What explains regional inequality in Uganda? The role of infrastructure, productive assets, and occupation Isis Gaddis, University of Goettingen Welfare Congress 2011, OECD, Paris Introduction • While poverty has fallen in Uganda since 1992, inequality has increased • Analysis in World Bank (2009) show that halting the trend in increasing inequality while sustaining growth is important if Uganda is to reach its poverty targets • But what explains high and rising inequality in Uganda? • One of the simplest ways to see what factors are driving inequality is to perform a between-within decomposition • Bivariate decomposition (theil-t or theil-l) • This shows that regional inequality is unusually high in Uganda, and it has been growing over time Introduction Regional and International Comparison of and Within-Group Inequality (theil-t) Rural-urban decomposition Regional Decomposition Number of Between Within groups Between Within 2005/06 2002 2000/01 2005/06 27 2 7 15 73 98 94 85 24 6 5 16 76 94 95 84 (8) (20) (4) (4) Other countries Benin 2003 Brazil 2004 Vietnam 1997/98 14 5 86 9 21 8 25 79 92 75 (12) (5) (61) East Africa Kenya Mozambique Tanzania Uganda Sources: East Africa: World Bank staff estimates. Other countries: World Bank (2003); Ferreira, Leite and Litchfield (2006); Minot, Baulch and Epprecht (2006). Introduction Inequality Decomposition (theil-t), 1992/93 - 2005/06 100% 80% Within urban within regions 60% Within rural within regions 40% Within urban between regions 20% Within rural between regions Between urban/rural 0% 1992 2002 2005 Introduction Poverty by Region, 2005/06 p0 p1 p2 National 0.311 0.087 0.035 Rural Central Eastern Northern Western 0.209 0.375 0.642 0.214 0.047 0.095 0.223 0.054 0.016 0.036 0.099 0.019 Urban Central Eastern Northern Western 0.055 0.169 0.397 0.093 0.011 0.044 0.115 0.020 0.005 0.015 0.045 0.006 Introduction • This paper seeks to understand which factors explain inequality between regions (Central, Northern, Western, Eastern) • Analyze differences between urban regions, and between rural regions (not urban-rural differential) • The welfare measure is consumption per adult • We focus on the following explaining factors: – Infrastructure (roads and electricity) – Productive assets (education and land) – Employment structure Methodology • Micro-simulation approach based on Bourguignon, Ferreira and Lustig (2005) – adapted to consumption data • Extension of the traditional Oaxaca-Blinder decomposition • Typically used to explain income-distribution dynamics • Simulates are series of counterfactual distributions to decompose the differences between actual distributions: – – – – Multivariate (unlike the bivariate Theil decompositions) Distinguishes between endowment and price effects (like OB) Can accommodate interdependencies between variables Simulates full distributions and can thus decompose any functional indicator (e.g. poverty and inequality indices) Methodology • Estimate a model of consumption (at the hh-level) by region (r) • XCONS,h,r includes: – productive assets: education of all hh members and (rural) size of land holdings – infrastructure: electricity access and (rural) distance to a trunk road – employment of the head and other hh members – demographic control variables (not used for simulation) • αc,r are county-specific intercepts Methodology • Price simulations: equalize returns to (specific) household endowments across regions (by importing the coefficient vector from the reference region) • Endowment simulations: use non-parametric and parametric approaches to equalize (specific) endowments across regions – Rank-preserving transformation for continuous or dichotomous variables (land holding size, years of education, road distance, electricity access) – Multinomial logit for categorical variables (occupation) – The endowment distribution simulated by importing the coefficients vector of the discrete choice models from the reference region • Reference: Central Uganda (keeps urban-rural differences) Methodology Results: price simulations (p0) Base region: Eastern Northern Western Eastern Northern Western rural urban Observed Base region Central region 0.372 Δ% Price simulations electricity Δ% rural roads Δ% education Δ% rural land Δ% infrastructure (electricity & rural roads) productive assets (education & rural land) occupation Δ% Δ% Δ% 0.214 0.173 -44% 0.641 0.21 -67% -2% -72% 0.371 0% 0.389 5% 0.275 -26% 0.383 3% 0.389 5% 0.294 -21% 0.404 9% 0.641 0% 0.648 1% 0.54 -16% 0.646 1% 0.647 1% 0.545 -15% 0.673 5% 0.214 0% 0.228 7% 0.186 -13% 0.219 2% 0.228 7% 0.194 -9% 0.225 5% 0.405 0.048 -88% 0.095 -49% (see infrastructure below) (not applicable) (see productive assets below) (not applicable) 0.187 0.405 0.097 8% 0% 2% 0.189 0.406 0.102 9% 0% 7% 0.262 0.472 0.086 51% 17% -9% Results: returns to education Rural Uganda 1.00 0.80 0.60 Central 0.40 Eastern 0.20 Northern 0.00 Western no education some primary completed primary some secondary completed secondary Urban Uganda 1.00 0.80 0.60 Central 0.40 Eastern 0.20 Northern 0.00 Western no education some primary completed primary some secondary completed secondary Results: endowment simulations (p0) Base region: Eastern Northern Western rural Observed Base region Central region 0.372 Δ% Endowment simulations