ and ic ε ˆ are drawn from their estimated distributions, and an estimate of per capita expenditure and the poverty indices are obtained. After k simulations, we can get the average and standard deviation over the k different simulated estimates of the expenditure and poverty indexes. In this study, we will estimate the poverty indexes of households with and without a disabled member at the regional, provincial and district levels. Using the data on disability in the 2009 VPHC, we can divide households into one group of households with a disabled member and another group of households without a disabled member. We can estimate the poverty indexes of the two groups of households, then compute the gap in poverty indexes between these two groups: NDDp PPG − = , (3) where p G is the gap in poverty indexes or mean expenditure, D P and ND P are the mean expenditure or poverty indexes of households with a disabled member and households without a disabled member, respectively. The gap in poverty can be regarded as a measure of the correlation between poverty and disability at the small areas. If there is no 4 Following Foster, Greer and Thorbecke (1984) the FGT class of poverty measures take the following form: ∑ ∑ −= α α ))/(1() 1 ()( zyw w FGT ii i Wherey i is per capita expenditure for those individuals with weight w i below the poverty line and zero for those above, z is the poverty line and ∑ i w is total population size. α is equal to 0 for the poverty rate, 1 for the poverty gap index (also called the poverty depth index), and 2 for the squared poverty gap index (also called the poverty severity index). 8 correlation between poverty and disability, we will expect a small difference in poverty between households with and households without disability. Regressions of poverty gaps between households with and households without disability We will examine several factors associated with the poverty-disability correlation. The poverty-disability correlationis measured by the gap in the poverty indexes between households with and without disabled members. We will run a regression of the gap in poverty indexes on several explanatory variables at the district level. Since the observations are districts and there can be a spatial correlation between dependent variables and error terms, we apply the following spatial model: dddd uXWGG + + + = β λ α (4) ddd Muu ε ρ + = (5) Where d G is the gap in poverty indexes between disabled and non-disabled households of district d, d X is a vector of explanatory variables of the district. W and M are spatial- weighting matrices (with zero diagonal elements). The dependent variables are allowed to be correlated with each other. The model is a type of spatial econometric model with the first-order spatial-autoregressive and first-order spatial-autoregressive disturbances (see, e.g., Haining, 2003; Drukker et al., 2010, 2011). W and M are spatial-weighting, which are set equal to each other and equal to the inverse-distance between centroids of districts. This matrix weight allows for the high correlation between close districts and low correlation between far districts. 4. Empirical results 4.1. Disability in Vietnam 9 Construction of an uncontroversial definition of disability is difficult. According to a measurement method suggested by the Washington Group, which was established by United Nations Statistical Division with the participation of over 100 National Statistical Offices and international agencies (Madans et al., 2010), disability is measured in household surveys by asking respondents about their difficulties in basic functional domains such as seeing, hearing, walking, self-care, cognition, and communication. (Schneider, 2009; Madans et al., 2010). The 2009 VPHC relies on a similar method suggested by the Washington Group on Disability Statistics to measure the disability. More specifically, interviewees are asked about their difficulties in the four basic functions including seeing, hearing, walking, and remembering. There are four multiple exclusive responses: (i) no difficulty, (ii) some difficulty, (iii) a lot of difficulty and (iv) unable to do (cannot do at all) 5 . Based on the availability of the 2009 VPHC data and following Loeb, Eide, and Mont (2008) and Mont and Nguyen (2011), we will define a person to be disabled if she or he has a little difficulty in at least two of the functional domains (seeing, hearing, walking, and remembering), or a lot of difficulty or unable to do at least one of the domains. The above measure of disability includes people with mild and moderate, as well as severe disabilities. In addition, we also conduct the analysis using a higherthreshold level fordisability, which is defined as having considerable difficulty (a lot of difficulty and unable to do) in at least one of the four functional domains. This measure of disability excludes those with only mild or moderate disabilities. Table 1 presents the proportion of people aged above five with difficulties in the four functional domains. There are 5.0 and 3.1 percent of respondents having difficulty in seeing and difficulty in hearing, respectively. The proportion of people having difficulty in walking and remembering is 3.7 and 3.5 percent, respectively. 