ETHIOPIA POVERTY ASSESSMENT Harnessing Continued Growth for Accelerated Poverty Reduction © 2020 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington, DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org SOME RIGHTS RESERVED This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. RIGHTS AND PERMISSIONS This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) http:// creativecommons. org/licenses/by/3.0/igo. Under the Creative Commons Attribution license, you are free to copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following conditions: ATTRIBUTION—Please cite the work as follows: World Bank (2020). Ethiopia Poverty Assessment: Harnessing Continued Growth for Accelerated Poverty Reduction. Washington DC. © World Bank. TRANSLATIONS—If you create a translation of this work, please add the following disclaimer along with the attribution: This translation was not created by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this translation. ADAPTATIONS—If you create an adaptation of this work, please add the following disclaimer along with the attribution: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by the World Bank. THIRD-PARTY CONTENT—The World Bank does not necessarily own each component of the content contained within the work. The World Bank therefore does not warrant that the use of any third-party owned individual component or part contained in the work will not infringe on the rights of those third parties. The risk of claims resulting from such infringement rests solely with you. If you wish to re-use a component of the work, it is your responsibility to determine whether permission is needed for that re-use and to obtain permission from the copyright owner. Examples of components can include, but are not limited to, tables, figures, or images. All queries on rights and licenses should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; e-mail: pubrights@worldbank.org. Photo Contributors: Arne Hoel, Dominic Chavez, Gelila Woodeneh, Sylwia Pecio, Shohei Nakamura, Daisy Demirag ETHIOPIA POVERTY ASSESSMENT Harnessing Continued Growth for Accelerated Poverty Reduction HARNESSING CONTINUED GROWTH FOR ACCELERATED POVERTY REDUCTION 1 Contents Abbreviations and Acronyms________________________________________________________________________ 6 Acknowledgements________________________________________________________________________________ 7 Executive Summary________________________________________________________________________________ 9 Introduction______________________________________________________________________________________ 29 Chapter I 2011-2016: Continued Growth and Poverty Reduction_________________________________ 35 Introduction___________________________________________________________________________________ 36 2. Significant household consumption growth and poverty reduction________________________________ 36 2.1 Solid reduction in the poverty headcount__________________________________________________ 36 2.2 Consumption growth concentrated in the upper parts of the distribution______________________ 45 2.3 A small increase in inequality____________________________________________________________ 47 2.4 No clear impact of the El-Niño drought___________________________________________________ 50 2.5 Testing the sensitivity of the poverty estimates_____________________________________________ 53 3. Gains in non-monetary dimensions of welfare as well____________________________________________ 54 4. How Did the Extreme Poor Fare?_____________________________________________________________ 57 5. Conclusions________________________________________________________________________________ 61 Chapter II Who Are the Poor?__________________________________________________________________________ 63 Introduction___________________________________________________________________________________ 64 2. The Poverty Profile__________________________________________________________________________ 64 2.1 Poverty increasingly concentrated in rural areas____________________________________________ 64 2.2 The poor live in large households with high dependency rates_______________________________ 65 2.3 Younger and female-headed households less likely to be poor_______________________________ 67 2.4 The poor are largely uneducated_________________________________________________________ 70 2.5 The poor depend on agriculture__________________________________________________________ 72 2.6 The poor are relatively remote and have lower access to services____________________________ 74 3. Correlates of poverty________________________________________________________________________ 78 3.1 Significant returns to education and a penalty on being divorced_____________________________ 78 3.2 Large effects of occupations____________________________________________________________ 80 3.3 Remoteness is correlated with lower consumption levels____________________________________ 81 3.4 Location matters – even after controlling for other factors___________________________________ 81 Conclusions___________________________________________________________________________________ 83 HARNESSING CONTINUED GROWTH FOR ACCELERATED POVERTY REDUCTION 3 Chapter III Drivers of Poverty Reduction in Ethiopia_________________________________________________ 85 Introduction___________________________________________________________________________________ 86 2. Decomposing poverty reduction between 2000 and 2016_______________________________________ 88 3. Drivers of poverty reduction: Evidence from a zonal panel dataset_________________________________ 95 3.1 How has growth contributed to poverty reduction?_________________________________________ 97 3.2 Agricultural growth and poverty reduction_________________________________________________ 99 3.3 Drivers of agricultural growth___________________________________________________________ 101 Conclusion___________________________________________________________________________________ 104 Chapter IV Household poverty dynamics and economic mobility________________________________ 107 Introduction__________________________________________________________________________________ 108 2. Describing household consumption dynamics_________________________________________________ 108 3. Poverty status and transitions, 2012-2016____________________________________________________ 110 4. Profiling chronic and transitory poverty________________________________________________________ 116 5. The dynamics of poverty transitions: Exit and entry_____________________________________________ 119 5.1 The characteristics associated with poverty transitions_____________________________________ 119 5.2 Regressions of poverty exit and poverty entry dynamics___________________________________ 120 Conclusion___________________________________________________________________________________ 125 Chapter V Urban Poverty in Ethiopia _______________________________________________________________ 127 Introduction__________________________________________________________________________________ 128 2. Spatial diagnostic of urban poverty, living standards, and labor markets___________________________ 130 2.1 A brief profile of the urban poor_________________________________________________________ 130 2.2 Dissecting urban poverty reduction______________________________________________________ 136 2.3 Access to services and amenities by city size_____________________________________________ 140 2.4 Urban labor market developments ______________________________________________________ 143 2.5 Integration of rural migrants into urban labor markets_________________________________________ 150 Conclusions__________________________________________________________________________________ 155 4 ETHIOPIA POVERTY ASSESSMENT Chapter VI Poverty and Social Protection __________________________________________________________ 157 Introduction__________________________________________________________________________________ 158 2. Ethiopia’s Social Protection System__________________________________________________________ 159 3. The Rural Productive Safety Net Project______________________________________________________ 164 3.1 The national picture___________________________________________________________________ 164 3.2 Highlands vs Lowlands________________________________________________________________ 175 3.3 A Regional Perspective________________________________________________________________ 178 4. Humanitarian Food Aid_____________________________________________________________________ 180 4.1 The national picture____________________________________________________________________ 181 4.2 Woreda vs household targeting__________________________________________________________ 184 4.3 Overlap and complementarity between HFA and PSNP____________________________________ 185 Conclusions__________________________________________________________________________________ 189 Chapter VII Inequality of Opportunity in Ethiopia ___________________________________________________ 191 Introduction__________________________________________________________________________________ 192 2. The Human Opportunity Index in Ethiopia_____________________________________________________ 193 2.1 Background and methodology__________________________________________________________ 193 2.2 Descriptive statistics and results_________________________________________________________ 195 3. Intergenerational educational mobility_________________________________________________________ 200 Conclusion___________________________________________________________________________________ 207 References ________________________________________________________________________________ 208 Annex 1: Additional material for Chapter 1__________________________________________________________ 213 Annex 2: Additional material for Chapter 2__________________________________________________________ 217 Annex 3: Additional material for Chapter 3__________________________________________________________ 221 Annex 4: Additional material for Chapter 4 _________________________________________________________ 223 Annex 5: Additional material for Chapter 5 _________________________________________________________ 228 Annex 6: Additional material for Chapter 7 _________________________________________________________ 230 HARNESSING CONTINUED GROWTH FOR ACCELERATED POVERTY REDUCTION 5 Abbreviations and Acronyms AE Adult equivalent BoLSA Bureau of Labor and Social Affairs CGE Computable General Equilibrium DHS Demographic and Health Survey ESS Ethiopia Socioeconomic Survey ETB Ethiopian Birr FDRE Federal Democratic Republic of Ethiopia GDP Gross Domestic Product GIC Growth Incidence Curve GoE Government of Ethiopia GTP Growth and Transformation Plan HCES Household Income and Consumption Expenditure Survey HFA Humanitarian Food Aid HH Household HOI Human Opportunity Index LFS Labor Force Survey NDVI Normalized Difference Vegetation Index PPP Purchasing Power Parity PSNP Productive Safety Net Program RAI Rural Accessibility Index RIF Recentered Influence Functions SNNPR Southern Nations, Nationalities and People’s Region TD Targeting differential TFR Total Fertility Rate UEUS Urban Employment and Unemployment Survey URRAP Universal Rural Roads Access Program USD United States Dollars WMS Welfare Monitoring Survey 6 ETHIOPIA POVERTY ASSESSMENT Acknowledgements This report was prepared by a core team consisting of Tom The report benefitted from inputs from officials of the Plan- Bundervoet (Task Team Leader, Poverty and Equity Practice), ning and Development Commission, the Ministry of Finance, Arden Finn (Co-Task Team Leader, Poverty and Equity GP), and the Central Statistics Agency. The team is indebted to Shohei Nakamura (Economist, Poverty and Equity GP), and the Central Statistics Agency (CSA) for making available Berhe Mekonnen Beyene (Economist, Poverty and Equity the datasets on which this report is based and to Professor GP), under the overall guidance of Pierella Paci (Practice Tassew Woldehanna for sharing the final consumption ag- Manager, Poverty and Equity GP), Nataliya Mylenko (Pro- gregate. Comments and inputs by participants at the 16th gram Leader), and Carolyn Turk (Country Director, AFCE2). Ethiopian Economics Association International Conference Nora Dihel, Zerihun Kelbore and Samuel Mulugeta contrib- (July 2018) are greatly appreciated. uted to the macro-economic analysis presented in the intro- The team thanks the report’s peer reviewers for thoughtful duction. Thomas Sohnesen (Consultant, Poverty and Equity inputs and comments. The peer reviewers were Kathleen GP), Lucian Bucur Pop (Senior Social Protection Specialist, Beegle (Lead Economist, GTGDR), Ruth Hill (Lead Economist, Social Protection and Jobs GP), Daisy Demirag (Consultant, Poverty and Equity GP), Johan Mistiaen (Program Leader, Social Protection and Jobs GP), Judith Sandford (Consul- AFCE2), Margaret Grosh (Senior Advisor, GSJD1), and tant, Social Protection and Jobs GP), and Abu Yadetta (Se- Alemayehu Seyoum Tafesse (Senior Research Fellow, IFPRI). nior Social Protection Specialist, Social Protection and Jobs The team also thanks Emily Schmidt and Mekamu Kedir at GP) contributed to Chapter 6 on social protection. Chapter IFPRI for making available spatial data used in this report. 7 was prepared in collaboration with Maude Cooper, Saori Iwamoto and Rewa Misra (Georgetown University) with guid- ance from Jacobus Cilliers (Georgetown University). Manex Bule Yonis (Consultant, Poverty and Equity GP) contributed to data analysis for several chapters in this report. HARNESSING CONTINUED GROWTH FOR ACCELERATED POVERTY REDUCTION 7 8 ETHIOPIA POVERTY ASSESSMENT Executive Summary This poverty assessment focuses on the evolution dimensions of living standards and examines the drivers of of poverty and other social indicators in Ethiopia be- observed trends, with a special focus on government pro- tween 2010/11 and 2015/2016 (henceforth referred to as grams. The aim of the poverty assessment is to provide pol- 2011 and 2016). Using data from a variety of sources, mainly icy makers and development partners with information and the twinned household living standards surveys (HCES and analysis that can be used to improve the effectiveness of WMS), the Ethiopia Socioeconomic Survey (ESS) and the their poverty reduction and social programs. Demographic and Health Surveys (DHS), the poverty as- sessment documents trends in monetary and non-monetary CONTINUED CONSUMPTION GROWTH AND POVERTY REDUCTION, ESPECIALLY IN URBAN AREAS Between 2011 and 2016, Ethiopia’s economy continued from 26 percent in 2011 to 15 percent in 2016 in urban Ethi- to grow rapidly, with an annual GDP growth rate in ex- opia, an 11-percentage-point decrease (Figure O 1). In rural cess of 9 percent. Fast economic growth translated into areas, poverty decreased by four percentage points, from 30 strong household consumption growth in urban areas but percent in 2011 to 26 percent in 2016. This reduction was not in rural areas. Consumption of urban households grew achieved in spite of the fact that the 2015/16 survey was at six percent per year, while the corresponding figure for conducted during the severe El-Nino drought. The national rural households was less than one percent. As a result, the poverty rate decreased from 30 percent in 2011 to 24 per- poverty rate, based on the national poverty line, decreased cent in 2016. Figure O 1 POVERTY  DECREASED IN BOTH RURAL AND URBAN AREAS Poverty headcount rate based on the national poverty line, 2011 and 2016 2011 2016 35 29.6 30.4 30 Percentage poor 25.7 25.6 25 23.5 20 14.8 15 10 5 0 National Urban Rural Source: HCES, 2011; 2016. World Bank staff calculations. EXECUTIVE SUMMARY 9 Human development indicators improved alongside year. This pattern was driven by rural areas, where the bot- the increase in consumption. Delivery in a health facility tom 20 percent experienced zero or negative consumption increased sharply from a low base, the share of fully immu- growth (Figure O 3). In contrast, growth across the urban nized children increased by 14 percentage points, and stunt- consumption distribution was always above 3 percent, even ing rates decreased from 44 percent in 2011 to 38 percent in for the poorest, and became increasingly strong towards the 2016 (Panel A of Figure O 2).1 Infant and child mortality rates upper end (Figure O 4). Given the largely rural nature of the decreased accordingly. Net enrolment in primary school in- Ethiopian population, the national pattern of growth close- creased, more children are completing primary school, and ly resembles the rural pattern presented in Figure O 3. This gross enrolment in secondary school was higher in 2016 than pattern is a continuation of the one observed between 2005 in 2011 (Panel B of Figure O 2). Despite these improvements, and 2011, when consumption of the bottom 15 percent of human development indicators remained low. In 2016, only 26 the population contracted. percent of births took place in a health facility (in the five years Given that the upper parts of the distribution experi- preceding the survey) and less than 40 percent of children had enced higher growth, inequality increased slightly. The received all basic vaccinations. Only one in three people be- Gini coefficient rose from 0.30 in 2011 to 0.33 in 2016 but tween 15 and 24-years-old had completed primary school. remains low in regional comparison. The increase in inequali- Consumption growth was higher for the upper parts ty is mainly due to the increasing disparity between rural and of the welfare distribution, while the poorest segment urban areas: Urban consumption, which was already higher of the population did not experience real consumption than rural consumption to begin with, grew rapidly, increas- growth from 2005 onwards. Between 2011 and 2016 ing the disparity in average consumption levels with rural ar- consumption did not grow for the bottom 15 percent of the eas. The share of total inequality that can be explained by population, in contrast to the top of the distribution where differences in welfare between urban and rural areas doubled growth rates reached a maximum of just under 6 percent per to 29 percent in 2016 (Figure O 5). Figure O 2 CHILD  HEALTH AND EDUCATION INDICATORS IMPROVED BETWEEN 2011 AND 2016 Selected health and education variables for children in 2011 and 2016 2011 2016 2011 2016 2011 2016 2011 2016 100 100 100 100 90 90 90 90 80 80 80 80 71.8 71.8 70 70 70 63.9 70 63.9 Percent Percent 60 60 Percent Percent 60 60 50 50 44.4 44.4 50 50 38.5 38.5 38.4 38.4 40 40 40 40 32.7 32.7 30.7 30.7 26.2 26.2 24.3 24.3 27.6 27.6 30 30 30 30 21.5 21.5 20 20 9.9 9.9 20 20 10 10 10 10 0 0 0 0 Health facility Health facility Fully Fully StuntedStunted Net primary Net primary school school PrimaryPrimary Gross Gross immunized deliverydelivery immunized childrenchildren school school completion completion (15- secondary (15- secondary childrenchildren enrolment 24) enrolment 24) school school enrolment enrolment Source: DHS 2011; 2016. 1 Children are fully immunized if they received all eight basic vaccinations: BCG, three doses of Polio, three doses of DPT, and one dose of MCV. This is calculated for the sample of children aged between 12 and 23 months. 10 ETHIOPIA POVERTY ASSESSMENT Figure O 3 WELFARE  OF THE POOREST Figure O 4 …WHILE  GROWTH WAS 20 PERCENT IN RURAL STRONG ACROSS THE AREAS DID NOT INCREASE URBAN WELFARE BETWEEN 2011 AND 2016… DISTRIBUTION Average annual growth rates of rural Average annual growth rates of rural consumption by percentile between consumption by percentile between 2011 and 2016 2011 and 2016 Rural Urban 8 8 −2 −1 0 1 2 3 4 5 6 7 −2 −1 0 1 2 3 4 5 6 7 Annual mean growth rate (%) Annual mean growth rate (%) 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Percentiles Percentiles 95% confidence bounds 95% confidence bounds Source: HCES 2011; 2016. World Bank staff calculations. Source: HCES 2011; 2016. World Bank staff calculations. EXECUTIVE SUMMARY 11 Figure O 5 INEQUALITY  INCREASED DUE TO THE INCREASING GAP BETWEEN URBAN AND RURAL AREAS Decomposition of the Gini coefficient into a between rural-urban component and a within-component 20.1 15.0 Relative contribution to Overlap Gini (%) 55.9 65.0 Within Between 29.1 14.9 2011 2016 Source: HCES 2011; 2016. World Bank staff calculations. Figure O 6 CHARACTERISTICS  OF THE POOR The poor are largely uneducated A.   ive in households with high dependency rates B. L Poverty rate by education of household head Poverty rate by household dependency rate 35 35 31.3 31.3 30 28.4 28.4 30 30 30 25.7 25.7 25 25 22.1 22.1 25 35 25 22.7 22.7 20.1 20.1 31.3 20 30 28.4 20 20 30 35 20 31.3 15.6 15.6 25.7 28.4 13.4 13.4 12.8 12.8 30 22.7 15 25 30 15 22.1 15 25 15 25.7 20.1 10 25 10 22.1 10 20 25 10 22.7 20 4.9 4.9 15.6 20.1 5 20 5 12.8 13.4 3.3 3.3 5 20 5 15 15 15.6 12.8 13.4 0 10 15 0 0 10 15 0 4.9 0.5-1 0.5-1 0-0.5 0-0.5 1.5-2 1.5-2 2- 1-1.5 1-1.5 2- 3.3 nl ed on im n y ary y da y ry nd y y y ry ry 10 10 ar imar on ar co ar ar ar pr atio 5 5 da da i 4.9 at prim ec im se nd d Dependency Dependency rate rate con n on 3.3 uc uc te pr pr e o o 0 5 0 5 c y -sec y ec ed se e e ylete ertylete s s y ar -s t 0-0.5 0.5-1 1-1.5 1.5-2 2- e e o o e e otse t ar d st pl 0 0 plary N N p rp im p ndlet t r o e Pe ndco Po tm om m pr om a elte I ry pla a i pl 0.5-1 0-0.5 Dependency 1-1.5 1.5-2 2- im nd p rate o atp co ca n am ec m rm om rn y ry ry ioC C nc oo ao In Iu co prs o a da dc ed ete nc C eonC im im Dependency rate se cose e prItn e on e ct e le o se t- u pl le c ec N et p e os ed t om pl m ps -s et P plp co mCo te st om em Are mainly engaged in farming C.   nd B. A 30more likely to be headed by a man 30 are N In Po cm o pl pl C Farming Farming Self-employed Self-employed nonfarm nonfarm Wage nonfarm Wage nonfarm co o In 26.6 26.6 m om C Occupation of household head by consumption quintile Poverty rate by gender of head of household In co C In 100 100 4.7 4.7 6.3 25 25 6.3 5.7 5.7 10 10 11.5 11.5 Farming Self-employed nonfarm Wage nonfarm 30 20.1 20.1 12.7 12.711.4 11.4 28.7 28.7 26.6 15.3 15.3 20 20 rate Poverty rate 80 80 Farming Self-employed nonfarm Wage nonfarm 30 100 25 15.9 15.9 26.6 4.7 5.7 6.3 10 15 13.9 13.9 Poverty Poverty 100 11.5 15 25 20.1 60 60 4.7 12.7 6.3 11.4 5.7 10 28.7 25.7 25.7 15.3 20 rate 80 11.5 11.4 20.1 12.7 28.7 15.9 15.3 10 10 20 Poverty rate 80 83.8 83.8 81 40 40 81 82.9 82.9 15 13.9 15.9 74.7 74.7 60 25.7 5 5 15 13.9 60 45.5 45.5 25.7 20 20 10 40 83.8 81 82.9 74.7 0 0 10 40 83.8 81 82.9 0 0 74.7 5 Female Male Male Female Female Male Male Female 20 45.5 Q1 Q1 Q2 Q2 Q3 Q3 Q4 Q4 Q5 Q5 5 Urban Urban Rural Rural 20 Quintiles 45.5 0 of consumption Quintiles expenditures of consumption expenditures 0 0Male Female Male Female 0Q1 Q2 Q3 Q4 Q5 Male Urban Female Male Rural Female Q1 of consumption Quintiles Q2 expenditures Q3 Q4 Q5 Urban Rural Quintiles of consumption expenditures Source: HCES 2011; 2016. World Bank staff calculations. 12 ETHIOPIA POVERTY ASSESSMENT Rapid urban poverty reduction means that the poor are largely reflect the regional population shares, with Oromia increasingly concentrated in rural areas. Though the ru- and Amhara accounting for the bulk of the poor given their ral share of the population decreased by two percentage large populations (Table O 1). Disparities in poverty are high- points, the rural share of poverty increased to 88 percent in er across agro-ecological zones. Using the “five Ethiopias” 2016. The poor in Ethiopia have the following characteris- classification, an agro-ecological classification based on alti- tics: They tend to live in rural areas, in large households with tude, rainfall, and predominant livelihoods, poverty is highest high dependency rates, headed by an older and little-edu- in the drought-prone lowlands (the lowland areas of Oromia cated household head. They mainly engage in agriculture and SNNPR and a small part of Afar – poverty rate of 32 per- and casual labor for their livelihood, and are relatively isolat- cent) and lowest in the drought-prone highlands (the east- ed from key infrastructure worse access to basic services. ern parts of Tigray and Amhara and north-eastern parts of Poor households are less likely to be headed by a woman, Oromia – poverty rate of 21 percent). The depth and severity and households in pastoral areas are less likely to be poor of poverty is also highest in the drought-prone lowlands. It (Figure O 6) . 2 is important to note that despite popular perceptions to the contrary, the pastoral areas actually have relatively low mon- Regional disparities in consumption levels and pover- etary poverty rates (22 percent in 2016). People in pastoral ty remain limited. Differences in consumption levels across areas however are lagging on non-monetary dimensions of regions explained a mere two percent of total inequality in welfare, such as education, health and basic infrastructure. 2016. As such, the regional contributions to overall poverty Table O 1 POVERTY  RATES, POVERTY SHARES, AND POPULATION SHARES BY REGION AND AGRO-ECOLOGICAL ZONE, 2016 POVERTY RATE POVERTY SHARE POPULATION SHARE BY REGION Tigray 27.0% 6.6% 5.8% Afar 23.6% 1.9% 1.9% Amhara 26.1% 25.5% 23.0% Oromia 23.9% 38.3% 37.8% Somali 22.4% 5.5% 5.8% Benishangul Gumuz 26.5% 1.3% 1.1% SNNPR 20.7% 17.5% 19.9% Gambella 23.1% 0.4% 0.4% Harari 7.1% 0.1% 0.3% Addis Ababa 16.8% 2.6% 3.6% Dire Dawa 15.4% 0.3% 0.5% BY AGRO-ECOLOGICAL ZONE Moisture-reliable highlands 23.6% 58.5% 58.4% Drought-prone highlands 20.8% 19.9% 22.5% Moisture-reliable lowlands 25.4% 4.7% 4.3% Drought-prone lowlands 31.7% 7.5% 4.7% Pastoral areas 21.9% 6.9% 7.4% Note: Poverty share denotes the contribution of the region to overall poverty. Source: HCES, WMS, 2016. World bank staff calculations. 2  The pastoral population does however have far worse access to public services and basic infrastructure. EXECUTIVE SUMMARY 13 Characteristics of the very poor – the bottom 10 per- The extreme poor are also more likely to be rural (compared cent of the population that did not experience any real to the poor) and more isolated from markets. Geographically, consumption growth since 2005 – resemble those of the extreme poor are more likely to be located in SNNPR the poor, only more extreme. Whereas the poor are char- and Somali regions. Despite the stagnation in consumption acterized by large households, high dependency rates, and of the extreme poor, indicators of their non-monetary living a lack of education, the extreme poor have yet larger house- conditions improved between 2011 and 2016 but remain low holds, higher dependency rates, and even less education. (Figure O 7). Figure O 7 LIVING  CONDITIONS OF THE BOTTOM 10 PERCENT IMPROVED BETWEEN 2011 AND 2016 Trends in selected indicators from the bottom 10 percent, 2011 and 2016 2011 2016 100 80 60 % 40 20 0 Children Improved Improved Net primary Primary school fully water roof school completion immunized (%) source (%) material (%) enrolment (%) (15-24, %) Source: WMS 2011, 2016; DHS, 2011, 2016. World Bank staff calculations. Strong Urban Poverty Reduction capita GDP between 2000 and 2005, when urban poverty levels also stagnated, strong and sustained economic growth was Driven by Small and translated into robust consumption growth and poverty re- Medium-Sized Towns and Increased duction at the household level (Figure O 9). The contribution of Self-Employment urban areas to poverty reduction will further increase in com- ing years as improved rural education levels and land scarcity Urban areas are becoming increasingly important for speed up rural-urban migration and the ongoing reforms cre- poverty reduction. One third of poverty reduction between ate more job opportunities in the urban private sector. 2011 and 2016 was attributable to urban areas, up from 15 percent in the 2005-2011 period (Figure O 8). Population The strong reduction in urban poverty between 2011 shifts from rural to urban areas did not contribute to poverty and 2016 can mainly be accounted for by small and reduction because rural-to-urban migration, while increasing, medium-sized towns, by households with an unskilled is still relatively low. Strong consumption growth and poverty head, and by households engaged in trade, services, reduction in urban Ethiopia has been very much related to and urban agriculture. Small and medium-sized towns strong economic growth: After a period of stagnation in per accounted for over half of urban poverty reduction between 14 ETHIOPIA POVERTY ASSESSMENT 2011 and 2016 (Panel A of Figure O 10) and close to 60 per- same between 2011 and 2016 and there was little move- cent of poverty reduction happened in households that were ment of labor between different cities. Improvements in the engaged in trade and agriculture (Panel B of Figure O 10). education levels of the urban labor force contributed substan- Structural transformation and labor mobility did not contrib- tially, accounting for 1.8 percentage points of the 11-percent- ute to poverty reduction because there was so little of it: The age-point reduction in urban poverty (Panel C of Figure O 10). sectoral occupational structure in urban areas remained the Figure O 8 THE  CONTRIBUTION OF Figure O 9 SUSTAINED  ECONOMIC URBAN AREAS TO POVERTY GROWTH HAS LIFTED REDUCTION IS INCREASING MANY URBAN HOUSEHOLDS OUT Rural-urban decomposition of the reduction in poverty OF POVERTY GDP per capita and urban poverty rates, 0 40 300 Poverty rate (%) Change in poverty headcount 2000-2016 -1 35 -2 36.9 250 35.1 0 40 30 300 Poverty rate (%) Per capita GDP (2000=100) -3 Change in poverty headcount -1 200 -4 35 25 -2 36.9 250 25.7 35.1 -5 -3 30 20 150 200 -6 -4 25 15 25.7 -7 -5 20 150 100 14.8 -6 10 -8 15 -7 100 50 5 14.8 -9 -8 10 50 -10 -9 5 0 0 -10 2005-2011 2011-2016 0 2000 2005 2011 0 2016 2005-2011 2011-2016 2000 2005 2011 2016 Rural Urban Population shift Poverty rate (urban) GDP per capita Rural Urban Population shift Poverty rate (urban) GDP per capita Note: The population shift effect estimates the change in poverty due to a shift in population from rural to urban areas. Source: HCES 2011; 2016. World Bank staff calculations. Source: HCES 2011; 2016. World Bank staff calculations. EXECUTIVE SUMMARY 15 Figure O 10 SMALL  AND MEDIUM TOWNS, TRADE AND AGRICULTURE, AND THE LOW- SKILLED ACCOUNTED FOR THE BULK OF URBAN POVERTY REDUCTION Contribution to urban poverty reduction, 2011-2016, percentage points Contribution by city size, percentage points A.  Addis Ababa -2.4 Major towns -1.8 Medium towns -3.2 Small towns -2.9 Population shifts -0.5 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 Note: the population shifts estimate the change in poverty due to a shift in population across towns (migration between towns). Contribution by economic sector of household head, percentage points B.  Agriculture and mining -2.3 Manufacturing -1.2 Construction -1.1 Trade -3.5 Infrastructure -0.4 Services -1.9 Public administration -0.6 Inter-sectoral shifts 0.3 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 Note: The intersectoral shifts measure the change in poverty due to a shift of workers from one sector to another. Contribution by education of household head, percentage points C.  No education -2.6 Incomplete primary -2.7 Complete primary -0.6 Incomplete secondary -0.8 Complete secondary -1.8 Post secondary -1.1 Increased education levels -1.8 -3 -2.5 -2 -1.5 -1 -0.5 0 Source: HCES 2011; 2016. World Bank staff calculations. 16 ETHIOPIA POVERTY ASSESSMENT Increased returns to and engagement in self-employ- engage in self-employment increased from 13 percent in ment explain a fair share of the reduction in poverty 2011 to 19 percent in 2016, and this was especially import- in urban Ethiopia. Returns to self-employment (relative to ant in raising consumption levels of the poorest households wage-employment) increased across the welfare distribution (Figure O 12). Although the returns to having a self-employed between 2011 and 2016, except for the poorest (Figure O household head were not particularly strong for poor ur- 11). Poor households benefited from an increased share of ban households, take-up of self-employment by non-head household members taking up self-employment. The share household members was a strong driver of consumption of urban households in which non-head household members changes over time. Figure O 11 THE  PREMIUM OF SELF- Figure O 12 TAKE-UP  OF SELF- EMPLOYMENT OVER EMPLOYMENT WAS WAGE EMPLOYMENT MOST IMPORTANT FOR INCREASED OVER THE CONSUMPTION GROWTH URBAN CONSUMPTION OF THE POOREST URBAN DISTRIBUTION HOUSEHOLDS Returns to self-employment versus wage The effect of additional household employment in urban areas 2011-2016 members in self-employment in urban areas 2011 to 2016 .04 .05 .03 .04 Log difference Log difference .02 .03 .01 .02 .01 0 −.01 0 10 20 30 40 50 60 70 80 90 10 20 30 40 50 60 70 80 90 Consumption percentiles Consumption percentiles Premium: Self employment vs wage employment Effect of increasing share of self employment Source: HCES 2011, 2016. World Bank staff calculations. Source: HCES 2011, 2016. World Bank staff calculations. EXECUTIVE SUMMARY 17 Though small towns experienced rapid poverty reduc- that small towns are expected to add much of the urban tion, infrastructure and access to services and ameni- population (Figure O 13) and given their importance as local ties did not keep up. The share of substandard housing 3 centers of demand and employment for surrounding rural slightly increased in small towns between 2011 and 2016, areas, the large investments currently happening in urban and the share of households with access to improved sani- local governments would need to be expanded to smaller tation and an improved solid waste management system re- towns as well. mained low at less than 10 percent of the population. Given Figure O 13 SMALL  TOWNS AND SECONDARY CITIES WILL ACCOUNT FOR THE BULK OF URBAN POPULATION GROWTH Urban population trends and projections, 2007-2035 52.6 50 5.81 40 20.242 31.1 Less than 50,000 Population (million) 30 50,000 to 100,000 4.561 5.593 100,000 to 500,000 9.225 20 17.5 Addis Ababa 11.9 3.273 3.284 3.488 10 2.74 1.782 20.929 2.276 14.002 1.139 8.916 5.708 0 2007 2015 2025 2035 Source: HCES 2011, 2016. World Bank staff calculations. Figure O 14 AGRICULTURE  REMAINS THE LARGEST CONTRIBUTOR TO POVERTY REDUCTION Sectoral decomposition of changes 2005 to 2016 Change in poverty headcount 2011-2016 2005-2011 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 Pop. Shift Other Serv. Const. Manu. Agri. Source: HCES 2005, 2011, 2016. World Bank staff calculations. 3  Defined as lack of access to piped water and improved sanitation, and overcrowding (more than three persons per room). 18 ETHIOPIA POVERTY ASSESSMENT Although the contribution of urban areas to overall returns to characteristics. Improvements in household poverty reduction is increasing, the rural nature of characteristics – also called “endowments” – explain most Ethiopia means that the agricultural sector remains of the changes in consumption between 2011 and 2016 and crucial. Two thirds of the reduction in poverty between are particularly important for the bottom 40 percent (Figure 2011 and 2016 can be explained by agriculture, down from O 15). Changes in the returns to these characteristics are the contribution over previous period (Figure O 14). Growth only positive for the top quartile. Essentially, this means that of agriculture will remain critical for poverty reduction, giv- the wealthiest households were best equipped to translate en its share of employment and GDP. Changes within the increased asset holdings into increased consumption, likely services sector accounted for about 15 percent of poverty due to their higher levels of education. The endowment effect reduction between 2011 and 2016. The role of structural of asset accumulation between 2011 and 2016 was partic- transformation – shifts in the population out of agriculture ularly strong for the bottom 40 percent. The assets underly- and into manufacturing or services – was very limited over ing this effect include land ownership, livestock ownership, the last period, reflecting the familiar growth with structural and ownership of various household durables. The effects transformation narrative on Ethiopia. The role of this factor is of increasing urbanization were very muted across the entire likely to increase in the future. distribution. This is because the overall share of the urban population in Ethiopia grew slowly between 2011 and 2016, Changes in household consumption can be decom- even though the absolute number of people migrating to ur- posed into a part that is due to changes in household ban areas was large. characteristics and a part that is due to changes in the Figure O 15 CHANGES  IN Figure O 16 ASSET  ACCUMULATION CHARACTERISTICS OF EXPLAINS THE BIGGEST HOUSEHOLDS EXPLAIN PART OF THE CONSUMPTION MOST OF THE INCREASE IN INCREASE CONSUMPTION SINCE 2011 Characteristics’ contributions to The contributions of endowments consumption changes, 2011-2016 and returns to consumption growth 1.2 2011-2016 1.2 1.2 1.2 1.2 1.2 Contribution to change in consumption 1 Contribution to change in consumption Contribution to change in consumption Contribution to change in consumption Contribution to change in consumption Contribution to change in consumption 1 1 1 1 1 0.8 0.8 0.8 0.8 0.8 0.8 .2 .2 .2 .2 .2 .2 0.6 0.6 0.6 0.6 0.60.6 Log difference in consumption Log difference in consumption Log difference in consumption in consumption in consumption Log difference in consumption 0.4 0.4 0.4 0.4 0.40.4 .1 .1 .1 .1 .1 .1 0.2 0.2 0.2 0.2 0.2 0.2 0 difference 0 difference 0 0 0 0 0 0 0 0 0 0 -0.2 -0.2-0.2-0.2 -0.2 −.1 −.1 −.1 −.1 −.1 -0.2 Log Log -0.4 -0.4-0.4-0.4 -0.4 −.1 -0.6 -0.6-0.6-0.6 -0.4 -0.6 −.2 −.2 −.2 −.2 −.2 10 10 20 20 103010 2040 30 20 40 103050 204060 30 50 40 305070 50 60406080 60 70 507090 80 706080 90 7090 90 80 80 90 −.2 Consumption Consumption percentiles percentiles Consumption Consumption percentiles percentiles Consumption percentiles -0.8 -0.8-0.8-0.8 -0.6 -0.8 10 20 Difference Difference30 Difference 50 40 Endowments Difference Endowments Difference 60Returns Endowments Endowments 70Returns Endowments 80Returns Returns 90 Returns Poorest PoorestPoorest 2 PoorestPoorest 2 2 2 3 2 3 3 3 3 Richest Richest Richest RichestRichest Consumption percentiles -0.8 Consumption Consumption quartiles Consumption quartiles Consumption Consumption quartiles quartiles quartiles Poorest 2 3 Ric Difference Endowments Returns Education Education Demographics Education Education Demographics Demographics Demographics Education Urbanization Urbanization Urbanization Demographics Urbanization Urbanization Sector Assets AssetsAssets Assets Home Assets Sector Sector conditions Home Sector Home conditions Sector Home Income conditions Home Income conditionssources Income sources conditions Income Consumption Income sourcessources sources quartiles Source: HCES 2011, 2016. World Bank staff calculations. Education Source: Demographics HCES 2011, Urbanization 2016. World Bank staffAssets Sector calculations. Home conditions Income EXECUTIVE SUMMARY 19 The total effect (endowments plus returns) of chang- improvements were so small for this group. In contrast, ed- es in assets is prominent for all four quartiles of the ucation was the second main driver of consumption growth distribution. For the poorest quartile, approximately corre- for the wealthiest quartile of households. sponding to the poor population, changes in home condi- Poverty fell fastest in the zones that had the strongest tions (higher access to electricity and improved water, better agricultural growth between 2000 and 2016, highlight- dwelling materials) had an overall larger positive effect on ing the continued importance of agriculture in improv- consumption than did asset accumulation (Figure O 16). Even ing the living standards of the poorest. Improved access though the accumulation of assets was an important endow- to towns and urban centers, as measured through decreased ment effect for poor households, decreasing returns to those travel times, was associated with strong poverty reduction, assets meant that the total effect of assets was more muted, indicative of the complementary nature of agricultural and though still positive. This means that the consumption gain non-agricultural growth. Expansions in the use of improved associated with acquiring more assets was smaller for poor seeds was an important driver of increased agricultural pro- households than it was richer households. The total effect of duction and therefore poverty reduction, while the role of the asset accumulation is highest for the top quartile, indicating expanded use of fertilizer is less clear. With improved seeds that this group also had the highest returns to asset accumu- being used on only about six percent of cultivated land, there lation. Improvements in educational attainment did not play is scope for expanding its use in the future. an important role for poor households, mainly because the Figure O 17 ASSET  ACCUMULATION EXPLAINS THE BIGGEST PART OF THE CONSUMPTION INCREASE Characteristics’ contributions to consumption changes, 2011-2016 1.2 1 Contribution to change in consumption 0.8 Education 0.6 Demographics 0.4 Urbanization Assets 0.2 Sector 0 Home conditions -0.2 Income sources -0.4 -0.6 -0.8 Poorest 2 3 Richest Consumption quartiles Source: HCES 2011, 2016. World Bank staff calculations. 20 ETHIOPIA POVERTY ASSESSMENT Figure O 17 GROWTH  IN AGRICULTURAL OUTPUT AND PSNP PARTICIPATION WERE ASSOCIATED WITH POVERTY REDUCTION AT THE ZONAL LEVEL Sectoral growth, safety nets, infrastructure and poverty reduction 2000 to 2016 Effect on poverty of a 1% change in Agricultural output per capita Manufacturing output per capita Services output per capita Proportion in PSNP Distance to primary school Distance to large town −0.6% −0.4% −0.2% 0 0.2% 0.4% Change in poverty headcount Source: HCES, WMS, 2000, 2005, 2011, 2016. World Bank staff calculations. In contrast to earlier periods, cash crops were more Poverty is Mostly Transitory, though important for poverty reduction between 2011 and 2016 than cereals. There was a shift away from the produc- 16 Percent of the Rural Population tion of cereal crops in favor of cash crops. This shift had sig- Was Chronically Poor Between 2012 nificant poverty-reducing effects, largely because of the rapid and 2016 relative gains in the prices of cash crops over these years. Based on the longitudinal ESS surveys, most lon- Within cash crops there was a shift towards the production ger-term poverty in Ethiopia is transient, but there are of khat, particularly in the SNNPR region. The relative price of notable shares of households trapped in chronic pov- khat increased sharply within the group of cash crops, likely erty. Around 16 percent of the Ethiopian population in rural explaining its progressive take-up. While the overall increase areas and small towns was chronically poor over the 2012- in crop prices helped net producers, and will likely continue 2014-2016 period. Just under one third experienced transi- to do so, there are also potential losers from these changes. tory poverty between 2012 and 2016 (Figure O 18). Taken Policy should be nimble enough to ensure that the effects together, almost half of the population experienced at least of rising prices on vulnerable households and parts of the one spell of poverty between 2012 and 2016, reflecting the population are effectively mitigated. high extent of consumption variability and vulnerability in rural Ethiopia. Chronic poverty is mainly concentrated in SNNPR, while transitory poverty is highest in Amhara. Relative to the transitory poor, the chronic poor have larger households and dependency rates, less land and fewer assets, and less ed- ucation. The chronic poor are more likely to benefit from the Government’s social protection programs. EXECUTIVE SUMMARY 21 Figure O 18 CHRONIC  POVERTY IS Figure O 19 MOST  OF THE CHRONIC HIGHER IN SNNPR AND POOR LIVE IN SNNPR AND TRANSITORY POVERTY AMHARA IN AMHARA Regional shares of each poverty Chronic and transitory poverty over the category 2012-2016 period, rural areas and small 100% towns 100% Chronic poor 80% Chronic poor 80% 60% 60% Transient poor Transient poor 40% 40% 20% 20% Never poor Never poor 0% 0 10 20 30 40 50 60 0% Rural and Amhara Oromia SNNPR Tigray Others 0Amhara 10 20 30 40 50 60 Rural and small towns Amhara Oromia SNNPR Tigray Others Oromia SNNPR Tigray Others small towns Chronic poor Transient poor Never poor Amhara Oromia SNNPR Tigray Others Chronic poor Transient poor Never poor Source: ESS 2012, 2014, 2016. World Bank staff calculations. Source: ESS 2012, 2014, 2016. World Bank staff calculations. Upward economic mobility between 2011 and 2016 PSNP at any stage between 2012 and 2016 were less likely was higher in towns and cities than in the rural hinter- to exit poverty than the overall average, as were round three land. Downward mobility - the risk of falling into poverty - was food aid households. This is a finding that potentially bears higher in rural areas: 26 percent of the non-poor population out the effective targeting of the PSNP, as this is a pattern in rural areas had fallen into poverty by 2016, compared to we would expect to see, especially is the baseline consump- 14 percent in towns and four percent in cities. Other factors tion levels of PSNP participants are very low. Male-headed associated with a higher probability of escaping poverty are a households, household with large dependency rates and higher level of education of the household head, and location households living in the drought-prone lowlands were more in pastoral areas. Households that had participated in the likely to fall into poverty. 