1. ETHIOPIAN DEVELOPMENT RESEARCH INSTITUTE Decomposing the Impact of Poverty Covariates in Ethiopia Ibrahim Worku, Bethelehem Koru and FanayeTadesse IFPRI ESSP-II 1 9th International Conference on the Ethiopian Economy, Ethiopian Economic Association (EEA) July 21, 2011 Addis Ababa
8. Objective Previous study made by MoFED using HICE and WMS showed the dynamics of growth and poverty in Ethiopia (1995/96-2004/05) using three survey conducted by CSA. The study indicated that there is decline in national poverty across multiple poverty indicators. According to literature on poverty, if there appears to be a decline in poverty the factor which accounts most should be the focal point of policy intervention Thus, this study attempts to identify which of the covariates of poverty take significant share over these survey periods
9. Data and Estimation CSA HICES and WMS data are used for the analysis Nationally representative; Periods considered: 1995/96 and 2004/05 HICES: Consumption Expenditures and WMS: on various dimensions: from demographic, infrastructure, to public utilities together capture various dimensions Decomposition methods introduced by the seminal papers of Oaxaca (1973) and Blinder (1973) will be employed to analyze the effect of various covariates of poverty Due to the limitations of the standard Blinder-Oaxaca decomposition, we have attempted to support/modify the findings using Fairlie (2003) decomposition method
10. Methodology Descriptive tables Classification : National, Rural and Urban Households * Poor/non-poor-National classification (based on quintiles of real per capita expenditure groups) In light of the sated objective: Econometrically support the findings of the previous study, MoFED (2008), using regression analysis
11. Econometric Analysis: Model Specification Outcome variable : two dimensions a) Real per capita expenditure groups: 1996-2004 b) The probability of being non-poor groups: 1996-2004 The idea is to identify the role of the covariates
13. Non-linear specification (cont.) Fairlie (2003) decomposition method: It decomposes the mean outcome variable , in stead of the mean of each of the dependent variables, the first term in brackets - the part that is due to group differences in distributions of X (covariates), and the second term - the part due to differences in the group processes determining levels of Y to the changes in the distribution of the covariates It also allows to work on discrete dependent variables: Accordingly, we used dependent variable with poor/non-poor classification 40% bottom per capita expenditure are considered to be poor
14. Result and discussion Real per capita expenditure: wellbeing indicator has shown to be increasing, in particular for rural areas. Average age slightly declined Single head in urban areas is increasing Demographic features somehow remained static
17. Substantial increment in the percentage of households in secondary school is observed in the rural area than the Urban. For the Urban area higher level of education is found to improve significantly from 1996 to 2004.Percentage change of household head under different education categories over 1996- 2004
18. Distributional patterns of real per capita expenditure When we look at the distributional pattern of per capita expenditure in rural areas, there appears to be significant improvement
19. Cont. Urban per capita expenditure has shown marginal transition
20. Descriptive on per capita expenditure National per capita expenditure over 1996- 2004 has shown an improvement
23. Linear decomposition results Endowment effect: Asset ownership, education and income source diversification contribute positively for welfare improvement The returns to variables (the unexplained): the returns of being unemployed depresses consumption. The returns of being located in rural areas accounts a lot for welfare improvement; i.e. urban poverty, as indicated in previous studies, is persistent. The returns of having productive members plays a role for welfare improvement too. But the linear decomposition has the following pitfall : a) over states the impact of each of the covariates b) Total impact exceeds 100%, which is difficult to explain Thus, resort to non-linear decomposition suggested by Fairlie (2003): which decomposes only the explained part.
24. Results for the non-linear Accordingly, the difference in outcome variable significantly declined Result of the overall decomposition of the non-linear model
25. Estimation of Fairlie decomposition method Demographics: difficult to explain Asset ownership contributed a lot to get out of poverty Education and endowment of productive individuals contribute for the decline in poverty
26. Cont. Slight Regional convergence is exhibited for Amhara region with Addis There is divergence with SNNPR in comparison with Addis In general, there is no pronounced difference with the reverence group (Addis)
27. Conclusion In general, over the survey periods (i.e. 1995/6 and 2004/5), Poverty has shown a declining trend Endowment: ownership of assets , education and household size contributes a lot for the decline in poverty. Returns of being participant in the labor force helps for wellbeing improvement Non-linear: the results are still consistent (only for endowment effect) education, asset ownership and endowment of productive labor force play a lot for probability of exit There has been a slight convergence(Amhara, Afar and Gambella regions) but pronounced divergence (SNNP region) in comparison with Addis Regarding education level, statistical difference is observed with individual whose highest level of education is secondary school over the two year period. This holds true for both Urban and rural area.
28. Caveats Non-linear decomposition using gdecomp; it decomposes both the explained and unexplained components Rural regression model need to be modified Extend the analysis with upcoming data