5 The WG recommended six census questions, but set the minimum useful set as four questions, recognizing that space on censuses is often tight and some countries were resistant to including all six questions. Vietnam was one such country – and as such there is probably an underestimation of the rate of disability. [...]... expenditure of districts and the disability-poverty correlation The disability-poverty correlation as the mean expenditure increases The magnitude of the squared mean expenditure is very small, and there is no data on the right-hand side of the U-shape in which the disability-poverty correlation increases as the mean expenditure increases Table 5: Regression of difference in the poverty indexes between households... 0.480 The above table uses the sample of rural districts The poverty rate and the poverty gap indexes are measure in percent The difference in the poverty indexes is measured in percentage point Standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% Source: Estimates from the 2009 VPHDC and the 2010 VHLSS 5 Conclusions Increasing attention is being paid to the. .. 0.609 0.575 The poverty rate and the poverty gap indexes are measure in percent The difference in the poverty indexes is measured in percentage point Standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% Source: Estimates from the 2009 VPHDC and the 2010 VHLSS Table 7 shows regressions using variables linked to infrastructure and services, namely the quality... 0.330 0.346 The poverty rate and the poverty gap indexes are measure in percent The difference in the poverty indexes is measured in percentage point Standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1% Source: Estimates from the 2009 VPHDC and the 2010 VHLSS 19 In Table 6, we add regional and demographic variables There is a clear difference in the poverty-disability...Table 1: The proportion of people aged above five with difficulties in functional domains (in percent) Having difficulty in seeing Northern Mountain Red River Delta Having difficulty in walking Having difficulty in remembering 4.92 3.42 3.67 3.53 (0.07) Region Having difficulty in hearing (0.04) (0.04) (0.04) 5.08 3.91 (0.06) (0.06) 4.10 4.81 4.64... OLS Explanatory variables Spatial regression Difference in poverty rate (percentage point) Difference in poverty gap index (percentage point) Difference in poverty severity index (percentage point) Difference in poverty rate (percentage point) Difference in poverty gap index (percentage point) Difference in poverty severity index (percentage point) District are capitals of provinces (yes=1, no=0) -0.3150... Unfortunately, the lack of panel data sets prevents researchers from disentangling these effects and seeing which, if any, predominates However, data from Vietnam allows us to examine this interrelationship in a way not done previously Using small area estimation techniques, we found that disability rates vary across Vietnam – not just at the provincial level, but at the district level, as well Moreover, the. .. Explanatory variables thousand/km2) Difference in poverty rate (percentage point) Difference in poverty gap index (percentage point) Spatial regression Difference in poverty severity index (percentage point) Difference in poverty rate (percentage point) Difference in poverty gap index (percentage point) Difference in poverty severity index (percentage point) (2.9064) (1.2958) (0.8244) (3.7849) (1.1647)... members In the South East – which is much more economically developed – the respective poverty rates are about 10.8 percent and 6.6 percent Keeping in mind however , that the census only used the 4 WG questions and not the full 6 questions (thus missing some disabled people), and that these data do not account for the additional costs of living with a disability, these gaps probably understate the poverty... in the poverty-disability correlation between regions even when the mean expenditure is controlled for The gap in poverty between households with and without disability is highest in the Northern Mountains The gap is large in districts with a large proportion of ethnic minorities This corresponds with the hypothesis that less infrastructure and access to services strengthens the disability poverty association . difficulties in functional domains (in percent) Region Having difficulty in seeing Having difficulty in hearing Having difficulty in walking Having difficulty in remembering Northern. data on the right-hand side of the U-shape in which the disability-poverty correlation increases as the mean expenditure increases. Table 5: Regression of difference in the poverty indexes. 0.346 The poverty rate and the poverty gap indexes are measure in percent. The difference in the poverty indexes is measured in percentage point. Standard errors in parentheses * significant