The Caste/Ethnic Bases of Poverty Dynamics: A Longitudinal Analysis of Chronic and Structural Poverty in Nepal

Abstract

This paper examines poverty dynamics and their socioeconomic determinants between 1996 and 2011 in Nepal. With chronic and structural poverty headcount ratios of around 17 per cent, poverty is mostly transient and stochastic affecting up to four-fifths of the population. Descriptively, indigenous Janajatis and lower caste Hindus exhibit the highest rates of chronic and structural poverty. Panel data models suggest significant roles of human capital and household assets in determining poverty, however, with the evidence of caste/ethnic penalty limited mostly to Janajatis. Findings provide important insights into the structure and determinants of poverty dynamics, helping to rethink policies to address them.

Acknowledgements

This analysis is based on data from the Nepal Living Standard Surveys and the author expresses commitment to making available the data and codes of analysis upon request. The author appreciates the comments received from anonymous referees which have improved the paper greatly. The remaining errors and omissions are those of the author.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. This index focuses on such empowerment measures as knowledge and awareness, participation, confidence and comfort, social networks, and efforts to influence as well as other social inclusion indicators including self-perception of status, access and public intimidation, political influence, and services and opportunities (World Bank & DFID, 2005).

2. These years are reported as shorthand, with actual data collection spanning over Nepali calendar years approximating the middle of 1995/1996, 2003/2004, and 2010/2011.

3. This closeness is particularly salient in the case of the balanced panel when weighting is used. The average nominal per capita consumption estimates reported by the CBS (2011b) were NPR 6802, 15,848, and 34,829 for 1996, 2004, and 2011 respectively.

4. These estimates do not make adult equivalent adjustments by household size. While these adjustments can address the economies of scale for larger households, the small costs of housing relative to other consumption expenses suggest that this may not be very useful. It is important to note, however, that poverty estimates attenuate when adult equivalence scales are used (see Table A.6 in the Online Appendix).

5. Since less than 13 per cent of households from the first panel survey are included in the balanced panel, the differences are greater for 1996 than for 2004 and 2011.

6. These six major caste/ethnic categories are aggregated from over 100 identified groups in Nepal. The upper, middle, and lower caste Hindus all are Hindu groups facing very different degrees of privileges and disadvantages throughout Nepal’s history. Janajatis are not exclusively identified as Hindus although they were forcibly assimilated into the Hindu caste system. While a part of Janajatis, the Newars have enjoyed economic privilege afforded by their dominance in business professions. Muslims and others, albeit not fully coherent in socioeconomic privilege, are grouped to create a category that includes all others.

7. While the assets or endowments used to assess economic hardships can vary from human capital and infrastructure to social networks and relationships (Carter & May, 2001), household assets provide a reasonable proxy measure of them (Carter & Barrett, 2006). Assuming that the typical return to household assets would be about one third the investment – not in a literal sense because assets are not the only investment or endowment, with human capital making perhaps an even more important investment – households with assets valued at over three times the applicable consumption poverty line are expected to escape poverty more easily. These household assets include farmland, housing, and other fixed assets that can generate return on a regular basis. The cut-off of assets at three times the consumption poverty line is arbitrary and using observed assets may not fully capture the expected return which can vary depending on not just household endowments but caste/ethnic discrimination or penalty as well. But this still gives a reasonable basis using the framework of structural poverty (Carter & Barrett, 2006; Carter & May, 2001).

8. Theoretically, this indirect form of discrimination is quite significant since lower castes and other marginalised groups may have been prevented historically from developing comparable levels of human capital and assets. But this would require a deeper analysis of the determinants of intergroup differences in human capital, assets, and even opportunities involving enormous endogeneity. Fully accounting for this endogeneity concern would be beyond the scope of this paper.

9. While household fixed effects would be more consistent in estimating the impact, they would not be feasible given the time-invariant measurement of many household characteristics. But the model will be estimated with heteroskedasticity-robust standard errors to achieve greater consistency.

10. 1Estimates are reported with cluster-robust standard errors. Empirically, an ideal strategy would be to estimate the model in a purely panel data framework just like in the case of ordinary least squares regressions. But since this is not available on Stata, the software package used in this analysis, the second best strategy that is used here is to apply clustering so that the data are grouped around 434 clusters, with the three years serving as iterations within clusters.

11. For chronic poverty, this categorisation disregards the detailed, eight-way poverty transitions matrix presented in Figure 1 and uses outcomes from Table 2 applied to the balanced panel. For structural poverty, it uses the stochastic and structural categories from Table 3.

12. This approach decomposes intergroup differences in outcomes like consumption into one that is explained by some observed (productive) characteristics or endowments and another that is attributable to belonging to the groups themselves by using counterfactuals (Blinder, 1973; Oaxaca, 1973). Counterfactuals are conditions of one group that are applied to another in order to estimate the unexplained variation. While this procedure is generally applied to linear regressions, these estimates were obtained by using techniques extended for nonlinear models presented in Table 5 (Sinning, Hahn, & Bauer, 2008).