22 ETHIOPIA POVERTY ASSESSMENT Figure O 20 MORE  EDUCATED HOUSEHOLDS, HOUSEHOLDS HEADED BY WOMEN, AND PASTORALIST HOUSEHOLDS WERE MORE LIKELY TO EXIT POVERTY Probability of exiting poverty by baseline characteristics 80 70 Percentage 60 50 40 30 e e u. u. u. ia ay ra er 3 P 3 3 nd nd nd nd t is PR al al W W W N m ed ed ed ha th al gr la la la la M m PS ro N O P od n or gh gh w w Ti Am Fe o y y io SN N O lo lo st ar ar N fo er hi hi ns PS pa im nd e ne Ev ee e ne te bl bl Pr co ro nd Ex lia Fr ro lia −p Se la re −p re w ht e− ht e− Lo ug ur ug ur ro st ro st D oi D oi M M Household head Household Above average Below average Note: Dashed line is the average probability of exiting poverty of 57.91%. Source: ESS 2012, 2014, 2016. World Bank staff calculations. The PSNP Contributed to Poverty PSNP targeting is progressive both in the highlands and lowlands. While the share of beneficiaries that is drawn Reduction, and its Contribution from the bottom quintile is substantially higher in the high- Could be Further Enhanced lands, the share that is drawn from the bottom 40 percent is higher in the lowlands. Inclusion of households in the top The Productive Safety Net Program (PSNP) significantly quintile is higher in the highlands. On the regional level the contributed to poverty reduction. The PSNP provides con- data show that, relative to what would be possible in case ditional (on work) or unconditional cash or food transfers to of perfect targeting, Afar obtains the best targeting perfor- targeted poor rural households during the lean season. At the mance when targeting is evaluated against consumption zonal level, a one percent annualized increase in PSNP cov- poverty. This counter-intuitive outcome is explained by the erage was associated with a 0.1 percent annualized decrease absence of first-stage woreda targeting: In Afar, all woredas in the poverty rate (Figure O 17). This implies that the PSNP are included in the PSNP, and hence there is no exclusion of was overall well targeted. Indeed, in 2016 34 percent of PSNP the poor because of selection of woredas. beneficiaries were in the bottom welfare quintile and over 60 percent were drawn from the bottom 40 percent (Figure O 21). EXECUTIVE SUMMARY 23 Figure O 21 MOST  OF PSNP BENEFICIARIES ARE IN THE LOWER CONSUMPTION QUINTILES Share of beneficiaries by quintile, 2011 and 2016 2011 2016 40 33 33.8 30 27 23.6 18.8 18.1 20 14.6 11.6 10 9.5 10 0 Q1 Q2 Q3 Q4 Q5 Quintiles of pre-transfer consumption expenditures Source: ESS 2012, 2014, 2016. World Bank staff calculations. The PSNP’s contribution to poverty reduction can be Just like the PSNP, Humanitarian Food Aid (HFA) was further increased. The analysis in this Poverty Assessment reasonably well-targeted in 2016. As per the design, highlights three main issues. First, the number of beneficia- PSNP and HFA reach different types of households: PSNP ries at regional level bears little relation to the prevalence of households share many of the typical characteristics of the poverty or self-reported food insecurity, with the number of poor (few assets and little livestock, remote, little education), beneficiaries exceeding the number of poor and food-inse- while HFA households are similar to the average household cure people in certain regions and falling far short in others in rural areas, with the difference that their calorie intake is (Figure O 22). Second, geographical targeting (selection of substantially lower, hinting at a recent exposure to a nega- woredas) adds little to the PSNP’s targeting performance, tive shock. There are however substantial inclusion errors in which is largely due to poverty and food-insecurity not being HFA targeting, with 30 percent of beneficiaries in the top two geographically concentrated in Ethiopia (Figure O 23). Third, consumption quintiles. These inclusion errors are due to HFA under-coverage remains an issue, with only 13 percent of targeting in woredas where PSNP is not active. Further har- Ethiopia’s poor covered by the PSNP in 2016. Better aligning monizing the PSNP and HFA is likely to improve performance regional caseloads to regional needs and expanding PSNP and targeting of the joint programs. to more woredas but with smaller beneficiary numbers per woreda are likely to increase PSNP’s coverage of the poor and its contribution to poverty reduction. 24 ETHIOPIA POVERTY ASSESSMENT Figure O 22 DISPARITIES  BETWEEN Figure O 23 WOREDA  SELECTION DOES REGIONS’ SHARES IN NOT ADD MUCH TO OVERALL OVERALL CHRONIC TARGETING PERFORMANCE POVERTY AND REGIONS’ Decomposition of the targeting SHARES IN OVERALL PSNP differential into a “woreda-selection” CASELOAD and a “within-woreda household selection” component Regions’ contribution to national chronic Oromia poverty and Amhara Tigraycaseload, national PSNP SNNPR Somali Afar 2016 Selection of woredas Household selection 0.4% 2.2% 100% Oromia 2.9%SNNPR Tigray Amhara 7.0% Afar Somali 10 90% Selection of woredas 9 Household selection 80% 0.4% 2.2% 20.9% 100% Targeting differential 2.9% 7.0% 10 8 70% 90% 39.7% 80% 12.7% 20.9% 9 7 60% Targeting differential 70% 39.7% 8 6 50% 13.0% 12.7% 7 60% 5 8.4 40% 6 50% 33.6% 13.0% 4 4.9 23.7% 30% 5 8.4 40% 3 4.9 20% 33.6% 23.7% 4 30% 2 10% 21.7% 3 20% 18.2% 2 1 1.7 0% 10% 18.2% 21.7% 0.8 1 0 1.7 0% Share in national chronic Share in national PSNP 0.8 0 Poverty Food insecurity poverty Share in national bene ciaries chronic Share in national PSNP Poverty Food insecurity poverty bene ciaries Note: Targeting differential is the difference between coverage of the poor (food-insecure) and that of the non-poor (food-secure) Source: ESS 2012, 2014, 2016. World Bank staff calculations. Equitable Access to Opportunities is Access to opportunities for Ethiopian children im- proved between 2011 and 2016 and disparities in ac- Increasing, but the Rural Poor are at cess have narrowed, leading to an increase in the Risk of Being Left Behind Human Opportunity Index. Household location and household wealth, circumstances that are largely outside Inequality in welfare between households and individ- children’s control, are the main factors determining access uals is partly the result of inequities in access to basic to key opportunities: 40 percent of children aged 15 to 18 opportunities earlier in life. If, for instance, education pro- in urban areas were enrolled in secondary school in 2016, duces significant returns, an adult who had the opportunity compared to 10 percent of rural children of the same age. to complete schooling when she was young will have higher Half of children aged 15 to 18 in households from the top welfare levels than an otherwise comparable person who did consumption quintile had completed primary school, com- not have the opportunity to go to school. The resulting in- pared to less than 20 percent of children in the bottom con- equality can be considered unfair: It is not the result of differ- sumption quintile (Figure O 24). ences in talent or hard work, though rather of circumstances early in life. In an equitable society, an individual’s circum- Education is a particularly important childhood oppor- stances at birth (such as being born a girl or a boy, in a rural tunity. The extent to which parental education influences or an urban area, in a poor or a better-off household, etc.) children’s education weakened between 2011 and 2016, should not influence the individual’s access to a set of im- reflecting an increase in intergenerational mobility. The prob- portant opportunities (such as education, health care, clean ability of a child being enrolled in primary school has become water, etc.). less dependent on parental education, but the opposite has happened for enrolment in secondary school. Improvements EXECUTIVE SUMMARY 25 Figure O 24 LOCATION  AND HOUSEHOLD WEALTH ARE THE MAIN SOURCES OF INEQUITY IN ACCESS TO OPPORTUNITIES Coverage of basic opportunities, urban vs rural and poorest quintile vs richest quintile, 2016 100% 90% 80% Coverage of access 70% 60% 50% 40% 30% 20% 10% 0% Primary Completed Attending Electricity Improved Within 5km enrolled primary secondary water of health post Urban Rural Richest Poorest Note: Primary enrolled refers to children between 7 and 14 years of age. Completed primary and attending secondary refers to persons between 15 and 18-years-old. Access to electricity, improved water and a health post refers to children between 7 and 18-years-old. Source: WMS, HCES 2016. World Bank staff calculations. in access to education took place for children with poorly children of extremely poor households in rural areas accu- educated parents in urban areas and children with relative- mulate more schooling, which may require the introduction ly higher educated parents in rural areas. There is a large of additional policy instruments. education-effect of living in an urban area: Relative to rural children of the same age, the average urban child had com- Perspectives on Continued Poverty pleted 1.44 more grades of education in 2016. Reduction Going Forward Household consumption levels have a large influence on whether the household’s children go to school, and Given its large share in employment and livelihoods, this has implications for poor rural children who are especially for the poor, the agricultural sector still being left behind. While the effect of household consump- holds the key for sustained poverty reduction in the tion on access to primary school did not change between short-to-medium term. Bar large climatic shocks, the 2011 and 2016, its effect on enrolment in secondary school agricultural sector will continue to drive national poverty re- increased (Figure O 25). The effect of household welfare lev- duction, though its contribution will progressively decrease. els on children’s schooling is significantly stronger in rural ar- The analysis presented in this Poverty Assessment suggests eas, indicating greater scope for upward mobility in urban there is room to further increase yields, mainly through pro- areas (where access to schooling is far less dependent on moting the use of improved seeds and further increasing household wealth). The implication of these results is that access to markets. children of poor households and poorly-educated parents The agricultural sector will however not to able to ab- in rural areas are in danger of being left behind. Breaking sorb the rapidly growing labor force, at least not in the intergenerational transmission of poverty will require that its current form. The Ethiopian working-age population is 26 ETHIOPIA POVERTY ASSESSMENT Figure O 25 ENROLMENT  IN SCHOOL BECAME MORE DEPENDENT ON HOUSEHOLD WELFARE BETWEEN 2011 AND 2016 School enrollment probabilities by household consumption, 2011 and 2016 Primary school enrollment Secondary school enrollment Probability of secondary enrollment .8 .8 Probability of primary enrollment .6 .6 .4 .4 .2 .2 0 0 6 7 8 9 10 11 6 7 8 9 10 11 Log of real consumption per adult equivalent Log of real consumption per adult equivalent 2011 2016 2011 2016 Source: WMS 2011, 2016. World Bank staff calculations. projected to grow at two million per year in the coming de- (such as the risk of losing land ownership in rural areas) holds cade. The increasing scarcity of agricultural land in the high- significant promise for continued poverty reduction. lands means that an ever-larger share of young people will Effective safety nets will remain essential. Given that not inherit sufficient land to make an independent living and currently only a small share of the poor population in Ethi- will need to transition to livelihoods off the farm4. Given the opia is covered by safety nets, further expanding and better relatively low education levels of the labor force, especially targeting existing safety nets will be necessary. One option in rural areas, the bulk of the newcomers will not qualify for to reach more of the poor with the existing safety nets is to modern wage employment in the formal economy. This im- expand safety to all woredas in the country, though with a plies that most of the growing labor force will try to make a smaller number of beneficiaries per woreda to manage the living in the informal or semi-formal sector, both in wage- and fiscal implications. The safety net should be flexible enough self-employment. The projected increase in the size of the to scale up and down depending on the particular state of agri-food sector as Ethiopia urbanizes and urban incomes national and local economies. grow could be a large generator of employment for young people leaving the farm5. Finally, improving the welfare of the bottom 10 percent of the population will require more investments in the The relative shift out of farming into off-farm and non- human capital of children. In an ideal scenario, children of farm occupations will also require a spatial shift in the extremely poor households would accumulate more educa- distribution of the population. Though accurate numbers tion and be able to move out and diversify into more produc- do not exist due to the outdated nature of the Population tive activities, breaking the intergenerational transmission of Census (2007), it is estimated that about 80 percent of the poverty. This however is not taking place: Children from poor population is still rural, making Ethiopia under-urbanized giv- and poorly educated parents in rural areas severely lag on en its income level. At the same time, urban areas of all sizes education and the effect of household wealth on child educa- are growing substantially and poverty in cities and towns is tion has strengthened since 2011. Devising policies or inter- falling quickly. Given the close linkages between small towns ventions to keep children from poor rural families in school for and their surrounding rural hinterland, and the lower skills re- longer, potentially through income-incentives, will be crucial quirements for jobs in small towns, investing in small towns in any attempt to share the benefits of growth more widely. and facilitating migration from rural areas by removing barriers 4 This is already happening now: Expectations of land inheritance are a main driver of migration decisions and non-agricultural em- ployment (Kosec et al, 2017). 5  Minten et al (2018). EXECUTIVE SUMMARY 27 28 ETHIOPIA POVERTY ASSESSMENT Introduction The most recent Poverty Assessment for Ethiopia was strong. GDP grew at an average rate of over nine percent per published in 2015. This Poverty Assessment covered the year, resulting in a 38 percent increase in per capita GDP lev- period 1996-2011. As a result, this current Poverty Assess- els. Economic growth was mainly driven by services, which ment mainly focuses on the 2011-2016 period, although explained 42 percent of the expansion in GDP between 2011 certain chapters also take a longer-term view and describe and 2016. Agriculture contributed 25 percent to the growth trends and conduct analyses spanning a longer time period. in GDP over the same period, while industry accounted for A widely-quoted fact about Ethiopia is that growth has been 34 percent (Figure 1). The growth in the industrial sector was strong and sustained over the last decade and a half. The mainly driven by the construction subsector, which account- recent and ongoing economic reforms in the country mean ed for 77 percent of industrial growth contribution to the real that not only is the nature of growth likely to change, but the GDP. Manufacturing contributed on average 22 percent to relationship between economic growth and poverty reduc- industrial growth contribution to the real GDP. Between 2011 tion will also shift. In order to provide some broader context and 2016, the contribution of manufacturing to real GDP before getting to poverty numbers, this introduction briefly growth doubled from a low base, increasing from 0.5 to 1.0 reviews the main macro-economic developments between percentage point. 2011 and 2016. Economic output has shifted from agriculture to indus- Between 2011 and 2016 Ethiopia continued its develop- try since 2010/11. Agriculture’s share in GDP decreased mental state model, characterized by a strategic focus from 46 percent in 2011 to 38 percent in 2016, while the on agriculture and industrialization coupled with large share of industry increased from 14 percent to 24 percent public infrastructure investments facilitated by hetero- over the same period. Services remained fairly constant at dox marco-financial policies. Economic growth remained 39 to 40 percent of GDP. Changes in sectoral employment Figure 1 SERVICES  AND INDUSTRY HAVE BEEN DRIVING GROWTH Sectoral contribution to GDP growth Agriculture Industry Services GDP 12 11.3 10.6 10.3 10.4 10.2 10 5.2 3.2 3.6 8.1 8 4.7 6.3 6 4.4 1.9 3.9 4.2 4 3.1 2.1 2 4.1 2.9 3.1 2.6 2.2 2.3 0.9 0 2010/11 2011/12 2012/13 2013/14 2014/15 2015/16 Source: National Planning Commission INTRODUCTION 29 Figure 2 THE  SHARE OF INDUSTRY IN GDP IS INCREASING Sectoral shares in GDP, % 100 90 80 38.7 39.2 38.7 39.4 39.2 39.7 70 60 50 13.5 15.1 17.4 18.8 21.6 23.7 40 30 20 46.3 44.7 43.4 41.3 39.8 37.5 10 0 2010/11 2011/12 2012/13 2013/14 2014/15 2015/16 Agriculture Industry Services Source: National Planning Commission shares have been less dramatic. Agriculture’s share of em- After significant spikes in 2011 and 2012, inflation ployment modestly decreased to 74 percent in 2013 (year moderated and was largely contained to single-digits of the latest Labor Force Survey), down from 78 percent in during 2013/14 – 2016/17. Inflation rapidly rose in 2010/11 2005. Most workers shifted towards services (employment and 2011/12 reaching a peak of 40.7 percent in August 2011 share of 18 percent in 2013) and, to a lesser extent, industry before starting to come down and reaching single-digit levels (share of 9 percent in 2013). in March 2013 at 7.7 percent. Food inflation which peak- ed at 51.7 percent in October 2011 was the major driver Spending on the pro-poor sectors6 was maintained at of the spike in inflation although increases in nonfood infla- around 12 percent of GDP during 2010/11 – 2015/16. tion also contributed. After moderating to single digits during Of the pro-poor sectors, allocation to the education sector 2013/14 – 2016/17, inflation again picked up to double digits reached 4.7 percent of GDP in 2016/17 before declining to in 2017/18 (Figure 4). 4 percent in 2017/18. The share spend on roads was larger at the beginning of the period than at the end, and stood at Exports have been on a downwards trend between 1.8 percent in 2017/18. Spending on water and agriculture 2011 and 2016. Exports as a share of GDP decreased from have largely shown increases through 2014/15 but declined close to 17 percent in 2011 to eight percent by 2016. The over the past couple of years. The reallocation in government poor performance of exports means that Ethiopia’s debt lev- expenditures over the past couple of years affected the pro- els need to be monitored carefully. While external debt levels poor sectors which declined to 11.1 percent in 2016/17 and are not unusually high at 54 percent of GDP, low export levels 9.5 percent in 2017/18 (Figure 3). have led to the deterioration of two key measures of capacity to repay, the debt-to-export and debt service-to-export ra- tios, pointing to vulnerability to debt distress. 6  The government’s definition of pro-poor sectors includes education, health, agriculture, roads and water. 30 ETHIOPIA POVERTY ASSESSMENT Figure 3 SPENDING  ON PRO-POOR SECTORS STAYED LARGELY UNCHANGED BETWEEN 2011 AND 2016 Spending on pro-poor sectors, % of GDP Agriculture Water Roads Education Health Total pro-poor sectors 14 12 10 8 6 4 2 0 2010/11 2011/12 2012/13 2013/14 2014/15 2015/16 Source: Ministry of Finance INTRODUCTION 31 Under the current reform agenda, the economic pros- and poverty reduction between 2011 and 2016 and assess- pects for FY2019 and the medium term should remain es whether non-monetary indicators of welfare corroborate stable. Annual real GDP growth is projected to be around the consumption and poverty trends. Chapter II presents the 7.9 percent in FY2019 and 8.2 percent in the medium term. poverty profile, while Chapter III examines the drivers of pov- The reform agenda is expected to address some macro- erty reduction, focusing mainly though not exclusively on the economic imbalances, while moderate fiscal deficits and most recent period (2011-2016). Chapter IV exploits the lon- prudent monetary policy are expected to reduce the rate gitudinal nature of the Living Standards Measurement Study of inflation and keep it in the single digits. Merchandise ex- to assess household economic trajectories over time and ports could recover in the medium term, as large investment quantify chronic and transitory poverty. The large decrease in projects, such as the railway to the Port of Djibouti and large urban poverty is dissected in Chapter V. Chapter VI focuses power dams (with potential for electricity exports), become on targeting of the Government’s two main social protection operational. programs, the Productive Safety Net Program and Human- itarian Food Aid. Finally, Chapter 7 focuses on inequality of To summarize, the period on which this Poverty As- opportunity and intergenerational economic mobility to ex- sessment focuses (2011-2016) was characterized by amine the extent to which exogenous circumstances affect strong economic growth and sustained spending on access to key services and opportunities in Ethiopia. pro-poor sectors. The year during which the last poverty survey was implemented, 2015/16, was however a severe Throughout this Poverty Assessment, when doing drought year, which may have influenced the pace of pov- cross-country comparisons, we will benchmark Ethiopia erty reduction. based on the countries that were identified as “structural peers” in the 2016 Systematic Country Diagnostic. These This Poverty Assessment proceeds as follows: Chapter countries are Rwanda, Burkina Faso, Uganda, Tanzania, Mo- I summarizes the trends in household consumption growth zambique, and Myanmar. Figure 4 INFLATION  WAS MOSTLY CONTAINED BETWEEN 2013 AND 2016 Inflation, year-on-year percentage 60 50 40 30 20 10 0 Jul-10 Oct-10 Jan-11 Apr-11 Jul-11 Oct-11 Jan-12 Apr-12 Jul-12 Oct-12 Jan-13 Apr-13 Jul-13 Oct-13 Jan-14 Apr-14 Jul-14 Oct-14 Jan-15 Apr-15 Jul-15 Oct-15 Jan-16 Apr-16 -10 2010/11 2011/12 2012/13 2013/14 2014/15 2015/16 General Food Nonfood Source: CSA 32 ETHIOPIA POVERTY ASSESSMENT Figure 5 EXPORTS  DECLINED BETWEEN 2011 AND 2016 Exports as a share of GDP Goods exports Service exports 18 16 14 8.1 12 10 6.5 6.0 5.7 8 4.8 6 4.1 4 8.6 7.3 6.5 5.9 2 4.6 4.0 0 2010/11 2011/12 2012/13 2013/14 2014/15 2015/16 Source: National Bank of Ethiopia INTRODUCTION 33 34 ETHIOPIA POVERTY ASSESSMENT CHAPTER I 2011-2016: Continued Growth and Poverty Reduction Despite adverse circumstances related to the 2015/16 El-Nino drought, poverty reduction continued between 2011 and 2016. The share of the population below the poverty line decreased from 30 percent in 2011 to 24 percent in 2016. The reduction in poverty was particularly strong in urban areas, where the poverty headcount decreased from 26 percent to 15 percent. Poverty reduction in rural areas was relatively subdued and statistically significant only in three regions. The reduction in poverty is robust to the use of alternative deflators. Consumption growth between 2011 and 2016 was higher in the upper parts of the welfare distribution, leading to a modest increase in inequality. The increase in inequality was mainly driven by the increased welfare disparity between urban and rural areas. Consumption levels of the bottom 10 percent did not increase between 2011 and 2016, a continuation of the pattern since 2005. Given that the bottom 10 percent is mainly rural, poverty severity in rural areas was higher in 2016 than in 2005. Trends in non-monetary dimensions of welfare confirm the positive consumption trend. Ownership of durables and housing quality improved between 2011 and 2016, as did access to an improved water source. Human development indicators increased from a low base, both for the general population and the bottom 10 percent. Human development outcomes remain however poor, as reflected by Ethiopia’s low rank in the new Human Capital Index. Contrary to popular perception, poverty reduction between 2011 and 2016 was strongest in the pastoral areas, which largely overlap with the regions of Somali and Afar. Of the five main agro-ecological zones, only the moisture-reliable lowlands (mainly overlapping with the regions of Gambella and Benishangul Gumuz) did not experience poverty reduction between 2011 and 2016. The drought-prone lowlands, mainly consisting of the lowland parts of Oromia and SNNPR, experienced a reduction in poverty but remained by far the poorest ecological zone in the country. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 35 Introduction As shown in the introduction, Ethiopia continued to household living standards in Ethiopia between 2011 post strong economic growth between 2010/11 and and 2016. The chapter has three parts. In the first part we 2015/16. Between 2010/11 (henceforth referred to as 2011) will offer a detailed description of the evolution of poverty and 2015/16 (henceforth referred to as 2016), Ethiopia’s using data from the two most recent Household Consump- economy grew at a rate of over 9 percent per year, resulting tion Expenditure Surveys (henceforth HCES) implemented in in a 38 percent increase in per capita GDP levels over the 2010/11 and 2015/16. In the second part we will examine same period. Agricultural production grew by 31 percent. whether non-monetary indicators corroborate the trend in Poverty headcount followed the general trend, with the share household living standards as measured by the HCES. Data of population living below the national poverty line dropping on these non-monetary indicators are provided by the Wel- from 30 percent in 2011 to 24 percent in 2016. Given that fare Monitoring Surveys (WMS), implemented at the same the latest survey was implemented during an exceptional time and on the same sample as the HCES surveys, and the drought year (2015/16), this is a remarkable achievement. Demographic and Health Surveys (DHS), also implemented in 2011 and 2016. The third part will discuss the practice of This chapter will use several sources of data to paint poverty monitoring and measurement in Ethiopia and formu- a detailed picture of the evolution of poverty and late options to further strengthen that practice. 2. SIGNIFICANT HOUSEHOLD CONSUMPTION GROWTH AND POVERTY REDUCTION Household consumption growth and poverty reduction continued between 2011 and 2016, despite adverse circumstances linked to the 2015/16 El-Nino drought. Poverty reduction was strong in urban areas but weak in rural areas. Growth was stronger in the upper parts of the welfare distribution, leading to a modest in- crease in inequality from a low base. The bottom 20 percent in rural areas and the bottom 10 percent at the national level did not experience real consumption growth, replicating a pattern that was also found in the earlier 2005-2011 period. As a result, poverty severity did not decrease much despite the reduction in the poverty headcount. The reduction in poverty is robust to the use of alternative methods of deflating, though the officially-used method tends to overestimate the reduction in poverty in the recent period. 2.1 Solid reduction in the poverty growth rate of 6 percent. In contrast, growth among rural households was sluggish at 1 percent per year, perhaps due headcount to the influence of the 2015/16 El Nino drought (Table 1). Continued economic growth translated into improved Average household consumption increased in all re- living standards at the household level, particularly in gions.7 The highest consumption growth was observed in urban areas. Household consumption expenditures per Harari (increase of 59 percent), Dire Dawa (increase of 55 adult equivalent, the welfare metric used in this Poverty As- percent), and Gambela (increase of 34 percent). From ob- sessment (see Box 1), increased by 14 percent in real terms serving Table 1, it is clear that the increase in consumption between 2011 and 2016, translating into an annual growth within regions was largely driven by urban areas (except for rate of 2.6 percent. The increase in household welfare was Tigray). While the increase in urban consumption is statis- particularly strong in urban areas with an annual consumption tically significant in all regions and cities except for urban 7 The sample coverage for Afar and Somali improved in 2016 where more zones were included and the results in Table 1 and subse- quent tables/figures in this chapter and other chapters that compare 2011 and 2016 don’t include households from these new zones (see Box 1). 36 ETHIOPIA POVERTY ASSESSMENT Somali, the change in rural consumption is only statistically levels rather than means, we find that median consumption significant in Tigray, Gambella and Harari. Rural households in decreased in Amhara and Afar, pointing to different dynamics other regions did not experience an increase in consumption at different parts of the welfare distribution in those regions between 2011 and 2016. Looking at median consumption (Annex Table 1). Table 1 MEAN  ANNUAL HOUSEHOLD CONSUMPTION PER ADULT IN DECEMBER 2015 PRICES, 2011 AND 2016 TOTAL URBAN RURAL % % % 2011 2016 CHANGE 2011 2016 CHANGE 2011 2016 CHANGE National 11,009 12,500 13.5 13,901 18,649 34.2 10,434 11,014 5.6 Tigray 11,630 14,108 21.3 17,691 20,536 16.1 10,074 12,038 19.5 Afar 10,641 12,902 21.2 13,945 18,645 33.7 9,298 10,512 12.5 Amhara 10,944 12,340 12.8 14,325 21,879 52.7 10,464 10,557 0.9 Oromia 10,947 12,022 9.8 13,891 18,080 30.2 10,504 11,022 4.9 Somali 10,565 11,714 10.9 12,942 14,470 11.8 10,004 9,242 10.6 Benishangul-Gumuz 11,435 13,373 17.0 15,124 18,524 22.5 10,832 12,112 11.8 SNNPR 10,725 12,204 13.8 13,391 18,049 34.8 10,414 11,157 7.1 Gambella 10,334 13,855 34.1 12,477 17,945 43.8 9,325 11,745 26.0 Harari 13,264 21,059 58.8 15,344 24,028 56.6 11,397 17,479 53.4 Addis Ababa 12,831 16,237 26.5 12,831 16,237 26.5 - - - Gambella 11,268 17,428 54.7 11,617 20,718 78.3 10,532 11,393 8.2 Note: The increase in mean consumption is statistically significant in all regions except for Somali. For rural areas, the increase in mean consumption is statistically different from zero only in Tigray, Gambella and Harari. Source: HCES, 2011; 2016. World Bank staff calculations. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 37 Box 1 Consumption aggregation and poverty measurement Most of the analysis presented in this Poverty Assessment is based on detailed consumption data included in the Household Consumption Expenditure Surveys (HCES) (2010/11 and 2015/16). All consumption of food and nonfood items is included, regardless of whether these items are purchased on the market, come from own production, or were received as gifts. For own-consumption and gifts, the quantities consumed are valued at prevailing prices in the enumeration area. Although consumption is expressed on an annual basis, the reference period used during data collection varies based on the nature of the consumption items. For example, information on food and food-related items was asked twice a week using the “last three days” and “last four days” as reference periods (households are visited twice during the HCES). For house rent, durable goods, clothing, health and education expenditures, and so forth, the “last three months” and “last 12 months” were used as references. Imputed rent is included in the consumption aggregate8. To capture the effect of seasonal variations, the data were collected over a 12-month span (Hamle 1 to Sene 30/July 8 to July 7), by randomly allocating sampled households to different months. To adjust for price variations across time and space, spatial and temporal price deflators are used. First, nominal consumption is adjusted for price differences across reporting levels, by using the spatial deflators provided by FDRE (2012, 2017). Second, spatially-deflated consumption levels are expressed in December prices (December 2010 and December 2015), by using the food and nonfood Consumer Price Indexes provided by the Central Statis- tics Agency. The food and nonfood Consumer Price Indexes are also used to bring the December 2010 consump- tion expenditure (2010/11 HCES) to December 2015 prices. Finally, to adjust for variations in household size and composition, consumption expenditure is divided by the officially-used adult-equivalent scales, which are based on calorie requirements and vary by age and sex. This exercise should result in a consumption aggregate that can be compared through space and time. The consumption aggregates used in this Poverty Assessment are the official ones used by the National Planning Commission. The poverty rates presented in this chapter are based on the national poverty line. The poverty line is based on a food basket that is required to achieve the minimum daily calorie requirement – 2,200 kilo-calories per adult in Ethi- opia – and adjusted upwards to include non-food consumption. The food basket was determined in 1996 and has not been changed since. Accordingly, the poverty line in 1996 was 1,075 Birr per adult equivalent per year in 1996. In 2011, the poverty line was updated by costing the items in the original food basked at prevailing prices and doing a similar adjustment for non-food consumption. The updated poverty line was 3,781 Birr per adult equivalent per year (in December 2010 prices). For the most recent poverty measurement the 2011 poverty line was inflated using the GDP deflator, resulting in a poverty line of 7,184 Birr per adult equivalent per year in December 2015 prices. Given the peculiar nature of using the GDP deflator to update a poverty line, this chapter will also conduct an analysis of the sensitivity of the poverty estimates to different deflators. All temporal comparisons in this Poverty Assessment exclude a number of zones in Somali and Afar regions. In 2011, the HCES covered only two zones in Afar and three in Somali. In 2016, the coverage was improved – five zones and eight zones were covered respectively in Afar and Somali. While the 2016 data is more representative of pastoral areas and the two regions, to maintain comparability, in this chapter (and subsequent chapters – when comparison is made between 2011 and 2016), households sampled from the new zones are excluded from the analysis. Sources: FDRE (2012, 2017); Central Statistical Agency 2018. 8 Imputed rent was calculated by the Central Statistics Agency (CSA) and was included in the consumption aggregate that was shared with the Bank team. 38 ETHIOPIA POVERTY ASSESSMENT Consumption increased across the major agro-eco- largely with Gambella and Benishangul-Gumuz), and the logical zones, except in the drought-prone lowlands. pastoral areas (Somali and almost all of Afar). Annex 1 plots Using a classification based on altitude and rainfall, we define the agro-ecological zones on a map. Average household five distinct agro-ecological zones: The drought-prone high- consumption increased everywhere except in the drought- lands (mainly eastern parts of Amhara and Tigray, but also prone lowlands (increase not statistically significant). Perhaps north-eastern Oromia), the drought-prone lowlands (the low- counter-intuitive, consumption increased most in the pasto- land areas of Oromia and SNNPR and western parts of Afar), ral areas and the moisture-reliable lowlands, driven by large the moisture-reliable highlands (large parts of Oromia, SN- consumption gains in the cities (Table 2). NPR, Amhara), the moisture-reliable lowlands (overlapping Table 2 MEAN  ANNUAL HOUSEHOLD CONSUMPTION PER ADULT IN DECEMBER 2015 PRICES, 2011 AND 2016 NATIONAL URBAN RURAL 2011 2016 DIFF (%) 2011 2016 DIFF (%) 2011 2016 DIFF (%) Drought prone highlands 11,260 13,422 19.2 14,495 20,342 40.3 10,690 11,860 10.9 Drought prone lowlands 10,605 11,128 4.9 13,973 20,914 49.7 10,081 9,310 -7.6 Moisture reliable highlands 10,885 12,136 11.5 13,664 17,421 27.5 10,319 10,839 5 Moisture reliable lowlands 12,193 14,705 20.6 15,844 24,872 57 11,410 12,082 5.9 Pastoral areas 10,814 12,907 19.4 12,458 17,660 41.8 10,112 10,880 7.6 Note: Changes in bold are statistically significant. Source: HCES, 2011, 2016; World Bank staff calculations. Household consumption growth translated into solid 30.4 percent to 25.6 percent (Figure 6). This is a departure poverty reduction. At the national level, the percentage of from the pattern of poverty reduction observed between people whose consumption was below the national poverty 2005 and 2011, when urban and rural areas experienced line (ETB 7,184 per adult equivalent per year) decreased from similar magnitudes of poverty reduction9. While poverty in 29.6 percent in 2011 to 23.4 percent in 2016, a statistically Ethiopia is fairly low in comparison to a selection of compar- significant decline of six percentage points. In urban areas, ator countries, the degree to which growth has translated poverty decreased strongly from 25.6 percent in 2011 to into poverty reduction has also been fairly low (Box 2). 14.8 percent in 2016, while in rural areas it decreased from Figure 6 POVERTY  DECREASED IN BOTH RURAL AND URBAN AREAS Poverty rate based on national poverty line, 2011 and 2016 2011 2016 35 30.4 29.6 30 Percentage poor 25.7 25.6 23.5 25 20 14.8 15 10 5 0 National Urban Rural Source: HCES; 2011, 2016. World Bank staff calculations. 9  See Ethiopian Poverty Assessment 2014. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 39 Box 2 A low poverty rate, but also a low transmission from growth  to poverty reduction Compared with its structural peers, Ethiopia has a fairly low poverty rate. With a poverty rate of 27 percent based on the international poverty line (US$1.9 per person per day in purchasing power parity), poverty in Ethiopia is lower than all the structural peers except for Myanmar (Figure 7). However, the extent to which growth (in GDP per capita) in Ethiopia translates into poverty reduction has also been low: The “poverty-elasticity of growth”, a measure of the extent to which GDP growth decreased poverty, amounted to -0.33 between 1997 and 2016, which means that a one percent increase in per capita GDP was accompanied with a 0.33 percent decrease in poverty rates. Among the sample of comparators, Mozambique and Rwanda had a lower responsiveness of poverty to growth (Figure 8). The semi-elasticity, which measures the percentage point change in poverty for a one percent change in per capita GDP, was lowest in Ethiopia: Between 1997 and 2016, a one percent increase in per capita GDP was accompanied by a 0.19 percentage point reduction in poverty10. Tanzania’s semi-elasticity, for instance, was close to four times higher. Figure 7 ETHIOPIA  HAS LOWER POVERTY THAN ALL ITS COMPARATORS EXCEPT MYANMAR Trends in poverty in Ethiopia and its comparators using the 1.9 USD PPP poverty line: 1997 -2016 100 80 60 40 20 0 1998 2009 2014 2001 2011 2016 2002 2008 2014 2015 2000 2010 2014 2000 2007 2011 1999 2012 2016 Burkina Faso Ethiopia Mozambique Rwanda Tanzania Uganda Myanmar Source: HCES; 2011, 2016. World Bank staff calculations. The reasons behind Ethiopia’s relatively low conversion rate between growth and poverty reduction are not entirely clear. Research shows that countries with low levels of initial development (high initial poverty rates) tend to have lower growth-poverty elasticities, as do countries with high levels of inequality (Bourguignon, 2003; Ravallion, 2012). While Ethiopia definitely had low levels of initial development, it also had among the lowest levels of inequality. It is possible that the baseline level of development in Ethiopia was so low that growth has increased incomes of the poor but not yet to the level of pulling them above the poverty line. If that hypothesis were true, continued economic growth could lead to much more poverty reduction in the future.  he growth elasticity of poverty is notoriously sensitive to the baseline level of development. If initial levels of consumption are 10 T low, growth rates in consumption will be relatively high for a same absolute change, which will lead to an underestimation of the growth-elasticity of poverty. As such, growth elasticities tend to be higher in richer countries. The semi-elasticity partly avoids this pattern and does not automatically increase when a country grows richer (Klasen and Misselhorn, 2008). 40 ETHIOPIA POVERTY ASSESSMENT Figure 8 BUT  THE RATE AT WHICH GROWTH HAS TRANSMITTED TO POVERTY REDUCTION IN ETHIOPIA IS AMONG THE LOWEST Growth elasticity of poverty for Ethiopia and its comparators, 1997 - 2016 Tanzania Burkina Faso Uganda -0.19 Ethiopia -0.33 Mozambique Rwanda -1.00 -0.90 -0.80 -0.70 -0.60 -0.50 -0.40 -0.30 -0.20 -0.10 0.00 Semi-elasticity Elasticity Note: The elasticities are estimated by taking the first and last years between 1997 and 2016 when data on poverty is available. Myanmar is not included because poverty data is available only for 2015. Source: World Development Indicators. World Bank staff calculations. The significant decrease in the poverty rate also led increased strongly in the sample, but is only marginally signif- to a modest decrease in the absolute number of poor icant from a statistical point of view. people. At national level, the number of poor people de- Depth and severity of poverty also decreased between creased from 22.5 million to 20.2 million. It decreased from 2011 and 2016, with substantial spatial differences. 3.2 million to 2.4 million in urban areas and from 19.3 million The depth of poverty, which measures how far on average to 17.7 million in rural areas. the consumption of the poor is from the poverty line (also Poverty decreased in most regions, particularly in ur- called the poverty gap), modestly dropped at the national ban areas. Poverty rates decreased significantly in all re- level, reflecting a sharp decrease in urban areas and a weak gions expect for Tigray, Benishangul-Gumuz and Harari, one in rural areas. The severity of poverty, which measures where the decrease observed in the sample cannot be gen- the average poverty gap for the poor but attaches more eralized to the population (Table 3). Poverty reduction was weight to the poorest, decreased strongly in urban areas but especially strong the city administrations of Addis Ababa and remained constant in rural areas (Figure 9). At the regional Dire Dawa, and in the regions that had the highest poverty level, poverty severity decreased strongly in Afar, Benishan- rates in 2011 (Somali and Afar – see Box 3). While the re- gul-Gumuz and Gambella, and in the city administrations duction in urban poverty was statistically significant in almost (Addis Ababa and Dire Dawa), and increased sharply from a all region, poverty reduction in rural areas only happened in low base in Harari. Afar, Somali and SNNPR. Poverty rates in rural Dire Dawa CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 41 Table 3 POVERTY  DECREASED IN MOST REGIONS, ESPECIALLY IN URBAN AREAS11 Percentage of people below the national poverty line, by region and urban vs rural TOTAL URBAN RURAL % % % 2011 2016 CHANGE 2011 2016 CHANGE 2011 2016 CHANGE Tigray 31.8 27.0 -4.8 13.7 14.2 0.5 36.5 31.1 -5.3 Afar 36.1 25.6 -10.5 23.7 10.6 -13.1 41.1 31.9 -9.2 Amhara 30.5 26.1 -4.4 29.2 11.6 -17.6 30.7 28.8 -1.9 Oromia 28.7 23.9 -4.8 24.8 15.3 -9.5 29.3 25.3 -4.0 Somali 32.8 16.8 -16.0 23.1 19.4 -3.7 35.1 16.3 -18.8 Ben.-Gumuz 28.9 26.5 -2.4 21.3 17.7 -3.6 30.1 28.7 -1.4 SNNPR 29.6 20.7 -8.9 25.8 14.4 -11.3 30.0 21.9 -8.1 Gambella 32.0 23.0 -9.0 30.7 16.6 -14.1 32.5 26.4 -6.1 Harari 11.1 7.1 -4.0 11.7 6.0 -5.7 10.5 8.5 -2.0 Addis Ababa 28.1 16.8 -11.3 28.1 16.8 -11.3 - - - Dire Dawa 28.3 15.4 -12.9 34.9 11.1 -23.8 14.2 23.3 9.1 Note: Changes in bold are statistically significant. Source: HCES, 2011, 2016; World Bank staff calculations. In rural areas, the reduction in poverty headcount com- severity is remaining constant (and in some regions increas- bined with a stagnation in poverty severity suggests ing) while headcount is falling means that whoever remained that the fewer people who are poor are increasingly fall- in poverty between 2011 and 2016 became ever poorer. This ing behind. All else equal, a reduction in poverty headcount pattern will be explored in more detail in the next section. would lead to a reduction in poverty severity. The finding that Figure 9 REGIONAL  VARIATIONS IN THE TREND IN POVERTY SEVERITY Poverty severity based on national poverty line, 2011 and 2016 2011 2016 0.045 0.040 0.035 0.030 0.025 0.020 0.015 0.010 0.005 0.000 Rural Urban Tigray Afar Amhara Oromiya Somali Benshangul SNNPR Gambella Harari Addis Ababa Dire Dawa Urban / Rural Region Source: HCES; 2011, 2016. World Bank staff calculations. 11 If households from the new zones of Afar and Somali are included in the 2016 analysis, total (urban and rural combined) poverty be- comes 23.6% and 22.4% and rural poverty will be 26.5% and 22.3% in Afar and Somali, respectively. Urban poverty in Afar remains the same (10.6%) while it went up to 19.4% in Somali. 42 ETHIOPIA POVERTY ASSESSMENT Box 3 W  hat’s behind the strong poverty reduction in Somali and Afar? Somali and Afar experienced exceptionally strong poverty reduction between 2011 and 2016. Poverty in Somali dropped by 16 percentage points, while poverty in Afar decreased by 11 percentage points. This finding is at odds with the common perception of Afar and Somali, both predominantly pastoral regions, as being the most destitute regions of Ethiopia. It also seems at odds with the persistently high number of people in these regions that are deemed in need of emergency food aid. And it is also at odds with the believe that the 2015/16 El Nino drought was particularly devastating in the pastoral lowlands. Is there a way to reconcile the data with the perceptions? Figure 10 AFAR  WAS HEAVILY AFFECTED BY THE EL-NINO DROUGHT Soil moisture anomalies, September 2015 vs 1981-2014 Source: FEWS NET, NASA, 2015. Box continue on next page. