Abstract
Estimating poverty measures for disabled people in developing countries is often difficult, partly because relevant data are not readily available. We extend the small-area estimation developed by Elbers, Lanjouw and Lanjouw (2002, 2003) to estimate poverty by the disability status of the household head, when the disability status is unavailable in the survey. We propose two alternative approaches to this extension: Aggregation and Instrumental Variables Approaches. We apply these approaches to data from Tanzania and show that both approaches work. Our estimation results show that disability is indeed positively associated with poverty in every region of mainland Tanzania.
Similar content being viewed by others
Notes
The proportion of disabled people would also depend on the way it is defined and measured. See Online Appendix for a related discussion on this issue.
See Mitra et al. (2017) for a review of literature on extra cost of disability (i.e., additional income that disabled people will need to achieve a given level of standards of living).
For example, community-level indicators are often merged into the census and survey. Once covariates from auxiliary datasets are merged into the census and survey, we regard these covariates to be part of the census and survey datasets for our purpose.
Both Mont and Cuong (2011) and Loyalka et al. (2014) provide poverty estimates by disability status only for urban and rural areas in Vietnam and China, respectively. Hoogeveen (2005) discussed below reports poverty rates by the disability status only for four regions. Mont and Nguyen (2018) provide poverty measures at the level of six regional estimates. We report poverty rates by the disability status of the household heads for 21 regions.
Mont and Nguyen (2018) also apply the ELL method to derive poverty measures for people with and without disability at a spatially disaggregated level. However, it does not allow for the difference in model parameters between these two groups.
Note that the regression in the second step is unweighted, since the accuracy of \({\hat{u}}^2_k\) as an estimate of \(\sigma ^2_{u,k}\) does not systematically improve as the number of observations in each cluster k grows. This is because the error term \(\eta \) does not go away by aggregation. Weighting can actually make the estimate of \(\sigma ^2_{\eta }\) less accurate.
If either \({\tilde{\sigma }}^{2,(r)}_{\eta }\) or \({\tilde{\sigma }}^{2,(r)}_{\epsilon }\) is negative, we redraw until both are positive.
Some regions were split after the survey was conducted. In 2002, the Manyara region broke out of the Arusha region and the census data allow us to distinguishes between Manyara and Arusha regions, but the survey data do not.
In Online Appendix Table A1, we report the distribution of types of disability.
Online Appendix Table A2 also shows that the head’s economic disability status is more strongly correlated with poverty than the presence of an economically disabled member, showing the importance of household head’s disability status relative to other member’s disability status. Online Appendix also provides further discussions.
In Online Appendix Table A4, we also report the summary statistics and correlations of \({\hat{u}}^2_k\), \({\bar{A}}_{k, \text {non-disabled}}\), and \({\bar{A}}_{k, \text {disabled}}\).
Online Appendix Table A6 reports the point estimates and standard errors by disability status in each region.
Note, however, that there is an ongoing effort to improve the data collection standards and data availability (Abualghaib et al. 2019; Groce and Mont 2017). The methodology proposed in this paper will remain relevant because it takes time to improve data availability and because it is still useful to see the change from the past situation.
In addition, we also implemented a version of Hoogeveen (2005), where we use the same set of covariates as those in Table 2, except that we replace the individual disability status with the cluster-level prevalence of disability. The point estimate (standard error) of \(P_0\) for non-disabled households, disabled, and mainland Tanzania are 0.490 (0.009), 0.554 (0.010), and 0.491 (0.008), respectively. Hence, poverty appears to be overestimated when a comparable model is used. While this apparent overestimation may be driven by a particular specification we use, the standard errors also appear to be too low, particularly for non-disabled households. This is also the case when the Instrumental Variables Approach is used. This results is expected, since Hoogeveen (2005) does not take into consideration the fact that the cluster-level prevalence of disability is just a proxy for the individual-level disability status.