electricity Δ% rural roads Δ% education Δ% rural land Δ% infrastructure (electricity & rural roads) productive assets (education & rural land) occupation Δ% Δ% Δ% Eastern Northern Western urban 0.214 0.173 -44% 0.641 0.21 -67% -2% -72% 0.353 -5% 0.372 0% 0.345 -7% 0.387 4% 0.353 -5% 0.362 -3% 0.384 3% 0.623 -3% 0.638 0% 0.592 -8% 0.646 1% 0.62 -3% 0.598 -7% 0.64 0% 0.204 -5% 0.213 0% 0.198 -7% 0.224 5% 0.203 -5% 0.206 -4% 0.216 1% 0.405 0.048 -88% 0.095 -49% (see infrastructure below) (not applicable) (see productive assets below) (not applicable) 0.115 0.247 0.066 -34% -39% -31% 0.136 0.315 0.069 -21% -22% -27% 0.149 0.396 0.103 -14% -2% 8% Results: combined simulations Combined simulations Northern rural Combined simulations Western rural 40 percent 20 100 20 40 60 80 100 0 20 percentile 40 60 80 100 0 20 percentile difference to Central rural infr., assets, occup. percent all Combined simulations Western urban 0 20 40 60 80 100 percentile difference to Central urban infr., assets, occup. -20 -50 -20 0 0 0 50 percent 100 40 Combined simulations Northern urban 100 60 40 20 80 difference to Central rural infr., assets, occup. all 150 80 Combined simulations Eastern urban 60 percentile difference to Central rural infr., assets, occup. all 40 20 0 0 0 0 50 20 40 percent 150 60 200 60 250 Combined simulations Eastern rural 0 20 40 60 80 100 percentile all difference to Central urban infr., assets, occup. 0 20 40 60 80 100 percentile all difference to Central urban infr., assets, occup. all Some caveats • • • • • • • No a causal model, no clear identification of effects Potential endogeneity problems (esp. for electricity access) Accounting exercise No general equilibrium effects No standard errors/confidence intervals County-effects (unobservables) play a huge role Not all simulations have a clear policy implication (e.g. equalizing land holding sizes) • Simulations do not necessarily reduce total regional inequality (because the urban-rural gap may even get larger) Conclusion • The simulations show that the following factors come out as determinants of regional inequality in Uganda – – – – Educational attainment (urban and rural) Access to electricity (urban and rural) Returns to education (rural) Returns to non-agricultural activities (urban and rural) • This suggests policies to invest in education and electricity and increase profitability of non-agricultural employment in lagging areas • However, inequality considerations need to be balanced with overall growth considerations Thank you! References • Bourguignon, François, Francisco H. G. Ferreira and Phillippe G. Leite (2008). “Beyond Oaxaca-Blinder: Accounting for Differences in Household Income Distributions.” Journal of Economic Inequality Vol. 6: 117-148. • Bourguignon, François, Francisco H. G. Ferreira and Nora Lustig (eds.) (2005). The Microeconomics of Income Distribution Dynamics in East Asia and Latin America. Washington DC: World Bank and Oxford University Press. • Ferreira, Francisco H. G. (2010). “Distributions in Motion: Economic Growth, Inequality and Poverty Dynamics.” World Bank Policy Research Working Paper No. 5424, Washington DC: World Bank. • Leite, Phillippe G., Alan Sanchez and Caterina R. Laderchi (2009). “The Evolution of Urban Inequality in Ethiopia.” Draft version March 2009, World Bank HDNSP and AFTP2. Results: simulated education Rural Urban Central Eastern Northern Western Central Eastern Northern Western Actual educational attainment (%) no formal educ. 12.2 17.1 24 22.7 3.6 9.8 14.9 9.4 some primary 46.2 49.5 53.1 47.3 24.2 33.5 41.1 29.2 compl. primary 16.2 14.8 12.5 15.1 18.7 18.5 16.9 19.9 some secondary 15.8 12.2 6.6 9.3 24.3 20.2 14.5 16.5 compl. secondary 9.5 6.4 3.8 5.6 29.2 18 12.8 25 Total 100 100 100 100 100 100 100 100 Average years 5.7 5 4.2 4.5 8.0 6.7 5.7 7.1 Simulated educational attainment (%, rank-preserving transformation) no formal educ. 12.2 12.1 12.1 12.1 3.6 3.6 3.7 3.6 some primary 46.2 46.3 46.4 46.4 24.2 24.2 24.3 24.2 compl. primary 16.2 16.2 16.2 16.2 18.7 18.8 18.7 18.9 some secondary 15.8 15.8 15.8 15.8 24.3 24.3 24.3 24.2 compl. secondary 9.5 9.5 9.5 9.5 29.2 29.1 29.1 29.1 Total 100 100 100 100 100 100 100 100 Average years 5.7 5.7 5.7 5.7 8.0 8.0 8.0 8.0 Results: simulated electricity Rural Central Eastern Northern Western Actual electricity access percent 10.8 2.7 0.2 2.0 Urban Central Eastern Northern Western 55.1 28.2 9.2 26.8 Simulated electricity access (rank-preserving transformation) percent 10.8 11.2 11.4 11.3 55.1 58.0 56.7 56.7 (I) Actual 2005/06 Price simulations: infrastructure productive assets occupation Endowment simulations: infrastructure productive assets occupation Combined simulations: infrastructure, productive assets and occupation all (incl. county FE and demographic prices) (II) between urban theil-t and rural 0.321 15% (III) (IV) (V) Share of inequality … within urban within regions 25% within urban between regions 4% within rural within regions 48% (VI) (VII) total inequality within between rural regions between =(II)+(IV)+(VI regions ) 8% 27% 0.322 0.301 0.342 15% 14% 16% 24% 26% 25% 4% 4% 5% 49% 51% 46% 8% 6% 8% 27% 24% 28% 0.320 0.320 0.322 18% 16% 16% 26% 25% 25% 2% 3% 4% 47% 50% 47% 7% 6% 8% 27% 24% 28% 0.302 15% 25% 3% 53% 5% 22% 0.272 14% 27% 0% 59% 0% 14%