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 43 Box continued from previous page. In Afar, the answer is at least partly related to aid. According to remote sensing data on soil moisture, Afar was heavily affected by the 2015/16 drought (Figure 10). In response to the drought, the Productive Safety Net Program (PSNP) was scaled up and complemented by Humanitarian Food Aid (HFA). Overall, over half of the population of Afar was covered by either PSNP or HFA (or both) during the 2016 HCES survey. This high level of coverage trans- lated into higher consumption expenditures and lower poverty rates. A simulation that attempts to remove benefits (both in-cash and in-kind) from the consumption aggregate suggests that poverty rates in 2016 in Afar would have been a lot higher in absence of PSNP and HFA12: 34 percent instead of the actual poverty rate of 24 percent (Figure 13). PSNP and aid in Afar were successful in alleviating the effects of the drought. In Somali the picture is less straightforward. One zone in northern Somali (Shinile zone) was heavily affected by the drought, while the largest part of Somali had normal or wetter than usual soil moisture. Rainfall in most of Somali region was better than average during the drought, except for Shinile zone (which experienced a large rainfall deficit; FEWS NET, 2015). Coverage of PSNP and HFA in Somali remained however high, at 38 percent of the total popula- tion during the 2016 HCES. In contrast to Afar, this high level of coverage does not explain the strong poverty reduc- tion in Somali: Simulated pre-benefit poverty rates are similar to actual post-benefit poverty rates ((Figure 11). This either means that benefits were directed to households that were already above the poverty line to begin with or that benefits went to extremely poor households who, despite the assistance, remained below the poverty line. Chapter 6 in this Poverty Assessment finds most evidence for the first explanation. Figure 11 PSNP  AND HFA WERE HIGHLY EFFECTIVE IN AFAR Actual poverty rates in 2011 and 2016; simulated pre-benefit poverty rate 2016 Poverty rate - 2011 Poverty rate - 2016 Pre-bene t poverty rate - 2016 40 36.1 34.2 32.8 30 23.6 22.4 22.9 20 10 0 Afar Somali Note: The pre-benefit poverty rate is an estimate of what poverty would have been in 2016 in the absence of PSNP and HFA. Source: HCES; 2011, 2016. World Bank staff calculations. 12 The simulation subtracted 100 percent of PSNP and HFA benefits from the final consumption aggregate. Less extreme simulations (where only part of the benefits is subtracted-suggesting a marginal propensity to consume of lower than 1) would result in a lower effect of PSNP and aid. The marginal propensity to consume is set at 1 because PSNP and HFA are targeted to extremely poor households, who would likely consume a big part of any extra income. The reader should keep in mind that this simulation is only that: A simulation. 44 ETHIOPIA POVERTY ASSESSMENT Poverty decreased in all agro-ecological zones ex- consumption levels significantly increased between 2011 cept for the moisture-reliable lowlands. The strongest and 2016. This points towards a highly unequal pattern of poverty reduction took place in pastoral areas, where the growth in these regions, with the upper parts of the distribu- headcount decreased from 32 percent in 2011 to 18 percent tion growing fast while the poorest segments stagnate13. The by 2016 (Table 4). The moisture-reliable lowlands, which is drought-prone lowlands, which include the lowlands parts of mainly composed of Benishangul-Gumuz and Gambella, did Oromia and SNNPR and parts of Afar, remain the poorest not experience a reduction in poverty even though average despite a significant reduction in poverty. Table 4 POVERTY  DECREASED IN ALL AGRO-ECOLOGICAL ZONES EXCEPT THE MOISTURE-RELIABLE LOWLANDS FGT poverty indicators by agro-ecological zone, 2011 and 2016 HEAD COUNT DEPTH OF POVERTY SEVERITY OF POVERTY 2011 2016 2011 2016 2011 2016 Drought prone highland 28.0 20.8 6.1 5.1 2.1 2.0 Drought prone lowland 38.9 31.7 11.6 10.9 5.2 5.0 Moisture reliable highland 29.4 23.6 8.2 6.8 3.3 2.8 Moisture reliable lowland 24.6 25.4 6.0 5.9 2.1 2.0 Pastoralist 31.8 17.8 8.5 5.2 3.2 2.0 Source: HCES, 2011; 2016. World Bank staff calculations. 2.2 Consumption growth concentrated in the upper parts of the distribution The patterns of consumption growth between 2011 and 2016 were different at the bottom and at the top of the distribution. Figure 12 shows the average annual percentage change in consumption between 2011 and 2016 for each percentile of the distribution, ranging from the poor- est one percent to the richest one percent. Growth for the bottom 15 percent was not statistically different from zero, in contrast to the top of the distribution where growth rates reached a maximum of just under 6 percent per year be- tween 2011 and 2016. The overall average increased by 2.4 percent per year, while the median (50th percentile) grew at 2 percent per year. The absence of gains for the poorest segment of the population was driven by rural areas. The bottom 20 percent in rural areas did not experience an increase in con- sumption between 2011 and 2016 (Figure 13). While growth in rural areas was higher for the upper parts of the distribu- tion, annual growth rates did not exceed 3 percent, even for 13 The growth incidence curve for the moisture-reliable lowlands is strongly upward-sloping, with annual growth rates around 10 percent for the upper percentiles and zero growth for the bottom decile. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 45 the richest percentile. In contrast, growth across the urban the mean in urban areas was almost two-and-a-half times consumption distribution was always above 3 percent, even the national average, at 5.9 percent per year. The slight con- for the poorest, and became increasingly strong towards the traction in consumption for the poorest 20 percent in rural upper end (left-hand panel of Figure 13). The growth rate of areas explains the stagnation in rural poverty severity. Figure 12 CONSUMPTION  GROWTH WAS ZERO FOR THE POOREST, AND STRONGLY POSITIVE FOR THE RICHEST Average annual growth rate of consumption by percentile between 2011 and 2016 6 Annual mean growth rate (%) 0 1 2 −1 −2 3 4 5 0 10 20 30 40 50 60 70 80 90 100 Percentiles 95% confidence bounds Source: HCES 2011; 2016. World Bank staff calculations. Figure 13 GROWTH  WAS STRONG AND POSITIVE IN URBAN AREAS, BUT LOWER AND VARIABLE IN RURAL AREAS Average annual growth rates of consumption by percentile and urban/rural location between 2011 and 2016 Rural Urban 8 8 −2 −1 0 1 2 3 4 5 6 7 −2 −1 0 1 2 3 4 5 6 7 Annual mean growth rate (%) Annual mean growth rate (%) 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 Percentiles Percentiles 95% confidence bounds 95% confidence bounds Source: HCES 2011; 2016. World Bank staff calculations. 46 ETHIOPIA POVERTY ASSESSMENT The bottom 10 percent of the population has not ex- accounted for three percent of total national consumption, perienced real consumption growth since 2005. The slightly down from 3.5 percent in 2011. At the other side, the previous poverty assessment showed that consumption of wealthiest 10 percent accounted for 27 percent of total con- the bottom 15 percent contracted between 2005 and 2011, sumption, up from 24 percent in 2011. In other words, while both in rural and urban areas. The contraction in consump- the average consumption of the richest 10 percent was about tion for the bottom 15 percent continued in the more recent seven times larger than the average consumption of the poor- period (2011-2016), but only in rural areas. As a result, the est 10 percent in 2011, it was nine times bigger in 2016. severity of poverty in rural areas was higher in 2016 (3.1) than The different consumption dynamics at both tails of in 2005 (2.7)14. It is important to understand that, given the the distribution led to a modest increase in inequality. cross-sectional nature of the HCES data, this does not nec- The Gini coefficient, which is especially sensitive to changes essarily imply that households who were poor to begin with in the middle of the distribution, increased only slightly from (in 2005) have increasingly become more impoverished15. It 0.3 in 2011 to 0.33 in 2016. The Gini remains low in regional means that whoever was in the bottom 15 percent of rural comparison (Figure 15). Other indicators of inequality show welfare in 2016 had lower monetary living standards than a sharper increase: The ratio of consumption at the 90th whoever was in the bottom 15 percent in 2005. percentile and consumption at the 10th percentile increased from 3.6 in 2011 to 4.3 in 2016, reflecting strong consump- 2.3 A small increase in inequality tion growth for the richest and stagnating consumption for the poorest (Table 5). The Atkinson index of inequality, which The very poor experienced zero or negative consump- is more sensitive to changes at the bottom of the income tion growth between 2011 and 2016. As a result, the share distribution, increased from 0.25 in 2011 to 0.29 in 2016. of total consumption accruing to the bottom 10% of the Inequality remains however low in comparative perspective population decreased, while the share captured by the top (Figure 15). 10% increased (Figure 14). In 2016, the poorest 10 percent Figure 14 THE  CONSUMPTION SHARE OF THE POOREST DECLINED WHILE THAT OF THE RICHEST INCREASED BETWEEN 2011 AND 2016 Consumption share of the bottom and top 10% and 20% between 2011 -2016: national, urban and rural 0.100 0.500 Consumption share Consumption share 0.080 0.400 0.060 0.300 0.040 0.200 0.020 0.000 0.100 Bottom 10% Bottom 20% Bottom 10% Bottom 20% Bottom 10% Bottom 20% 0.000 Top 10% Top 20% Top 10% Top 20% Top 10% Top 20% National Urban Rural National Urban Rural 2011 2016 2011 2016 Source: HCES 2011; 2016. World Bank staff calculations. 14  FDRE, 2017 15 In fact, Chapter 4 of this Poverty Assessment will show that there is a substantial amount of mobility in Ethiopia, with poor house- holds exiting poverty and non-poor households falling back. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 47 Table 5 INEQUALITY  INCREASED BETWEEN 2011 AND 2016 2011 2016 Gini Index 0.30 0.33 Atkinson Index 0.25 0.29 p90/p10 ratio 3.62 4.32 Note: Poverty share denotes the contribution of the region to overall poverty. Source: HCES, 2011; 2016; World bank staff calculations. Figure 15 INEQUALITY  IN ETHIOPIA REMAINS LOW IN REGIONAL COMPARISON Gini coefficient and consumption share of bottom 40%, Ethiopia and comparators, latest data Gini coef cient Share of bottom 40% 0.60 0.54 0.50 0.45 0.43 0.40 0.38 0.38 0.33 0.35 0.30 0.20 0.19 0.19 0.20 0.18 0.16 0.16 0.12 0.10 0.00 Ethiopia Burkina Faso Tanzania Myanmar Uganda Rwanda Mozambique Source: HCES 2011; 2016. World Bank staff calculations. Figure 16 INEQUALITY  INCREASED DUE TO THE INCREASING GAP BETWEEN URBAN AND RURAL AREAS Decomposition of the Gini coefficient into a between rural-urban component and a within-component 20.1 15.0 Relative contribution to Overlap Gini (%) 55.9 65.0 Within Between 29.1 14.9 2011 2016 Source: HCES 2011; 2016. World Bank staff calculations. 48 ETHIOPIA POVERTY ASSESSMENT The increase in inequality was mainly driven by the poverty into a growth and redistribution component using the increasing disparity between urban and rural areas. Datt-Ravallion decomposition technique. It shows the parts Households in urban areas, who were already better-off in of the poverty change due to (i) the change in average con- 2011, experienced strong consumption growth between sumption and (ii) the change in the distribution of consump- 2011 and 2016, while households in rural areas experi- tion (changes in inequality). At country level, the growth effect enced fairly weak consumption growth. As a result, the “be- is negative (poverty-reducing) while the redistribution effect tween-share” of inequality -the part of inequality that is due is positive (poverty-increasing) offsetting part of the pover- to differences in average welfare levels between urban and ty reduction due to growth in consumption. If consumption rural areas- increased from 15 percent in 2011 to 29 percent growth would have been distribution-neutral (if inequality had in 2016 (Figure 16). Inequality across regions remained low not changed between 2011 and 2016), poverty would have in 2016, with the between-share accounting for a mere two decreased by over seven percentage points rather than the percent of overall inequality in 2016. observed six percentage points. The increase in inequality however partly offset the reduction in poverty, leading to a Given the increase in inequality, all poverty reduction six percentage-point poverty reduction. A similar pattern is between 2011 and 2016 was driven by growth in house- observed in rural and urban areas, though both the growth hold consumption. Figure 17 decomposes the change in and redistribution components were larger in urban areas. Figure 17 CONSUMPTION  GROWTH DROVE POVERTY REDUCTION Datt-Ravallion decomposition of poverty change between 2011 and 2016 2 1.2 1 0 -1 -2 -3 -4 -5 -6 -7 -6 -8 -7.3 Poverty reduction Consumption growth Increase in inequality Source: HCES 2011; 2016. World Bank staff calculations. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 49 2.4 No clear impact of the El-Niño and/or vegetation anomalies on household consumption17. Similalry, Hirvonen and others (2018) found that the drought drought did not lead to a widespread increase in child undernutrition Despite the severity of the 2015/16 El-Nino drought, it in the country, but that there was an adverse impact in areas is hard to identify its aggregate effects in national sta- with a limited road network. Bachewe and others (2017) found tistics. In 2015/16, Ethiopia was hit by the El-Nino drought, that real cereal prices decreased during the drought, consis- labeled the worst drought in five decades. Failure of two con- tent with a story of limited agricultural impacts of the drought. secutive rainy seasons in 2015 led to a sharp increase in hu- Chronic malnutrition decreased between 2011 and 2016, manitarian requirements with more than 10 million Ethiopians acute malnutrition remained unchanged, and overall ag- in need of humanitarian food aid, on top of the chronically ricultural production in 2015/16 decreased only margin- food-insecure PSNP caseload of eight million16. Government ally (and remained higher than overall production levels and development partners mounted a large-scale humani- two years earlier in 2013/14)18. The food security indicators tarian response, which was credited with having averted in the Welfare Monitoring Survey (WMS) tell a largely similar any loss of life due to starvation. Available evidences fail to story: The share of the Ethiopian population who experienced identify the effect of the drought on households’ welfare as a food shortage in the 12 months prior to the survey decreased discussed below. from 22 percent in 2011 to 10 percent in 2016 (Table 6). The Using data on a sample of Ethiopian households ob- food gap -the number of months a household experienced served before (2014) and after/during the drought food shortages – remained the same, but given that the gap (2016), researchers fail to find a clear negative impact only refers to those households who actually experienced food of the drought on household welfare (measured by con- shortages, it also decreased on the aggregate level. On the sumption). Using drought indicators based on remote sens- regional level, only Benishangul-Gumuz experienced a self-re- ing data, Sohnesen (2018) does not find an impact of rainfall ported increase in food insecurity from a low base. Table 6 FOOD  SECURITY IMPROVED BETWEEN 2011 AND 2016 Incidence of food shortages and average duration of food shortages – food gap 2011 2016 FOOD SHORTAGE FOOD GAP FOOD SHORTAGE FOOD GAP (%) (MONTHS) (%) (MONTHS) Tigray 13.2 3 11.9 2.5 Afar 7.7 5.2 9 3.8 Amhara 23.2 3.1 10.4 3 Oromia 16 3.1 10.5 3.6 Somali 30.3 4.4 6 3 Benishangul-Gumuz 5.6 2.1 8.5 2.8 SNNPR 35 3.4 12.6 3.2 Gambella 31.6 2.6 3.8 1.2 Harari 8 3.2 0 - Addis Ababa 7.8 4 1.1 3.6 Dire Dawa 13.5 1.6 7.7 2 National 21.6 3.3 10.2 3.3 Source: WMS, 2011; 2016. World Bank staff calculations. The food gap is only calculated for those households who reported a food shortage. 16  Based on the 2016 Humanitarian Requirements Document 17 When using a self-reported indicator of drought exposure, there is a large negative impact of (self-reported) drought exposure on con- sumption. This is likely due to endogeneity: People who have had a bad year are more likely to report having been exposed to shocks. 18  Based on the Ethiopian Demographic and Health Surveys (2011 and 2016) and the Agricultural Sample Surveys (2011-2016). 50 ETHIOPIA POVERTY ASSESSMENT Poverty rates decreased even in those areas that ex- to 30.9 percent for those in the first quartile - woredas that perienced the most severe rainfall shocks. An analysis experienced positive rainfall shocks in 2015. It is noteworthy of poverty trends by quartiles of woreda level average rainfall that the areas that experienced the highest rainfall shocks shocks during the months of the main rainy season (June, during the El-Nino drought are mostly found in central Ethi- July, August and September) in 2015 shows that rural pov- opia and are not those that are dry and considered to be erty decreased in all the quartiles (Table 7)19. For the wore- more vulnerable for rainfall shocks (See map in Annex I). das that are in the fourth quartile – those that experienced Most of the woredas in Somali region are in the first quartile the highest rainfall shock in 2015, rural poverty fell from 22.1 and experienced positive rainfall shocks in 2015 (more rain percent to 18.8 percent while it dropped from 34.4 percent than usual). Table 7 POVERTY  DECREASED EVEN FOR THE AREAS THAT EXPERIENCED THE HIGHEST RAINFALL SHOCK Percentage of poor people in 2011 and 2016 by quartiles of 2015 rain fall shocks QUARTILE OF RAINFALL SHOCKS 2011 2016 First (lowest shock) 34.4 30.9 Second 36.4 28.4 Third 29.5 23.7 Fourth (highest shock) 22.1 18.8 Source: HCES, 2011; 2016; World bank staff calculations. Several explanations are possible for the apparent humanitarian food aid drastically expanded its coverage and non-effects of such a major drought. First, the drought appears to have been well targeted to the most drought-af- mainly affected areas that contribute little to overall agricul- fected areas (see Chapter 6), likely cushioning the effect of the tural production, resulting in only a marginal decline of na- drought on consumption levels. Finally, it is likely that the avail- tional production levels. Second, thanks to widespread land able national household survey data are not granular enough management and restoration practices, agriculture may have to detect localized impacts of the drought: It is likely that become more resilient to droughts: While 2015/16 was clearly the drought had large negative effects confined to relatively a meteorological drought with rainfall levels far below average small geographical units that are not manifested in large scale (Figure 42), vegetation conditions were actually above aver- household survey datasets. Thus, while further study may be age during the drought – the Ethiopian landscape was green- required, one of the above reasons or a combination of them er than average (right-hand-side graph in Figure 18). Third, might have led to an overall muted effect of the drought. 19 To construct the rainfall shocks at woreda level, deviations in rainfall from the long-run average (2000-2016) were first calculated for the months of June, July, August and September in 2015 separately and expressed as z-scores. The average z-score for the four months was then constructed and used to divide the woredas into four quartiles. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 51 Figure 18 BAD  RAINS, BUT GOOD VEGETATION… Rainfall deviation from long-term averages Vegetation Condition Index – VCI 100 4 Vegetation anomalies growing season 80 2 Rain anomalies 60 0 40 −2 20 −4 0 00 02 04 06 08 10 12 14 16 00 02 04 06 08 10 12 14 16 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 Note: z-scores below 0 mean negative rainfall deviations (less rain than normal). VCI measures the state of vegetation in a given year compared to similar periods in the previous years. Source: Sohnesen, 2018, based on data from NASA’s National Oceanic and Atmospheric Administration. 52 ETHIOPIA POVERTY ASSESSMENT 2.5 Testing the sensitivity of the official (ETB7,436 versus ETB7,184 – per adult equivalent per year) – we call this Alternative 1. Second, following the official poverty estimates practice prior to 2011, we use the CPI deflator to express Poverty measures can be sensitive to the way the pov- both 2011 and 2016 consumption in 1996 prices and then erty line is adjusted over time, particularly during times apply the original poverty line of ETB 1,075 – this is termed as of inflation. In Ethiopia, the official poverty line was set in Alternative 2. The results are discussed below. 1996 based on a food basket that is enough to get the min- The national poverty rate decreased between 2011 and imum daily calorie requirement for a healthy life (see Box 1). 2016 under the two alternative methods also, though Different adjustment techniques have then been used to ac- the magnitude becomes weaker, particularly in rural count for price changes over time. The approach followed be- areas. Under Alternative 1, poverty at national level de- fore 2011 was to express consumption values in 1996 prices creased from 29.6 percent to 25.1 percent – compared to using the CPI and then use the original 1996 poverty line. But 23.5 percent according to the official poverty line. In Alterna- in 2011, the original basket was re-costed using current pric- tive 2, poverty decreased from 23.4 percent in 2011 to 20.5 es of each of the items in the original food basket. In 2016, the percent in 2016. The low poverty rate in 2011 in case the 2011 poverty line was adjusted to 2016 prices by using the CPI deflator is used was also noted in the previous poverty GDP deflator. The choice of the GDP deflator instead of CPI assessment. In both alternative scenarios, poverty reduc- is not clear and is a departure from past practices. Annex I tion is strong in urban areas and weak in rural areas (Figure discusses in more detail Ethiopia’s poverty monitoring system 20). Poverty severity remains unchanged at the national level and possible ways to further strengthen it. under the alternative scenarios. For urban areas, depth and To assess whether the poverty trends between 2011 severity of poverty fell under the two methods. In rural areas, and 2016 are sensitive to the deflator used, we apply they stayed the same under Alternative 1 and slightly rose in two alternative methods. Frist, we apply the CPI instead Alternative 2 (Figure 19). Regional patterns of poverty reduc- of the GDP deflator to bring the 2011 official poverty line to tion remain largely the same under the alternative deflating 2016 prices. This gives a slightly higher poverty line than the scenarios (Annex Figure 2). Figure 19 SENSITIVITY  OF POVERTY MEASURES TO DIFFERENT POVERTY LINE DEFLATORS: URBAN AND RURAL Sensitivity of poverty measures to different poverty line deflators: urban and rural 2011 2016 40.0 Percentage of poor 30.0 20.0 10.0 0.0 Head count Depth Head count Depth Head count Depth Head count Depth Head count Depth Head count Depth Severity Severity Severity Severity Severity Severity National Urban Rural National Urban Rural Case 1 Case 2 (2011 PL - of cial, (1996 PL brought to 2011 and 2016 using CPI ) 2016 PL - 2011 PL adjusted uisng CPI) Source: HCES 2011; 2016. World Bank staff calculations. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 53 3. GAINS IN NON-MONETARY DIMENSIONS OF WELFARE AS WELL Trends in non-monetary indicators of living standards largely confirm the reduction in poverty. Between 2011 and 2016, ownership of durables increased, housing quality modestly improved, and more people gained access to an improved water source. Human development indicators improved as well though most remain low. In 2016, only one in three youth in the 15-24 age-cohort had completed primary education. The growth in household consumption between 2011 to electricity (Figure 20). In 2016, 61 percent of Ethiopian and 2016 was accompanied by improvements in households lived in a house with an improved roof20, up from non-monetary dimensions of welfare. The share of 47 percent in 2011. Having improved walls (cement, stone households with a television, mobile phone, refrigerator, an- with cement, bricks, cement blocks, covered adobe, wood imal cart, and motorcycle increased, often from a low base, planks/shingles) remains however rare. Access to electricity as did the use of improved housing materials and access increased from 23 percent in 2011 to 26 percent in 2016. Figure 20 ASSET  HOLDINGS AND LIVING CONDITIONS IMPROVED BETWEEN 2011 AND 2016 Selected household characteristics in 2011 and 2016, % of households with asset/characteristic A. Household Durables B. Housing/Energy 70 70 70 70 70 70 70 70 60 60 60 60 60 60 2016 2016 2011 2011 60 60 2016 2016 2011 2011 50 50 2016 2016 2011 2011 50 50 2016 2016 2011 2011 50 50 50 50 Percent Percent Percent Percent 40 40 40 40 Percent Percent Percent Percent 40 40 40 40 30 30 30 30 30 30 30 30 20 20 20 20 20 20 20 20 10 10 10 10 10 10 10 10 0 0 0 0 0 0 0 0 ile ile e ne ge gene ne rigfrig or or cra r im im rt rt or or rt rt ot ot le le e e ov ov ll ll ov ov ll ll oooo vrov of of s/ of of ce ce s/ s/ nt nt ce ce el el nt nt tri tri eleele ty ty tri tri r or ity ity r cr c g g ng ng M MelTel on on n n atlo to pr pr wawa pr pr wawa cl cl fo fo kinkin ot ot ca ca le io io M M yc yc ec ec o o icirici o fri friho ho ReRe ratrat p ro ro il ro ro AcAc tiletile eme AcAc to to meme f c c ki ki r ra si si cy cy obob on h bi isvis ty ri ri c c AnAn l ca ty ty o o oooo M M php ReRe p p MM l l vi vi Im Im ed ed Im Im ed ed nie e Im Im ed ed r: r: ed ed t ci ct ct a ce ss ss ce ce or or aa El El s t s t ctrc ci ci co co r: r: cem le le e m s s e e ov ov a v Te Te oboe im ile oooo /e pr pr AnfA Fl Fl pro Fl Fl tilets T Im Im ec ec fo El El city C. Sanitation D. Water 100 100 100 100 100 100 100 100 80 80 38.3 38.3 32.3 32.3 80 80 35.3 35.3 80 80 38.3 38.3 32.3 32.3 80 80 46.3 46.3 35.3 35.3 46.3 46.3 Percent Percent Percent Percent 60 60 60 60 11.7 11.7 Percent Percent Percent Percent 60 60 60 60 11.7 11.7 9 9 40 40 43.5 43.5 52.2 52.2 40 40 9 10.19 17 17 10.1 17 17 40 40 43.5 43.5 52.2 52.2 40 40 10.1 10.1 23.3 23.3 21.1 21.1 20 20 4.6 4.6 20 20 21.1 21.1 20 20 1 1 20 20 23.3 23.3 4.6 4.6 10.5 10.5 1 10.8 10.8 1 14.3 14.3 2.8 10.5 2.8 2.8 10.8 10.8 2.8 11.1 11.1 0 0 10.5 0 0 11.1 11.1 14.3 14.3 0 2.8 0 2011 2011 2.8 2.8 2.8 0 0 2016 2016 2011 2011 2016 2016 2011 2011 2016 2016 2011 2011 2016 2016 toilet toilet Flush Flush Improved Improved pit latrine pit latrine PipedPiped waterwater - compound - compound waterwater PipedPiped - outside - outside compound compound Flush toilet Flush Composting Composting toilet toilet toilet Improved pit pit Improved Unimproved latrine pit Unimproved latrine pit latrine latrine Piped water Piped Protected - water Protectedcompound - compound well/tubewell PipedPiped water well/tubewell Protected - outside water Protected spring - outside spring compound compound Composting Composting OtherOther toilet unimproved toilet unimproved Unimproved No toilet toilet toilet Unimproved No toilet pit facility latrine pit facility latrine Protected Protected improved OtherOther well/tubewell well/tubewell improved Protected Protected Unimproved spring spring sources Unimproved sources unimproved OtherOther unimproved No toilet toilet toilet facility No toilet facility OtherOther improved improved Unimproved Unimproved sources sources Source: DHS, 2011; 2016 20  An improved roof is defined as a finished roof: Corrugated iron, wood, cement fiber, ceramic tiles, cement, roofing shingles. 54 ETHIOPIA POVERTY ASSESSMENT Access to improved water improved as well. In 2016, from a low base, the share of fully immunized children in- 35 percent of Ethiopians used an unimproved water source, creased by 14 percentage points, and stunting rates de- down from 46 percent in 2011. The type of sanitation used creased from 44 percent in 2011 to 38 percent in 2016 (Panel changed too, though unimproved toilet facilities remain the B of Figure 21)21. Infant and child mortality rates decreased norm (the use of improved pit latrines remained unchanged accordingly (Panel A of Figure 21). Net enrolment in prima- at about 11 percent). A third of households still do not use ry school increased, more children are completing primary any toilet facility (Panel C of Figure 20). school, and gross enrolment in secondary school was higher in 2016 than in 2011 (Panel C of Figure 21). Human development indicators followed the overall positive trend. Delivery in a health facility sharply increased Figure 21 CHILDREN’S  MORTALITY DECREASED, AND THEIR HEALTH IMPROVED BETWEEN 2011 AND 2016 Selected health and education variables for children in 2011 and 2016 A. Child and infant mortality rates mmunization, health facility delivery and stunting B. I 100 100 100 100 90 90 80 80 80 80 1000 per1000 70 70 60 60 60 60 deathsper Percent Percent 50 50 44.4 44.4 ofdeaths 40 40 38.5 38.5 38.4 38.4 40 40 30 30 26.2 26.2 24.3 24.3 #of 20 20 20 20 # 9.9 9.9 10 10 0 0 0 0 Infant Infant Child Child Under ve Under ve Health facility Health facility Fully Fully Stunted Stunted mortality mortality mortality mortality mortality rate mortality rate delivery delivery immunized immunized children children rate rate rate rate children children 2011 2011 2016 2016 2011 2011 2016 2016 C. Education 80 80 71.8 71.8 70 70 62.4 62.4 60 60 50 50 Percent Percent 40 40 32.7 32.7 30.7 30.7 27.6 27.6 30 30 21.5 21.5 20 20 10 10 0 0 primary Primary Net primary Net school Primary school Gross Gross school school completion completion secondary secondary enrolment enrolment (15-24) (15-24) school school enrolment enrolment 2011 2011 2016 2016 Source: DHS, 2011; 2016 21 Children are fully immunized if they received all eight basic vaccinations: BCG, three doses of Polio, three doses of DPT, and one dose of MCV. This is calculated for the sample of children aged between 12 and 23 months. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 55 Despite the sharp improvements, human development To summarize, the evolution of asset and human cap- indicators in Ethiopia remained low. In 2016, only 26 ital indicators corroborate the poverty and consump- percent of births happened in a health facility (in the five tion dynamics set out earlier. Between 2011 and 2016, years preceding the survey) and less than 40 percent of chil- household consumption levels increased, asset holdings dren had received all basic vaccinations. Only one in three increased, and physical housing conditions modestly im- children 15-24-years-old had completed primary school, proved. Indicators of child health improved substantially and and over 25 percent of age-eligible children were not in access to education broadened from a low base. school. As a result, Ethiopia ranks relatively low on the Hu- man Capital Index (Box 4). Box 4 Human Capital in Ethiopia During the October 2018 Annual Meetings, the World Bank launched the Human Capital Index. The Human Capital Index (HCI) is designed to capture the amount of human capital a child born today could expect to attain by age 18. The HCI has three components: (i) Survival, measured by the under-five mortality rate; (ii) Expected years of learn- ing-adjusted school, measured by the quantity of education a child can expect to attain by age 18, corrected by a measure of learning quality-proxied by student achievement tests; and (iii) Health, measured by the stunting rate of children under five and the probability of a 15-year-old surviving until age 60. The health and education components of the index are combined in a way that reflects their contribution to worker productivity, based on evidence from rigorous micro-econometric empirical studies. The resulting index ranges between 0 and 1. A country in which a child born today can expect to achieve both full health (no stunting and 100 percent adult survival) and full education potential (14 years of high-quality school by age 18) will score a value of 1 on the index. Therefore, a score of, for instance, 0.5 signals that the productivity as a future worker for a child born today is 50 percent below what could have been achieved with complete education and full health. Given the strong correlation between per capita GDP and human capital, low-income countries tend to score poorly on the HCI. Ethiopia is no exception. With a HCI of 0.38, Ethiopian children born today can expect, as future workers, to attain 38 percent of their potential productivity. With a score of 0.38, Ethiopia ranks 135th out of 157 countries. Relative to the comparator countries, Ethiopia scores at par with Uganda, better than Mozambique (0.36) and Rwanda (0.37), and worse than Tanzania (0.40) and Myanmar (0.47). Relative to its overall rank (135), Ethiopia scores lower on learning-adjusted years of school (4.5 years) and share of children not stunted (62 percent). Ethio- pia however overperforms relative to its income level: Given GDP per capita, the human capital index in Ethiopia is higher what would be expected, reflecting the Government’s large investments in the health and education sectors. Source: World Bank, 2018. 56 ETHIOPIA POVERTY ASSESSMENT 4. HOW DID THE EXTREME POOR FARE? Though the extreme poor, here defined as the poorest 10 percent of the population, did not experience any real growth in consumption, they have experienced some improvements in non-monetary welfare. Health and education indicators for children from extremely poor households improved from a low base, household calorie intake increased and the share of extremely-poor households that experienced food shortages de- creased. The extreme poor are lagging on the fertility transition, with total fertility rates that were marginally higher in 2016 than in 2000. Children from extremely poor households face substantial barriers in school, with extremely poor 14-year-olds being more likely to be attending Grade 2 rather than Grade 8, which would be his/her age-eligible grade. As mentioned in Section 2, the bottom 10 percent of Living conditions of the bottom 10 percent improved the population did not experience any growth in con- from a low base between 2011 and 2016. Figure 22 sum- sumption between 2005 and 2016. A logical follow-up marizes the evolution of living conditions of the bottom 10 question is then whether or not the extreme poor have been percent along four dimensions: Housing, water, health, and stagnating on other dimensions of welfare as well. This sec- education. The share of the extreme poor (here defined as tion will explore this question based on data from the WMS the bottom 10 percent) living in a house with an improved and DHS surveys. roof increased sharply from a low base, from 1 percent in 2011 to 10 percent in 2016, while access to an improved water source22 increased from 16 percent to 26 percent over Figure 22 LIVING  CONDITIONS OF THE BOTTOM 10 PERCENT IMPROVED BETWEEN 2011 AND 2016 Trends in selected indicators from the bottom 10 percent, 2011 and 2016 2011 2016 100 90 80 70 60 50 % 40 30 20 10 0 Children Improved Improved Net primary Primary fully water roof school school immunized source material enrolment completion (%) (%) (%) (%) (15-24, %) Source: WMS; 2011, 2016; DHS, 2011, 2016. World Bank staff calculations. 22 Following the Joint Monitoring Programme’s definition, improved water is defined as piped water (both in dwelling and public tap), protected well, protected sprig, tube well, borehole, and rainwater. No distance criterion is applied. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 57 the same period. The share of children in extremely poor sharply increased from a low base, daily per adult equiva- households that received all basic vaccinations increased lent calorie intake increased, and the incidence of self-re- from 15 percent in 2011 to 24 percent in 2016, while net pri- ported food shortages sharply decreased. Other indicators mary school enrolment reached 66 percent in 2016, up from remained unchanged: The share of extremely poor children 56 percent in 2011. Completion of primary school remained stunted or wasted remained stable at a high level, as did unchanged at about 17 percent of the 15-24 age-cohort, household size and indicators of fertility (Table 8). In general, which means that while there are more children from ex- the extreme poor are lagging on Ethiopia’s fertility transition. tremely poor households in school, they are not completing While the Total Fertility Rate (TFR)23 decreased substantially primary school at higher rates than before. in the third, fourth, and fifth wealth quintiles, TFR decreased only modestly in the second quintile (a decrease of 0.6 in 16 Looking at trends in other indicators shows a mixed years) and did not change at all in the bottom quintile (TFR picture of progress and stagnation. The share of ex- of 6.4 - Figure 23). tremely poor children that were born in a health facility Shouldn’t this be Table 8? or needs to re-number? 23 The TFR is the average number of children that would be born to a woman over her lifetime if she was to experience the exact current age-specific fertility rates through her lifetime. 58 ETHIOPIA POVERTY ASSESSMENT Table 8 A  MIXED PICTURE OF PROGRESS FOR THE EXTREME POOR Means of selected variables for the bottom 10 percent, 2011 and 2016 MEAN 2011 2016 DIFFERENCE Household head literate (%) 32.6 35.7 3.1 Household size 7.3 7.2 -0.1 Dependency ratio 1.38 1.41 0.03 Cumulative fertility 3.6 3.7 0.1 Births in health facility (%) 2.7 7 4.3 Children stunted (%) 47.1 45.2 -1.9 Children wasted (%) 13.9 14.4 0.5 Average annual household expenditures per AE (2015 Birr) 3,827 3,762 -65 Daily calorie intake per AE 1,633 1,777 144 Food shortage (%) 31 20 -11 Source: WMS, 2011; 2016. World Bank staff calculations. The food gap is only calculated for those households who reported a food shortage. Differences in bold are statistically significant. Figure 23 FERTILITY  RATES AMONG THE EXTREME POOR ARE NOT DECREASING Total fertility rate by quintile, 2000 and 2016 2000 2016 7 6 5 4 TFR 3 2 1 0 Q1 Q2 Q3 Q4 Q5 Source: DHS, 2000, 2016. World Bank staff calculations. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 59 Despite improvements, the extent to which public ser- Access to education for the extreme poor is crucial but vices reach the extreme poor remains low. In 2016, a remains beset by challenges. In an ideal scenario, children mere one in four extremely poor children had received all of extremely poor households would accumulate more edu- basic vaccinations, about one in four had access to an im- cation and be able to move out and diversify into more pro- proved water source, and only seven percent of children were ductive activities, breaking the intergenerational transmission born in a health facility (in the five years preceding 2016). of poverty. While access to education has improved across Only 17 percent of the 15-24 age cohort had completed pri- the board, close to 17 percent of 14 to 16-year-old children mary school (in the bottom 10 percent). While these figures from extremely poor households never see the inside of a represent real progress from 2011 and earlier, they highlight classroom. For those who are in school, grade progression the enormous efforts that lie ahead in making access to ba- tends to be difficult: A 14-year-old from an extremely poor sic public services less dependent on location and wealth household is most likely to be either in grade 3 or grade 6, (Chapter 7 will expand on this). and is more likely to be in grade 2 than in grade 8, the grade a 14-year-old would attend in case of perfect progression (Figure 24). In the overall population, 14-year-olds are most likely to be attending grades 6, 7, or 8. Figure 24 GRADE  PROGRESSION FOR BOTTOM 10% CHILDREN IS DIFFICULT Probability of a 14-year-old to attend any of the grades of primary school, 2016 Bottom 10% Overall population 20 18 16 14 12 % 10 8 6 4 2 0 Grade 1 Grade 2 Grade 3 Grade 4 Grade 5 Grade 6 Grade 7 Grade 8 Source: WMS, 2016. World Bank staff calculations. 60 ETHIOPIA POVERTY ASSESSMENT 5. CONCLUSIONS Despite adverse climatic circumstances, the reduction Human development indicators sharply improved from a low in poverty continued between 2011 and 2016. In contrast base. Despite the improvements in human development in- to the earlier periods, poverty reduction was especially strong dicators, access to key health and education services re- in urban areas, while rural areas experienced fairly weak pov- mained low in 2016. erty reduction. The increasing disparity in household welfare One of the main finings in this chapter is the contin- between rural and urban areas pushed inequality up, though ued stagnation of the bottom 10 percent. Between 2005 inequality remains low in regional comparison. Within rural and 2016, consumption levels of the poorest 10 percent did and urban areas, the better-off generally experienced faster not increase, leading to levels of rural poverty severity that consumption growth. were higher in 2016 than in 2005. Non-monetary welfare out- Trends in non-monetary dimensions of welfare con- comes of the bottom 10 percent did improve, though mod- firm the improvement in household living standards. estly, and human development indicators for this segment Ownership of durables increased, housing conditions mod- remain exceptionally low. The next chapter will look more in estly improved, as did access to an improved water source. detail at this bottom segment of the Ethiopian population. CHAPTER I. 2011-2016: CONTINUED GROWTH AND POVERTY REDUCTION 61 62 ETHIOPIA POVERTY ASSESSMENT CHAPTER II Who Are the Poor? Ethiopia has a traditional poverty profile. The poor tend to live in rural areas, in large households with high dependency rates, headed by an older and little-educated household head. They mainly engage in agriculture and casual labor for their livelihood. The poor are relatively isolated from key infrastructure and have worse access to services. In contrast to conventional wisdom, poor households are less likely to be headed by a woman than non-poor households. Regional disparities in poverty are relatively low, meaning that the regional distribution of poverty largely reflects the regional distribution of the population. Differences in poverty rates are higher across agro- ecological zones, with the lowland parts of Oromia and SNNPR having the highest poverty rates. The ultra-poor, defined as the poorest 10 percent of the population, have similar characteristics as the poor, but are more likely to reside in rural Somali and SNNPR. Education of the household head displays the strongest correlation with household consumption levels. Returns to education have increased between 2011 and 2016, both in urban and rural areas. In rural areas, completing primary school is associated with 21 percent more consumption, though only few rural Ethiopians attain that much education. In urban areas, returns to post-secondary education increased by over 40 percent. Occupation is the second main correlate of consumption, with households that engage in non-farm self-employment having the highest consumption levels. In contrast to common perception, the pastoral areas are not the poorest in the country. Controlling for other factors, households in pastoral regions had significantly higher consumption levels in 2016. The pastoral regions are however lagging on virtually all non-monetary human development indicators, reflecting the difficulty of providing basic services to sparsely populated areas with mobile populations. CHAPTER II. WHO ARE THE POOR? 63 Introduction The previous chapter presented the poverty and wel- mainly focus on 2016, though comparisons with 2011 will fare trends between 2011 and 2016. The aim of this also be made. The next section summarizes the characteris- chapter is to provide a poverty profile by comparing the char- tics of the poor focusing on demographic, geographic, and acteristics of the poor and the non-poor and presenting the socioeconomic characteristics. Section 3 estimates the main correlates of consumption. The analysis in this chapter will correlates of consumption, while the final section concludes. 2. THE POVERTY PROFILE Ethiopia has a traditional poverty profile. The poor tend to live in rural areas, in large households with high dependency rates, headed by an older and little-educated household head. They mainly engage in agricul- ture and casual labor for their livelihood. The poor are relatively isolated from key infrastructure and have worse access to services. In contrast to conventional wisdom, poor households are less likely to be headed by a woman than non-poor households. Regional disparities in poverty are relatively low, meaning that the regional distribution of poverty largely reflects the regional distribution of the population. Differences in pov- erty rates are higher across agro-ecological zones, with the lowland parts of Oromia and SNNPR having the highest poverty rates. 2.1 Poverty increasingly increased by two percentage points (Figure 25). This pattern reflects the stronger poverty reduction in urban as opposed concentrated in rural areas to rural areas. In 2016, close to 90 percent of the poor lived Between 2011 and 2016, poverty became more con- in rural areas, compared to a rural population share of 80 centrated in rural areas. While the rural population share percent. Future poverty reduction will need to happen mainly decreased by three percentage points, from 83 percent through improvements in rural areas and through increased in 2011 to 80 percent in 2016, the rural share of poverty mobility to urban areas. Figure 25 THE  POOR BECAME MORE CONCENTRATED IN RURAL AREAS Population versus poverty share in 2011 and 2016: urban and rural Urban Rural 85.6 83.4 87.9 80.5 14.4 16.6 12.1 19.5 Poverty share Population share Poverty share Population share 2011 2016 Source: HCES, 2011; 2016. World Bank staff calculations. 