References
Abualghaib O, Groce N, Simeu N, Carew MT, Mont D (2019) Making visible the invisible: why disability-disaggregated data is vital to “leave no-one behind’’. Sustainability 11(11):3091
Angrist J, Krueger A (1992) The effect of age at school entry on educational attainment: an application of instrumental variables with moments from two samples. J Am Stat Assoc 87(418):328–336
Araujo MC, Ferreira FHG, Lanjouw P, Özler B (2008) Local inequality and project choice: theory and evidence from Ecuador. J Public Econ 92(5–6):1022–1046
Bjorklund A, Jantti M (1997) Intergenerational income mobility in Sweden compared to the USA. Am Econ Rev 87(5):1009–1018
Braithwaite J, Mont D (2009) Disability and poverty: a survey of World Bank Poverty Assessments and implications. ALTER Eur J Disabil Res 3:219–232
Collier P, Radwan S, Wangwe S, Wagner A (1986) Labour and poverty in rural Tanzania: Ujamaa and rural development in the United Republic of Tanzania. Oxford University Press, Oxford
Currie J, Yelowitz A (2000) Are public housing projects good for kids? J Public Econ 75:99–124
de Weerdt J (2010) Moving out of poverty in Tanzania: evidence from Kagera. J Dev Stud 46(2):331–349
Dee TS, Evans WN (2003) Teen drinking and educational attainment: evidence from two-sample instrumental variable estimates. J Law Econ 21(1):178–209
Demombynes G, Özler B (2005) Crime and local inequality in South Africa. J Dev Econ 76(2):265–292
Elbers C, Lanjouw JO, Lanjouw P (2002) ‘Micro-level estimation of welfare.’ Policy Research Department Working Paper 2911, The World Bank
Elbers C, Lanjouw JO, Lanjouw P (2003) Micro-level estimation of poverty and inequality. Econometrica 71(1):355–364
Elbers C, Lanjouw JO, Lanjouw P (2005) Imputed welfare estimates in regression analysis. J Econ Geogr 5(1):101–118
Elbers C, Lanjouw P, Mistiaen A, Özler B, Simler K (2004) On the unequal inequality of poor communities. World Bank Econ Rev 18(3):401–421
Elbers C, Fujii T, Lanjouw P, Özler B, Yin W (2007) Poverty alleviation through geographic targeting. J Dev Econ 83(1):198–213
Elwan A (1999) ‘Poverty and disability: a survey of the literature.’ SP Discussion Paper 9932, The World Bank
Feige EL, Watts HW (1972) An investigation of the consequences of partial aggregation of micro-economic data. Econometrica 40(2):343–360
Filmer D (2008) Disability, poverty and schooling in developing countries: results from 14 household surveys. World Bank Econ Rev 22(1):141–163
Foster J, Greer J, Thorbecke E (1984) A class of decomposable poverty measures. Econometrica 52(3):761–766
Fujii T (2008) How well can we target aid with rapidly collected data?: empirical results for poverty mapping from Cambodia. World Dev 36(10):1830–1842
Fujii T (2010) Micro-level estimation of child undernutrition indicators in Cambodia. World Bank Econ Rev 24(3):520–553
Fujii T (2019) Regional prevalence of health worker absenteeism in Tanzania. Health Econ 28(2):311–316
Grech S (2016) Disability and poverty: complex interactions and critical reframings. In: Grech and Soldatic (eds) Disability in the Global South: the Critical Handbook. Springer. chapter 14, pp 217–236
Groce NE, Mont D (2017) Counting disability: emerging consensus on the Washington Group questionnaire. Lancet 5:E649–E650
Haveman R, Wolfe B (2000) The economics of disability and disability policy. In: Culyer AJ, Newhouse JP (eds) Handbook of Health Economics: Volume 1B, pp 995–1051
Hoogeveen J (2005) Measuring welfare for small but vulnerable groups poverty and disability in Uganda. J Afr Econ 14(4):603–631
Inoue A, Solon G (2010) Two-sample instrumental variable estimators. Rev Econ Stud 92(3):557–561
Ivaschenko O, Lanjouw P (2010) A new approach to producing geographic profiles of HIV prevalence: an application to Malawi. World Med Health Policy 2(1):235–266
Kisanji J (1995) Interface between culture and disability in the Tanzanian context: part I. Int J Disabil Dev Educ 42(2):93–108
Kisanji J (1995) Interface between culture and disability in the Tanzanian context: part II. Int J Disabil Dev Educ 42(2):109–124
Loeb ME, Eide AH (2004) Living conditions among people with activity limitations in Malawi: a national representative study. SINTEF Report STF78 A044511, SINTEF Health Research, Oslo, Norway
Loeb ME, Eide AH (2006) Living conditions among people with activity limitations in Zambia: a national representative study. SINTEF Report SINTEF A262, SINTEF Health Research, Oslo, Norway
Loyalka P, Liu L, Chen G, Zheng X (2014) The cost of disability in China. Demography 51:97–118