64 ETHIOPIA POVERTY ASSESSMENT In contrast to many countries, there is no strong re- shares of regions largely reflect their population shares, with gional concentration of poverty in Ethiopia. Poverty the most populated regions accounting for the bulk of the rates among the largely rural regions vary from 21 percent poor. Oromia region accounts for 38 percent of the poor in in SNNPR to 27 percent in Tigray.24 Poverty in the two city Ethiopia, Amhara for 26 percent and SNNPR for 18 percent. administrations (Addis Ababa and Dire Dawa) and the largely In contrast to conventional wisdom, the pastoral regions of urban region of Harari is significantly lower, at 15 percent in Afar and Somali are not poorer than the average: Poverty Dire Dawa, 17 percent in Addis Ababa, and seven percent in rates in 2016 amounted to 22 percent in Somali and 24 per- Harari (Table 9). Given small regional disparities, the poverty cent in Afar.25 Table 9 POVERTY  RATES, POVERTY SHARES, AND POPULATION SHARES BY REGION AND AGRO-ECOLOGICAL ZONE, 2016 POVERTY RATE POVERTY SHARE POPULATION SHARE BY REGION Tigray 27.0% 6.6% 5.8% Afar 23.6% 1.9% 1.9% Amhara 26.1% 25.5% 23.0% Oromia 23.9% 38.3% 37.8% Somali 22.4% 5.5% 5.8% Benishangul Gumuz 26.5% 1.3% 1.1% SNNPR 20.7% 17.5% 19.9% Gambella 23.1% 0.4% 0.4% Harari 7.1% 0.1% 0.3% Addis Ababa 16.8% 2.6% 3.6% Dire Dawa 15.4% 0.3% 0.5% BY AGRO-ECOLOGICAL ZONE Moisture-reliable highlands 23.6% 58.5% 58.4% Drought-prone highlands 20.8% 19.9% 22.5% Moisture-reliable lowlands 25.4% 4.7% 4.3% Drought-prone lowlands 31.7% 7.5% 4.7% Pastoral areas 21.9% 6.9% 7.4% Note: Poverty share denotes the contribution of the region to overall poverty. Source: HCES, WMS, 2016; World bank staff calculations. Different agro-ecological zones have fairly similar The moisture-reliable highlands account for the bulk of the poverty rates, except for the drought-prone lowlands. poor (close to 60 percent), not because they are particularly The drought-prone lowlands, which include the eastern and poor but because the population is concentrated in these southern parts of Oromia and the southern parts of SNN- highlands. PR (but do not include pastoral areas of Afar and Somali), had the highest poverty rate in 2016, at 32 percent. The 2.2 The poor live in large households drought-prone highlands, which include the eastern parts of Tigray and Amhara, had the lowest poverty rates (21 per- with high dependency rates cent). Pastoral areas had relatively low monetary poverty Both in rural and urban areas, poor households tend rates (in contrast, access to services and human develop- to be larger and have more children and higher depen- ment outcomes tend to be much worse in pastoral areas). dency rates. The average poor household contains about 24  The largely rural regions are Tigray, Afar, Amhara, Oromia, Somali, Benishangul-Gumuz, SNNPR, and Gambella. 25  The results for Afar and Somali include households sampled from the new zones. CHAPTER II. WHO ARE THE POOR? 65 1.5 more members than non-poor households, both in urban areas26. In rural areas, the average poor household contains and rural areas (Table 10). The larger household sizes for the 1.4 dependents for every working-age adult, compared to poor are mainly driven by a larger number of children (defined 1.1 for non-poor households. The strain on household re- as less than 15-years-old). As a result, dependency ratios sources is significantly higher for poor households. are far higher for poor households, both in urban and rural Table 10 POOR  HOUSEHOLDS TEND TO BE LARGER, WITH MORE CHILDREN AND WITH MORE ILLITERATE MEMBERS Demographic and socio-economic characteristics by poverty status: urban and rural URBAN RURAL NON-POOR POOR NON-POOR POOR Household size 3.5 5.2 4.6 6.1 Number of children 1.1 2.0 2.1 3.1 Dependency ratio 0.55 0.92 1.06 1.40 Note: * Out of those who are 21 and above. ** Out of those who are 10 and above Source: HCES, WMS, 2016; World bank staff calculations. Dependency ratios are strongly correlated with pov- While dependency rates have decreased since 2011, erty. Households with less than 0.5 dependents per work- the poorest are lagging. The average dependency ratio ing-age adult have an average poverty rate of 16 percent, decreased from 1.1 to 1.0 between 2011 and 2016, mainly while households with 2 or more dependents for working-age driven by significant decreases in the upper quintiles (Figure adult have poverty rates in excess of 30 percent (Figure 26). 27). The dependency rate in the bottom quintile remained 20 percent of the poor in Ethiopia live in households with constant at 1.5 dependents per working-age adult. This dependency rates higher than 2. reflects the persistently high fertility rates for women in the Figure 26 HOUSEHOLDS  WITH HIGHER DEPENDENCY RATES ARE POORER Poverty rate by dependency ratio, 2016 35 31.3 30 25.7 25 22.7 20.1 20 15.6 15 10 5 0 0-0.5 0.5-1 1-1.5 1.5-2 2+ Dependency rate Source: WMS, HCES, 2016. World Bank staff calculations. 26 The dependency ratio is defined as the number of dependent children (younger than 15 years) and elderly persons (65 years or over) divided by the number of working-age adults (15-64 years of age). The dependency ratio indicates the number of dependents for each working-age adult in the household. 66 ETHIOPIA POVERTY ASSESSMENT Figure 27 DEPENDENCY  RATIOS ARE STUCK AT A HIGH LEVEL IN THE BOTTOM QUINTILE Dependency ratio by consumption quintile, 2011 and 2016 Q1 Q2 Q3 Q4 Q5 1.60 1.47 1.51 1.40 1.31 1.40 Dependency ratio 1.20 1.21 1.26 1.00 1.06 0.80 0.97 0.60 0.71 0.40 0.54 0.20 0.00 2011 2016 Source: WMS, HCES; 2011, 2016. World Bank staff calculations. bottom quintile: While the total fertility rate (TFR) dropped poor households. The difference in the age of the house- from 5.5 children per woman in 2000 to 4.6 in 2016, TFR in hold head is relatively smaller in rural areas: 45 years for the bottom quintile was the same in 2016 (6.4) as in 2000 the non-poor versus 47 years for the poor. This age-effect (6.3 ). 27 is partly explained to natural dynamics of household forma- tion and composition: young households usually have lower dependency rates (family formation is only just beginning) 2.3 Younger and female-headed and hence low poverty rates, while households with a head households less likely to be poor aged 38-58 typically have many mouths to feed. Dependen- cy rates decrease again, and poverty rates decrease again, Poverty is lower for households with younger heads. for older households as their children get married and form In urban areas, the average age of the household households of their own (Figure 28). head is 39 for non-poor households and over 44 for 27  Based on DHS, 2000; 2016. CHAPTER II. WHO ARE THE POOR? 67 Figure 28 POVERTY  IS LOWEST FOR HOUSEHOLDS WITH YOUNG HEADS Poverty rate by age of the household head, 2016 35.0% 29.6% 30.0% 27.6% 25.0% 22.0% 23.0% 20.0% 15.0% 9.0% 10.0% 5.0% 0.0% 14-29 30-37 38-47 48-57 58- Age of household head Source: HCES, 2011; 2016. World Bank staff calculations. In contrast to conventional wisdom, poverty is lower The finding that female-headed households are less for female-headed households than for male-headed likely to be poor should not obscure the fact that wom- ones. This is not a new finding but was also documented in en lag in many other dimensions. The recent Ethiopia the previous poverty assessment. Poverty for female-head- Gender Diagnostic Report of the World Bank (2019) shows ed households amounted to 19 percent in 2016, significantly that female farmers are less educated and have lower ac- lower than the 25 percent for male-headed households. The cess to land and finance compared to their male counter- lower poverty rates for female-headed households are driv- parts. They are also less likely to attend extension programs en by rural areas (Figure 29). In urban areas, female- and and use agricultural inputs like fertilizers, pesticides and her- male-headed households have similar poverty rates. The bicides, and as a result agricultural productivity is lower for poverty rates of female-headed households are however female farmers. In the labor market, women are less likely to influenced by the specific reason why the household is fe- be in more desirable types of employment and earn substan- male-headed (Box 5). tially lower wages for similar characteristics. Figure 29 FEMALE-HEADED  HOUSEHOLDS IN RURAL AREAS ARE LESS LIKELY TO BE POOR Poverty rate by sex of household head in 2016: rural and urban 30.0 26.6 25.0 20.1 20.0 Poverty rate 15.9 15.0 13.9 10.0 5.0 0.0 Male Female Male Female Urban Rural Source: HCES, 2011; 2016. World Bank staff calculations. 68 ETHIOPIA POVERTY ASSESSMENT Box 5 P  overty rates for female-headed households increase if there was a man involved Households can be female-headed for various reasons. Of the 27 percent of households in Ethiopia that were fe- male-headed in 2016, the bulk (44 percent) was female-headed because of the death of the husband. 26 percent were female-headed even though the female head was married (most likely due to work migration of the husband). 22 percent was because of divorce or separation while almost eight percent of female heads were never married. Poverty rates vary depending on why a household is female-headed. Household headed by a woman who has never been married have the lowest poverty rates, while households headed by a divorced/separated or widowed woman have the highest poverty rates (within the group of female-headed households - Figure 30). Needless to say, these finding are far from causal but driven by several selection effects: Never-married female heads of household, for instance, are on average more educated, younger and live in smaller families than the average household head. They are also more likely to live in urban areas. The second part of this chapter will examine whether marital status still affects consumption when controlling for other influences. Figure 30 POVERTY  THE LOWEST FOR NEVER MARRIED FEMALE HEADED HOUSEHOLDS Poverty by for female headed households by marital status in 2016 25 20.4 20.1 20 18.4 15 10 7.6 5 0 Never married Married Divorced or Widow separated Source: HCES, 2011; 2016. World Bank staff calculations. CHAPTER II. WHO ARE THE POOR? 69 2.4 The poor are largely uneducated school. Households whose head have completed secondary or more have low poverty rates, though there are only few of A salient characteristic of the poor all around the world those households. is their general lack of education and skills. In Ethiopia, lack of education is a general characteristic of household The low educational attainment of household heads is heads, not only of the poor: 82 percent of all household heads a legacy effect of earlier times when few people had did not complete primary education, and this increases to 94 the opportunity to gain an education. In recent decades percent for poor households. There is a strong link between access to education has expanded dramatically and gross education and poverty: Poverty is highest among house- enrolment rates in primary school now exceed 100 percent holds with a head who never went to school and decreases (DHS, 2016). Net primary enrolment rates have increased with each extra level of education (Figure 31). A complete to 71 percent in 2016. Early dropout is however rife and cycle of primary education seems to have the biggest returns few young people complete a full cycle of primary school in terms of poverty reduction: Households whose head has (eight grades in Ethiopia). In 2016, only one in three youth completed primary education have poverty rates that are less aged between 15 and 24 had completed primary education. than half of those of households whose head never went to Completion is tightly linked to household welfare and remains low for the bottom four welfare quintiles. Even in the highest Figure 31 EDUCATION  IS A MAIN CORRELATE OF POVERTY Poverty rates by educational attainment of the household head, 2016 30 28.4 25 22.1 20 12.8 13.4 15 10 4.9 5 3.3 0 n y y y y ry ar ar ar ar io da at im im nd nd on uc pr pr co co ec ed se se e e -s et et o e e pl pl st N et et m om Po pl pl co m om C In co C In Source: HCES, 2016. World Bank staff calculations. 70 ETHIOPIA POVERTY ASSESSMENT quintile though, only half of youth (15-24) had completed pri- The relatively weak relationship between consumption mary education in 2016 (Figure 32). Low primary completion and education indicators up until the fifth quintile is rates of the poor result in a steep economic gradient in sec- less of a surprise when looking at expenditures in ab- ondary school enrolment, with youth from the upper quintile solute terms. Average household expenditures per capita being more than five times more likely to attend secondary amount to US$1.3 a day in the poorest quintile (in PPP terms) school relative to youth from the bottom quintile (Figure 33). and increase slowly to US$3.6 per day in the fourth quin- The strong link between household welfare and educational tile (Figure 33). Expenditures double from the fourth to the attainment of the household’s youth suggests a high degree fifth quintile. Overall, household consumption expenditures of intergenerational transmission of poverty (Chapter 7 will remain relatively low in the bottom four quintiles. analyze this in more detail). Figure 32 EDUCATIONAL  ATTAINMENT AND ENROLMENT IS LOW AND STRONGLY LINKED TO HOUSEHOLD WELFARE Share of youth 15-24 that completed primary Gross secondary school enrolment by asset quintile, education, by consumption quintile, 2016 2016 100 100 90 90 80 80 70 70 56 60 49.6 60 50 50 40 32.2 40 30.5 25.3 26.8 30 20.1 30 19.8 15 20 20 10.4 10 10 0 0 Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 Quintiles of consumption per AE Quintiles of asset index Source: HCES, WMS, DHS 2016. World bank staff calculations. Figure 33 HOUSEHOLD  CONSUMPTION REMAINS FAIRLY LOW UNTIL THE TOP QUINTILE Average household expenditures per capita per day by quintile, 2016, PPP USD 8 Expenditures per capita per day 7.2 7 6 5 4 3.6 2.8 3 2.2 2 1.3 1 0 Q1 Q2 Q3 Q4 Q5 Quintiles of consumption per capita Source: HCES, WMS, 2016; World bank staff calculations. CHAPTER II. WHO ARE THE POOR? 71 2.5 The poor depend on agriculture sector, households engaging in services have a lower pover- ty rate (13 percent) than households earning a livelihoods in While agriculture is the mainstay of employment in the secondary sector (20 percent). Ethiopia as a whole, it is particularly important for the livelihoods of the poor. Overall, 70 percent of household Households that depend on casual labor and the pri- heads in Ethiopia have their main occupation in agriculture mary sector for their income are most likely to be poor or livestock, but this increases to over 83 percent for poor (Figure 8). Households that derive their livelihood through households (Figure 28). Beyond agriculture, poor households self-employment in the services sector or through regular engage in non-farm self-employment but do not have the salaried employment have the lowest poverty rates. House- necessarily education or skills to access non-farm wage em- holds that state their main income source as donations or ployment. The occupational structure of the bottom 80 per- remittances also have below-average poverty rates (Figure cent of households is fairly similar: 75 percent of households 35). There are some notable differences between urban and in the fourth quintile still engage in agriculture as a main oc- rural areas. In urban areas, the poorest households are those cupation (Figure 34). The top quintile is structurally different: who depend on casual labor, with poverty rates in excess of Over 70 percent of households in the top quintile have a main 30 percent, while the least poor are those who live off dona- occupation outside agriculture and almost half is engaged tions28. In rural areas, the poorest are the crop and livestock in non-farm wage employment (Figure 34). In the non-farm producers, while poverty rates are lowest for households who report remittances as their main source of livelihood. Figure 34 AGRICULTURE  REMAINS THE MOST COMMON OCCUPATION, ESPECIALLY FOR THE POOR Main occupation of household head, by quintile, 2016 Farming Self-employed nonfarm Wage nonfarm 100 4.7 5.7 6.3 10 90 11.5 12.7 11.4 15.3 28.7 80 70 60 25.7 50 40 83.8 81 82.9 74.7 30 20 45.5 10 0 Q1 Q2 Q3 Q4 Q5 Quintiles of consumption expenditures Source: HCES, WMS, 2016. 28 Overall, three percent of households report donations as their main income source. In urban areas however, this increases to seven percent. 72 ETHIOPIA POVERTY ASSESSMENT Figure 35 CASUAL  LABORERS AND CROPS AND LIVESTOCK PRODUCERS HAVE THE HIGHEST POVERTY RATES Poverty rates by household’s main income source, 2016 35 28.6 30 26.2 25.7 25.7 25 20.5 19.1 20 14.8 14.8 15 11.3 11.2 10 5 0 r g n k g ns il s t es en o ta ce oc tio in rin ab ic tio re ep m an t uc u v es l na ct oy er d ke al itt od liv an fa -s Do u pl m ck as pr u em d Re le an lf an to C p Se sa -m ro es d n le rie -c tio iv ho lf la -l uc Se -w lf Sa Se lf od Se lf pr Se p ro -c lf Main livelihood of household Se Source: HCES, WMS, 2016. CHAPTER II. WHO ARE THE POOR? 73 2.6 The poor are relatively remote compared to 45 percent of the non-poor. Rural connectivity has improved considerably since 2010, linked to the Uni- and have lower access to services versal Rural Roads Access Program (URRAP), resulting in The poor tend to live in more remote and poorly-con- decreases in travel time and improved market access (World nected areas. Relative to the rural non-poor, the rural poor Bank, 2018). URRAP roads have however been construct- live further away from roads, health facilities, and urban ed in places that were already less isolated to begin with, centers (Table 11). For instance, 57 percent of the poor live leaving the most remote and difficult-to-reach places uncon- more than three kilometers away from an all-weather road, nected (Box 6). Table 11 IN  RURAL AREAS, THE POOR ARE LOCATED FURTHER FROM HEALTH FACILITIES, ROADS AND URBAN CENTERS Distance to health facility and road connectivity by poverty status in 2016: rural areas NON-POOR POOR % located less than 2 kms 46.3 38.6 % located between 2 and 3 kms 19.6 18.3 Distance to health facility % located between more than 3 kms 34.1 43.1 Average distance in km 3.4 4.2 % located less than 2 kms (close) 43.1 33.2 % located between 2 and 3 kms (far) 11.5 10.1 Distance to all weather road % located more than 3 kms (very far) 45.4 56.7 Average distance in km 6.3 8.0 % located less than 1 hr (close) 46.4 33.8 Distance to nearest town % located between 1 hr and 2 hrs (far) 26.4 29.7 (in walking time) % located more than 2 hrs (very far) 27.2 36.5 Average distance in km 1.8 2.2 Source: HCES, WMS, 2016 Better connectivity is correlated with lower poverty, 93 percent of the urban non-poor uses electricity as a source the more so the better connected a place is. The Rural of light while the percentage for the poor is only 82 percent. Accessibility Index (RAI) measures the share of the popula- The patterns are similar for rural areas though the proportion tion of a woreda that lives within two kilometers of a road in of households who have access to these services is general- good or fair condition. As the RAI increases, poverty rates ly significantly lower (Figure 38). decrease, but the strong dent in poverty only happens when To summarize, the poor in Ethiopia are largely con- RAI exceeds 50 percent (when at least half of the popula- centrated in rural areas and are likely to live in large tion of the woreda lives within 2km of a road-see Figure 37). households with high dependency rates. Poor house- However, only a small fraction of the rural population lives holds are likely to be headed by uneducated and older heads in woredas where RAI exceeds 50 percent. This reflects the of households and are more likely to have a male household general point hat while connectivity has increased in recent head. The poor largely depend on agriculture and casual years, large parts of rural Ethiopia remain poorly connected. labor for their livelihood and are relatively isolated from key Access to key services is also lower for the poor, both infrastructure such as roads and markets. While 24 percent in rural and urban areas. In urban areas, 97 percent of of households fall below the poverty line, the characteristics non-poor households have access to improved water sourc- of the bottom 60 percent of the population are remarkably es while the proportion for the poor is 93 percent . Similarly, 29 similar. The ultra-poor, here defined as the bottom 10 percent 29  Improved water sources include tap water, protected well and rain water. 74 ETHIOPIA POVERTY ASSESSMENT Box 6 R  ural Roads, Poverty, and Resilience Under the umbrella of the first Growth and Transformation Plan, the government launched the Universal Rural Road Access Program (URRAP) in 2010. URRAP aims to ensure all rural communities of the country have an all-weather road connection. URRAP has led a major expansion of the rural road network, adding 56,000 kilome- ters of road since 2010, while an additional 29,000 kilometers were constructed by the public works under the Productive Safety Net Program (PSNP). URRAP implementation for the first five years (2010-2015) cost 28 billion Ethiopian Birr, or US$$1.4 billion. A URRAP review study was conducted in 2018 by the World Bank and the Government of Ethiopia (World Bank, 2018). The study found that rural roads have improved connectivity and accessibility of rural communities. About six percent of rural Ethiopians (close to five million people) became newly connected to all-weather roads. The average travel time to the nearest town reduced by 30 minutes between 2010 and 2016. However, the study also found that rural roads development had so far taken place mainly in communities that were already better con- nected to begin with, leaving remote communities largely unconnected to the road network. Rural roads supported welfare and resilience of rural Ethiopians amid the recent severe droughts, and its impact was highest in the more remote communities. The econometric analysis reported in the study suggests that rural roads increased household consumption by 16 percent between 2012 and 2016, increasing to 28 percent in newly connected remote communities (remote defined as more than 2 hours away from nearest town before road construction – see Figure 36). Relative to drought-affected communities where no road development took place, drought-affected communities that had been connected between 2010 and 2014 were better able to cope with the effects of the drought. Further analysis also found increased crop sales and non-farm income-generating activity in communities that benefited from rural road development. Figure 36 RURAL  ROADS INCREASED HOUSEHOLD CONSUMPTION AND RESILIENCE % increase in consumption due to rural road development, 2012-2016 HHs in remote HHs in drought-hit (Impact on HH consumption - %) All households communities communities 50 40 30 27.9 20 21.7 16.1 10 0 Source: Nakamura, Bundervoet, and Nuru (2019). The study highlighted the fundamental role of rural roads in poverty reduction and resilience by connecting rural residents to markets. Despite such benefits, the implementation of the URRAP has been slowing down due to the lack of funding, which also make it challenging to provide proper maintenance to the vast amount of already de- veloped rural roads. It will be important to connect the remaining rural communities, particularly in remote areas, to accelerate poverty reduction and enhance the resilience of rural populations against shocks. CHAPTER II. WHO ARE THE POOR? 75 Figure 37 POVERTY  DECREASES AS RURAL CONNECTIVITY INCREASES Poverty rates by rural accessibility, 2016 0.73 0.5 P(Poor ) 26.9 11.9 4.7 0 20 40 60 80 100 Rural accessibility Index Source: HCES, WMS, 2016. Figure 38 THE  POOR HAVE LESS ACCESS TO IMPROVED WATER AND ELECTRICITY BOTH IN RURAL AND URBAN AREAS Access to improved water and electricity by poverty status in 2016: urban and rural 100.0 97.7 93.5 93.2 83.1 80.0 56.0 Percent 60.0 47.6 40.0 20.0 9.0 5.1 0.0 Non-poor Poor Non-poor Poor Non-poor Poor Non-poor Poor Improved Electricity as Improved Electricity as water source a source of light water source a source of light Urban Rural Source: HCES, 2011; 2016. World Bank staff calculations. of the population, has largely similar characteristics as the characteristics that may be more prevalent there (such as low poor but tend to be concentrated in different places (Box 7). education levels, poor connectivity, etc.). Similarly, the finding that poverty rates are lower in female-headed households All figures presented so far are merely bivariate cor- may not be due to a pure gender effect, but due to women relations and do not control for the influence of other having other characteristics that correlate with lower poverty characteristics. For instance, it is possible that house- (such as smaller households). The next section will provide holds in the drought-probe lowlands are not more likely to a more robust analysis of the correlates of poverty by con- be poor just because of their location, but because of other trolling for the effects of confounding factors in a regression. 76 ETHIOPIA POVERTY ASSESSMENT Box 7 T  he characteristics of the ultra-poor The previous chapter showed that the ultra-poor, defined as the bottom 10 percent of the population, did not experience any real increase in consumption since 2005. Do the ultra-poor have particular characteristics that can be used to identify them, and are these characteristics different from those of households that also fall below the poverty line but not in the bottom 10 percent? The ultra-poor have similar characteristics as the overall poor population, only extremer. Whereas the poor are characterized by large households, high dependency rates, and a lack of education, the ultra-poor have yet larger households, higher dependency rates, and still less education (Table 12). The ultra-poor are also more likely to be rural (compared to the poor) and more distanced from markets (lower score on the market accessibility index). There are no discernable differences in occupations and livelihoods across the poor and ultra-poor, and both groups are equally likely to be in the Productive Safety Net Project (PSNP). Table 12 THE  ULTRA-POOR LOOK LIKE THE POOR, ONLY EXTREMER ULTRA-POOR POOR SIGNIFICANCE Household size 7.2 6.5 *** Number of children 3.7 3.3 *** Dependency rate 1.6 1.4 ** Female head of HH (% yes) 15.8 16.5 HH head married (% yes) 86.9 87 HH head completed primary education (% yes) 3.1 7.3 *** Any HH member completed primary education (% yes) 18.8 26.6 *** Market access index 0.018 0.024 *** Rural (%) 90.4 85.9 *** In PSNP (% yes) 12.3 10.3 Agriculture main livelihood (% yes) 86.9 84.3 Note: Ultra-poor defined as people in the bottom 10 percent of the consumption distribution. ***: Statistically significant at 1%; **: Statistically significant at 5%. Source: HCES, WMS, 2016; World bank staff calculations. Relative to the poor and the overall population, the ultra-poor are more likely to live in SNNPR and Somali. While SNNPR accounted, in 2016, for 18 percent of the poor (Table 3), it accounted for 24 percent of the ul- tra-poor. Somali accounted for nine percent of the ultra-poor, compared to five percent of the poor. While Amhara region is characterized by high levels of poverty, accounting for 26 percent of national poverty in 2016, its share of the ultra-poor is relatively low (18 percent), pointing towards a lower severity of poverty (or, more transitory poverty in Amhara, as will be argued in Chapter 4). The drought-prone lowlands, the lowland belt in eastern and southern Oromia and southern SNNPR, are significantly overrepresented among the ultra-poor: The drought-prone low- lands account for seven percent of the population, 10 percent of the poor, and 14 percent of the ultra-poor. In terms of absolute numbers though, the bulk of the ultra-poor live in Oromia (39 percent) and the moisture-reliable highlands (59 percent). CHAPTER II. WHO ARE THE POOR? 77 3. CORRELATES OF POVERTY The multivariate analysis largely confirms the descriptive findings from the previous section. Education, occupation, and demographics are the main correlates of household consumption expenditures, with large households with many children and lower educated heads and with a main livelihood in casual labor having the lowest consumption levels. In rural areas, female-headed households have substantially higher con- sumption levels, but there is a substantial penalty on being divorced. Both in urban and rural areas, house- holds engaging in non-farm self-employment, mainly in the services-sector, have the highest consumption levels. Returns to education increased between 2011 and 2016, both in rural and urban areas. There are substantial effects of location, with the drought-prone lowlands and the moisture-reliable highlands having lower consumption levels. This section presents the correlates of poverty– how different 3.1 Significant returns to education variables affect monetary living standards controlling for the and a penalty on being divorced influence of other variables. Consumption expenditures per adult equivalent is taken as an independent variable while Education is a main correlate of poverty, both in ur- the explanatory variables includes three broad categories, ban and rural areas. Relative to households with an un- namely, household head characteristics, household-level educated head, households with heads who have any level socio-economic characteristics and geography/location vari- of education have higher consumption and the difference ables. Household head characteristics include marital sta- increases with the level of education (Figure 39 and Figure tus, sex, age, employment status and education level, while 40). While the association between education and consump- household level socio-economic characteristics includes tion is stronger in urban areas, returns to education in rural livelihood, asset ownership, demographic composition, and areas are also substantial: Relative to a rural household with education and employment of members other than the head. an uneducated head, a household headed by someone who For rural areas, PSNP participation is also included as part of completed primary school has consumption levels that are household level characteristics. Finally, geographic variables 21 percent higher. This increases to 34 percent for house- include agro-ecological groups and whether the household holds whose head is secondary-educated, though this is is located in a zone that borders other countries. For rural rare in rural areas. In urban areas, returns are highest at the areas, distance to all weather roads and the closest town post-secondary levels, with households headed by someone are included as additional geographic/location variables. Be- with post-secondary education having consumption lev- cause of the difference in socio-economic structure between els that are 64 percent higher, all else equal. Education of rural and urban areas, separate regressions are run for each household members other than the head matters too: The – this allows to see if the same covariate has different effect more illiterate household members in the household, the low- on consumption in rural and urban areas. The results are re- er consumption. ported in Figure 39 (urban) and Figure 40 (rural). 78 ETHIOPIA POVERTY ASSESSMENT Returns to education in terms of consumption have straightforward to explain, but are consistent with the pat- increased, both in urban and in rural areas. Whereas in terns presented in Chapter 1, whereby especially the bet- urban areas post-secondary education had a 45 percent re- ter-off (and more educated) farmers in rural areas have been turn in terms of household consumption in 2011 (relative to doing well. a household with an uneducated head), it had a 64 percent While having a female head of household is not cor- return in 2016. This may reflect the increase in real hourly related with lower living standards, there is a signifi- wages in urban Ethiopia between 2011 and 2016 as well cant penalty on being divorced. The divorce penalty is as the modest decrease in unemployment30. In rural areas, particularly high in rural areas, where it is correlated with over the return of having a head who completed primary educat- 15 percent lower consumption, all else equal (Figure 40). As ed doubled from 10 percent in 2011 to 21 percent in 2016. only very few male heads of household are divorced, this The increasing returns to education in rural areas are not Figure 39 CONSUMPTION  CORRELATES IN URBAN AREAS IN 2016 AND 2011 Correlates of consumption per adult equivalent in urban areas: 2011, 2016 Never married (refernce: status of married) Marital head Divorced Widowed Household head characteristics Adult education Education of head Primary incomplete (reference: no education) Primary complete Secondary incomplete Secondary complete Post secondary Age of head Female head Employed head Casual - labor Household level: socio-economic variables household (reference: Self - crop production salary employment) Main livelihood of Self - livestock production Self - crop & livestock Self - manufacturing Self - service Others Household size Dependency ratio # Non-head employed members # Non-head illiterate members Household owns a house Drought prone lowland Geography/loc drought prone ation related Ecological (reference: highland) Moisture reliable lowland zones Moisture reliable highland Pastoralist -30.0 -20.0 -10.0 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 Percent 2011 2016 Note: Effects that are statistically significant (at least at 5 percent) are in dark yellow for 2016 and in dark green for 2011. Source: HCES 2011, 2016. World Bank staff calculations. 30 According to own calculations on the Urban Employment and Unemployment Surveys, real hourly wages in urban Ethiopia in- creased by 9.6 percent between 2011 and 2016. The unemployment rate marginally declined from 18.3 percent to 17.3 percent. CHAPTER II. WHO ARE THE POOR? 79 penalty applies to women-headed households. Households percent less consumption in rural areas and six percent less headed by widows also have lower consumption levels: consumption in urban areas. Larger households also have Eight percent lower in urban areas and 10 percent in rural lower consumption levels. areas. In contrast, there is no adverse consumption effect of having a female head of household. In urban areas, there is 3.2 Large effects of occupations no association whatsoever between consumption and sex of the household head. In rural areas, being female-headed is The main livelihood of the household matters for con- associated with significantly higher consumption levels. Age sumption though differently in rural and urban areas. of the household head is not correlated with consumption Those who mainly depend on service sector self-employ- expenditures. ment do relatively better both in urban and rural areas, and more so in rural areas. In urban areas, self-employment in Controlling for other variables, dependency ratios are the service sector is associated with 13.1 percent more significantly correlated with consumption levels. Each consumption compared to the reference group of salaried 0.1 increase in the dependency ratio is associated with five Figure 40 CONSUMPTION  CORRELATES IN RURAL AREAS IN 2011 AND 2016 Correlates of consumption per adult equivalent in rural areas: 2011, 2016 Never married head (reference: (refernce: status of no education) married) Marital Household head characteristics head Divorced Widowed Adult education Education of Primary incomplete Primary complete Secondary incomplete Secondary complete or above Age of head Female head Employed head Salary - employment Household level: socio-economic variables Main livelihood of (reference: crop Casual - labor production) household Self -livestock production Self - crop & livestock Self - service/manufacturing Others Household size Dependency ratio # Non-head employed members # Non-head illiterate members Owns land Owns livestock Safetynet participation Drought prone lowland tance Ecological (reference: Geography/location related close) highland) drought Moisture reliable lowland zones prone Moisture reliable highland Pastoralist Far (between 2 to 3kms) wethe (refer- r road ence: to all Dis- Very far (more than 3kms) Far (b/n 1 & 2hrs) group: (refer- close) tance town ence Dis- to Very far (more than 2hrs) Border zone -30.0 -20.0 -10.0 0.0 10.0 20.0 30.0 40.0 50.0 2011 2016 Percent Note: Effects that are statistically significant (at least at 5 percent) are in dark yellow for 2016 and in dark green for 2011. Source: HCES 2011, 2016. World Bank staff calculations. 80 ETHIOPIA POVERTY ASSESSMENT employment. Those who depend on crop and livestock in ur- walk from a town has a consumption penalty of 10 percent, ban areas have less consumption while those depending on rising to 14 percent for people living more than 2 hours away casual labor are typically the poorest. In rural areas, self-em- (Figure 40). ployment in the non-farm sector (either services or manu- facturing) is associated with 15 percent higher consumption 3.4 Location matters – even after (relative to own crop production) while those depending on casual labor have significantly lower consumption levels. controlling for other factors While depending on livestock for a livelihood was associated Controlling for other factors, location still exhibits a with significantly lower consumption levels in 2011, this was statistically significant association with household no longer the case in 2016 (Figure 40). consumption levels. Relative to households in the drought- Being employed has a different correlation with welfare prone highlands, rural households in the drought-prone low- levels in urban vs rural areas. In rural areas of low-income lands and moisture-reliable highlands are poorer, while rural countries, labor force participation rates tend to be high and households in the moisture-reliable lowlands have higher unemployment rates low. This does not indicate well-func- consumption levels. This result, which was also found in the tioning rural labor markets, though rather the necessity to previous Poverty Assessment, cautions against the reliance work to survive in the absence of formal unemployment in- on moisture- or rainfall-based indicators to make statements surance schemes. Ethiopia is no exception: According to the about poverty levels (moisture-reliable highlands for instance latest national Labor Force Survey, labor force participation in have lower consumption levels than the drought-prone high- rural Ethiopia was 88 percent and unemployment a mere two lands). While severe adverse rainfall shocks within geograph- percent31. This is also borne out of the analysis in Figure 16: ical units are typically related to lower welfare outcomes, Whether the head of household is employed or not does not differences in rainfall or dryness across geographical units matter for the household’s consumption levels. In urban ar- are poorly correlated with welfare. eas there is a significant correlation between employment of The persistent notion that the pastoral areas of Ethio- the head and consumption levels, though the magnitude of pia are the poorest of the country is also not confirmed the relation is relatively weak (six percent higher consumption by the data. Figure 40 shows that living in pastoral areas is levels if head is employed). Employment of non-head house- not correlated with consumption levels. If we also include the hold members is also correlated with higher consumption in pastoral zones that were added to the HCES in 2016 (the urban areas. nomadic zones that were not interviewed in 2011), then there is a positive association between living in pastoral areas and 3.3 Remoteness is correlated with consumption (Annex Figure 4). This is not a new result: The pastoral areas, which cover most parts of Afar and Somali lower consumption levels region, have typically had average or below average poverty In rural areas, remoteness from roads and population rates. Monetary poverty is hence not more of a problem in centers is associated with lower welfare levels. Relative pastoral areas than elsewhere (see for instance the poverty to households that are close to an all-weather road (close de- rates in Table 4). Pastoral areas are however significantly lag- fined as within 2 km), rural households who live between 2km ging on human development outcomes. Education, health, and 3km from an all-weather road have five percent lower and other social indicators tend to be much worse in the consumption, and households who live more than 3km away pastoral regions (Afar and Somali). Other regions with a sig- have eight percent lower consumption. The impact in terms nificant pastoral population, such as Oromia, also tend to of consumption becomes higher the higher the distance to perform below average on human development outcomes the road. Distance to towns and small population centers (Table 13), reflecting the difficulty of providing public services has a higher magnitude: Living between a 1 and 2-hours’ in low-density areas with mobile populations. 31  Based on the last national Labor Force Survey in 2013. CHAPTER II. WHO ARE THE POOR? 81 Table 13 PASTORAL  REGIONS DO NOT HAVE HIGHER RATES OF MONETARY POVERTY, BUT LAG ON HUMAN DEVELOPMENT OUTCOMES NET ATTENDANCE NET ATTENDANCE DELIVERY IN IN PRIMARY SCHOOL IN PRIMARY SCHOOL UNDER-5 HEALTH FACILITY – BOYS (%) – GIRLS (%) MORTALITY RATE (%) Tigray 78.1 86.3 59 56.9 Afar 62.6 63.7 125 14.7 Amhara 73.4 79.9 85 27.1 Oromiya 67.4 65.2 79 18.8 Somali 60.6 57.2 94 17.9 BSG 81.3 72.8 98 25.7 SNNPR 75.9 76.3 88 25.5 Gambella 86.9 89.6 88 45 Harari 80 76.2 72 50.2 Addis Ababa 98.2 92.1 39 96.6 Dire Dawa 82.2 75.3 93 56.2 Note: Under-5 mortality rate is expressed as number of deaths per 1,000 live births. Delivery in health facility refer to women who gave birth in the five years prior to the survey. Source: DHS, 2016. Living in zones that border another country is correlat- lower. Several of the neighboring countries are plagued by ed with lower consumption levels. This was already the chronic instability (Somalia and South Sudan) which may have case in 2011 but seems to have increased in magnitude spillover effects on the other side of the border, and relations since. In 2016, living in a border zone was, all else equal, as- between Ethiopia and the northern neighbor, Eritrea, have sociated with 17 percent lower consumption levels. Explain- been tense until the recent rapprochement and re-opening of ing this effect is difficult, but certain elements could potentially the border. Finally, it may also be the case that the border contribute: First, most of the border zones are located in the variable is picking up omitted variables, such as the 2015/16 Developing Regional States (Somali, Afar, Gambella, and Ben- drought (though, surprisingly, research has failed to find an ishangul-Gumuz) where administrative capacity tends to be adverse welfare effect of the drought at household level). 82 ETHIOPIA POVERTY ASSESSMENT CONCLUSIONS The poverty profile for Ethiopia is largely unsurprising. the poor are characterized by remoteness, large households, As in most low-income countries in the world, the poor are high dependency rates, and a lack of education, the ul- poorly educated, depend on agriculture and./or casual la- tra-poor are yet more remote, have yet larger households, bor, live in large households with high dependency rates, and higher dependency rates, and still less education. Compared tend to be further away from key infrastructure. In contrast to the overall distribution of the poor, the ultra-poor are more to common perceptions, female-headed households are less likely to be located in rural areas of Somali and SNNPR. likely to be poor. There is however a substantial penalty in Returns to education have increased between 2011 terms of consumption of being divorced: Households head- and 2016. In urban areas, the increasing returns to education ed by divorced women have substantially lower consump- are likely related to increasing real hourly wages and decreas- tion levels. Controlling for other factors, households in the ing unemployment between 2011 and 2016. The significant drought-prone lowlands (the lowland areas of Oromia and increase in education returns in rural areas is more difficult to SNNPR) and moisture-reliable highlands have lower con- explain but may reflect the increase in use of technology from sumption levels. a low base. Use of agricultural technology such as improved The ultra-poor, the bottom 10 percent whose con- seeds and fertilizers and herbicides have increased in recent sumption has not grown since 2005, have largely sim- years, and are mainly taken up by more educated farmers. ilar characteristics of the poor, but extremer. Whereas CHAPTER II. WHO ARE THE POOR? 83 84 ETHIOPIA POVERTY ASSESSMENT CHAPTER III Drivers of Poverty Reduction in Ethiopia The first two chapters documented the decline in poverty in Ethiopia between 2011 and 2016, and also presented profiles of the poor population. This chapter generally focuses on a longer period of time – 2000 to 2016 and investigates what the drivers behind poverty reduction in Ethiopia have been. It does so in two distinct sections. The first uses household survey data to decompose poverty changes so that the roles of different characteristics can be uncovered. It also analyzes how important changes in endowments versus changes in returns to these endowments were over time. The second section updates the zonal panel dataset that was used in the previous Poverty Assessment in order to investigate the roles of agricultural productivity, investments in public infrastructure and sectoral changes in reducing poverty between 2000 and 2016. The analysis finds that the contribution of urban areas to poverty reduction is increasing and is likely to grow. Nevertheless, poverty reduction overall was still heavily concentrated in rural areas and in the agricultural sector. Productivity growth in agriculture will remain critical to poverty reduction, given its share of employment. The role of structural transformation – shifts out of agriculture and into manufacturing or services – was very limited in reducing poverty over the period. Increases in endowments, and particularly asset accumulation, played a large role in lowering poverty rates. Poverty fell fastest in the zones that had the strongest agricultural growth between 2000 and 2016, while the expansion of the PSNP between 2011 and 2016 also played a crucial role in reducing poverty rates. Improved access to large towns, as measured through decreased travel times, was associated with strong poverty reduction, indicative of the complementary nature of agricultural and non-agricultural growth. There was a shift away from the production of cereal crops in favor of cash crops between 2011 and 2016. This shift had significant poverty-reducing effects, largely because of the rapid relative gains in the prices of cash crops, especially khat, over these years. While this increase in crop prices helped net producers, and will likely continue to do so, there are also potential losers from these changes. Policy should be nimble enough to ensure that the effects of rising prices on vulnerable households and parts of the population are effectively mitigated. CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 85 Introduction This chapter builds on the analysis of Chapters 1 and 2 The chapter proceeds as follows. Section 2 decomposes by focusing on what the key drivers behind Ethiopia’s the reduction in poverty between 2000 and 2016 into shifts poverty reduction over the 2000 to 2016 period have within sectors versus shifts between sectors, and also in- been. The decade and a half that this chapter covers saw vestigates whether changes in endowments or changes in sustained strong growth in the country, though that growth the returns to these endowments were more important for was not necessarily spread evenly over the distribution of poverty reduction. Section 3 extends the zonal panel analysis consumption. The headcount poverty rate at the start of the contained in the last Poverty Assessment and assesses the period stood at 47.4 percent. By 2016 this had been reduced contributions of overall growth, sectoral growth, agricultur- by 24 percentage points to 23.4 percent. Understanding al productivity and investments in public services and infra- which factors drove this decline is the focus of this chapter. structure to poverty reduction. An important question dealt with in this chapter is how different the nature of poverty reduction between 2011 and 2016 was, compared to the 2000 to 2011 period. As noted in the last Poverty Assessment for the country World Bank (2015) the provision and expansion of rural safety nets, large investments in infrastructure and increasing market in- tegration all served to compliment overall economic growth in reducing poverty. However, increasing rates of urbaniza- tion, an ever more educated population and shifts in sectoral occupations could mean that there is a changing nature to poverty reduction. A longitudinal zonal data set is used in combination with household survey data to explore what the most important factors that drove welfare improvements were. Among these factors are changes in the sectoral com- position of households, different rates of sectoral productivity growth over time, changes in agricultural productivity and prices, and the expansion of safety nets, particularly in ru- ral areas. This zonal panel dataset is constructed using data from a variety of sources including the Household Income and Expenditure Survey (HCES), the Welfare Monitoring Survey (WMS), the Agricultural Sample Survey (AgSS), the manufacturing census, the survey of trade and distributive services, administrative data from the Productive Safety Net Programme (PSNP), and the Livelihoods, Early Assessment and Protection project (LEAP) datasets from various years. 