Lusardi A (1996) Permanent income, current income, and consumption: evidence from two panel data sets. J Bus Econ Stat 14(1):81–90
Masset E, White H (2004) Are chronically poor people being left out of progress towards the Millennium Development Goals? a quantitative analysis of older people, disabled people and orphans. J Hum Dev 5(2):279–297
Mitra S (2004) Disability and social safety nets in developing countries. SP Discussion Paper 0509, The World Bank
Mitra S (2018) Disability, health and human development. Palgrave, New York
Mitra S, Palmer M, Kim H, Mont D, Groce N (2017) Extra costs of living with a disability: a review and agenda for research. Disabil Health J 10:475–484
Mont D, Nguyen C (2018) Spatial variation in the poverty gap between people with and without disabilities: evidence from Vietnam. Soc Indic Res 137:745–763
Mont D, Cuong NV (2011) Disability and poverty in Vietnam. World Bank Econ Rev 25(2):323–359
National Bureau of Statistics (2002) Household Budget Survey 2000/01 (Dar es Salaam, Tanzania: National Bureau of Statistics, United Republic of Tanzania)
National Bureau of Statistics (2003a) 2002 Population and Housing Census: General Report (Dar es Salaam, Tanzania: National Bureau of Statistics, United Republic of Tanzania)
National Bureau of Statistics (2003b) 2002 Population and Housing Census: Volume I Methodology Report (Dar es Salaam, Tanzania: National Bureau of Statistics, United Republic of Tanzania)
Polinsky AM (1977) The demand for housing: a study in specification and grouping. Econometrica 45(2):447–461
Sohnesen TP, Ambel AA, Fisker P, Andrews C, Khan Q (2017) Small area estimation of child undernutrition in Ethiopian woredas. PLoS ONE 12(4):e0175445
Taylor HR, Katala S, Muñoz B, Turner V (1991) Increase in mortality associated with blindness in rural Africa. Bull World Health Organ 69(3):335–338
UNICEF (1999) Children in Need of Special Protection Measures. Dar es Salaam, Tanzania
United Nations (2019) Disability and Development Reprot 2018: Realizing the Sustainable Development Goals by, for and with persons with disabilities. New York: United Nations
Welsch RE, Kuh E (1976) The variances of regression coefficient estimates using aggregate data. Econometrica 44(2):353–363
White H (1984) Asymptotic theory for econometricians. Academic Press, London
World Bank (2015) Tanzania Mainland Poverty Assessment. Washington DC, USA: World Bank. Available from http://hdl.handle.net/10986/22021
World Health Organization (2011) World Report on Disability. World Health Organization
Yeo R, Moore K (2003) Including disabled people in poverty reduction work: “nothing about us, without us’’. World Dev 31(3):571–590
Zaidi A, Burchardi T (2005) Comparing incomes when needs differ: Equivalization for extra costs of disability in the U.K. Rev Income Wealth 51(1):89–114
Funding
An earlier version of this study was supported by SMU Research Grant (07-C208-SMU-007).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
Fujii declares that he has no conflict of interest.
Ethical approval
All procedures performed in studies involving human participants were in accordance with the ethical standards of the Singapore Management University and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This research grew out of my consulting work for the World Bank. I benefited from discussion with Chris Elbers, Hans Hoogeveen, Denis Leung, Wietze Lindeboom, Roy van der Weide, and an anonymous referee. An earlier version of this study was supported by SMU Research Grant (07-C208-SMU-007). Usual caveats apply.
Supplementary Information
Below is the link to the electronic supplementary material.
Appendices
Appendix
Appendix A: Comparison of regional poverty estimates
Table 5 provides the survey-only estimates of poverty rates at this level in column (1) and their corresponding SAE estimates in column (2) at the level of 20 survey regions. The survey-only and SAE estimates are generally close and have a strong positive correlation of 0.71. The null hypothesis for a two-sided z-test of equality of the survey-only and SAE estimates is not rejected at a five percent level of significance in all regions. The null hypothesis for the joint \(\chi ^2\)-test of the equality of the two estimates in all regions was also not rejected. Therefore, on balance, even though the difference between survey-only and SAE estimates can be large in absolute value for a few regions, there is no evidence that survey-only and SAE estimates are systematically different.
In column (3), we report the share of people living in disabled households. As the comparisons of column (3) with columns (1) and (2) show, there is a strong positive correlation between the poverty rate and share of people in disable households. The correlation coefficient between columns (1) and (3) [(2) and (3)] is 0.66 [0.68]. Because the calculation of the correlation between columns (1) and (3) does not involve any imputation, the positive correlation cannot be attributed to imputation.
Appendix B: Proofs
Proof of Proposition 1
We first establish the consistency of the cluster-level OLS estimator \({\hat{\varvec{\beta }}}_{COLS}\equiv ({\bar{\varvec{X}}}^T_C{\bar{\varvec{X}}}_C)^{-1}{\bar{\varvec{X}}}_C{\bar{\varvec{Y}}}_S\). Letting \(j=0\) in Assumptions A3–A5 and using \(K_S\rightarrow \infty \) as \(K_C\rightarrow \infty \), we have:
This means that we can obtain consistent estimates of \(\sigma ^2_\eta \) and \(\sigma ^2_{\epsilon ,g}\) by running a regression of the squared OLS residual \({\hat{u}}^2_k\) on a constant and \({\bar{A}}_{k,1},\cdots ,{\bar{A}}_{k,G}\). Hence, we can obtain a consistent estimate \({\hat{\varvec{\Omega }}}_A\) of \(\varvec{\Omega }_A\) by replacing \(\sigma ^2_\eta \) and \(\sigma ^2_{\epsilon ,g}\) with their consistent estimates in the definition of \(\varvec{\Omega }_A\). Then, letting \({\hat{\varvec{J}}}_1\equiv {\hat{\varvec{\Omega }}}_A\varvec{W}_S{\hat{\varvec{\Omega }}}_A\), we have \({\hat{\varvec{J}}}_1\xrightarrow {p}\varvec{J}_1\) as \(K_C\rightarrow \infty \). From this and assumptions A1–A5 and using \(K_S\rightarrow \infty \) as \(K_C\rightarrow \infty \), we have the following results as \(K_C\rightarrow \infty \):
where \(\Xi _0\equiv \lim _{K_S\rightarrow \infty } K^{-1/2}_S {\bar{\varvec{X}}}^T_C{\hat{\varvec{J}}}_1({\bar{\varvec{X}}}_S-{\bar{\varvec{X}}}_C) \varvec{\beta }\varvec{\beta }^T({\bar{\varvec{X}}}_S-{\bar{\varvec{X}}}_C)^T{\hat{\varvec{J}}}_1{\bar{\varvec{X}}}_C\). Notice that the between-cluster variations in the regressors are already differenced out in \(({\bar{\varvec{X}}}_S-{\bar{\varvec{X}}}_C)\). Therefore, \(({\bar{\varvec{X}}}_S-{\bar{\varvec{X}}}_C)\) is influenced by the within-cluster variation in the regressor. With a sufficiently large fraction of households sampled in each cluster, the contribution of \(\Xi ^{-1}_1\Xi _0\Xi ^{-1}_1\) to the asymptotic variance of \({\hat{\varvec{\beta }}}_{AGG}\) is small relative to \(\Xi ^{-1}_1\Xi _2\Xi ^{-1}_1\). Hence, the asymptotic variance of \({\hat{\beta }}_{AGG}\) is approximately equal to \(\Xi ^{-1}_1\Xi _2\Xi ^{-1}_1\). \(\square \)
Proof of Proposition 2
First, note that \(N_D\) is approximately proportionate to \(K_D\) and that \(N_D\rightarrow \infty \) as \(K_D\rightarrow \infty \) for \(D\in \{C,S\}\). The consistency and asymptotic normality of the TS2SLS estimator can be derived from a variant of the standard argument (White 1984, Chap. 5). Under assumptions A1-A2 and A6-A9 and using \(K_S\rightarrow \infty \) as \(K_C\rightarrow \infty \), we have the following results as \(K_S\rightarrow \infty \):
Notice that \(\varvec{Q}_1\) does not show up in the final expression because \((N_C^{-1}\varvec{Z}^T_C\varvec{W}_C \varvec{Z}_C)(N_S^{-1}\varvec{Z}^T_S\varvec{W}_S \varvec{Z}_S)^{-1}\xrightarrow {p} \varvec{Q}_1\varvec{Q}_1^{-1}=I\) as \(K_C\rightarrow \infty \). \(\square \)
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Fujii, T. Spatial disaggregation of poverty and disability: application to Tanzania. Empir Econ 66, 705–734 (2024). https://doi.org/10.1007/s00181-023-02478-8
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00181-023-02478-8