86 ETHIOPIA POVERTY ASSESSMENT Box 8 E  vidence on the drivers of poverty reduction in Ethiopia There is a growing body of literature aimed at uncovering the drivers of the recent poverty reduction in Ethiopia. Dercon (2006) presents early analysis of the impact of economic reform between 1989 and 1995 on consumption poverty in Ethiopia. The most important drivers behind consumption changes over the period were how relative prices changed – these had strong effects on the returns to physical and human capital, as well as to location. Changes to the terms of trade, particularly strong real increases in producer prices, increasing returns to road infrastructure and location are consistent with economic changes that took place under the market-oriented re- forms that took place in that period. This complements the analysis in Dercon (2004) which looks at consumption changes specifically, rather than at the impacts of economic reforms. Dercon (2004) found that there were signifi- cant and persistent negative impacts of rainfall shocks on household consumption, and that the lack of insurance and social protection coverage exacerbated how low these effects were felt for. Dynamically, a drop in rainfall of 10 percent that occurred 4 to 5 years ago reduced current growth rates by one percentage point. Dercon et al (2012) use a decade and a half of longitudinal data from Ethiopia to show that although chronically poor popula- tions benefit from some of the key drivers of growth such as better connectivity and market access, the growth rate of their consumption is lower than that of other groups, mainly because their initial conditions are severely lacking in both human and physical capital. This reflects earlier work in Dercon (2009) that showed the key roles played by agricultural extension services and an expansion of the road network between 1994 and 2004. Hill and Tsehaye (2018) highlight the central role of higher agricultural productivity in reducing poverty. This was particularly true in the period between 2005 and 2011, during which agricultural output growth accounted for a 2.2 percent drop in poverty annually. However, the authors note that the increase in productivity was conditional on being in close proximity to urban centers. Another condition for poverty reduction through agricultural growth was that the adoption of productivity enhancing technologies such as fertilizer and improved seeds were dependent on good rainfall and the maintenance of high prices. Vandercasteelen et al (2018) investigate the proximity of rural teff famers to primary and secondary cities impacts the prices of agricultural produce, adoption of improved technology and farming intensification. They find that positive effects on prices and technology adoption is far higher for farmers that live close to cities compared to those that live close to secondary towns. This dynamic occurs in concert with evidence from other countries showing that the poverty reduction associated with secondary towns is driven by rural migrants being more likely to engage in non-farm sector work after they migrate (Christaensen et al 2016). A study by Dorosh et al (2018) looks forward and uses a CGE model to simulate which kinds of investments are most likely to lead to the strongest poverty reduction in Ethiopia. The results suggest that relatively higher invest- ments into urban development will lead to stronger poverty reduction in the medium-to-long-term, while immedi- ate investments in the rural non-farm sector are likely to be relatively more pro-poor. The relationship between the expansion of Ethiopia’s social protection programs and poverty reduction is dis- cussed in detail in Chapter 6. A study that complements that chapter (Hirvonen et al 2016) looks at the compara- tive effectiveness of social protection and income taxation in driving poverty reduction in the country. The authors find that the PSNP has been effective in reducing poverty by 0.9 percentage points, and simulate that perfect targeting of the program in its current form would reduce aggregate poverty by about 1.5 percentage points. CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 87 2. DECOMPOSING POVERTY REDUCTION BETWEEN 2000 AND 2016 The decomposition method most suited to attributing into an intra-sectoral effect (changes in poverty levels within changes in poverty to intra-sectoral versus population each group), and a population shift effect (changes in the shift effects is due to Ravallion and Huppi (1991). This characteristics of the population). Decomposing poverty approach exploits the additive decomposability of the stan- changes in this way allows us to determine which of the two dard FGT measures of poverty in order to generate the de- factors was more responsible for driving poverty changes composition. The aim is to decompose changes in poverty over a given time period. Box 9 Decomposing poverty changes over time The poverty decompositions used in this chapter are based on Ravallion and Huppi (1991).32 The decompositions will reflect changes in the poverty headcount rate, but extend naturally to the poverty gap and poverty gap squared mea- sures. Poverty at time t is given as Pt . The change in poverty between t and t+1 is composed of the following effects: n Pt+1 – Pt = ∑ i=1 sit ( Pi,t+1 – Pi,t ) Intra-sectoral effect n +∑ i=1 Pit (si,t+1 – si,t ) Population shift effect n +∑ i=1 (Pi,t+1 – Pi,t )(si,t+1 – si,t ) Interaction effect For the purposes of this chapter, i represents the specific sector, and n is the number of sectors. These include: urban vs rural; the five main sectors of occupation; self-employed versus not self-employed. Pi,t is the poverty rate of group i in period t. Finally, si,t is the population share of group i in period t. An illustrative example, adapted from Valderrama and Viveros (2014) may be useful in explaining the counterfactual assumptions underlying this decomposition. Consider that in time period 1 there are three sectors – agriculture, manufacturing and services in which a household head could be employed. Each of these sectors has a different poverty rate associated with it, with the highest poverty rate in the agriculture sector and the lowest poverty rate in the services sector. Consider a simple change in which a group of households shifts from the agriculture sector to the services sector. If, after the shift, the within-sector poverty rates remain the same, then it must be that the national poverty rate went down. This is a pure population shift effect – the drop in national poverty was driven entirely by population shift from agriculture to services. Consider now a situation in which the poverty rate in the agriculture sector goes down, but no households change sectors.33 Again, the national poverty rate would decrease. In this situation, however, the decrease would have been driven entirely by the intra-sectoral effect. In practice, as shown in this chapter, the overall poverty change will be a combination of both the population shift effect and the intra-sectoral effect, along with an interaction effect to balance out the accounting exercise. 32 Poverty is defined at the household level, and for this reason the chapter uses the occupational characteristics of the household head as the characteristic that defines the entire household. Finally, the caveat should be added that the decomposition is purely a statistical exercise that should be used to understand past changes, rather than a tool that should be used to estimate future trends in poverty. 33 This could also be the case if the poverty rate decreases in one sector but there is no net mobility across sectors. 88 ETHIOPIA POVERTY ASSESSMENT Figure 41 THE  URBAN CONTRIBUTION TO POVERTY REDUCTION IS INCREASING Rural-urban decomposition of poverty changes 2000 to 2016 Rural Urban Population shift Interaction Change in poverty headcount 2 0 -2 -4 -6 -8 -10 2000-2005 2005-2011 2011-2016 Source: HCES; 2000; 2005; 2011, 2016. World Bank staff calculations. The absolute magnitude of poverty reduction was sim- At the same time, the poverty share of agricultural house- ilar for the 2000-2005 and 2011-2016 periods, but the holds increased slightly over the same time period, reflecting rural-urban composition of this reduction was very faster poverty reduction in other sectors. As shown in Figure different. Figure 41 shows that in the first period the ap- 42 the contribution of manufacturing to poverty reduction proximately 6 percentage point drop in poverty was driven was generally quite small, in line with the share of house- almost entirely by changes within rural areas. In the 2011 to holds that were engaged primarily in that sector. Changes 2016 period, about one third of the fall in the national pov- within the services sector accounted for about 15 percent of erty rate could be explained by changes in urban areas. The poverty reduction between 2011 and 2016. Structural trans- result of this, as shown in Chapter 2, was that the share of formation, here proxied by the population shift effect across urban households in the overall number of poor households economic sectors, played a very minor role, explaining only 5 declined. Poverty reduction within urban areas was driven percent of the poverty reduction in the last period. primarily by changes within households that were engaged The contributions of the self-employed versus wage in the trade sector, as discussed in Chapter 5. At no point workers to poverty reduction were very similar in the did population shifts from rural to urban areas contribute sig- first and final time periods. This is unsurprising, given that nificantly to the overall reduction in poverty over the period. the self-employed are generally agricultural, and agriculture’s However, it should be noted that this decomposition anal- contribution to poverty reduction was also quite consistent ysis does not consider spillover effects of urbanization for over the same time periods. Annex Figure 5 breaks down the rural economies, such as increasing demand for agricultural poverty changes by employment type. Comparing the mid- products, improved access to better agricultural inputs, re- dle and final windows of time reveals that even though the mittances, etc. share of the population that was not self-employed increased The agriculture sector remained the largest contrib- between 2011 and 2016, the absolute and relative contribu- utor to poverty reduction while the role of structural tions of this group to poverty reduction went down. The small change remains very limited. Agriculture’s share in ex- population shift effects were in fact poverty enhancing in the plaining national poverty reduction dropped from about three first and second windows, but were poverty reducing in the quarters (2000 to 2005) to about two thirds (2011 to 2016). 2011 to 2016 period. CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 89 Figure 42 THE  AGRICULTURE SECTOR REMAINS THE LARGEST CONTRIBUTOR TO POVERTY REDUCTION Sectoral decomposition of poverty changes 2000 to 2016 Agriculture Manufacturing Construction Services Other Population shift Interaction 2 Change in poverty headcount 0 -2 -4 -6 -8 -10 2000-2005 2005-2011 2011-2016 Source: HCES; 2000; 2005; 2011, 2016. World Bank staff calculations. Improvements in the characteristics of individuals of the household residents. All of the positive consumption and households – endowments – explain most of the change in rural households can be explained by an increas- changes in welfare and are particularly important for ing level of endowments, as this line is always above zero, the bottom 40 percent or rural households. The blue while the returns line is always negative. line in Figure 43 shows the change in consumption between For urban households the roles are reversed, with re- 2011 and 2016 over the distribution of consumption itself for turns generally outweighing endowments as drivers rural households. This is analogous to the growth incidence of consumption change. Figure 44 shows that for urban curves that were presented in earlier chapters. The red line households the endowment effect is larger than the return and the green line show the roles played by endowments effect only for the bottom quartile. Neither of the effects is and returns to endowments respectively across the distribu- negative at any point of the distribution, and so their sum – tion. Details on how overall changes were decomposed into the blue line – is always above them. Returns to assets were endowment and returns contributions are outlined in Box 11. particularly strong drivers of positive consumption change at Changes in endowments contributed positively to the bottom of the urban distribution, while returns to chang- changes in consumption for households across the ing household demographics (smaller household sizes and distribution. The characteristics used in the decomposition lower dependency ratios) are the most important drivers of include education, demographics, household location, con- consumption change for the top 40 percent of the urban ditions (access to electricity, improved water, dwelling unit distribution. materials), assets, and the sector and main income source 90 ETHIOPIA POVERTY ASSESSMENT Figure 43 CHANGES  IN CHARACTERISTICS OF RURAL HOUSEHOLDS EXPLAIN MOST OF THE INCREASE IN CONSUMPTION SINCE 2011 The contributions of endowments and returns to consumption growth, rural households 2011 to 2016 .3 Log difference in consumption .2 .1 0 −.1 −.2 10 20 30 40 50 60 70 80 90 Consumption percentiles Difference Endowments Returns Source: HCES 2011, 2016. World Bank staff calculations. Figure 44 WHILE  CHANGES IN RETURNS EXPLAIN MOST OF THE CONSUMPTION INCREASE FOR URBAN HOUSEHOLDS SINCE 2011 The contributions of endowments and returns to consumption growth, urban households 2011 to 2016 .3 Log difference in consumption .2 .1 0 −.1 −.2 10 20 30 40 50 60 70 80 90 Consumption percentiles Difference Endowments Returns Source: HCES 2011, 2016. World Bank staff calculations. CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 91 Box 10  ecomposing changes in consumption into endowment and D returns components Changes in household consumption levels over time can be decomposed into two components. The first component is due to changes in personal/household characteristics or endowments – more education, more household assets, better infrastructure and so on. The second component is due to changes in the returns to those endowments – returns to education, returns to assets, increased land productivity and so on. These components can then be further analyzed to identify the contributions of each specific attribute to changes in consumption, over 34 the entire welfare distribution. The decomposition procedure takes place over two steps. In the first step a counterfactual distribution is es- timated to show what consumption in 2011 would have looked like if that society had 2016’s characteristics. The difference between the actual consumption distribution in 2016 and this counterfactual distribution is change in consumption that can be explained by the changes in the characteristics or endowments of the population between 2011 and 2016, with returns held constant. The difference between the actual consumption distribution in 2011 and the counterfactual distribution is then the part of the change in consumption that is due to changes in returns to those characteristics/endowments over the period (endowments are held constant). There is also an interaction term that captures changes in the correlation between endowments and returns over time. In the second step, RIF regressions are used to decompose the explained and unexplained parts into the contribution of each individual covariate at different percentiles of the distribution. By far the biggest contribution to positive consump- distribution, as can be seen in Figure 46. For the top 20 per- tion change was the accumulation of assets. The effect cent of urban households increasing years of education was of asset accumulation between 2011 and 2016 was the larg- the most significant driver of consumption changes between est factor explaining positive consumption changes for every 2011 and 2016, although the education endowment effect percentile of the rural distribution, but was particularly strong was uniform over the distribution. For a pooled model that for the bottom 40 percent. This is highlighted in Figure 45 decomposes changes for the whole country together (not which shows endowment effects for rural households. The shown) the effects of increasing urbanization were very mut- assets underlying this effect include land ownership, live- ed across the entire distribution. This is because the overall stock ownership, and ownership of various durables includ- share of the urban population in Ethiopia grew slowly be- ing a cellphone, television and bicycle. tween 2011 and 2016, even though the absolute number of people migrating to urban areas was large. As noted in The accumulation of assets was important for urban Chapter 5, although the urban population increased by about households, as it was for rural households, but the ef- 4.1 million people between 2011 and 2016, the urban share fects of education were far more prominent. For urban of the population increased by only 2.5 percentage points households the accumulation of assets was also import- over the same time period. ant, but this effect declined steadily over the consumption 34  The endowment and returns decompositions in this chapter are based on the Recentered Influence Function (RIF) and Uncon- ditional Quantile Regression (UQR) methodology contained in Firpo, Fortin et al. (2009) which generalizes the method detailed in Oaxaca (1973) and Blinder (1973). 92 ETHIOPIA POVERTY ASSESSMENT Figure 45 ASSET  ACCUMULATION EXPLAINS THE BIGGEST PART OF THE CONSUMPTION INCREASE FOR RURAL HOUSEHOLDS The contributions of endowments to consumption growth, rural households 2011 to 2016 .1 Log difference in consumption .075 .05 .025 0 10 20 30 40 50 60 70 80 90 Consumption percentiles Education Demographics Assets Sector House cond. Income sources Source: HCES 2011, 2016. World Bank staff calculations. Figure 46 WHILE  INCREASING EDUCATION AND CHANGING DEMOGRAPHICS ARE MORE IMPORTANT FOR URBAN HOUSEHOLDS The contributions of endowments to consumption growth, urban households 2011 to 2016 .1 Log difference in consumption .075 .05 .025 0 10 20 30 40 50 60 70 80 90 Consumption percentiles Education Demographics Assets Sector House cond. Income sources Source: HCES 2011, 2016. World Bank staff calculations. CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 93 The total effect (endowment plus returns) of changes The total effect of asset accumulation is highest for in assets is prominent for all four quartiles of the dis- the top quartile, indicating that this group also had the tribution. Figure 47 adopts a slightly different approach and highest returns to asset accumulation. Figure 45 showed groups the overall effect of the different components for each that the asset endowment effect decreased over the distribu- of the four consumption quartiles. For the bottom quartile, tion, while Figure 46 shows that the overall effect increased. approximately corresponding to the poor population, chang- This means that the returns to assets became stronger over es in home conditions (access to electricity, improved water, the distribution of consumption. Although the total effect of dwelling unit materials) had an overall larger positive effect education also increased across the consumption distribu- on consumption than did asset accumulation. The total ef- tion, the overall effect was driven by endowment changes fect of improving home conditions then becomes more and rather than by returns. The effect of education was weak for more muted as households become richer. Even though the the poorest households, reflecting the poor performance of accumulation of assets was an important endowment effect education in rural areas. for poor households (Figure 44), decreasing returns to those Demographic changes contributed negatively to con- assets meant that the total effect of assets was more muted, sumption changes between 2011 and 2016. This was though still positive (Figure 45). true for all except the second quartile. This overall effect was driven by changes in returns, as the endowment effects from demographic changes were generally small and positive, as reflected in Figure 46. Figure 47 ASSET  ACCUMULATION EXPLAINS THE BIGGEST PART OF THE CONSUMPTION INCREASE BETWEEN 2011 AND 2016 Overall contributions to consumption changes 2011 to 2016 1.2 Contribution to change in consumption 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 Poorest 2 3 Richest Consumption quartiles Education Demographics Urbanization Assets Sector Home conditions Income sources Source: HCES 2011, 2016. World Bank staff calculations. 94 ETHIOPIA POVERTY ASSESSMENT 3. DRIVERS OF POVERTY REDUCTION: EVIDENCE FROM A ZONAL PANEL DATASET In order to better understand the drivers behind the over time, and there has also been variation in sectoral output changes in poverty, a longitudinal zone-year dataset growth and in the roll-out of public infrastructure and support ranging from 2000 to 2016 is used. The dataset and find- programs over the period covered. 50 zones are covered in ings serve as a direct update to the previous Ethiopia Poverty the analysis, and this covers almost all of Ethiopia’s popula- Assessment World Bank (2015), and also to Hill and Tsehaye tion. Box 11 provides a brief overview of the data sources (2018). The analysis makes use of the fact that poverty has used in this part of the chapter, as well as the methodology been reduced at different rates in different zones in Ethiopia used in the assessment. Box 11 Datasets and methodology for estimating poverty changes  using a zonal panel This chapter presents an update to the results of the zonal panel estimates of growth, safety nets, infra- structure and poverty reduction presented in World Bank (2015) and Hill and Tsehaye (2018). As such, the description of the methodology briefly presented here draws on the exposition in these texts. Poverty estimates, population estimates, and the number of people engaged in work in each sector at the zonal level come from HICES datasets for the years covered. Data from the Central Statistical Agency’s Agri- cultural Sample Survey is used for estimates of zonal agricultural production, the area of land cultivated, the value of agricultural output, the proportion of land using improved seeds and fertilizer, and a weighted crop price index. Data on manufacturing output is derived from an annual census of large and medium manufacturing en- terprises. In the estimation the share of manufacturing output that is produced by small firms is allowed to vary across zones and over time, but a constant output growth rate of these firms is assumed over the 2000 to 2016 period. Data on zonal service sector output are calculated by using the HICES datasets to generate the number of service and trade workers and then multiplying by nationally published estimates of value added per worked in this sector. This yields a measure of service sector output per worker at the zonal level. Distance variables to schools, roads and large towns are constructed using data from the Welfare Mon- itoring Survey and the Rural Accessibility Index. Household-level distances are averaged to arrive at zonal level estimates. Administrative data on the number of beneficiaries of the PSNP are aggregated to the zonal level. Finally, rainfall shocks and associated crop losses are derived at the zonal level from data from the Livelihoods, Early Assessment and Protection (LEAP) project. There are two main regressions that are estimated in this chapter. The first examines the effect of aggregate output growth, public good provision and infrastructure on changes in the poverty rate. The following is estimated: Δlnp zt = b 0 + b Y ΔlnY zt + b N ΔlnN zt + b E ΔlnE zt + b D ΔlnD zt + u z + e zt where p zt is the poverty rate in the zone at time t, Y zt is zonal output at time t, N zt is the proportion of people in the zone who are PSNP participants at time t, E zt is a measure of increased access to primary schools in the zone at time t, and D zt is a measure of infrastructure in the zone at time t. The second kind of equation that is estimated investigates poverty reduction and sectoral output growth and includes the change in the log poverty as the dependent variable with the same explanatory variable as above, but now also with the shares of agriculture, manufacturing and service sector outputs interacted with their growth rates to allow for the impact on poverty to be in line with the size of the sector. CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 95 Hill and Tsehaye (2018) find that between 1996 and manufacturing and services respectively. Thus, even though 2011 poverty reduction was highest in zones that had the employment share of agriculture trended downwards, it the highest growth in agricultural output. This effect was is still by far the largest sector in the Ethiopian economy, and particularly strong from 2005 onwards. Growth in agricultural changes in output and prices in this sector will be key to un- output per capita contributed to poverty reduction, but there derstanding overall poverty changes at the zonal level. This was no discernable effect to growth in manufacturing or ser- has already been alluded to in the chapter in the sectoral vices output per capita to poverty reduction at the national decompositions of poverty that were presented in Figure 42. level. Within the agricultural sector the production of cereal The expansion of public infrastructure variables saw a crops was central to poverty reduction, but this effect relied higher proportion of households covered by the PSNP, on the intersection of good rainfall, high prices and access while access to school, public transport and nearby to markets. large towns increased substantially. The zonal average The strong reduction in poverty measures at the zonal of PSNP coverage rose from eight percent in 2011 to almost level between 2000 and 2016, the key dependent vari- 14.5 percent by 2016. At the same time the average distance able in this part of the chapter, is evident in Table 14.35 to public transport dropped by over four kilometers, and the The poverty headcount, gap and severity measures were travel time taken to get to the nearest large town dropped all about half in 2016 what they were in 2000. The over- substantially as well. The predicted level of crop loss due to all employment share in agriculture reduced by eight per- low rainfall increased between 2011 and 2016, though this centage points over the period, while there were increases took place against increased use of improved seeds and in- of three percentage points and five percentage points for creased coverage of fertilizer, on average. Table 14 DESCRIPTIVE STATISTICS OF VARIABLES USED IN ZONE PANEL REGRESSIONS  2000 2005 2011 2016 POVERTY Poverty rate 47.4 40.3 28.1 22.8 Poverty gap (%) 13.6 8.6 7.4 6.8 Poverty severity (%) 5.4 2.8 2.9 2.8 SECTORS Agriculture emp. share 80.1 79.5 77.0 72.0 Manufacturing emp. share 2.2 5.6 4.8 5.0 Services emp. share 17.7 14.9 18.3 23.0 SAFETY NETS, SERVICES, INFRASTRUCTURE Share of households in PSNP 0 0 8.3 14.4 Distance to nearest primary school (km) 4.1 4.1 2.7 1.9 Distance to public transport (km) 20.5 17.5 13.6 9.3 AGRICULTURE VARIABLES Predicted crop loss due to rainfall (%) 22.4 26.6 15.7 20.4 Land planted with improved seeds (%) 1.4 2.3 4.1 5.9 Land using fertilizer (%) 9.6 16.7 27.6 33.9 Index of crop prices (Birr per kg) 3.2 4.1 12.3 23.3 Source: Calculations from datasets described in Box 3 35 It is important to note that while these numbers are similar to the officially published poverty numbers, they should not be treated as such. The numbers in the table are zonal averages of poverty, for the 50 zones only that are followed over the time period. 96 ETHIOPIA POVERTY ASSESSMENT Figure 48 THE  POOREST ZONES EXPERIENCED THE FASTEST POVERTY REDUCTION The rate of zonal poverty reduction and initial poverty level 50 Change in poverty 2000 to 2016 )(% 0 −50 −100 20 40 60 80 100 Poverty rate in 2000 Source: Calculations from HCES, 2000; 2005; 2011, 2016. 3.1 How has growth contributed to detailed in Table 14. This results in a weighted elasticity of -0.13, meaning that a 1 percent increase in agricultural out- poverty reduction? put per capita is associated with a 0.13 percent reduction in The zones that were initially the poorest experienced the overall poverty rate. the fastest poverty reduction between 2000 and 2016. The poverty-reducing effect of increases in manufac- Figure 48 plots the percentage change in poverty at the zonal turing output per capita is significant post-2005, and level against initial poverty in 2000. The same pattern is evi- particularly in urban areas. Neither of the manufacturing dent when the percentage point change is used as the y-axis or services coefficients are statistically significant in column rather than the percentage change in poverty. A regression two, reflecting the same result described in World Bank of the change in poverty headcount on the change in zonal (2015). These two output measures are considerably more output per capita yields a coefficient of -0.34, which is sig- imprecise than the agricultural output estimates, and so it nificant at the 1 percent level. This suggests that a 1 percent cannot be ruled out that the magnitude of the coefficients in increase in growth resulted in a 0.34 percent reduction in the overall sample is this small because of attenuation bias. poverty over the period. This measure of the growth elasticity Perhaps for this reason the contribution of the services sec- of poverty using household consumption data is higher than tor to poverty reduction, as presented in Figure 42 earlier in the growth elasticity of poverty reported in Chapter 1, which the chapter is not reflected in the regression analysis. Never- used real changes in GDP. theless, when the time period is restricted to 2005 and later, Poverty fell fastest in the zones in which agricultur- and each zone is weighed by the proportion of its popula- al growth was strongest. The second column of Table tion that is urban, growth in manufacturing output per capita 15 shows the results of regression analysis assessing the has a poverty-reducing effect. The standard error is relatively roles of sectoral output growth and investments in public large, likely because of the level of imprecision associated infrastructure on poverty reduction. The implied elasticity of with this variable, but the effect is significant at the 10 per- poverty to agricultural growth can be calculated by multiply- cent level. The zone-level correlation between growth in ag- ing the coefficient by the average employment share of the ricultural employment and growth in services employment is sector for each of the years 2000, 2005, 2011 and 2016 stronger than the zone-level correlation between growth in CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 97 agricultural employment and manufacturing employment, as Improving access to large towns played an important can be seen in Figure 8 in the appendix. Agricultural growth role in explaining poverty reduction. In the analysis a thus seems to have positive spillovers on the services sector. town is considered to be large if it has a population of at least 50,000. Access is measured by the number of min- The expansion of the PSNP had a significant effect utes taken, on average, from each zone to the nearest large on reducing poverty at the zonal level, particularly in town36. On average an increase in the travel time to a large the 2011 to 2016 period. As shown in the second column, town by four percent increases the zonal poverty rate by one a one percent annualized increase in PSNP coverage was percent. The effect is particularly strong for locations that associated with a 0.1 percent annualized decrease in the are three hours of more away from a large town (confirming poverty rate. This effect is more significant (both economi- the results from the URRAP analysis presented in Chapter cally and statistically) than what is reported in Hill and Tse- 2). Unsurprisingly the effect disappears when the sample is haye (2018), which covered 1996 to 2011 and therefore only weighted by the urban population, as in columns 3 and 4 worked with one period of PSNP expansion. The effect of the of Table 15. The final column of the table presents outputs PSNP remains significant when observations are weighted from an instrumental variables regression in which growth by the urban population share between, but not when these in agricultural output has been instrumented with weather weights are applied in the 2005 to 2016 window. Table 15 SECTORAL  GROWTH, SAFETY NETS, INFRASTRUCTURE AND POVERTY REDUCTION 2000 TO 2016 1 2 3 4 5 WEIGHTED BY URBAN POP. IV ANNUALIZED PERCENTAGE 2000-2016 2000-2016 2000-2016 2005-2016 2000-2016 CHANGE IN POVERTY RATE ANNUALIZED PERCENTAGE CHANGE IN Output per capita -0.12* (0.07) Agricultural output per capita -0.17** -0.22*** 0.15 -0.03 (0.07) (0.07) (0.21) (0.12) Manufacturing output per capita -0.05 -0.15 -0.46* 0.00 (0.11) (0.10) (0.24) (0.09) Services output per capita 0.01 0.01 0.22 -0.02 (0.16) (0.35) (0.36) (0.12) Proportion in PSNP -0.12*** -0.10*** -0.08** -0.05 -0.11** (0.03) (0.03) (0.04) (0.03) (0.03) Distance to primary school -0.21 -0.19 -0.18 0.02 -0.09 (0.17) (0.17) (0.17) (0.14) (0.14) Distance to nearest large town 0.24** 0.25** 0.11 0.22 0.34*** (0.10) (0.10) (0.11) (0.19) (0.09) Constant -0.00 -0.01 -0.03*** -0.03*** -0.01 (0.01) (0.01) (0.01) (0.01) (0.01) Observations 168 168 154 124 166 R-squared 0.40 0.42 0.50 0.12 Number of zones 50 50 46 46 50 Note: Zone fixed effects are included but not shown. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Source: Calculations based on data sources described in Box 3. 36 Travel time for years before 2016 is based on the model and data in Schmidt and Kedir (2009). Access and connectivity for 2016 data are based on HICES 2015/16 and the Rural Accessibility Index. 98 ETHIOPIA POVERTY ASSESSMENT shocks. Although the sign on the agriculture variable is in the results below suggest that the expanded use of fertilizers did expected direction, the effect is not statistically significant. not appear to have a significant effect on poverty reduction, The fact that this relationship is not significant is in contrast in either good or bad conditions. Although the interaction be- to previous findings (Hill and Tsehaye (2018)) and may indi- tween agricultural growth and being close to a large town is cate a degree of reverse causality in that households that statistically significant in the results below, the magnitude of were less poor were better able to increase their agricultural the coefficient is far smaller: -0.82 in this study versus -3.40 income over the period. Four maps showing the rapid and in previous findings. However, this is still suggestive of the im- substantial increase in connectivity across the country can portance of the complementarities between agricultural and be found in Annex Figure 7 in Annex 3. non-agricultural growth on poverty reduction. The increased use of improved seeds is associated 3.2 Agricultural growth and poverty with significant poverty reduction – this is potentially reduction important given the relatively low existing use of im- proved seeds. As has been shown earlier in this chapter, The key factors driving the relationship between coverage of improved seeds grew from 1.4 percent of cul- growth in agriculture and poverty reduction are now tivated land in 2000 to 5.9 percent in 2016. This is far low- explored. Table 16 maintains the annualized percentage er than the 34 percent fertilizer coverage in the same year. change in the poverty rate as the dependent variable, but According to the results below, a one percent increase in now adds several new explanatory variables. These include the use of improved seeds is associated with a 0.14 per- interaction terms between growth in agricultural output per cent reduction in poverty, on average. The flip in the relative capita and: whether the travel time to the nearest large town importance of fertilizer versus improved seeds in this versus is less than or more than three hours, and growth in cereal previous studies suggests that there may be relatively higher and cash crop output per capita. Also included are the pro- returns to investing in expanding the latter. portion of cultivated land using improved seeds and fertiliz- er, and interactions between fertilizer use in bad conditions Changes in the output of cash crops had a larger pov- (drought and slower than average rises in crop prices) and erty-reducing effect than changes in the output of ce- good conditions. Finally, a weighted index of crop prices and reals. The effect was large in economic terms – a 1 percent an indicator for rainfall-induced crop losses are also included increase in cash crop output per capita was associated with in the estimation. a 0.58 percent reduction in the poverty rate, on average. Once again this is in contrast to previous findings (Hill and Proximity to the markets of large towns, increasing use Tsehaye (2018)) which highlighted the relative importance of of improved seeds, and rising crop prices were the key cereals over cash crops in reducing poverty. The determi- drivers behind agriculture’s role in reducing poverty be- nants of what drove the increase in agricultural output per tween 2000 and 2016. There has been a noticeable shift in capita will be explored shortly in this chapter. Increases in the drivers behind the relationship between agricultural growth crop prices may also have influenced individual decisions to and poverty reduction when data from 2016 are added to the farm with cereals versus cash crops. Gains accruing to net zonal panel. In the previous Poverty Assessment (World Bank producers can be seen by the fact that the crop price index (2015)) and in Hill and Tsehaye (2018), the most important coefficient is negative and statistically significant, in contrast explanatory variables for the equivalent regressions reported to studies covering the 1996 to 2011 period in which it was in Table 15 were proximity to a large town, growth in cereals smaller and not statistically significant. output per capita, and fertilizer use in good conditions. The CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 99 Table 16 AGRICULTURAL  GROWTH AND POVERTY REDUCTION ANNUALIZED PERCENTAGE CHANGE IN POVERTY RATE 1 2 3 4 Growth in agricultural output per capita: Close to large town -0.82** (0.33) Far from large town 0.08 (0.12) Cereal output per capita -0.13 (0.22) Cash crop output per capita -0.58** (0.26) Manufacturing output per capita -0.21 -0.07 0.03 0.03 (0.38) (0.11) (0.11) (0.11) Services output per capita -0.04 0.02 0.03 0.04 (0.17) (0.16) (0.16) (0.16) Proportion of land with improved seeds -0.14** -0.14** (0.06) (0.07) Proportion of land with fertilizer 0.02 (0.01) Fertilizer*bad conditions 0.02 (0.02) Fertilizer*good conditions 0.01 (0.02) Weighted crop price index -0.23* -0.23* (0.12) (0.12) Change in predicted rainfall-induced crop loss 0.00 0.00 (0.00) (0.00) Constant -0.04** -0.01 -0.00 -0.00 (0.02) (0.01) (0.01) (0.01) Observations 135 168 162 162 R-squared 0.10 0.49 0.48 0.48 Number of zones 50 50 49 49 Note: Zone fixed effects are included but not shown. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Source: Calculations based on data sources described in Box 3. 100 ETHIOPIA POVERTY ASSESSMENT 3.3 Drivers of agricultural growth The expansion of cash crops meant that there was a large increase in the share of zone-level output accru- Shifting the focus to agricultural growth rather than ing to cash crops rather than cereal crops. Growth in poverty reduction confirms that there were very large cash crop cultivation was especially strong for coffee, khat, movements into cash crop production between 2011 oil seeds such as sesame and nueg, and kocho. The zones and 2016. The analysis now moves away from using chang- that showed particularly strong shifts towards cash crops es in poverty as the dependent variable, and now uses the were in Oromia (Jimma, West Hararge and East Hararge), change in the log of agricultural revenue per capita at the SNNPR (Sidama) and Harari. Figure 49 shows changes in the zonal level. As a result of this change there are more years predominant type of crop revenue at the zonal level. This is that can be used for analysis, since there is agriculture data determined by which crop type had the higher revenue per for most years between 2000 and 2016, while there are only capita. Zones in green represent those in which predominant four years with poverty numbers over that period. The other revenue per capita switched from cereals to cash crops be- implication of this approach is that the analysis is now on tween 2011 and 2016. In the eastern parts of Oromia this shift an overall revenue effect across the entire distribution, rather was largely towards khat. Red zones represent a switch from than only on a poverty-reducing effect. cash crops to cereals, while zones in yellow did not change. Figure 49 STRONG  SHIFTS FROM CEREALS TO CASH CROPS IN THE WEST AND EAST OF THE COUNTRY Changes in the predominant kind of crop cultivation as measured by revenue per capita at the zonal level, 2011 to 2016 Source: Calculations from Agricultural Sample Surveys. CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 101 Within the category of cash crops there was a rapid price index resulted in a 0.7 percent increase in revenue from shift in the relative prices of khat and coffee. Figure 50 cereals. The corresponding number for cash crops, shown in shows an index of the price of khat divided by the price of column 6, was 1.39. It is likely that increases in crop prices coffee over the decade from 2007 to 2017, with 2007 as the induced farmers to cultivate more land, to work longer hours base. The dotted line shows the fitted linear trend over the se- in the field or to hire additional workers. As can be seen in the ries. The trend line indicates that the relative price of khat over final row of coefficients, expanding the area of cultivated land coffee grew by more than 40 percent over the ten years. This had very high returns, especially for cash crops. For these, may to a certain extent explain the switch in which cash crops a one percent expansion in the area of land cultivated was are produced in different parts of the country. West and East associated with a 2.6 percent increase in revenue per capita. Hararge zones, where khat is cultivated, both experienced There were important differences in the application of very sharp drops in poverty between 2011 and 2016. The fertilizer in bad versus good conditions. Applying fertiliz- chapter is agnostic about the direction of the relationship be- er to cereals in both good and bad conditions had a positive tween the shift to cash crops and poverty reduction in these impact on revenue growth (see column 2), though the effect zones, but they are outliers on both counts. was larger in good conditions. In contrast, column 6 shows The impact of crop price changes on revenue is about that the application of fertilizer on cash crops in bad condi- twice as large for cash crops that it is for cereals. Ta- tions is associated with a reduction in revenue from these ble 17 presents regression output on the drivers of reve- crops. Revenue from cash crops also appears to be more nue growth of cereals and cash crops separately. Column sensitive to drought conditions than revenue from cereals. 3 shows that a one percent increase in the weighted crop Figure 50 THE  RELATIVE PRICE OF KHAT OVER COFFEE JUMPED BY 40 PERCENT OVER A DECADE The ratio of khat prices to coffee prices between 2007 and 2017 2 1.8 1.6 Price ratio (2007=1) 1.4 1.2 1 0.8 0.6 0.4 07 08 09 10 11 12 13 14 15 16 17 Year Source: Figure provided by IFPRI. 102 ETHIOPIA POVERTY ASSESSMENT Table 17 DETERMINANTS  OF REVENUE GROWTH IN CEREALS AND CASH CROPS 2000 TO 2016 VARIABLES: CHANGE OR GROWTH IN 1 2 3 4 5 6 REVENUE GROWTH: CEREALS REVENUE GROWTH: CASH CROPS Rainfall-induced crop loss -0.005*** -0.006*** -0.001 -0.017* -0.018** -0.002 (0.002) (0.002) (0.002) (0.009) (0.009) (0.008) Proportion of land -0.039 -0.040 -0.020 -0.149 -0.151 -0.078 with improved seeds (0.039) (0.039) (0.034) (0.176) (0.176) (0.159) Proportion of land with fertilizer -0.039 -0.040 -0.020 -0.149 -0.151 -0.078 (0.039) (0.039) (0.034) (0.176) (0.176) (0.159) Land with fertilizer in bad conditions 0.116*** 0.040 -0.136 -0.426** (0.044) (0.039) (0.199) (0.183) Land with fertilizer in good conditions 0.170*** 0.108** 0.112 -0.122 (0.050) (0.043) (0.223) (0.203) Crop prices 0.686*** 0.688*** 0.700*** 1.341*** 1.349*** 1.393*** (0.034) (0.034) (0.029) (0.153) (0.154) (0.138) Area of land cultivated 0.679*** 2.565*** (0.061) (0.289) Constant -0.076*** -0.076*** -0.046** -0.316*** -0.319*** -0.205* (0.026) (0.026) (0.022) (0.116) (0.116) (0.105) Observations 377 377 377 377 377 377 Number of zones 38 38 38 38 38 38 (0.02) (0.02) (0.02) Note: Zone fixed effects are included but not shown. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Given the focus on agricultural production not all zones are included. Source: Calculations from Agricultural Sample Surveys and LEAP data. CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 103 CONCLUSION This chapter focused on what the key drivers behind manufacturing or services – was very limited over the last Ethiopia’s recent poverty reduction have been. House- period. The role of this factor is likely to increase in the future. hold survey data was used to decompose this poverty Increases in endowments do a better job of explain- reduction into contributions by sector, before the roles of en- ing poverty reduction than increases in the returns to dowments and returns to these endowments was analyzed. those endowments. Within endowments, the effect of in- The chapter then switched focus to a zonal panel in order creasing educational attainment was uniformly important for to assess the contributions to poverty reduction of overall urban households. The education endowment effect was growth, sectoral growth, agricultural productivity, and invest- smaller, but still positive for rural households, in particular the ments in social programs and public infrastructure. best off households in rural areas. In rural areas the asset Rural areas were responsible for most of the recent accumulation effect was particularly large for the bottom 40 poverty reduction in the country, but the contribution percent of households, while the top quarter of the distribu- of urban areas is increasing and is likely to continue tion experienced the highest returns to these assets. to grow. One third of poverty reduction between 2011 and Poverty fell fastest in the zones that had the strongest 2016 was attributable to poverty reduction in urban areas, agricultural growth between 2000 and 2016, while the representing a large increase in its overall contribution. At the expansion of the PSNP between 2011 and 2016 also same time, the share of agriculture in explaining poverty re- played a crucial role in reducing poverty rates. Improved duction stood at two thirds between 2011 and 2016, down access to large towns, as measured through decreased from three quarters in the 2000 to 2005 period. Growth of travel times, was associated with strong poverty reduction, agriculture will remain critical for poverty reduction, given its indicative of the complementary nature of agricultural and share of employment and GDP. The role of structural trans- non-agricultural growth. formation – shifts in the population out of agriculture and into 104 ETHIOPIA POVERTY ASSESSMENT Expansions in the use of improved seeds was an im- net producers, and will likely continue to do so, there are portant driver of increased agricultural production also potential losers from these changes. Policy should be and therefore poverty reduction, while the role of the nimble enough to ensure that the effects of rising prices on expanded use of fertilizer is less clear. With improved vulnerable households and parts of the population are effec- seeds being used on only about six percent of cultivated tively mitigated. land, there is scope for expanding its use in the future. Poverty reduction in Ethiopia over the last two de- There was a shift away from cereal crops towards cades has largely been based on public investments in cash crops between 2011 and 2016, and changes in agricultural, social protection and infrastructure pro- the output of cash crops had a larger poverty-reduc- grams. Given the recent shift in the country’s development ing effect than changes in the output of cereal crops. focus it is crucial that the growth of private markets, along The rapid increase of crop prices over the period meant that with their potentially poverty-reducing effects are made to revenue growth expanded rapidly, and this was particularly complement the progress that has already been made. true of cash crops. While this increase in crop prices helped CHAPTER III. DRIVERS OF POVERTY REDUCTION IN ETHIOPIA 105 106 ETHIOPIA POVERTY ASSESSMENT CHAPTER IV Household poverty dynamics and economic mobility The previous chapters showed that the poorest households in 2016 were worse off than the poorest households in 2011. Due to the cross-sectional nature of the HCES data, it was not possible to assess whether the households that were poor in 2016 were also the ones that were poor in 2011. This chapter uses data from a different survey, the Ethiopian Socioeconomic Survey (ESS), to examine consumption dynamics at the household level. The ESS is a longitudinal survey that interviewed a representative sample of households in rural areas and small towns of Ethiopia in 2012, 2014 and 2016. Larger towns and cities were also covered in the last two rounds of the survey. The ESS confirms that households that were poor in 2016 were on average poorer than in 2012. But these were not the same households. Households that were poorest in 2012 experienced the fastest rate of consumption growth, while consumption of households that were initially in the upper part of the distribution contracted between 2012 and 2016. While the large extent of upward mobility is a positive finding, the other side of the coin is that resilience was limited: About as many households escaped poverty as fell into it. Upward mobility was higher in towns and cities than in the rural hinterland. 58 percent of the rural population who were poor in 2012 had managed to escape poverty by 2016. This probability amounted to 62 percent in towns and 69 percent in cities. The risk of falling into poverty was also higher in rural areas: 26 percent of the non-poor population in rural areas had fallen into poverty by 2016, compared to 14 percent in towns and four percent in cities. Other factors positively influencing the probability to escape poverty are education of the household head and living in pastoral areas. Male-headed households, household with large dependency rates and households living in the drought-prone lowlands were more likely to fall into poverty. Exploiting the longitudinal nature of the data, we find that 16 percent of the Ethiopian population in rural areas and small towns were chronically poor over the 2012-2014-2016 period. An additional 31 percent experienced transitory poverty between 2012 and 2016. Taken together, almost half of the population experienced at least one spell of poverty between 2012 and 2016, reflecting the high extent of consumption variability in rural Ethiopia. Chronic poverty is mainly concentrated in SNNPR, while transitory poverty is highest in Amhara. Relative to the transitory poor, the chronic poor have larger households and dependency rates, less land and fewer assets, less education, and are more likely to be living in the moisture-reliable highlands. The chronic poor are more likely to benefit from the Government’s social protection programs. CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 107 Introduction The first two chapters in this Poverty Assessment have final round of data. Transitions amongst the urban population made the point that although consumption growth in were less dramatic, albeit over a shorter time period, with Ethiopia has been strong over the last decade, not just under 4 percent of the urban population entering poverty everyone has benefited, with the poorest 15 percent between 2014 and 2016. of people not experiencing real consumption growth. The longitudinal nature of the Ethiopia Socioeconomic While cross-sectional datasets allow us to see the parts of Survey (ESS) data means that households can be sep- the distribution in which growth has been relatively stronger arated into those that were chronically poor, and those or weaker, they are unable to inform us whether the initial- that experienced transitory poverty. The importance of ly worse off gained or lost, or whether the initially better off this for policy purposes is highlighted in Lipton and Raval- gained or lost. In other words, while we do know that the lion (1995), who note that an appropriate policy response to bottom decile did not become better off since 2005, we do chronic poverty would focus on increasing the attainment of not know whether this decile consists of the same people and returns to the assets (both human and physical) of the in different survey years. With longitudinal data we are able poor, while transient poverty would be better tackled through to tackle this question. Exploiting the time dimension of the initiatives that focus on insurance and income stabilization. data allows us to see who escaped poverty and why, who Accordingly, conclusions about longer run welfare depend on remained trapped in poverty and why and, crucially, what the how much mobility is present over time in society. profiles are of those who were chronically poor. This kind of approach is potentially particularly useful for informing policy This chapter is divided into several sections. The first actions in the future. describes changes in the welfare of a panel Ethiopian house- holds over time using ESS 2012, 2014 and 2016 data. Along with the changes in the poverty rate in rural The chapter then focuses on the kind of poverty that these parts of Ethiopia between 2012 and 2016, came a lot of households faced over the period and outlines the character- upward and downward economic mobility.37 60 percent istics and differences of households in chronic poverty and of those who lived in rural areas and were poor in 2012 were transitory poverty. The final section analyzes the dynamics of measured as non-poor in 2016, but more than a quarter who welfare in Ethiopia, with particular attention being paid to the started out as non-poor were measured as being poor in the determinants of poverty exit and entry. 2. DESCRIBING HOUSEHOLD CONSUMPTION DYNAMICS Chapter I showed that household consumption growth consumption level in 2012, and then comparing each house- was highest at the upper parts of the distribution and hold’s consumption in 2016 to its consumption in 2012, that the bottom 10 percent did not grow in real terms. regardless of where in the final distribution the household Due to the cross-sectional nature of the HCES, it was not ended up. Holding the initial ordering of households constant possible to ascertain whether this pattern was driven by in this way shows that annual growth rates in consumption poor household remaining poor, or by a large extent of mo- were highest for those who were initially the worst off. Growth bility across the consumption distribution. The longitudinal rates were in fact positive, on average, for those who were in ESS data suggests the second explanation: Consumption the bottom half of the 2012 distribution. Growth rates turned levels of the baseline poor grew relatively faster than those negative for those who were in the 6th decile and above in of the non-poor. Figure 51 is a simplified version of the full 2012, and are -14 percent for those who were initially the non-anonymous GIC that is given as Annex Figure 10. It was best off. created by ordering the rural and small-town population by 37 It is important to note that the poverty rates referred to in this chapter do not correspond to the official national poverty rates that are issued by Ethiopia’s Central Statistical Agency, which are estimated using a non-comparable consumption expenditure module in the Household Income Consumption and Expenditure Survey (HICES). 108 ETHIOPIA POVERTY ASSESSMENT Figure 51 THE  BASELINE POOR GREW FASTEST BETWEEN 2012 AND 2016 Growth rate of consumption conditional on decile in 2012 (non-anonymous quasi-GICs) 40% 34% 30% Annual percentage change 20% 11% 10% 8% 2% 1% 0% -1% -4% -5% -10% -8% -14% -20% Poorest 2 3 4 5 6 7 8 9 Richest 2012 consumption deciles Source: Own calculations from ESS 2012, 2014 and 2016. The consumption growth rates for the bottom three Another way of thinking about changes over the distri- deciles were high, but they grew from a very low base. bution is to compare the proportion of households that The absolute year-on-year increase in consumption for many experienced increases in consumption by baseline of the poor was not enough to lift them over the poverty line. decile. The outcome of this exercise is shown in Annex Fig- In other words, the consumption growth of the baseline poor ure 11. The share of households that experienced consump- was impressive in percentage terms but was modest in real tion increases between 2012 and 2016 fell monotonically ETB terms (Figure 52). On average, real growth was ETB over the deciles – from 83 percent in the poorest decile to 569 a year for the poorest 10 percent of rural Ethiopians, only 7 percent in the richest decile. Only in the poorest three and this decreased to a very small ETB 47 a year for the 5th deciles did more than half of households see an increase in decile. The drop in consumption for the richest 10 percent real consumption. was very significant. CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 109 Figure 52 THE  GROWTH IN CONSUMPTION OF THE POOREST WAS SMALL IN ABSOLUTE TERMS Year-on-year changes in consumption levels by consumption decile in 2012 – rural and small-town households only 569 500 320 287 81 47 -48 Annual change in ETB -500 -277 -365 -797 -1500 -2500 -3500 -4011 -4500 Poorest 2 3 4 5 6 7 8 9 Richest 2012 consumption deciles Source: Own calculations from ESS 2012, 2014 and 2016. 3. POVERTY STATUS AND TRANSITIONS, 2012-2016 There was a lot of mobility in and out of poverty in Table 18 TRANSITION MATRICES - END-YEAR  Ethiopia. Table 18 shows 2016 poverty status conditional POVERTY STATUS CONDITIONAL ON on 2012 status. Transitions into poverty were far more likely BEGINNING-YEAR POVERTY STATUS in rural areas than in small towns or urban areas. 26 percent of the rural non-poor in 2012 had entered poverty in 2016, Rural 2016 compared to 14 percent of those living in small towns. The Non-poor Poor same holds true in reverse, with rural areas having the low- Non-poor 74.6 25.5 100 2012 est conditional poverty escape rates of the three localities. A Poor 57.8 42.2 100 very high 58 percent of the initially poor in rural areas were non-poor in the final year, compared to almost two thirds Small towns 2016 in small towns, and 69 percent in urban areas. This overall Non-poor Poor high level of economic mobility echoes the findings of other Non-poor 86.3 13.7 100 2012 studies using ESS data in Ethiopia, the results of which are Poor 62.2 37.8 100 summarized in Box 12. Urban 2016 Non-poor Poor Non-poor 96.1 3.9 100 2014 Poor 69.3 30.7 100 Source: Own calculations from ESS 2014 and 2016. 110 ETHIOPIA POVERTY ASSESSMENT Box 12 F  indings from other studies of poverty dynamics using the ESS The main aim of a paper by Seff and Jolliffe (2016) is to use the first two rounds of ESS data to investigate the dynamics of multidimensional poverty in Ethiopia. These dynamics are then compared to the dynamics of money-metric (consumption expenditure) poverty. The authors find that there is very little overlap between those households that are in the bottom of the multidimensional poverty (MDP) distribution, and those who are at the bottom of the money-metric distribution. The rates of multidimensional poverty in the first two rounds of the ESS are far higher than the rates of consumption poverty, and changes in household consumption have a very low correlation to changes in multidimensional wellbeing. The analysis reveals that nonmonetary measures of poverty are better at picking up the shocks that households experience compared to money-metric measures of wellbeing. For policymakers, this implies that changes in both money-metric and multidimensional welfare should be considered carefully when designing programs. An analysis of the factors driving transitions into and out of multidimensional poverty reveals that the most important roles are played by education variables, access to an improved water source, and the ability to accumulate assets. Kafle, McGee et al. (2016) also use the first two rounds of the ESS data to explore both consumption and asset-based poverty in Ethiopia. Similar to Seff and Jolliffe (2016) this means that the analysis is restricted to ru- ral areas and small towns. The authors set the consumption poverty line at the 30th percentile of the consumption distribution during the first round of the ESS. This choice of poverty line is slightly higher than the one that is used in this chapter. An asset-based poverty line is derived analogously, based on the distribution of assets in the first round of the ESS. The analysis reveals that although the cross-sectional measures of consumption poverty were stable in both periods, there was a lot of economic mobility into and out of poverty. These large shifts into and out of poverty were not reflected as dramatically when an asset-based measure of welfare is used. A more detailed investigation of spending patterns shows that there was a general shift towards higher expenditure on relatively more nutritious food items, and away from staples. Like Seff and Jolliffe (2016) the authors of this paper find that consumption and asset-based poverty measures are largely uncorrelated. Asset-based poverty rates, however, fell more steeply between the two rounds of the ESS than consumption poverty did. Many of the more recent studies of poverty dynamics in Ethiopia build on early work by Dercon and Krishnan (2000) who use three rounds of data from 1 450 rural Ethiopian households and finds that although cross-sectional poverty estimates are rela- tively stable, there is a great deal of variation in household consumption that is driven by seasonality effects. The implication is that cross-sectional estimates of poverty that ignore the seasonal dimension of consumption data are likely to underestimate vulnerability to poverty. A more comprehensive picture of poverty dynamics the population. There are a number of ways in which repeat- emerges if the second wave (2014) of the ESS is added ed observations of welfare can be used to define various to the analysis. Now, instead of having four possible pov- inter-temporal categories of poverty and non-poverty. Box erty state combinations (PP, PN, NP, NN)38 there are eight 13 outlines some of these and provides the definitions of possible combinations. These can be used to identify the chronic and transitory poverty that are used in the remainder kind of poverty that was experienced by different groups in of this chapter. 38  P=poor; N=non-poor. CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 111 Box 13 Defining chronic and transitory poverty  A number of definitions for distinguishing chronic and transitory poverty have been proposed in the literature. Two of the most commonly-found ones are the spells approach (McKay and Lawson (2003)) and the components approach (Jalan and Ravallion (1998). The spells approach defines chronic poverty according to how many times (or how long) a household or individual has been below the poverty over a particular period of time. For example, a household could be defined as being in chronic poverty if it is poor in at least two out of the three waves of ESS. The components approach involves estimating the chronic and transitory components of some measure of permanent welfare. The fluctuating nature of household welfare over time is thought of as containing both a transitory compo- nent and a permanent component. The transitory component is generated by variability in household consumption levels, while the permanent component gives the poverty level if consumption does not stray from its average value. One application of the components approach is to think about the classification of poverty into chronic and transito- ry groups over three waves of data by defining several characterizations (adapted from Hulme and Shepherd (2003): ● Always poor: Consumption expenditure is below the poverty line in all three rounds of ESS. ● Usually poor: The average of consumption expenditure over the three rounds of ESS is below the poverty line, but the household is not poor in all three rounds. ● Occasionally poor: The average of consumption expenditure over the three rounds of ESS is above the pov- erty line, but the household is poor in at least one round. ● Never poor: Consumption expenditure is above the poverty line in all three rounds of ESS. In this chapter chronic poverty status is assigned to households/individuals who are always poor or usually poor. Transitory poverty is associated with those household/individuals who are occasionally poor. The share of the rural and small-town population that There were large regional differences in the extent of was in chronic poverty was about half of the share that chronic poverty.40 Chronic poverty rates were highest in was in transitory poverty between 2012 and 2016. 16 SNNPR and Amhara, at 28 percent and 20 percent respec- percent of the population in rural areas and small towns was tively.41 6 percent of the population in Oromia was in chronic chronically poor, according to the definition outlined in Box poverty between 2012 and 2016, while the share in Tigray 13. This was significantly lower than the 31 percent who was 8 percent (Figure 53). Rates of transitory poverty were experienced transitory poverty over the period. The high more evenly spread through the different regions and ranged mobility rates described earlier in this report go some way to between 24 percent in Oromia and 43 percent in Amhara. explaining why the transitory component is so high for these households.39 39 Fuje (2018) constructs a different measure of chronic poverty by defining this category as containing only those who were poor in all three rounds of data. This yields lower chronic poverty rates of 12 percent in rural areas, and 6 percent in small towns. The corresponding transitory poverty rate for rural areas in Fuje (2018) is 20 percent. 40  In this context, the region that a household is located in is the region that was recorded in 2012. 41 Earlier chapters showed that the poverty rate in rural SNNPR was around 22 percent in 2016. The fact that the chronic poverty rate in the rural parts of the region is 28 percent reflects the fact that while HCES offer a snapshot of welfare, chronic poverty is calculat- ed over a longer period of time. The average consumption level of a household over time can still be below the poverty line even if the 2016 consumption level is above the line. 112 ETHIOPIA POVERTY ASSESSMENT Figure 53 CHRONIC  POVERTY WAS HIGHEST IN SNNPR, WHILE TRANSITORY POVERTY WAS HIGHEST IN AMHARA Chronic and transitory poverty over ESS 2012 to ESS 2016 – rural and small-town households Chronic poor Transient poor Never poor 100% 90% 80% 37.2 40.9 52.8 53.3 70% 58.2 70.1 60% 50% 43.0 31.5 40% 30% 31.4 29.8 33.7 20% 23.6 27.6 10% 19.9 17.0 15.7 6.2 8.1 0% Rural and small Amhara Oromia SNNPR Tigray Others towns Source: Own calculations from ESS 2012, 2014 and 2016. Figure 54 breaks down the composition of poverty The regional differences in the nature of poverty over type by region. This is in contrast to Figure 52 which took time are highlighted once again in Figure 55. Given that each region and then broke down region-specific poverty there are three time periods and two possible poverty states types. Decomposing the distribution of chronic and transi- in each period, there are 8 possible poverty and non-poverty tory poverty in this way allows for a better understanding of configurations ranging from poor in all three periods (PPP), to where the majority of the chronically poor are concentrat- non-poor in all three periods (NNN). Overall, just 5 percent of ed. 45 percent of all the chronically poor in rural areas and the rural and small town population was poor in all three time small towns live in SNNPR. The share living in Amhara is also periods. This number, however, hides variations by region high and stands at 31 percent. These substantially outweigh which range from 9 percent in SNNPR to 2 percent in Oro- the population shares of SNNPR and Amhara which are 20 mia. The shares of this in Tigray and Amhara that were poor percent and 23 percent, respectively (shown in Table 9). 15 in all three rounds were 3 percent and 6 percent, respectively. percent of the chronically poor live in Oromia, while around 5 Almost three quarters of the population in Oromia was percent each are in Tigray and other regions. non-poor in 2012, 2014 and 2016, compared to just 37 The dynamic changes somewhat when the focus percent in Amhara. The share of the population that was shifts to transient poverty, with Amhara taking up the non-poor in 2012 and 2014 but fell into poverty in 2016 (the largest overall share of this category. 34 percent of the NNP category) was very evenly across regions, at between transient poor in Ethiopia live in Amhara, which is a similar 12 and 14 percent. There were fairly substantial proportions share of the chronically poor. The share of the transient poor of the population that escaped poverty after 2012 and re- living in Oromia is 28 percent – almost double the chronic mained non-poor (PNN). The shares in the PNN category share. In contrast, one quarter of the transient poor live in were 14 percent in Amhara, 11 percent in Tigray, 4 percent SNNPR, compared to 45 percent of the chronically poor. in Oromia, and 6 percent in SNNPR. CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 113 Figure 54 MOST  OF THE CHRONIC POOR LIVE IN SNNPR AND AMHARA Regional shares of each poverty category Chronic poor Transient poor Never poor 0 10 20 30 40 50 60 Amhara Oromia SNNPR Tigray Others Source: Own calculations from ESS 2012, 2014 and 2016. Figure 55 THE  LIKELIHOOD OF A HOUSEHOLD BEING POOR IN ALL 3 ROUNDS IS HIGHEST IN SNNPR Regional shares of each poverty category 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Non-urban Amhara Oromia SNNPR Tigray Others PPP PPN PNP NPP PNN NPN NNP NNN Source: Own calculations from ESS 2012, 2014 and 2016. 114 ETHIOPIA POVERTY ASSESSMENT Box 14 P  overty dynamics and measurement error Figure 53 showed that the prevalence of transient poverty was about double the prevalence of chronic poverty in rural areas and small towns in the three rounds of ESS data. One serious concern about the robustness of this finding, however, is that the high share is driven by measurement error pushing households above and below a poverty line in a particular period, rather than by real changes in consumption. Measurement error is likely to arise because it is difficult to measure all the variables that comprise consumption expenditure, along with imputing values for the household’s consumption of its own production (Baulch and Hoddinott, 2000). Furthermore, there is a significant burden placed on respondents who are asked to recall past expenditures. The sensitivity of consumption-based measures of welfare to measurement error are exacerbated when using longitudinal data. Even if measurement error is random, it exaggerates the real changes in consumption experi- enced by the household, resulting in an overestimation of poverty transitions (Glewwe and Gibson, 2005). As was noted in Box 12, Seff and Jolliffe (2016) find that there is a very low correlation between consumption poverty and multidimensional poverty in the first two rounds of the ESS. Furthermore, shocks are picked up more reliably in non-monetary than monetary measures of welfare. The correlation between monetary welfare over the three rounds of the ESS and the household’s asset index in round 1 is low. The correlation of average consumption over three rounds with a count asset index (a simple sum of the number of assets the household owns) is 0.31. The correlation between this welfare measure and an asset index derived using principal components analysis is about half of this. Another way of investigating this relationship is to construct a non-monetary poverty indicator so that the rate of asset poverty is the same as the rate of consumption poverty. The correlation of asset poverty to chronic monetary poverty is very low, at 0.13. This suggests significant measurement error in consumption. There are several ways in which the impact of measurement error on poverty dynamics can be estimated and possibly mitigated. An early example is a paper by Alderman and Garcia (1993) that regresses changes in assets on changes in incomes. If changes in incomes were nothing more than measurement error, then there should be no statistically significant relationship between changes in the asset and income variables. McCulloch and Baulch (2000) adopt an instrumental-variables approach in which consumption is used as an instrument for income. Regressions with and without the instrument are used to derive a noise-to-signal ratio of the welfare measure that is then used to scale down the variability of observed incomes. Glewwe (2012) and Aguero et al (2007) use health-related instrumental variables to estimate how much mobility is due to measurement error in Vietnam and South Africa respectively. Both find that at least 15 percent of reported mobility is due to measurement error. Although this chapter does not separately estimate the impact of measurement error on the estimation of transient poverty, assuming a similar contribution of 15 percent would imply that the true share of chronic poverty is closer to 20 percent than to the 15.7 percent shown in Figure 54. The implication is that the share of transient poverty reported in this chapter should be interpreted as an upper bound of the true share, while the share of chronic poverty should be interpreted as the lower bound of the true share. CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 115 4. PROFILING CHRONIC AND TRANSITORY POVERTY There are many significant differences in the compo- Somewhat surprisingly, chronically poor households sition and characteristics of chronic poor, transitory are not generally much remote than those households poor and never poor households. Table 19 presents dif- that did not experience poverty between 2012 and ferences in characteristics that are measured at the level of 2016. The average chronically poor household was located the household and the household head in 2012. The final about 14km from the nearest road, while the corresponding two columns of the table provide a test of the statistical sig- distances for the transitory poor and the never poor were nificance of the difference between the chronic and transitory 17km and 14km, respectively. As shown in Figure 56, zones poor, and the chronic and never poor, respectively. 42 in which chronic poverty exceeded transitory poverty did not appear to be significantly more remote than zones in which Chronically poor households in Ethiopia are larger, transient poverty dominated. Proximity to the nearest pop- contain more young children, and have higher de- ulation center of 20 000 people or more also did not vary pendency ratios than transitory poor households and too much between the groups. This proximity, measures in those that were never poor. The average household size the number of minutes taken to travel to the nearest town, of the chronically poor was 5.9, compared to 5.3 for the tran- decreased for all three groups between 2012 and 2016, and sitory poor, and 4.9 for the never poor. Differences between the differences between the groups are never significantly dif- the chronically poor and the other two categories were sta- ferent. The rural accessibility index was lowest for chronically tistically significant at the 1 percent level for this variable. This poor households at about 52, up from a score of 49 in 2012. difference is driven in part by the higher number of children aged under 14 in chronically poor households. These house- There was a significant difference in the proportion of holds had, on average, 0.3 more children than the transitory chronically and transitory poor households that were poor, and 0.6 more that the never poor. Unsurprisingly, this PSNP beneficiaries in the first round of the ESS (19 also meant that dependency ratios were highest in chronical- percent and 14 percent). The relatively good targeting of ly poor households.43 the PSNP in this round can be seen by the fact that never poor households were far less likely to have been part of the Chronically poor households owned about a quarter of program in 2012. a hectare less land per adult than never poor house- holds, on average. Households that were in transitory pov- Almost half of chronically poor households reported erty between 2012 and 2016 owned about 0.43 hectares per being food insecure over the last 12 months, though adult on average, compared to the 0.57 hectares per adult only 9 percent were recipients of food aid. In contrast to owned by never poor households. Differences in the asset in- the targeting of the PSNP, it appears that more transitory and dex between these groups of households are also significant never poor households benefited from the scaling up of food in both economic and statistical senses.44 The normalized aid programs between 2012 and 2016 than chronically poor asset index score of the chronically poor was -0.5, compared households did. to -0.2 for transitory poor households and 0.1 for households that were never poor. 42 Fuje (2018), using different definitions of chronic and transitory poverty, finds that there are generally very few differences in the baseline characteristics of the chronic poor and the transitory poor. 43 The definition of the dependency ratio comes from Central Statistical Agency of Ethiopia (2017) and is, “…the population that is not of working age (<15 and >64) divided by total number of working age persons (15-64 years). The value is then multiplied to express it in percent. Households with no working persons were excluded in the dependency ratio computation.” 44 The asset index used in this table is a share index which is calculated by first multiplying an indicator variable (for example: house- hold owns a fridge) by the proportion of households that own the variable (for example: proportion of households that own a fridge). These products are then summed over each component at the household level to generate the share index. The components of the index are: Refrigerator, sewing machine, radio, bicycle, car, cellphone, television, electric stove, kerosene stove, sofa, wardrobe, mattress, animal-pulled card 116 ETHIOPIA POVERTY ASSESSMENT Households that were in chronic poverty were less a member who had no education, compared to around 57 likely to have had a female head than those that were percent of never poor households. Although secondary and in transitory poverty or never poor, and more likely to post-secondary education attainment levels were very low in have a head with no education. 22 percent of never poor Ethiopia, 4 percent of never poor households were headed households were headed by a woman, while just over 17 by someone with a secondary or post-secondary education, percent of chronically poor households were. In contrast, compared to 0.4 percent of the chronically poor, and 2.4 per- 70 percent of chronically poor households were headed by cent of the transitory poor. Figure 56 CHRONIC  POVERTY IS CONCENTRATED IN THE SOUTH AND DOES NOT SEEM TO BEAR MUCH RELATION TO PROXIMITY OF MAJOR ROADS The locations of chronically poor and transient poor households in relation to major roads Chronic poverty > transient poverty Transient poverty > chronic poverty Major road Source: Own calculations from ESS 2012, 2014 and 2016. CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 117 Table 19 CHARACTERISTICS  OF THE CHRONIC POOR, TRANSITORY POOR AND THE NON-POOR (1) (2) (3) CHRONIC TRANSITORY NEVER POOR POOR POOR (1) VS (2) (1) VS (3) HOUSEHOLD Household size 5.9 5.3 4.9 *** *** Number of children <14 2.0 1.7 1.4 *** *** Dependency ratio 101.3% 88.5% 74.0% * *** Share of expenditure on food 82.3% 82.3% 81.2% Land per adult (hectares) 0.32 0.43 0.57 *** *** Asset index -0.5 -0.2 0.1 *** *** PSNP household in W1 (2012) 19.3% 14.2% 11.4% * *** Ever PSNP HH (W1 to W3) 25.2% 20.8% 19.0% * Not enough food last 12 months 47.4% 35.6% 27.8% *** *** Free food recipient W1 9.0% 8.9% 4.2% *** Free food recipient W3 9.3% 15.0% 10.4% ** Extension program W1 34.4% 30.7% 35.7% Extension program W3 45.5% 40.4% 47.2% HOUSEHOLD HEAD Age 43.8 46.0 44.4 ** Female 17.5% 18.2% 22.0% * No education 70.3% 71.5% 57.3% *** Primary 29.2% 25.7% 35.6% * Secondary 0.4% 2.4% 4.2% *** *** Post-secondary 0.0% 0.5% 2.9% ** *** ACCESSIBILITY Rural Accessibility Index 2016 52.18 56.47 59.23 ** Minutes to nearest town 2012 118 126 108 Minutes to nearest town 2016 106 111 103 Fare to urban center (Birr) 36.81 45.14 38.62 *** ACCESSIBILITY Moisture reliable highlands 74.0% 60.0% 61.9% *** *** Drought prone highlands 13.7% 26.7% 27.3% *** *** Moisture reliable lowlands 7.1% 5.3% 5.4% Drought prone lowlands 4.5% 6.1% 2.7% * Lowland pastoralist 0.6% 1.9% 2.6% *** *** Source: Own calculations from ESS 2012, 2014 and 2016, and from Rural Accessibility Index. 118 ETHIOPIA POVERTY ASSESSMENT 5. THE DYNAMICS OF POVERTY TRANSITIONS: EXIT AND ENTRY 5.1 The characteristics associated There are some useful insights that can come from plotting these different correlates together, not least with poverty transitions the fact that doing so can potentially uncover their Figure 57 and Figure 58 show the relationships be- ordering of importance. However, as noted in Dang, tween different household head and household char- Lanjouw et al. (2017), the major caveat is that there will be acteristics and poverty exit and entry transitions, overlap between different groups (for example, those with respectively. The dashed lines represent the average pov- higher education levels are more likely to live in urban areas erty exit and poverty entry rates of 57.9 percent and 24.8 where there is easier access to secondary and post-second- percent, corresponding to the combination of the rural and ary education). small-town transition matrices presented earlier in the report. Living in a household in which the head has attained The blue triangles above the line correspond to characteris- a secondary level of education is strongly associated tics that are associated with an above-average probability of with a higher probability of transitioning out of poverty. transition, while the red dots below the line are associated Over 80 percent of households with a head with second- with a below-average probability of transition. ary education that were poor in 2012 were no longer poor in 2016, compared to 50 percent of households in which the head had a primary level of schooling. The unconditional differences in the probability of exiting poverty were similar for male versus female headed households, with the latter having a slightly higher unconditional probability of escaping poverty. Households in Oromia and Tigray were most likely to exit poverty, compared to the overall rural and small-town average, while households in SNNPR were the least likely to (unconditionally) exit poverty between 2012 and 2016. Households that had participated in the PSNP at any stage between 2012 and 2016 were less likely to exit poverty than the overall average, as were round three food aid households. This is a finding that potentially bears out the effective targeting of the PSNP, as this is a pattern we would expect to see, especially is the baseline consump- tion levels of PSNP participants are very low. Unconditional poverty exit rates differed markedly by agro-ecological zone, with lowland pastoralist by far the most likely to have exited poverty between 2012 and 2016, and those living in the low- lands being the least likely. CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 119 Figure 57 HOUSEHOLDS  WITH SECONDARY-EDUCATED HEADS AND HOUSEHOLDS LIVING IN PASTORAL AREAS WERE MORE LIKELY TO ESCAPE POVERTY Poverty exit and unconditional transition rates for different characteristics 80 70 60 Percentage 50 40 30 Fe le u. ar u. Am u. SN ia ay 3 3 gh 3 d −p e lo d la e lo d N le er er PS NP st d t O a is PR r ee P W W an W n n n a a m ed d ed ha Ev Oth gr al la Lo ron wla a M m co ry e PS ro l l N e− xte od bl on or Ti gh w o y N E o ro relia nsi hi hi Se ma f pa nd e e l i ug ab ur ron Pr nd Fr ro reli p t− w ht e− h ug ur st st D oi D oi M M Household head Household Above average Below average Note: Dashed line is the average probability of exiting poverty of 57.91%. Source: Own calculations from ESS 2012, 2014 and 2016. The gender differences in poverty transitions are about 33 percent, compared to 22 percent in Tigray, 18 per- clearer when considering unconditional poverty entry. cent in Oromia, and 21 percent in the other regions. Lowland Individuals living in male-headed households were almost 7 pastoralists were by far the least likely of the agro-ecological percentage points more likely to have entered poverty be- groups to enter poverty, while those living in the drought- tween 2012 and 2016 than individuals living in female-head- prone lowlands were most likely to transition into poverty ed households. The strong effects of increasing educational over the period. attainment are once again present. Households in which the head had a secondary education entered poverty at a rate 5.2 Regressions of poverty exit and that was 8 percentage points lower than the average, while those in households in which the head did not have any edu- poverty entry dynamics cation entered poverty at a rate of about 27 percent. Similar In this final section of the chapter we exploit the lon- regional dynamics are present in this figure to the ones in gitudinal nature of the data by reporting the results of the previous figure. The probability of households in Amhara probit regressions for poverty exit and poverty entry and SNNPR entering poverty between 2012 and 2016 was between ESS1 and ESS3.45 Marginal effects from the probit 45 Full results can be found in the appendix, along with corresponding estimates for transitions between ESS 2012 and ESS 2014. 120 ETHIOPIA POVERTY ASSESSMENT Figure 58 HOUSEHOLDS  IN THE DROUGHT-PRONE LOWLANDS WERE MORE LIKELY TO FALL INTO POVERTY Poverty entry and unconditional transition rates for different characteristics 50 40 Percentage 30 20 Fe le N le . . Am . O a SN ia ay PR er er 3 gh 3 nd −p e lo d la e lo d PS NP 3 st d t du du u is r W W W n n pa an a a m ed ha th al gr la ug liab hla Lo ron wla M m Pr o e co ry e N PS ro l O Fr NP e− xte od bl on or w Ti y g ar E o ro relia nsi hi hi Se ma f nd Ev e e e l i e ur ron nd e −p r w ht e− M ght ur u ro st st D oi D oi M Household head Household Above average Below average Note: Dashed line is the average probability of entering poverty of 24.84%. Source: Own calculations from ESS 2012, 2014 and 2016. regressions are presented along with their 95 percent confi- the small number of poor secondary education households to dence intervals in Figure 59 and Figure 60. Marginal effects begin with. There was no difference in the probability of pov- to the left of the red vertical line are associated with lower erty exit between household heads with no education, and probabilities of poverty exit/entry, while those to the right are household heads with primary education. Table 19 showed associated with higher rates of exit/entry. that there were few differences between the chronically poor, the transitory poor, and the never poor in terms of distance to The largest effects associated with poverty exit were roads, population centers and markets. The regression out- the education level of the household head, and the lo- put confirms this, with the marginal effects of all three “dis- cation of the household. The other explanatory variables tance” variables being very close to zero, with small standard included in this specification are generally not statistically sig- errors. Households in the lowland pastoralist agro-ecological nificant at the 5 percent level. Households in which the head zone were, on average, about 30 percent more likely to exit had a secondary education (partial or completed) were about poverty than households in moisture-reliable highlands (the 31 percentage points more likely to exit poverty that house- base region), but there were no other statistically significant holds in which the head had no education (the base catego- zonal effects, while controlling for other variables. ry), though this was not particularly precisely measured, given CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 121 Figure 59 EDUCATION  AND LOCATION CORRELATED WITH THE PROBABILITY OF EXITING POVERTY Marginal effects associated with poverty exit between 2012 and 2016 Household Male head Primary edu. Secondary edu. No. of children Dep. ratio Land per adult (Ha.) Dist. to road Dist. to pop. cent. Dist. to market Household Extension program everpsnp Number of cows Access to bank/microlender Drought−prone highland Moisture−reliable lowland Drought−prone lowland Lowland pastoralist −40 −30 −20 −10 0 10 20 30 40 Marginal effect (percentage points) Figure shows marginal effect and 95% confidence interval. Base categories: Household head has no education; Household located in moisture−reliable highlands. Region and small town dummies included. Source: Own calculations from ESS 2012, 2014 and 2016. 122 ETHIOPIA POVERTY ASSESSMENT Figure 60 MALE-HEADED  HOUSEHOLDS WITH HIGH DEPENDENCY RATES AND SMALL LANDHOLDINGS MORE LIKELY TO FALL INTO POVERTY Marginal effects associated with poverty entry between 2012 and 2016 Household Male head Primary edu. Secondary edu. No. of children Dep. ratio Land per adult (Ha.) Dist. to road Dist. to pop. cent. Dist. to market Household Extension program Ever in PSNP Number of cows Access to bank/microlender Drought−prone highland Moisture−reliable lowland Drought−prone lowland Lowland pastoralist −30 −20 −10 0 10 20 30 Marginal effect (percentage points) Figure shows marginal effect and 95% confidence interval. Base categories: Household head has no education; Household located in moisture−reliable highlands. Region and small town dummies included. Source: Own calculations from ESS 2012, 2014 and 2016. The marginal effects of the agro-ecological in which into poverty. Once again, the differences in the distances the household was located are more strongly highlight- of households from roads, population centers and markets ed when considering the dynamics of poverty entry be- are not statistically different from zero.46 The differences be- tween 2012 and 2016. The marginal effects of the primary tween the agro-ecological zones are once again apparently, and secondary education categories for the household head with lowland pastoralists being far less likely to enter poverty, are negative, meaning that households in these categories on average, than the base category of households in mois- were less likely to enter poverty than the base category of ture-reliable highlands. In contrast, households in drought- a household head with no education. Once again, the very prone lowlands were the most likely to have entered poverty large standard errors of the secondary education catego- over the period – 20 percentage points more than the base ry are driven by the fact that there were relatively few sec- category. Of course, as with any poverty analysis, the results ondary education households that entered poverty over the are sensitive to the choice of the poverty line. Box 15 discuss- period. An additional hectare of land per adult is associated es how the results change when the line is shifted down and with a 7.5 percentage point power probability of transitioning then it is shifted up. 46  This is also true is distances are measured in block, for example less than 5km, more than 5km, or more than 10km. CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 123 Box 15  he sensitivity of poverty dynamics to the choice of the T poverty line Changing the analysis of dynamics using different poverty lines yields some interesting results. If the poverty line is set at the 20th, rather than the 25th percentile then the most significant driver of households exiting poverty was whether it participated in the extension program. Although this variable was positive in Figure 9, it was not statis- tically significant. The dynamics of poverty exit between the different agro-ecological zones are the same at this new poverty line, with pastoralist households far more likely to exit poverty than those living in any of the other four zones. There are also some interesting differences at this new lower poverty line when considering poverty entry. Al- though the male, secondary education, and dependency ratio coefficients are of a similar magnitude to those in Figure 10, the coefficient on land per adult is now even larger and is significant at the 1% level. Participation in the extension program is also negatively associated with transitioning into poverty – households that were participants were about five percentage points less likely to enter poverty over the period compared to non-participating households. Another option to test the sensitivity of the results is to raise the poverty line. If the line is set at the 40th, rather than the 25th percentile then there are some different results when considering the probability of households exiting poverty. While before there were very few significant variables, at a different poverty line things look different. The coefficient on land per adult is large and significant at the 1% level – each additional hectare of land per adult is associated with approximately a 25 percent higher likelihood of exiting poverty between round 1 and round 3. The coefficient on the variable measuring whether the household has access to a bank or a financial institution is also strongly associated with a higher probability of exiting poverty. Interestingly, at this higher poverty line, the significant effect of some of the policy variables (extension program and PSNP) is no longer statistically significant. In addition, there are no different effects in the probability of exiting poverty between the five agro-eco- logical zones, in contrast to the clear patterns that emerged at a lower poverty line. With the poverty line set at the 40th percentile there are also some different results when considering the probability of a household entering poverty. The coefficient on the land per adult variable is negative and large – about twice as large as was the case in Figure 10. At this poverty line households that are part of the extension program are 9 percent less likely to enter poverty than non-participating households, on average. Finally, and in con- trast to the poverty exit estimates, there are large differences between the agro-ecological zones, with pastoralist households being the least likely to enter poverty, and households in the drought prone lowlands the most likely to have become poor over the time period. Overall then, there are four variables that seem to be consistently associated with both a higher probability of exiting poverty and with a lower probability of entering poverty: ● Whether or not the head of the households has a secondary education ● Higher land per adult in the household ● Whether or not the household was a participant in the extension program ● Whether or not the household has access to a bank or financial institution 124 ETHIOPIA POVERTY ASSESSMENT Conclusion Households in rural areas and small towns did worse Chronically poor households tend to be larger, have than households living in urban areas between 2012 more children and less land per adult than transient and 2016. In particular, the losses experienced by the top poor and never poor households. Interestingly, chronically decile of households in rural areas between the first and third poor households were not substantially more remote than rounds of the ESS were relatively large. Overall there was a households that were never poor, as measured by distance large amount of mobility as measured by the consumption to the nearest road or population center with 20 000 people expenditure of households. Although only 10 percent of rural or more. The expansion of the extension program between households were observed as being poor in both 2012 and 2012 and 2016 is evident, with 34 percent of chronically poor 2016, almost one third experienced some kind of transition households covered in the first round of the ESS, and 46 into or out of poverty over the same period. percent covered in the third round. Although almost half of chronically poor households reported having experienced Most of the longer-term poverty in Ethiopia is transient, food shortages over the last 12 months, only 9 percent re- according to ESS data, but there are notable large ported being recipients of food aid. shares of households that are trapped in chronic pover- ty. The proportion of the population in chronic poverty is par- The ESS data analysis confirms much of the HCES ticularly high in the SNNPR region, in which nearly 30 percent data analysis presented in the previous chapters. Both of households were considered to be chronically poor. The the ESS and HCES show that (i) chronic poverty (called the share of the population in chronic poverty was lowest in Oro- “ultra-poor” in Chapter 1 and 2) is mainly concentrated in mia and Tigray regions. Overall, most of the chronically poor SNNPR and transitory poverty in Amhara, (ii) households in in the country live in the SNNPR region, followed by Amhara. pastoral areas have done relatively well, and (iii) households Most of the transient poor households are located in Amhara. in the drought-prone lowlands have fared relatively poorly. Both datasets also show that female-headed households have higher consumption levels and were less likely to fall into poverty. Finally, they also agree on urban areas perform- ing better than rural areas between 2011 (or 2012) and 2016. CHAPTER IV. HOUSEHOLD POVERTY DYNAMICS AND ECONOMIC MOBILITY 125 126 ETHIOPIA POVERTY ASSESSMENT CHAPTER V Urban Poverty in Ethiopia Strong consumption growth in urban Ethiopia led to an 11-percentage-point reduction in poverty between 2011 and 2016. The reduction in poverty happened across towns and cities of all sizes, though was mainly driven by small and medium-sized towns given their large contribution to the total urban population. Overall, poverty rates decrease with city size, with the exception of Addis Ababa which has a relatively high poverty rate. Households with little-educated heads working in trade, services or agriculture contributed most to urban poverty reduction. In addition, the rising education levels of the urban labor force contributed substantially to the reduction in urban poverty. Urban poverty reduction mostly took place within each industry rather than being driven by workers’ mobility across industries, pointing to limited structural transformation of employment. The main contributor to urban poverty reduction is the trade and service sector, while agriculture plays an important role in small towns, where the labor force is less educated and keeps close links with the rural hinterland. Consumption gains among households with self-employed heads and other employed household members have pushed down urban poverty. Households with a self-employed head accounted for 46 percent of the urban population and 53 percent of the reduction in poverty. Poverty reduction among these households was especially strong if other household members also took up self-employment. The importance of additional self-employment (additional in the sense of being led by a household member other than the head) holds regardless of the occupation of the head: Poverty reduction was strongest if other household members transitioned into self-employment, regardless the occupation of the household head. Though small towns have been important for urban poverty reduction, access to key services and amenities is not keeping up. The level of access to basic services is particularly low among low-income households in small towns. Thanks to large investments, access to services and amenities has been improving in pace with the urban growth in major towns. Investments in small towns will be required to improve living standards and reduce migration pressure on the bigger cities. The economic integration of rural migrants depends on city size. Recent migrants (less than three years) have substantially worse employment outcomes in Addis Ababa but not in large towns/secondary cities. In Addis however, migrants seem to catch up with the resident population as they stay longer in the city. Social integration seems to be more difficult: The education level of children of rural migrants is substantially worse than that of the resident population of the same age, even for migrants that have been in the city for long. CHAPTER V. URBAN POVERTY IN ETHIOPIA 127 Introduction Despite being one of the least urbanized Sub-Saharan Urban population growth will take place mainly in small countries, Ethiopia’s urban population has been grow- towns and secondary cities. Natural population increase ing fast. According to the latest rounds of HICES, Ethiopia’s accounted for the largest part of urban population growth in urbanization rate—the share of the country’s population in Ethiopia until recently (World Bank 2015). However, rural to urban areas—rose from 16.6 percent in 2011 to 19.1 percent urban migration is expected to outpace natural increase as in 2016.47 While its urbanization level is still among the lowest of 2018, contributing to more than 40 percent of urban pop- in Sub-Saharan Africa (Figure 61).urban population growth is ulation growth. Between 2015 and 2025, around 5 million rapid: The urban population increased by 6.2 percent annu- people are projected to be added in small towns with a pop- ally since 2011, which is much faster than rural population ulation of less than 50,000 (Schmidt et al. 2018). Secondary growth of 2.7 percent. This means that nearly 1 million peo- cities with a population of greater than 100,000 (such as the ple are added to the urban population every year. It is esti- regional capitals) will also grow at the similar scale, adding mated that there are 45 cities of at least 50,000 population 5.7 million people between 2015 and 2025. In the meantime, as of 2015 (Schmidt et al. 2018). Ethiopia’s urban population the contribution of Addis Ababa to the overall urban popula- is projected to reach 42 million by 2032 and its population tion will decline (though it will remain by far the biggest city). share to hit 30 percent by 2028 (World Bank, 2015). Figure 61 ETHIOPIA  REMAINS UNDER-URBANIZED Urbanization rates in Sub-Saharan African countries, 2010-2016 60 Ghana 50 Nigeria Senegal Urbanization rate (%) 40 SSA Tanzania 30 Kenya 20 Ethiopia 10 0 2000 2005 2010 2016 Year Source: World Development Indicators. 47 The Central Statistical Agency (CSA) defines urban areas as localities that satisfy one of the following criteria: 1) localities with 2,000 or more inhabitants; 2) all administrative capitals of regions, zones, and woredas; 3) localities with at least 1000 people who are primarily engaged in non-agricultural activities; and/or 4) areas where the administrative official declares the locality to be urban (Schmidt and Kedir 2009). 128 ETHIOPIA POVERTY ASSESSMENT Figure 62 SMALL  TOWNS AND SECONDARY CITIES WILL ACCOUNT FOR THE BULK OF URBAN POPULATION GROWTH Urban population trends and projections, 2007-2035 52.6 50 5.81 40 20.242 31.1 Less than 50,000 Population (million) 30 50,000 to 100,000 4.561 5.593 100,000 to 500,000 9.225 20 17.5 Addis Ababa 11.9 3.273 3.284 3.488 10 2.74 1.782 20.929 2.276 14.002 1.139 8.916 5.708 0 2007 2015 2025 2035 Source: Schmidt et al. (2018). Figure 63 HALF  OF THE URBAN POOR ARE LOCATED IN OROMIA AND ADDIS ABABA Regional share of the urban poor, 2016 Other regions, 5% Tigray, 7% Somali, 7% Oromia, 29% Amhara, 15% Addis Ababa, SNNPR, 15% 22% Source: HCES; 2011, 2016. World Bank staff calculations. CHAPTER V. URBAN POVERTY IN ETHIOPIA 129 Urban poverty has declined rapidly between 2011 and This chapter takes a closer look at the rapid urban 2016. As presented in Chapter 1, the share of the urban poverty reduction between 2011 and 2016 and at the population living below the national poverty line decreased constraints to be addressed to further accelerate ur- from 26 percent in 2011 to 15 percent by 2016. The abso- ban poverty reduction. The chapter proceeds as follows: lute number of poor also declined, despite rapid urban pop- The next section examines where urban poverty has been ulation growth. In 2016 there were an estimated 2.4 million rapidly declining and where it has been persisting, by looking people below the poverty line in urban areas, the majority at regions and the population size of cities. It also examines of which were located in cities in Oromia region and Addis various living conditions by household consumption and city Ababa (Figure 63). size. The chapter then takes a closer look at developments in urban labor markets to further explore what factors have made cities conducive places for poverty reduction. 2. SPATIAL DIAGNOSTIC OF URBAN POVERTY, LIVING STANDARDS, AND LABOR MARKETS 2.1 A brief profile of the urban poor Ababa and the 21 major towns), and rural areas in each re- gion (CSA 2017). For the analytical purpose of this chapter, SMALL TOWNS ACCOUNT FOR ONE-THIRD OF THE towns are classified as follows: (i) Addis Ababa (including URBAN POOR all the 10 sub-cities), (ii) major towns with a population of at least 100,000 (not including the sub-cities in Addis Aba- Analyzing poverty across towns of different sizes is ba), (iii) medium sized towns with a population ranging from useful, given their distinct demographic and economic 20,000 to 100,000, and (iii) small towns with a population of structures. The sampling design of the HICES/WMS gives fewer than 20,000. Rearranged in this way, the HICES/WMS poverty and other key statistics representative at 10 sub-cit- 2015/16 includes Addis Ababa, 20 major towns, 52 medium ies in Addis Ababa, 21 major towns (including regional cap- towns, and 54 small towns. itals), 8 other urban areas (i.e., urban areas except for Addis 130 ETHIOPIA POVERTY ASSESSMENT Figure 64 TOWNS  AND CITIES IN ETHIOPIA Legend Addis Ababa Major towns (population greater than 100,000) Towns with population 50,000 to 100,000 Other towns Source: CSA (2013). Poverty rates remain relatively high in small towns and medium-sized towns—received funding for infrastructure Addis Ababa. The previous Ethiopia Poverty Assessment development from the Urban Local Government Develop- (World Bank 2015a) shows that poverty rates tend to be low- ment Project (ULGDP) between 2008 and 2014 (Box 17). er in cities with a larger population, except for Addis Ababa. The second phase of the ULGDP targets additional 25 major This association seems to have remained in 2015/16 (Table and medium towns between 2014 and 2019. 2). The poverty headcount ratios in towns of a population of Even within the same regions, poverty varies by city less than 20,000 (20 percent) are higher than other bigger size. In Somali region, for example, the poverty rate is only towns and cities. A third of Ethiopia’s urban poor population 5.9 percent in Jijiga but 27.3 percent in the rest of urban ar- lives in these small towns. In contrast, major towns with a eas (Figure 66). Similarly, the poverty rate in the capital town population of greater than 100,000 have relatively low pov- of Gambella region is only 10 percent, while other urban ar- erty rates, such as Mekele (9.4 percent), Dire Dawa (11.1 eas in the region still have 22.8 percent of the population percent), Gonder (10.1 percent), Adama (14.1 percent), and in poverty. In contrast, gaps in poverty rates between major Hawassa (8.0 percent) (Figure 66). Addis Ababa is an excep- towns and other urban areas are relatively narrow in some tion with a poverty rate (16.8 percent) being slightly higher regions, such as Afar and Amhara. than the urban average (14.8 percent). Among the 21 ma- jor towns, 14 towns—as well as Addis Ababa and some CHAPTER V. URBAN POVERTY IN ETHIOPIA 131 Table 20 POPULATION  AND POVERTY DISTRIBUTION IN URBAN ETHIOPIA POPULATION POOR POPULATION POVERTY MILLION SHARE RATE MILLION SHARE (1) (2) (3) (4) (5) Small towns (20,000 or lower) 4.3 25.1% 20.0% 0.9 33.8% Medium sized towns (20,000 - 100,000) 5.9 34.8% 12.1% 0.7 28.4% Major towns (100,000 or greater) 3.6 21.1% 11.4% 0.4 16.2% Addis Ababa 3.2 19.0% 16.8% 0.5 21.5% Urban total 17.0 100.0% 14.8% 2.5 100.0% Note: Population is based on the 2016 population projection by the CSA Source: Staff calculations based on HICES 2015/16. Box 16 Where is poverty in Addis Ababa concentrated?  Based on the geo-localization included in the 2016 HCES, it is possible to identify clusters of low- and high-income households in Addis Ababa. Though Addis Ababa is a socially and economically mixed city, there are a number of distinct low-income neighborhoods. The largest cluster of poorer households (the green dots) is found surrounding Merkato in Addis Ketema and Lideta (Geja Sefer neighborhood), while a second and smaller low-income cluster is found in the east, between Goro and Gurd Shola (Figure 65). Small high-income clusters (red dots) are scattered across the city. Future work will look in more detail at the interplay of location, housing, and poverty in Addis Ababa. Figure 65 LOW-  AND HIGH-INCOME CLUSTERS IN ADDIS A. Low- and high-income clusters B. Low-income only ! ! !! ! !! ! ! ! ! ! ! ! ! !! ! ! ! ! ! !! ! ! ! !! ! !! ! !! !! ! ! ! ! ! !!!!! ! !! !! ! ! ! !!! ! ! ! ! !! !! ! ! !! !! ! ! !! ! !! ! ! ! ! ! ! !! ! !! ! ! ! ! !! ! ! ! ! !! !! !! ! ! ! ! ! ! ! ! ! !! !! !! ! !! !! !! ! ! ! ! ! !! !! ! ! !!! ! ! ! ! ! !! !! ! ! !! ! ! ! !!! !!! ! ! ! ! ! ! ! !! ! ! ! ! ! !! !!! !! ! !! ! ! !! ! ! !! ! ! !!! !! !! ! ! ! ! ! !!! ! ! ! ! ! ! ! !! ! ! !!! ! ! !! !! ! ! ! ! !! ! !!!! ! !! ! ! ! ! ! ! ! ! !! ! ! ! ! ! !!!!! !! !! ! ! !! ! !! ! !!! !! ! ! ! !! ! ! ! ! ! ! ! !! ! ! ! ! ! ! !! ! ! ! !! ! ! !!! ! ! ! ! !! ! ! ! ! ! ! !! ! ! ! ! !! !! ! !! ! !! ! ! ! ! !! ! !! ! !! ! ! !! ! ! ! !! !!! ! ! ! !! ! ! ! ! !! ! ! ! ! !! ! ! ! ! ! ! !! ! ! !! ! !! ! ! !! !! ! ! ! ! ! ! ! !! ! ! ! ! !! ! !!! ! ! !! ! ! ! ! !!! !! ! !! ! ! !! ! !! !! ! !!! ! !! ! ! ! !!!! ! !! ! ! ! !! ! ! !! ! ! !!! ! ! ! !!!! ! ! !! ! !! !! ! ! ! ! !! ! !! ! ! ! ! ! !! ! !! ! !! !!! !! ! ! !! !! !! !! ! ! ! ! !! ! ! !!! ! ! !! !! !! !!! !!!!! ! ! ! !! !! !!!!! !! !!! !! ! ! !! ! ! !!!! ! ! ! ! ! ! !! !! !! !! ! !!! !! ! ! !! !! ! ! !! ! !!! !!! !!!! ! ! !! ! ! ! ! ! !! !! ! ! ! !!! ! ! !! !!! ! !! ! ! ! ! ! ! !! ! !! !! ! ! ! ! ! ! ! !!! ! ! !!! ! !! ! ! ! ! ! ! ! !! !! !! ! ! !! ! ! ! !! ! ! ! ! !!!! ! ! ! !! ! ! ! !! !! !! ! ! ! ! ! ! !! ! ! ! ! ! !!!! !! ! !! ! ! ! ! ! !! !!! !! !! ! ! !! ! !! ! ! ! !! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! !! ! ! !! ! ! ! ! ! ! ! !!! ! !! ! ! ! !! !! ! ! ! !! ! ! !! !! ! ! ! ! ! !! !! ! !! ! !! ! !! !! ! ! !! ! !! !! ! ! !! ! ! ! ! ! ! ! !!! ! ! ! ! !! !! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! !! ! ! !!! ! ! ! !!! ! ! ! ! !! ! ! ! ! !! !!! ! ! ! ! ! !! ! ! ! !! !! ! !! ! ! !! ! ! Source: HCES, 2016. Low-income clusters in blue. 132 ETHIOPIA POVERTY ASSESSMENT Figure 66 POVERTY  RATES ARE GENERALLY LOWER FOR BIGGER TOWNS THAN SMALLER ONES Poverty rates by city size and region, 2016 30% ● Major towns 25% ▲ Other urban Poverty headcount ratio Dila Town 20% Arba Minch Addis Ababa 15% Bahir Dar Adama Asayta Shasheme Sodo Dire Dawa Jimma Assosa Mekele Gonder Debre Zeite 10% Gambella Dessie Nekemte Hawassa Debrebrehen Jigjga Hosaena Harari 5% 0% Tigray Afar Amhara Oromia Somali Benshangul SNNPR Gambella Harari Addis Dire Dawa Ababa Region Note: Towns targeted by the infrastructure development under the ULGDP I (2008-2014) are shown in bold font. Source: HCES; 2016. World Bank staff calculations. Box 17 T  he Urban Local Government Development Project (ULGDP) The World-Bank funded ULGDP aims to address institutional and fiscal gaps at the urban local govern- ment level by supporting improved institutional performance in the planning, delivery, and sustained pro- vision of urban services and infrastructure by local governments. The first phase of the program (2008-2014) provided capacity building grants to 37 local governments (US$7.5 million) and the Ministry of Urban Develop- ment, Housing, and Construction (MUDHCo) and performance-based grants to 19 local governments for urban infrastructure investments (US$403 million). The second phase of the ULGDP aims to provides US$557 million grants to additional 26 local governments between 2014 and 2019. During the first phase, 2.6 million people ben- efited from the infrastructure and services financed under the ULGDP, as 870 km of non-rural roads were built, 647 km of drainage lines were constructed, and 131 improved community water points were created. CHAPTER V. URBAN POVERTY IN ETHIOPIA 133 POOR HOUSEHOLDS TEND TO BE HEADED BY Poverty incidence among the households with completed AN OLDER, POORLY-EDUCATED AND SELF-EM- primary is relatively high in Addis Ababa, implying that prima- PLOYED HEAD ry education is not good enough to economically fare in the capital city. Many of these households with low education Households whose heads are older are more likely to levels are older as well. be poor in urban Ethiopia, and this pattern is observed across towns of different population sizes. Young Household heads’ employment status is also cor- households in Addis Ababa are least likely to be poor (pover- related with poverty as public employees are the least ty rate of less than three percent), poverty incidence among poor, and the pattern is not very different across ur- older households in the city is a lot higher (over 20 percent). ban sizes. In urban Ethiopia, households whose heads are In urban Ethiopia, the poorest age group is older households public employees are less likely to be poor across towns of living in small towns. Overall, female-headed households are all sizes (Panel D in Figure 67). Poverty incidence is higher not significantly more likely to be poor than male headed among self-employed households than private employees. households in urban areas (Panel A in Figure 67), though the This pattern is observed in Addis Ababa, major towns, me- association depends on city size. The poverty rates among dium towns, and small towns, respectively. While household female-headed households are 3 points higher in Addis Aba- head’s employment status is clearly correlated with poverty, ba and medium towns, though this gap is not observed in the employment status of other household members also major towns and small towns. potentially influences their poverty situation. This will be ex- amined in detail in the next section. Household heads’ education level is strongly associat- ed with poverty in urban Ethiopia. The higher the educa- Poverty status also depends on the economic sector tion level of household heads, the less poor the households of the household head’s primary job. Regardless of city are (Panel C). Poverty incidence among the households with size, poverty incidence is lowest among the households with secondary or higher education is low in all places. House- public administration jobs (Panel E). Poverty is highest in agri- holds with no education or incomplete primary are most like- culture, manufacturing, and construction jobs, sectors which ly to be poor, especially in Addis Ababa and small towns. unskilled workers tend to engage in. 134 ETHIOPIA POVERTY ASSESSMENT Figure 67 URBAN  POVERTY BY HOUSEHOLD CHARACTERISTICS A. By household head’s sex B. By household head’s age 25 30 28.0 20.8 19.1 25 20 23.5 25 30 21.3 (%) rate (%) 16.2 19.7 (%) rate (%) 20 20.2 28.0 14.4 19.3 15 20.8 30 17.6 25 19.1 25 17.4 20 15.2 11.5 16.1 14.1 15 14.4 23.5 28.0 21.3 Poverty 20.8 13.6 13.7 Poverty 16.2 19.7 13.0 10 19.1 25 20 20.2 12.3 20 11.2 14.4 11.0 19.3 10.9 23.5 10.6 15 10 10.3 rate 17.4 21.3 17.6 rate rate (%) 16.2 19.7 rate (%) 15.2 11.5 20 16.1 14.1 15 6.9 19.3 20.2 6.7 5 14.4 14.4 Poverty 5 13.6 13.7 5.6 Poverty 15 13.0 17.4 12.3 17.6 10 15.2 16.1 11.5 11.2 11.0 15 10 2.9 10.3 10.9 10.6 14.1 14.4 Poverty 0 13.6 13.7 Poverty 0 13.0 12.3 10 6.9 6.7 5 Urban AA Major 11.2 Medium 11.0 Small 10 5 Urban AA Major 10.3 Medium 10.9 5.6 Small 10.6 Male Female 6.9 <25 2.925-45 45-60 6.7 >60 5 5 5.6 0 0 Urban AA Major Medium Small Urban AA2.9 Major Medium Small 0 0 Male Female 25 <25 25-45 45-60 >60 35 Urban AA Major Medium Small Urban AA Major Medium Small head’s C. By householdMale education level Female D. By household <25 head’s45-60 25-45 occupation >60 type 22.8 21.0 30 20 19.4 25 (%) rate (%) 17.6 (%) rate (%) 25 35 16.7 17.2 15 22.8 14.1 14.6 13.9 21.0 20 30 25 20 13.7 35 Poverty 13.8 19.4 12.7 12.3 22.8 Poverty 15 10 10.2 17.6 25 30 16.7 17.2 21.0 9.5 20 15 19.4 7.9 rate 10 6.9 7.6 14.6 rate 20 14.1 13.9 rate (%) 13.7 17.6 rate (%) 25 17.2 5.8 5 16.7 13.8 12.7 5.0 12.3 Poverty 5 15 Poverty 15 20 10 14.1 14.6 10.2 13.9 13.7 9.5 13.8 12.7 7.9 12.3 0 0 7.6 Poverty 10 6.9 Poverty 15 10 10.2 5.8 9.5 Urban AA Major Medium Small 5 Urban AA Major Medium 5.0 Small 5 7.6 7.9 10 No education Primary incomplete 6.9 Public employee Private employee Self-employment Unemployed 5.8 0 5 5.0 5 Primary complete Secondary incomplete 0 Urban Secondary complete Major AA Medium Post-secondarySmall Urban AA Major Medium Small 0 0 No education Primary incomplete Public employee Private employee Self-employment Unemployed Urban AA Major Medium Small Urban AA Major Medium Small Primary complete Secondary incomplete 35 No education Secondary complete Primary incomplete Post-secondary Public employee Private employee Self-employment Unemployed Primary complete Secondary incomplete 30 Secondary complete Post-secondary Agriculture and mining 35 25 E. 35 30 By household head’s economic sector Manufacturing (%) rate (%) Agriculture and mining 20 Construction 30 25 Agriculture and mining Manufacturing Poverty Trade 15 25 Manufacturing Construction 20 rate Infrastructure rate (%) 10 Poverty 20 Trade Construction 15 Services 5 Poverty Trade Infrastructure 15 Public administration 10 0 Infrastructure Services 10 5 Urban AA Major Medium Small Public administration Services 5 0 Public administration Urban AA Major Medium Small 0 Urban AA Major Medium Small Note: Poverty rates are calculated at the household level. Major towns have populations greater than 100,000 (excluding Addis Ababa); medium towns have populations between 20,000 and 100,000; and small towns have populations less than 20,000. Source: HCES; 2016. World Bank staff calculations. CHAPTER V. URBAN POVERTY IN ETHIOPIA 135 2.2 Dissecting urban poverty to urban areas. Urban poverty fell by 11 percentage points between 2011 and 2016, over half of which can be explained reduction by poverty reduction in medium-sized and small towns (sec- ond panel of Figure 69). At the national level, small and me- SMALL AND MEDIUM-SIZED TOWNS ACCOUNT FOR THE BULK OF URBAN POVERTY REDUCTION dium-sized towns contributed one percentage point to the overall poverty reduction of six percentage points. Population Poverty rapidly decreased across cities of all sizes. shifts across towns of different sizes accounted for only a Headcount ratios have declined by around 10 percentage limited proportion of poverty reduction, since the population points in Addis Ababa, major towns, medium towns, and shares across Addis Ababa, major towns, medium towns, small towns, respectively (Panel A in Figure 68). While poverty and small towns have changed only marginally since 2011 rates remain relatively high in Addis Ababa and small towns, (pointing to limited inter-city migration). When it comes to the they have reduced poverty at a similar pace as major and reduction in vulnerability in urban areas, the contribution of medium towns since 2011. Poverty gap and severity mea- small and medium towns is even higher (Panel C). Vulner- sures have also decreased across the board (Panel B). Over- ability here is measured by using a higher poverty line: The all, these results demonstrate that urban poverty reduction official poverty line multiplied by 1.5. The reduction in urban between 2011 and 2016 has been robust and widespread. poverty has been accompanied by a move further away from the poverty line (in the right direction), suggesting that urban Small and medium-sized towns account for over half households would be able to deal with shocks without nec- of urban poverty reduction. Chapter 3 showed that a third essarily falling back into poverty. of Ethiopia’s poverty reduction since 2011 can be attributed Figure 68 STRONG  URBAN POVERTY REDUCTION ACROSS CITY SIZE Poverty trends by city size Poverty headcount ratio Poverty gap/severity 30 9 28.1 28.2 8 7.9 25 Poverty headcount ratio (%) 23.7 7 7.0 Poverty gap and severity 6.5 20.7 19.3 6 20 16.8 4.9 5 5.1 15 4.1 11.4 12.0 4 3.3 3.2 10 3 2.5 2.5 2.7 2.0 2 1.9 5 1.4 1.3 1 0.9 0 0 Poverty rate Poverty rate Poverty rate Poverty rate Severity Severity Severity Severity Gap Gap Gap Gap AA Major towns Medium Small towns towns AA Major towns Medium towns Small towns 2016 2011 2016 2011 Note: Major towns have populations greater than 100,000 (excluding Addis Ababa); medium towns have populations between 20,000 and 100,000; and small towns have populations less than 20,000. Source: HCES; 2016. World Bank staff calculations. 136 ETHIOPIA POVERTY ASSESSMENT Figure 69 SMALL  AND MEDIUM TOWNS DRIVE MUCH OF THE REDUCTION IN URBAN POVERTY AND VULNERABILITY Decomposition of poverty changes by city size, 2011-2016 Percentage point change in headcount ratio 0 -1 -2 -3 -4 -5 -6 -7 (A) National poverty Within rural -4.0 Within Addis Ababa -0.4 major towns -0.3 Urban total: -1.7 Within medium towns -0.5 Within small towns -0.5 Population shift -0.2 Interaction -0.2 (B) Urban poverty Within Addis Ababa -2.4 Within major towns -1.8 Within medium towns -3.2 Within small towns -2.9 Population shift -0.5 Interaction -0.2 (C) Urban vulnerability Within Addis Ababa -3.4 Within major towns -3.4 Within medium towns -6.5 Within small towns -5.6 Population shift -0.6 Interaction -0.5 Note: Numbers were calculated by excluding the zones that were covered only in the 2016 data for the purpose of comparison between 2011 and 2016. Major towns have populations greater than 100,000 (excluding Addis Ababa); medium towns have populations between 20,000 and 100,000; and small towns have populations less than 20,000. Source: HCES; 2016. World Bank staff calculations. In addition, the spillover effects of towns on surround- Bahir Dar and Adama in the case study) potentially benefits ing rural areas, which is not considered in the decom- rural farmers by raising agriculture output prices, facilitating position analysis above, has likely contributed to rural the uptake of modern inputs among farmers, and increasing poverty reduction. A growing literature argues the for im- yields on farms (Vandercasteelen et al. 2018). Chapter 3 also portance of secondary towns for poverty reduction (Chris- showed that agricultural growth is more poverty-reducing in tiansen and Kanbur 2017; Christiaensen and Todo 2014). In areas that are closer to towns. addition to direct impacts through rural to urban migration (Beegle, de Weerdt, & Dercon 2011; de Brauw, Mueller, & MOST OF THE REDUCTION IN URBAN POVERTY Woldehanna 2017), towns can support rural poverty reduc- CAME FROM HOUSEHOLDS WITH LITTLE-EDUCAT- ED HEADS tion by improving agricultural productivity (through increased availability of modern inputs and reduction in input prices) Most of the urban poverty reduction took place among and raising the demand for agricultural products. A recent households with little-educated heads. The contribution study suggests that proximity to secondary towns (e.g., to urban poverty reduction between 2011 and 2016 mainly CHAPTER V. URBAN POVERTY IN ETHIOPIA 137 came from households whose heads did not complete pri- important contributor to urban poverty reduction is the “pop- mary education (2.6 percentage points, or 23 percent) or ulation-shift effect” (yellow bars in Figure 70), which proxies never enrolled in school (2.7 percentage points, or 25 per- the overall improvements in education levels in urban areas cent) (Figure 70). This pattern is observed across towns with since 2011. The effect of increasing education levels was different population sizes. Compared to households with particularly large in Addis Ababa and small towns, where the heads who did not complete primary education, households gap in poverty between least educated households and oth- whose heads completed primary education have made ers is very wide (as shown in Panel C of Figure 67). only a moderate contribution to poverty reduction. Another Figure 70 URBAN  POVERTY REDUCTION MAINLY CAME FROM HOUSEHOLDS WITH LITTLE EDUCATED HEADS Decomposition of poverty changes by education of household head, 2011-2016 Contribution to poverty reduction (percentage points) 0 -1 -2 -3 -4 -5 -2.6 -2.7 -0.6 Urban Ethiopia -0.8 -1.8 -1.1 -0.2 -1.8 No education -3.5 -2.1 Primary incomplete -0.5 Addis Ababa -0.9 -0.8 Primate completed -0.4 -0.3 Secondary incomplete -2.0 Secondary completed -4.2 -4.0 -0.5 Post-secondary Major towns -0.4 -0.7 -0.8 Adult education -0.7 -0.5 Population shift -0.0 -4.2 -1.9 Medium towns -4.7 -0.9 -1.5 -1.5 0.3 -4.2 -2.0 -0.5 Small towns 0.0 0.1 -0.2 -0.1 -3.1 Note: Numbers were calculated by excluding the zones that were covered only in the 2016 data for the purpose of comparison between 2011 and 2016. Major towns have populations greater than 100,000 (excluding Addis Ababa); medium towns have populations between 20,000 and 100,000; and small towns have populations less than 20,000. Source: HCES; 2016. World Bank staff calculations. 138 ETHIOPIA POVERTY ASSESSMENT POVERTY REDUCTION MAINLY CAME FROM THE 11 percent), and in construction (1.1 percentage points, or TRADE, URBAN AGRICULTURE, AND SERVICES 10 percent) (Figure 71). The contribution of the manufacturing SECTORS sector to urban poverty reduction is relatively large in Addis Ababa and small towns. The construction sector was im- In terms of employment of the household head, urban poverty portant in major towns. Agriculture still accounted for a large reduction has taken place mainly in the trade and services contribution to poverty reduction in medium towns and small sector as well as agriculture. Urban poverty mainly declined towns, indicating a fair amount of urban agriculture in these among households that engage in jobs in the trade sector (3.5 towns. Poverty reduction due to the sectoral shifts has been percentage points, or 32 percent), in agriculture (2.3 percent- negligible in urban Ethiopia, in line with the familiar “growth age points, or 21 percent), in services (1.9 percentage points, without structural transformation” narrative on Ethiopia. or 18 percent), in manufacturing (1.2 percentage points, or Figure 71 URBAN  POVERTY REDUCTION MAINLY CAME FROM HOUSEHOLDS EMPLOYED IN TRADE AND AGRICULTUR Decomposition of poverty changes by economic sector of household head, 2011-2016 Contribution to poverty reduction (percentage points) 1 0 -1 -2 -3 -4 -2.3 -1.2 -1.1 Urban Ethiopia -3.5 -0.4 -2.0 -0.6 0.3 Agriculture and mining -0.6 -2.2 Manufacturing -1.1 Addis Ababa -3.7 -0.4 Construction -2.5 -0.7 -0.4 Trade -0.8 Infrastructure -0.5 -1.8 -4.0 Services Major towns -0.6 -0.9 -0.5 Public administration 0.1 Population shift -3.7 -0.5 -1.11 Medium towns -2.6 -0.3 -2.7 -1.0 0.5 -2.0 -1.6 -0.7 Small towns -3.5 -0.3 -1.2 0.3 0.9 Note: Major towns have populations greater than 100,000 (excluding Addis Ababa); medium towns have populations between 20,000 and 100,000; and small towns have populations less than 20,000. Interaction effects are not shown. The analysis does not include households whose heads are not active labor forces or unemployed. Source: HCES; 2016. World Bank staff calculations. CHAPTER V. URBAN POVERTY IN ETHIOPIA 139 2.3 Access to services and While electricity access has improved, other living con- ditions have not changed much in medium and small amenities by city size towns since 2010/11. Access to piped water is relatively LIVING STANDARDS ARE WORSE IN SMALL good in Addis Ababa (nearly 90 percent of households) (Pan- TOWNS, AND THE GAP WITH LARGER TOWNS HAS el A in Figure 73). Only about 70 percent of households have NOT BEEN CONVERGING access to piped water in major towns, medium towns, and small towns, and their situation did not change since 2011. The share of substandard housing is highest in medi- Compared to access to water, access to improved sanitation um and small towns. For the purpose of a broad assess- is much worse in urban Ethiopia: Only a fifth of urban house- ment, this chapter defines substandard housing based on holds have access to an improved sanitation facility (Panel B). three criteria: (i) lack of access to piped water (regardless of No clear improvement in sanitation was observed between private or public taps), (ii) lack of improved sanitation (flush 2011 and 2016. Within urban areas, people in smaller towns toilet or ventilated improved latrine), and (iii) overcrowding tend to have worse access to sanitation. Access to electricity (household size divided by the number of rooms is greater (Panel C) and improved solid waste management (Panel D) than 3). While housing units that lack all three criteria account have improved in urban areas since 2011. While the share of for only 7 percent in urban areas, 30 percent of houses lack households with electricity access has reached nearly 90 per- 2 criteria and 46 percent of houses lack 1 criterion (Figure cent in medium and small towns, solid waste management 72). The share of substandard housing is clearly correlated remains problematic. About 40 percent of urban households with city size, as smaller towns tend to have a larger share of rely on a waste disposable vehicle for waste management, substandard housing. The share of substandard housing has while many others still simply throw garbage away. slightly increased since 2011 due to a worsening housing sit- uation in medium and small towns, which were not included in the first-phase ULGDP. Figure 72 SUBSTANDARD  HOUSING IS MOST COMMON IN SMALL TOWNS Indicators of housing quality, 2011-2016 60 50 Share of households (%) 40 30 20 10 0 Score = 3 Score = 2 Score = 1 Score = 3 Score = 2 Score = 1 Score = 3 Score = 2 Score = 1 Score = 3 Score = 2 Score = 1 Score = 3 Score = 2 Score = 1 Urban Addis Ababa Major towns Medium towns Small towns 2016 2011 Note: Criteria are 1) access to improved water (piped water), 2) access to improved sanitation (flush toilet or VIP latrine), and 3) overcrowded (Household size divided by the number of rooms is greater than 3). Major towns have populations greater than 100,000 (excluding Addis Ababa); medium towns have populations between 20,000 and 100,000; and small towns have populations less than 20,000. Source: HCES; 2016. World Bank staff calculations. 140 ETHIOPIA POVERTY ASSESSMENT Figure 73 SANITATION  AND SOLID WASTE MANAGEMENT REMAINS PROBLEMATIC IN SMALL TOWNS Access to services/amenities by city size A. Piped water B. Improved sanitation (%) (%) 100 100 90 90 with access with access (%) (%) 80 100 80 100 70 90 70 90 with access with access 60 80 60 80 of households of households 50 70 50 70 40 60 40 60 households households 30 50 30 50 20 40 20 40 10 30 10 30 Share Share 0 20 0 20 Share of Share of Urban Addis Ababa Major towns Medium Small towns 10 Urban Addis Ababa Major towns Medium Small towns 10 Ethiopia towns Ethiopia towns 0 0 Urban towns 2011 Addis Ababa Major2016 Medium Small towns Urban Addis Ababa Major2016 towns 2011 Medium Small towns Ethiopia towns Ethiopia towns 2016 2011 2016 2011 (%) (%) 100 100 C. Electricity D. Waste collection by vehicle 90 with access with access 90 (%) (%) 80 100 80 100 70 70 90 with access with access 90 60 80 60 80 of households of households 50 70 50 70 40 60 40 60 households households 30 50 30 50 20 40 20 40 10 30 10 30 Share Share 0 20 0 20 Share of Share of Urban Addis Ababa Major towns Medium Small towns Urban Addis Ababa Major towns Medium Small towns 10 10 Ethiopia towns Ethiopia towns 0 0 Urban towns 2011 Addis Ababa Major2016 Medium Small towns Urban towns 2011 Addis Ababa Major2016 Medium Small towns Ethiopia towns Ethiopia towns 2016 2011 2016 2011 Note: Major towns have populations greater than 100,000 (excluding Addis Ababa); medium towns have populations between 20,000 and 100,000; and small towns have populations less than 20,000. Source: HCES; 2016. World Bank staff calculations. Access to basic services is overall better in larg- access to private piped water (within the dwelling) is consid- er towns, and richer households enjoy better access ered. Similar patterns are observed for improved sanitation across cities of different sizes. In case of access to piped (Panel B), access to electricity (Panel C), solid waste man- water, for example, 82 percent of the households in the top agement (Panel D), and overcrowding (Panel E). Medium and quintile have access to piped water, compared to 71 percent small towns are particularly lagging in providing solid waste of households in the bottom quintile (Panel A of Figure 74). management services and in access to improved sanitation. While this disparity is relatively small, it is important to note Addis Ababa fares relatively well: Even households in the that the type of piped water is not considered here. The gap poorest quintile in Addis Ababa have overall better access between richer and poorer households becomes wider if only than households in the richest quintile in major towns. CHAPTER V. URBAN POVERTY IN ETHIOPIA 141 Figure 74 ACCESS  TO AMENITIES WITHIN URBAN AREAS DEPENDS ON HOUSEHOLD WELFARE Access to services/amenities by city size and consumption quintile A. Piped water B. Improved sanitation 96.8 96.8 87.4 96.8 87.4 96.8 81.9 81.9 83.7 83.7 87.4 87.4 78.5 78.5 78.1 78.1 81.9 81.9 83.7 83.7 72.3 72.3 72.3 72.3 71.2 71.2 78.5 78.5 78.1 78.1 71.2 71.2 63.9 63.9 64.3 64.3 72.3 72.3 72.3 72.3 63.9 63.9 64.3 64.3 44.5 44.5 45.6 45.6 44.5 44.5 45.6 45.6 29.0 29.0 29.0 29.0 18.3 18.3 16.1 16.1 17.2 17.2 16.1 16.1 17.2 17.2 8.4 18.3 10.3 18.3 8.4 10.3 3.1 3.1 10.3 10.3 8.4 8.4 3.1 3.1 Urban Urban Addis Ababa Ababa AddisMajor towns Medium Major Medium towns towns Small towns towns Small towns Urban Addis Ababa Urban AddisMajor Ababa Major towns towns Medium Medium Small towns Small towns Urban Addis Ababa Urban AddisMajor Ababa towns towns towns Major Medium Medium Small towns towns Small towns Urban Urban Addis Ababa Ababa AddisMajor towns towns Medium Medium Major towns towns Small towns Small towns Q1 Q2 Q1 Q3 Q2 Q4 Q3 Q5 Q4 Q5 Q5 Q4 towns Q1 Q2 Q1 Q3 Q2 Q4 Q3towns Q5 Q1 Q2 Q1 Q3 Q2 Q4 Q3 Q5 Q4 Q5 Q1 Q2 Q1 Q3 Q2 Q4 Q3 Q5 Q4 Q5 C. Electricity D. Waste collection by vehicle 99.5 99.5 99.2 99.2 94.7 97.4 91.2 94.7 97.4 92.5 92.5 99.5 99.5 91.2 99.2 99.2 91.3 91.3 93.1 93.1 84.7 84.7 94.7 97.4 94.7 97.4 86.5 86.5 91.2 91.2 83.9 83.9 91.3 91.3 92.5 92.5 93.1 93.1 84.7 84.7 83.9 83.9 86.5 86.5 78.1 78.1 70.3 70.3 78.1 78.1 70.3 70.3 59.0 59.0 59.0 59.0 45.7 45.7 36.0 45.7 36.0 45.7 33.2 33.2 36.0 36.0 33.2 33.2 16.0 16.0 13.7 13.7 16.0 16.0 5.9 5.9 13.7 13.7 5.9 5.9 Urban Urban AddisMajor Addis Ababa Ababa Major towns towns towns Medium Medium towns Small Small towns towns Urban Urban Addis Ababa AddisMajor Ababa Major towns towns Medium Medium Small towns Small towns Urban Addis Ababa Urban AddisMajor Ababa towns towns towns Medium Major Small Medium towns towns Small towns Urban Addis Ababa Urban Ababa AddisMajor towns towns towns Medium Medium Major towns Small towns Small towns Q1 Q2 Q1 Q3 Q2 Q4 Q3 Q5 Q4 Q5 Q5 Q4 towns Q1 Q2 Q1 Q3 Q2 Q4 Q3towns Q5 Q1 Q2 Q1 Q3 Q2 Q4 Q3 Q5 Q4 Q5 Q1 Q2 Q1 Q3 Q2 Q4 Q3 Q5 Q4 Q5 Note: Major towns have populations greater than 100,000 (excluding Addis Ababa); medium towns have populations between 20,000 and 100,000; and small towns have populations less than 20,000. Consumption quintiles are calculated for each urban classification. Source: HCES; 2016. World Bank staff calculations. 142 ETHIOPIA POVERTY ASSESSMENT 2.4 Urban labor market even though more people are participating in the labor force indicates a solid pace of jobs growth in urban Ethiopia be- developments tween 2011 and 2016: The number of employed people in This subsection looks at developments in urban labor urban areas increased from 4.5 million in 2011 to 7 million by markets between 2011 and 2016 to provide context to 2016 (Table 21). Female unemployment remains high: 1 in 4 the rapid reduction in urban poverty. The analyses pre- women in urban Ethiopia is unemployed. sented in this section are based on the 2011 and 2016 HCES, The modest reduction in unemployment was accom- the 2010 to 2016 Urban Employment and Unemployment panied by an increase in wages. Real hourly wages in- Surveys (UEUS), and the Labor Force Survey (LFS) 2013, creased by 10 percent between 2011 and 2016. Public and each of which has different advantages and disadvantages.48 private sector wages increased at a similar pace, and the Between 2011 and 2016, labor force participation in- public sector wage remained substantially higher than the creased and unemployment decreased. In 2016, 74 per- private sector one (Figure 77). Real hourly wages increased cent of the working-age population participated in the labor most for the uneducated and the secondary-educated, but force, up from about 71 percent in 2011. LFP increased both decreased for workers with a postsecondary degree (Figure for men and women, though the gap remained large (Figure 78). The increase in wages for the uneducated is consistent 75). At the same time, unemployment decreased from 19.6 with the strong poverty reduction among households headed percent in 2011 to 17.3 percent in 2016, according to the by a little-educated head (although only a minority of the un- UEUS (Figure 76). The fact that unemployment decreased educated are wage-employment: 29% in 2016). Figure 75 LABOR  FORCE Figure 76 WHILE  UNEMPLOYMENT PARTICIPATION INCREASED RATES MODESTLY DECREASED Labor force participation in urban Ethiopia Unemployment rate in urban Ethiopia 85%85% 30 30 80%80% 28.528.5 25 25 Unemployment rate (%) Unemployment rate (%) 25.625.6 75%75% 20 20 70%70% 19.619.6 15 15 17.317.3 65%65% 10 10 60%60% 11.311.3 9.4 9.4 55%55% 5 5 50%50% 0 0 2011 2012 2011 2013 2012 2014 2013 2015 2014 2016 2015 2016 20102011 2010 20122014 20112012 20152016 20142015 2016 Overall Overall Women Women MenMen Female Female MaleMale All members All members Source: UEUS, 2011-2016. Source: UEUS, 2011-2016. 48 The HCES contains some basic labor-related information, yet it does not allow to identify unemployment based on a standard definition. Moreover, the HCES does not report income, which prevents any analysis about wages. The UEUS contain a variety of labor information, including unemployment and wages, on an annual basis. However, the UEUS does not allow to link such labor information to poverty due to the lack of comparability between the UEUS and the HCES. Another downside of the UEUS is its lack of migration-related information. Thus, this section relies on the LFS 2013, which contains detailed migration information, when analyzing migration and labor issues. Additional advantage of the LFS is its information about secondary jobs, which is not available in neither the HCES nor the UEUS. CHAPTER V. URBAN POVERTY IN ETHIOPIA 143 Table 21 EMPLOYMENT/UNEMPLOYMENT  TRENDS IN URBAN ETHIOPIA, 2010-2016 ALL HOUSEHOLDS HEADED BY UNSKILLED 2016 2011 CHANGE 2016 2011 CHANGE Number of employed (million) 7.03 4.52 2.51 2.76 2.16 0.61 Per household 1.58 1.56 0.02 1.44 1.42 0.02 Per working age 0.61 0.57 0.04 0.59 0.56 0.03 Unemployment rate Household head 8.5% 9.1% -0.7 10.5% 10.3% 0.2 Non-head member 25.6% 28.8% -3.2 26.6% 28.4% -1.8 All members 17.3% 19.6% -2.2 18.7% 19.5% -0.8 Note: Unskilled workers are defined in this chapter as those who did not complete primary education. Source: Staff calculations based on UEUS 2010-2016. Figure 77 REAL  HOURLY WAGES Figure 78 EXCEPT  FOR WORKERS WITH INCREASED A POST-SECONDARY DEGREE Real hourly wage, 2004 Birr Real hourly wage, 2004 Birr No No education education Secondary Secondary Overall Overall Private Private Public Public Primary Primary Post-secondary/Other Post-secondary/Other 3.50 3.50 5.00 5.00 3.00 3.00 Hourly wages (2004 Birr) Hourly wages (2004 Birr) Hourly wages (2004 Birr) Hourly wages (2004 Birr) 4.00 4.00 2.50 2.50 2.00 2.00 3.00 3.00 1.50 1.50 2.00 2.00 1.00 1.00 1.00 1.00 0.50 0.50 0.00 0.00 0.00 0.00 20112012 2011 20132014 20122013 20152016 20142015 2016 2011 2011 2012 2013 2014 2012 2013 2015 2016 2014 2015 2016 Source: UEUS, 2011-2016. Source: UEUS, 2011-2016. 144 ETHIOPIA POVERTY ASSESSMENT The education level of the urban labor force has sub- The composition of urban employment did not change stantially improved since 2011, with larger cities accom- much between 2011 and 2016. The most notable change, modating a larger share of better-educated workers. if any, is the increase in the share of self-employment in to- According to the UEUS, the share of the working-age pop- tal urban employment (increase by three percentage points ulation in urban Ethiopia without primary education declined - Figure 81). The sectoral composition of employment did from 20 percent in 2010 to 14 percent in 2016 (Figure 79). In not change at all Figure 82). This explains why the population tandem, the share of those who complete secondary educa- shifts across industries did not contribute to poverty reduction tion or more has increased from 23 percent to 28 percent. in Figure 71. Given the lack of mobility across industries, the Workers in small towns are on average less educated than strong urban poverty reduction between 2011 and 2016 took workers in larger towns and cities (Figure 80). The increasing- place within economic sectors. ly educated urban labor force reflect rapid improvements in urban education and has been a main correlate of decreasing poverty rates, explaining 16 percent of the reduction in urban poverty between 2011 and 2016 (Figure 70). Figure 79 EDUCATION  OF THE URBAN Figure 80 WORKERS  IN SMALL TOWNS LABOR FORCE IMPROVED ARE LESS EDUCATED Educational composition of the urban Educational composition of the urban labor force labor force by city size, 2016 100% 100% 100% 100% 11 13 11 13 90% 90% 15 16 15 16 90% 15 90% 17 15 18 18 18 18 19 17 20 19 20 80% 11 9 80% 11 9 2 2 8 8 80% 80% 6 6 8 8 88 8 8 8 8 70% 70% 70% 18 70% 18 13 13 22 21 22 22 21 22 23 24 23 24 60% 60% 25 25 60% 60% 24 24 7 7 23 23 50% 9 9 50% 9 9 50% 50% 26 26 9 9 9 9 10 26 10 10 10 10 40% 26 40% 10 10 10 40% 40% 26 26 26 26 10 10 30% 30% 27 27 30% 30% 26 26 26 25 24 25 24 23 26 23 20% 20% 20% 20% 21 21 29 29 10% 20 20 10% 20 18 20 15 18 15 10% 10% 14 14 14 14 14 14 14 14 8 8 0% 0% 0% 0% 2010 2011 2010 2012 2014 2011 2012 2015 2016 2014 2015 2016 Small Small Medium Major Medium Addis Major Addis towns towns towns towns towns towns Ababa Ababa No education No education Incomplete primary Incomplete primary No education No education Incomplete primary Incomplete primary Complete primary primary Complete Incomplete Incomplete secondary secondary Complete primary primary Complete Incomplete Incomplete secondary secondary Complete Post-secondary Complete secondary secondary Post-secondary Complete Post-secondary Complete secondary secondary Post-secondary Adult education Adult education Adult education Adult education Source: UEUS, 2011-2016. Source: UEUS, 2011-2016. CHAPTER V. URBAN POVERTY IN ETHIOPIA 145 Figure 81 SELF-EMPLOYED  Figure 82 NO  SECTORAL MARGINALLY INCREASED EMPLOYMENT SHIFTS Status in employment, urban areas Sectoral composition of employment, urban areas 100% 100% 100% 100% 10 9 8 10 9 9 8 8 8 9 8 8 90% 90% 90% 24 2490% 25 24 26 24 26 25 25 26 26 25 80% 80% 80% 80% 70% 4470% 44 43 46 44 70% 8 7 70% 8 8 5 7 6 8 6 5 6 6 43 46 44 46 46 46 46 4 4 4 4 4 7 7 4 7 7 7 7 60% 60% 60% 5 5 60% 5 5 5 5 5 5 6 6 6 6 50% 50% 50% 50% 28 29 28 27 29 27 30 40% 40% 40% 40% 30 28 27 27 28 20 21 24 20 21 21 21 24 21 21 21 21 30% 30% 30% 30% 8 8 8 8 9 8 8 8 7 9 8 7 20% 20% 20% 20% 12 12 12 12 12 10% 27 26 10% 24 27 24 26 25 24 25 24 25 25 10% 10% 13 12 13 12 12 12 12 9 8 9 7 8 6 6 6 7 0% 0% 0% 0% 6 6 6 2010 2010 2014 2011 2012 2012 2016 2011 2015 2014 2015 2016 2010 2011 20102014 2012 20122016 20112015 2014 2015 2016 Public Private Self-employment Public Private Self-employment Other Other Agriculture Mining Agriculture Mining Manufacturing Public utilities Manufacturing Public utilities Construction Commerce Construction Commerce Transport/ Finance/ Transport/ Finance/ Business services Communication Communication Business services Public administration Public administration Other services Other services Source: UEUS, 2011-2016. Source: UEUS, 2011-2016. A closer look at household job compositions suggests members had public wage-employment (Table 22). The share that urban poverty reduction between 2011 and 2016 of households with a self-employed head that also has other was mainly driven by households headed by self-em- employed household members increased from less than 23 ployed workers. People living in households headed by percent in 2011 to more than 28 percent in 2016, explaining self-employed heads accounted for 46 percent of the total a fair share of urban poverty reduction. Overall, looking at urban population in 2011 and 53 percent of all urban poverty Table 22, the reduction in poverty was largest when non- reduction between 2011 and 2016 (Table 22). The reduction head household members moved into self-employment, re- in poverty for households with a self-employed head was gardless of the head’s occupation. It seems, in other words, strongest when other (non-head) household members also that returns to self-employment increased between 2011 engaged in employment: A poverty decrease of 19 percent- and 2016 (see Figure 83 and Box 4). This is consistent with age points in case other household members were self-em- the finding in Figure 68 that the little-educated accounted for ployed (in addition to the self-employed head), 15 percentage most of the poverty reduction in urban Ethiopia (the unskilled points if other household members had private wage-em- are more likely to be self-employed). ployment, and 13 percentage points if other households 146 ETHIOPIA POVERTY ASSESSMENT Table 22 EMPLOYMENT  COMPOSITIONS WITHIN URBAN HOUSEHOLDS, 2011-2016 SHARE POVERTY RATE CONTRIBUTION TO POVERTY HEAD NON-HEAD MEMBER 2016 2011 CHANGE 2016 2011 CHANGE REDUCTION At least one public 2.8 3.9 -1.1 11.4 19.7 -8.4 2.9 No public but at least Public 5.6 5.7 -0.1 3.7 7.8 -4.1 2.1 one private employee Only self-emp 1.9 1.6 0.3 9.5 27.1 -17.7 2.5 Nobody employed 7.6 8.1 -0.5 6.4 12.4 -6.0 4.5 At least one public 1.9 2.7 -0.9 17.8 25.7 -7.9 1.9 No public but at least Private 0.9 1.0 0.0 10.6 17.6 -7.0 0.6 one private employee Only self-emp 3.6 2.8 0.8 17.1 27.6 -10.5 2.7 Nobody employed 5.8 5.2 0.6 11.6 22.2 -10.6 5.0 At least one public 18.0 16.2 1.8 19.6 32.4 -12.8 18.8 Own No public but at least 2.8 2.2 0.6 10.1 24.8 -14.8 2.9 account one private worker Only self-emp 7.3 4.4 2.9 14.4 33.4 -19.0 7.5 Nobody employed 16.6 23.6 -7.0 15.2 26.1 -10.9 23.3 At least one public 4.8 6.3 -1.5 24.2 36.4 -12.2 7.0 No public but at least 3.9 3.0 0.9 13.5 28.4 -14.8 4.0 Others one private Only self-emp 5.8 4.0 1.9 17.4 38.6 -21.2 7.6 Nobody employed 10.6 9.4 1.1 14.2 21.7 -7.5 6.4 Total 100.0 100.0 0.0 14.6 25.6 -11.0 100.0 Sources: Staff calculations based on HICE 2010/11 & 2015/16. Increasing returns to self-employment versus wage The positive effect of a higher proportion of house- employment were important drivers of consumption hold members being engaged in self-employment is changes in urban areas. Figure 83 shows the premium ac- strongest for the poorest urban households. A variable cruing to self-employment over wage employment in urban reflecting the share of household members that were en- areas between 2011 and 2016. The estimates are calculated gaged in self-employment activities was also included in the using the same RIF methodology outlined in Chapter 3 except RIF regressions. The role of that variables in explaining con- now the sample is restricted to urban households, and con- sumption changes between 2011 and 2016 is shown in Fig- trols for the type of employment of the household head are ure 84 below. The impact is strongest for the poorest urban added. The premium is positive across all but the very bottom households before declining to zero over the consumption of the distribution and is especially large for the wealthiest distribution. It therefore appears that although the returns to households. The small increase in the share of urban self-em- having a self-employed household head were not particularly ployment (Figure 81) combined with the increasing returns to strong for poor urban households, increasing the share of self-employment shown below, were both important in ex- household members who were self-employed was a strong plaining poverty reduction in Ethiopia’s cities and towns. driver of consumption changes over time. CHAPTER V. URBAN POVERTY IN ETHIOPIA 147 Figure 83 THE  PREMIUM OF SELF- Figure 84 INCREASING  THE SHARE EMPLOYMENT OVER OF HOUSEHOLD MEMBERS WAGE EMPLOYMENT IN SELF-EMPLOYMENT INCREASED OVER THE WAS MOST IMPORTANT URBAN CONSUMPTION FOR THE POOREST URBAN DISTRIBUTION HOUSEHOLDS Returns to self-employment versus wage The effect of increasing household employment in urban areas 2011-2016 shares of self-employment in urban areas 2011 to 2016 .04 .04 .05 .05 .03 .03 .04 .04 Log difference Log difference Log difference Log difference .02 .02 .03 .03 .01 .01 .02 .02 .01 .01 0 0 −.01 10 −.01 0 0 10 20 20 30 30 40 40 50 50 60 60 70 70 80 80 90 90 10 10 20 20 30 30 40 40 50 50 60 60 70 70 80 80 90 90 Consumption percentiles Consumption percentiles Consumption Consumption percentiles percentiles Premium: Premium: Self employment Self employment vs wage vs wage employment employment Effect of increasing Effect share of increasing of self share ofemployment self employment Source: HCES 2011, 2016. World Bank staff calculations. Source: HCES 2011, 2016. World Bank staff calculations. Larger cities have started to form agglomeration characteristics (such as age, sex, marital status, and edu- economies. Urban economic theory posits that cities with cational attainment), workers earn 5.7 percent more as the a large population size have productivity advantages due town population size doubles. The chance of finding a wage to agglomeration economies. A common way to measure job (or the share of workers with wage jobs) increases by 4.4 such agglomeration effects on labor productivity is to relate percentage points as the town population size doubles (Fig- nominal wages to the population size of towns. In the com- ure 85). Importantly, the estimated effects are larger among petitive markets where average wages reflect the average female (6.4 percentage points), the youth (7.7 percentage marginal product of labor, higher (nominal) wages in large points), and unskilled workers (7.4 percentage points). In cities indicate that workers earn more than they would oth- addition, unemployment is not necessarily higher in larger erwise do in smaller towns. The econometric analysis in Ka- towns once workers’ characteristics are considered, where- mei and Nakamura (2019) suggests that given their observed as underemployment is lower. 148 ETHIOPIA POVERTY ASSESSMENT Box 18 S  elf-employment and poverty transitions Panel data from the Ethiopia Socioeconomic Survey (ESS) can be used to investigate the relationship between poverty dynamics and the sector of employment of the head of the household. This analysis is restricted to urban households that were interviewed in both the second and third rounds (2014 and 2016) of the ESS. The poverty line is set at the 25th percentile of the national consumption distribution, which is consistent with what was done in the poverty dynamics chapter in this poverty assessment. There were substantial movements out of poverty for households with a self-employed head. As shown in the two panels in Table 23 within the self-employed category there were more transitions for those households in the informal sector (the majority of households) than there were in the formal sector. Of all the informally self-em- ployed households in urban Ethiopia that were below the poverty line in 2014, two-thirds had escaped poverty by 2016. This is far higher than the seven percent that fell below the poverty line over the same period. A similar pattern holds for the formally self-employed, who experienced far more positive (out-of-poverty) transitions than negative (into poverty) ones. Table 23 STRONG  MOVEMENTS OUT OF POVERTY FOR THE URBAN SELF-EMPLOYED Conditional poverty transition probabilities, 2014-2016 Formal self-employed (urban) 2016 Non-poor Poor Non-poor 98.2 1.8 100 2014 Poor 44.7 55.3 100 Informal self-employed (urban) 2016 Non-poor Poor Non-poor 93.3 6.7 100 2014 Poor 66.6 33.4 100 Source: ESS, 2014, 2016. World Bank staff calculations. CHAPTER V. URBAN POVERTY IN ETHIOPIA 149 Figure 85 CITY  POPULATION SIZE AND LABOR OUTCOMES A doubling of town popualtion size is associated with an increase of... Change (percent or percentage points) -12 -7 -2 3 8 5.7 Nominal wage (Addis Ababa included, %) -4.0 Nominal wage (Addis Ababa excluded, %) 4.4 Chance of wage employment (ppt) 6.4 Chance of wage employment among women (ppt) 7.7 Chance of wage employment among the youth (ppt) 7.4 Chance of wage employment among the unskilled (ppt) 2.5 Chance of formal employment (book account, ppt) 5.1 Chance of formal employment (license, ppt) -1.6 Labor force participation among married female (ppt) 5.6 Unemployment among married female (ppt) 6.8 Working hours (%) 2.1 Underemployment (ppt) Note: The numbers indicate the changes in labor outcomes (expressed in either percent or percentage points) associated with a 100 percent increase in the town population size, after workers’ characteristics are controlled for. The estimated two-stage least-square regression models include past population density as the instrumental variable to reduce endogeneity. See Kamei and Nakamura (2019) for details. Source: HCES 2011, 2016. World Bank staff calculations. 2.5 Integration of rural migrants and more rural youth to look for employment in urban areas. This is a good thing, as research shows ample welfare bene- into urban labor markets fits of increased internal migration (Box 19). For Ethiopia too, Though still limited, rural to urban migration is expect- De Brauw, Mueller and Woldehanna (2017) find that both ed to increase substantially in coming years. 49 Given rural to rural and rural to urban migration lead to substan- that increasing skills often drive migration, the higher overall tial gains in real consumption levels (adjusted for rural-urban education levels in Ethiopia can be expected to induce more differences in cost of living). 49 The latest survey with systematic information on migration was the 2013 LFS. In the five years prior to the 2013 LFS, six percent of Ethiopians changed zone of residence, a similar share as in 1999. The pattern of internal migration is however changing, with rural-to-urban becoming the dominant flow as of 2008. 150 ETHIOPIA POVERTY ASSESSMENT Box 19 F  rom advocating restrictions to letting them move Economic theory now recognizes that governments should not try to hold on to people Many governments around the world actively attempt to discourage internal population movements from rural to urban areas. In a 2013 UN survey on Population and Development, 148 out of the 185 surveyed countries with data (80 percent) had government policies aimed at reducing internal migration from rural to urban areas. Such efforts are particularly prevalent among countries in Africa, where 85 percent of countries have policies aimed at reducing rural-to-urban migration. Only five countries in the world had policies aimed at speeding up migration to urban areas: China, Sri Lanka, Poland, Tajikistan, and the Maldives.50 Governments’ efforts to control rural-urban migration have their roots in the early influential literature on the links between rural-urban migration and urban unemployment. In the well-known Harris-Todaro model (1970), differences in expected incomes between rural and urban sectors attract migrants from rural areas. Wages in the modern urban sector are fixed and exogenous, and jobs are rationed. Only a small fraction of rural migrants find employment in the modern urban sector, with the others unemployed or underemployed in an urban informal sector. Job creation programs in urban areas raise the expected urban income, stimulating further rural-urban migration and, if the labor demand elasticity in urban areas is large enough, increase the level of urban unemployment. This implication of the model was particularly important for policy because it argued against making cities attractive and implicitly endorsed measures to discourage or reverse migration (Commission on Growth and Development, 2009). Though the Harris-Todaro model has been and remains influential, evidence supporting the predicted link between migration and urban unemployment is weak. Many of the critical assumptions and predictions of the model have not been supported by subsequent empirical studies of labor markets in developing coun- tries. More robust and more plausible alternative models of migration have since emerged, with very different policy implications.51 In particular, increasing returns to scale in the modern sector (vs. constant returns in agriculture) and spillovers from clustering imply that movement from lagging to leading places could have sub- stantial growth and welfare payoffs. In addition, on average, migration brings sizable economic benefits to the migrant in terms of increased consumption levels. Also, migrants who move to cities tend to maintain strong links with their home communities, sending back remittances that boost consumption and investment in origin communities and help to converge living standards across space. In that sense, limiting migration comes down to slowing down development. As argued by the World Development Report on Reshaping Economic Geography, the policy challenge is not to keep people from moving, but how to keep them from moving for the wrong reasons. Agglomera- tion forces and economic opportunities will inevitably pull workers and families to cities, and the goal for policy is how to best accommodate these flows. To avoid migration for the wrong reasons, governments should work to eliminate or alleviate the factors that push people out of their origin areas, such as agricultural decline, due to pressures of population growth or environmental degradation, inefficient land tenure systems, and lack of adequate public services. Migration due to push factors is unlikely to add to agglomeration benefits but likely to exacerbate the urban congestion that policy-makers strive so hard to avoid. 50  World Population Policies Database (http://esa.un.org/poppolicy/about_policy_section.aspx). 51  See Lall, Selod and Shalizi (2006); Commission on Growth and Development (2009). CHAPTER V. URBAN POVERTY IN ETHIOPIA 151 Given that migrants will become a larger share of the and 10 years in the city) and the resident population. How- urban labor force in coming years, it is important to ever, as they stay in Addis longer, rural migrants increasing- assess how they perform in the urban labor market vis- ly find a chance to work in public employment and private à-vis the resident population. Migrants from rural areas permanent jobs, and their employment structure becomes to Addis Ababa and major towns are far less educated than more comparable to the resident population (Table 24). Ru- non-migrant residents of those towns (though more educat- ral migrants in other major towns outside Addis also have ed than non-migrants from the same origin zone). Those mi- lower educational levels than non-migrant household heads. grants tend to work for temporary or casual private jobs, as However, these lower education outcomes do not seem to well as informal self-employment jobs. There seems to be a affect their employment status much. The employment sta- structural difference in the time it takes migrants to catch up tus of recent migrants in major towns is fairly similar to that of with the resident population. In Addis Ababa, recent migrants older migrants and the resident population, suggesting that (less than three years in the city) have substantially worse economic integration of migrants in those towns is relatively employment outcomes relative to older migrants (between 3 smooth compared to the capital city. Table 24 MIGRANTS’  CHARACTERISTICS IN ADDIS ABABA AND MAJOR TOWNS, 2013 ADDIS ABABA MAJOR TOWNS <3 YRS 3-10 YRS NON-MIGRANT <3 YRS 3-10 YRS NON-MIGRANT Sex Male 86.2 78.5 63.6 75.2 78.2 70.9 Female 13.8 21.5 36.4 24.8 21.8 29.1 Education No education 20.1 14.9 3.2 20.9 17.2 9.3 Primary incomplete 32.9 39.2 12.3 25.6 33.8 21.4 Primary complete 7.2 12.3 9.3 10.0 9.0 11.3 Secondary incomplete 37.9 16.0 20.7 23.4 15.4 23.1 Secondary complete 2.0 3.2 23.6 2.9 4.0 15.5 Post-secondary 0.0 14.4 30.8 17.3 20.3 19.3 Adult education 0.0 0.0 0.1 0.0 0.4 0.1 Employment status Public employee 0.0 17.6 20.6 24.1 23.7 25.2 Private employee (permanent) 13.1 20.1 23.2 8.3 7.5 9.3 Private employee (temporary) 32.6 18.0 11.5 17.9 19.4 12.8 Private employee (contract) 2.3 6.2 5.3 3.8 2.7 5.1 Private employee (casual) 11.9 5.4 2.7 3.5 5.7 3.3 Self-employment (formal) 5.9 4.0 11.1 1.4 3.6 4.6 Self-employment (informal) 25.3 24.0 20.3 37.3 35.0 35.5 Others 9.0 4.6 5.4 3.9 2.4 4.1 Note: Only household heads of ages between 25 and 45 years are included. Migrants are only those who moved from rural areas. Sources: Staff calculations based on LFS 2013. 152 ETHIOPIA POVERTY ASSESSMENT However, the educational attainment of rural migrants’ individuals of the same age in non-migrant households (near- children does not catch up with those of non-migrants. ly 90 percent complete primary education). A similar pattern In the capital city, 70 percent of 18 to 20-year-old individu- is observed in major towns. It is important to better under- als in the households headed by rural migrants who arrived stand the challenges rural migrants’ face when it comes to during the last 10 years had not completed primary educa- education in their urban destination. Otherwise, increasing tion (Table 25). This share amounts to 45 percent among the intergenerational economic mobility—which is a key function households who arrived between 10 and 20 years ago. Nev- of cities—is undermined (see Chapter 7). ertheless, their educational levels are by far lower than the Table 25 RURAL  MIGRANTS’ CHILDREN ARE FAR LESS EDUCATED THAN THE CHILDREN OF THE RESIDENT POPULATION Characteristics of 18-20-year-olds, by parents’ migration status and duration ADDIS ABABA MAJOR TOWNS <10 YRS 10-20 YRS NON-MIGRANT <10 YRS 10-20 YRS NON-MIGRANT Sex Male 44.8 34.1 52.8 51.2 51.8 50.9 Female 55.2 65.9 47.2 48.8 48.3 49.1 Education No education 15.3 0.0 0.4 5.6 0.0 3.0 Primary incomplete 44.3 55.2 11.6 40.0 36.2 17.9 Primary complete 18.3 4.4 11.3 9.4 14.4 10.7 Secondary incomplete 22.1 40.4 52.0 41.8 49.5 48.6 Secondary complete 0.0 0.0 14.6 0.0 0.0 10.4 Post-secondary 0.0 0.0 10.1 3.3 0.0 9.5 Note: Only individuals of ages between 18 and 20 whose household heads are migrants from rural areas or non-migrants. Sources: Staff calculations based on LFS 2013. Qualitative research suggests that new rural migrants harassment by local authorities and law enforcement (and for face significant challenges in towns and cities. A quali- young women, sexual harassment by brokers and employ- tative study commissioned by the World Bank confirms that ers), the difficulty of obtaining ID cards (Box 20), and, related, most migrants move to urban areas in search for work and the difficulty of accessing different types of government sup- better opportunities (mainly education), and to escape rural port. Interviews with regional and city authorities revealed a areas they describe as bereft of hope and prospects.52 Young staunchly negative view of rural-urban migration, labeling it women also migrate to escape arranged marriages and tra- as “unacceptable” or “illegal” and exacerbating problems in ditional gender roles. The migration experience is described urban areas. SME bureaus and BOLSAs confirmed not pro- as tough and full of challenges, and risky for young women in viding services to migrants as they do not have ID cards. This particular. Despite the many challenges, the bulk of migrants restrictive context leads to a lose-lose situation: Rural Ethio- rate their migration as positive, saying that it opened up op- pian migrants are exposed to unnecessary risks and hard- portunities that were unthinkable in rural areas. Life in the ships while receiving towns and cities forego the increase in city however is hard, with migrants complaining of frequent economic output migrant workers can bring. 52 BDS Center for Development Research. “Employability of rural migrants in urban settings in Ethiopia: Final Report”. August 2017, Addis Ababa. CHAPTER V. URBAN POVERTY IN ETHIOPIA 153 Box 20 The correlates of having a Kebele ID  Ethiopia does not have a national ID, but works with so-called Kebele IDs. Kebele IDs are IDs issued by the Kebele (village) in which a person is residing. Having a Kebele ID is a prerequisite for being able to benefit from cer- tain government support services and programs in the Kebele. For instance, obtaining an unemployment ID and gain- ing access to government services for the unemployed requires having a Kebele ID. As a result, recent rural migrants that not have the ID from their new urban Kebele cannot access the full package of public services and support. Is not having an ID widespread? Among the 28,000 households surveyed in Addis Ababa for the screening pur- pose of the Urban Productive Safety Net Program (UPSNP), 14 percent of the households were not registered at the Kebele. Nearly 60 percent of non-registered households did not attempt to register at woreda, while 26 percent of non-registered households did attempt but their registration was unsuccessful (did not obtain the ID). Relative to households whose head was born in Addis, households headed by rural migrants are 84 percentage points less likely to be registered at the Kebele. Households with younger and less educated heads tend not to be registered (Figure 56). Interestingly, female-headed households are more likely to be registered, compared to male-headed households. Figure 86 YOUNG  RURAL MIGRANTS ARE LEAST LIKELY TO HAVE A KEBELE ID Correlates of having a Kebele ID from Addis Ababa Female 35 Age <30 - 96 Age 30-39 - 89 Age 40-49 - 75 Age 50-59 - 58 Age 60 or older (base) No education (base) Primary incomplete 8 Primary complete 27 Secondary incomplete 38 Secondary complete 57 Post-secondary 70 Adult education 78 Orthodox (base) Protestant - 31 Muslim - 18 Born in Addis Ababa (base) Moved from rural areas - 84 Moved from urban areas - 65 -100 -80 -60 -40 -20 0 20 40 60 80 100 Probability of woreda registration relative to base category (%) Source: UPSNP baseline survey. The results show that rural migrants are less likely to be registered. They however do not show whether this non-registration carries certain adverse consequences. For instance, does the mere fact of not being registered ex- plain the worse education outcomes for migrants’ children documented in Table 25? A focused study is warranted to examine the causes behind migrants’ worse outcomes and whether this is at all related to the ID system. 154 ETHIOPIA POVERTY ASSESSMENT Conclusions Ethiopia is rapidly urbanizing from a low base. Despite to be informally pegged to public sector wages, are low and fast urban population growth, the number of people below not sufficiently determined by their productivity. It is important the poverty line in urban areas decreased, reflecting a strong to remove any regulatory barriers which may be distorting drop in the poverty rate. The reduction in poverty was widely wages and job creation in the private sector. The ongoing shared across all types of cities. Given their large share in the reform agenda in the country is expected to contribute posi- urban population, small and medium-sized towns accounted tively to private sector development. for much of the reduction in urban poverty between 2011 Self-employment has been the main contributor to ur- and 2016. ban poverty reduction. Households with a self-employed Urban poverty reduction has been associated with head experienced fast poverty reduction, and even faster positive developments in the labor market. To further if other household members also transitioned into self-em- increase the labor market’s contribution to urban poverty re- ployment. The low base of private sector wage employment duction, it needs to be made more inclusive by designing in- implies that self-employment will absorb, in absolute terms, terventions and policies to increase labor force participation of most of the new urban labor force over the coming decade, women and youth. This is especially important for bigger cit- even though the private sector is expected to start growing ies: A doubling of population leads to a six percentage-point faster. To facilitate self-employment and poverty reduction, it increase in the unemployment rate for married women. would be wise to simplify and streamline the current onerous procedures to start and operate a micro- and small enterpris- Private sector wage employment has contributed little es (in terms of licensing and workspace requirements, annual to poverty reduction. As noted in the fifth Ethiopia Eco- license renewals, certificates of competence, etc.). nomic Update, wage levels in the private sector, which seem CHAPTER V. URBAN POVERTY IN ETHIOPIA 155 156 ETHIOPIA POVERTY ASSESSMENT CHAPTER VI Poverty and Social Protection The Productive Safety Net Program (PSNP) and Humanitarian Food Aid (HFA) are Ethiopia’s main social protection programs in terms of spending. While the PSNP addresses chronic food insecurity in rural areas, HFA responds to acute food insecurity as a result of severe shocks, mainly of climatic nature. This chapter assesses targeting, coverage, and beneficiary incidence of both programs as it relates to poverty. The analysis in this chapter shows that the PSNP is overall well targeted, with beneficiaries more likely to be poor and having food shortages, owning fewer assets, and living in more remote and drier places. Within an overall positive picture, three main issues emerge. First, in both 2011 and 2016 the number of beneficiaries by region bore little relation to poverty or food security needs by region, with caseloads exceeding poverty or food security indicators in certain regions and substantially falling short in other regions. Second, first-stage selection of districts (woredas) added little to overall targeting performance, suggesting a rethink on the merits of geographical targeting in a country with relatively small spatial welfare disparities in rural areas. And third, under- coverage remained an important issue, with only 13 percent of the poor population in Ethiopia covered by PSNP in 2016. Increasing the coverage of the poor will likely require an expansion of the PSNP to more woredas in rural Ethiopia and better aligning regional needs and beneficiary numbers, without necessarily increasing the total number of beneficiaries. Overall, HFA is reasonably well targeted as well. Inclusion errors nevertheless exist, with a substantial share of HFA beneficiaries being significantly better-off on a wide range of indicators. These inclusion errors are mainly due to HFA targeting in woredas where PSNP is not active: HFA is well targeted in woredas where PSNP is active, though poorly targeted where PSNP is not active. Results suggest HFA targeting could be improved by harmonizing PSNP and HFA. Finally, the data show that issues of food insecurity have become progressively less salient in Ethiopia, reflecting Ethiopia’s development success over the past decades. Poverty, while it has also decreased substantially, has remained stickier. In light of this, the Government could consider reorienting the focus of safety nets to poverty in general and target benefits to the poorest rather than the narrowly-defined food insecure. This shift would lead to a more inclusive social protection policy. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 157 Introduction The previous chapters showed that, despite substan- (PSNP) and the Humanitarian Food Aid (HFA) intervention. tial improvements, poverty and vulnerability in Ethio- Data on both projects are available from the 2011 and 2016 pia remain widespread. Over 21 million Ethiopians still live HCES. The Urban Productive Safety Net Program (UPSNP) in absolute poverty -are not able to meet their most basic only started operations in early 2017, and as such has not needs- and transitions in and out of poverty are common. been included in the most recent HCES. From the outset, Indeed, Chapter 4 suggests that about half of the Ethiopi- it is important to note that this chapter will not estimate im- an population experienced a spell of poverty between 2012 pacts of social protection interventions. Rather, the focus is and 2016, while the share of chronically poor households on coverage, benefit incidence, and targeting, with the aim to -households that are always poor- also remains significant at provide actionable suggestions on how to improve coverage about 15 percent. These findings highlight the need for social of the poorest and most vulnerable people in society. protection interventions that help increase income levels and, This chapter proceeds as follows: Section 2 summarizes equally important, protect households from seasonal fluctu- Ethiopia’s social protection system and its evolution in terms ations and shocks. of coverage and financing. This section also briefly summa- This chapter will analyze the targeting, coverage, and rizes the large literature estimating the impacts of the PSNP. benefit incidence of Ethiopia’s main social protection Section 3 focuses on targeting, coverage and benefit inci- interventions. For reasons of availability of comparable and dence of the PSNP, while Section 4 focuses on HFA. The final good quality data, the analysis will largely focus on the two section concludes. largest programs: The rural Productive Safety Net Project 158 ETHIOPIA POVERTY ASSESSMENT 2. ETHIOPIA’S SOCIAL PROTECTION SYSTEM For the Government of Ethiopia, social protection is protection instruments in each focus area. Focus area 1 rep- a key part of a policy framework focused on reducing resents the largest area of intervention, accounting for 71.4 poverty, social and economic risk, vulnerability and percent of social protection spending between 2012/13 and exclusion. The National Social Protection Policy comprises 2015/16.53 Spending in 2015/16 was increased because of an five focus areas: safety nets, tailored livelihoods support, so- increased emergency food intervention in response to the El cial security, increased access to basic services by vulnera- Nino drought; but even without this safety nets would remain ble groups and legal protection for those who are vulnerable the largest area of investment in social protection in Ethiopia. to abuse and violence. Table 26 identifies the principal social Table 26 PRINCIPAL  SOCIAL PROTECTION INSTRUMENTS BY FOCUS AREA FOCUS AREA 1: FOCUS AREA 2: FOCUS AREA 3: FOCUS AREA 4: FOCUS AREA 5: PRODUCTIVE EMPLOYMENT OP- SOCIAL ACCESS TO LEGAL PRO- SAFETY NETS PORTUNITIES AND INSURANCE HEALTH, EDUCA- TECTION AND LIVELIHOODS TION AND OTHER SUPPORT SOCIAL SERVICES a) Unconditional Technical support a)  Mandatory social a)  Social transfers a)  a) Communications transfers to on and off-farm insurance for human capital for prevention livelihoods development of abuse and exploitation b) Conditional b) Employment Index linked b)  Health fee waivers b)  Care for people b)  transfers services and weather insurance and health living outside standards insurance subsidies protective family environments c) Public works c) Financial services c) Life insurance Establishment of a c)  Protective legal and c)  social work system policy environment d) Scale-up Community based d)   ervices for d) S Support to d)  mechanisms for health insurance persons with survivors of abuse disaster response disabilities and exploitation e) School feeding Drop-in centres e)  and hotline f) Establishment of a network of specialized service providers Source: OECD, 2019. 53 Kefyalew Endale, Pick, A. and Tassew Woldehanna, 2019, Financing Social Protection in Ethiopia: A Long-term Perspective, OECD Development Policy Papers No. 15, OECD CHAPTER VI. POVERTY AND SOCIAL PROTECTION 159 PSNP and HFA account for the bulk of spending on so- 88). This decline in benefit spending (in real terms) reflects cial protection in Ethiopia. Since its inception in 2004/5, a significant fall in the value of benefits, which in turn has coverage of the PSNP has increased from 4.8 million (in reduced the programme’s effectiveness in reducing pover- 2004/5) to 8 million in 2015/16, though caseloads were on a ty (Devereux, Sabates-Wheeler and Slater, 2008). Benefits downwards trend since 2012 (Figure 87). Despite this scale were however increased in 2018 and 2019. up, financing for the PSNP has declined in real terms (Figure Figure 87 THE  NUMBER OF PSNP BENEFICIARIES HAS INCREASED BETWEEN 2005 AND 2016… 9 8 7.8 7.7 8 7.6 7.3 7.4 7.2 7.2 6.8 7 6 6 5.1 4.8 5 4 3 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 Source: OECD, 2019. Figure 88 WHILE  FUNDING, IN REAL TERMS, HAS DECREASED Cash transfers – left axis – and food assistance – right axis- per year Cash transfer, total Food assistance, total 6000 3000 ETB million in 2010/11 prices Metric ton, thousands 5000 2500 4000 2000 3000 1500 2000 1000 1000 500 0 0 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 6 /1 / / / / / / / / / / / 04 05 06 07 08 09 10 11 12 13 14 15 20 Source: OECD, 2019. 160 ETHIOPIA POVERTY ASSESSMENT As a country prone to increasingly frequent climatic pervasive in urban areas (FDRE, 2016).55 The prevalence and shocks, emergency drought relief plays a crucial role depth of food insecurity has decreased substantially over in supporting populations affected by these crises. the past decade. In 2005, close to one-third of Ethiopians HFA beneficiary numbers vary wildly from year to year, re- reported experiencing food shortages in the 12 months pre- flecting the vicissitudes of unpredictable weather patterns. 54 ceding the survey. This decreased to 22 percent in 2011 and The number of beneficiaries was fairly low in 2010/11 (ap- 10 percent in 2016 (Table 27). The food gap, defined as the proximately 2.8 million) but increased to 10 million during the number of months per year food insecure households report 2015/16 El Nino drought. Though accurate data on spending food insecurity, decreased from 3.9 months in 2005 to 3.3 on emergency relief is difficult to come by given its largely months in 2016. The progress becomes even more striking off-budget nature, it is estimated that over US$1 billion was when considering all households, and not only the ones who spent on HFA during the 2015/16 drought year. reported having experienced a food shortage: Overall, the number of months per year an Ethiopian citizen experienced Food security is at the heart of Ethiopia’s social protec- food insecurity decreased from 1.2 months in 2005 to 0.7 tion system. The rural PSNP and the HFA target chronically months in 2011 and 0.3 months in 2016. From the impact and acute food insecure people in rural areas, respectively, evaluations, it is clear that the PSNP has substantially con- while the UPSNP is part of a broader Urban Food Security tributed to the improvement in food security (Box 21). Strategy that aims to minimize food insecurity, which it deems Figure 89 EMERGENCY  RELIEF CASELOADS VARY FROM YEAR TO YEAR Emergency relief beneficiaries, millions 20 18 16 14 Millions 12 10 8 6 4 2 0 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 PSNP Relief minimum Relief maximum Source: OCHA, 2018. 54 HFA is defined as direct transfers to individuals or households for the purpose of increasing the quantity and/or quality of food con- sumption in anticipation of, during, and in the aftermath of a humanitarian crisis. It includes both in-kind food aid and cash transfers for smoothing consumption. 55 The 2016 WMS suggests food insecurity in urban areas in not that pervasive: Four percent of the urban population reported having experienced a food shortage in the 12 months preceding the survey. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 161 Table 27 AS  A SIGN OF PROGRESS, FOOD SHORTAGES ARE INCREASINGLY LESS PREVALENT Food shortage in the last 12 months -% yes- and number of months of shortage, 2005-2011-2016 2005 2011 2016 FOOD FOOD FOOD SHORTAGE FOOD GAP SHORTAGE FOOD GAP SHORTAGE FOOD GAP (%) (MONTHS) (%) (MONTHS) (%) (MONTHS) Tigray 32.5 3.5 13.2 3 11.9 2.5 Afar 33.1 3.3 7.7 5.2 9 3.8 Amhara 30.4 3.5 23.2 3.1 10.4 3 Oromia 36.8 4.3 16 3.1 10.5 3.6 Somali 28.9 3.8 30.3 4.4 6 3 Benishangul-Gumuz 22.5 3.1 5.6 2.1 8.5 2.8 SNNPR 26.6 3.9 35 3.4 12.6 3.2 Gambella NA NA 31.6 2.6 3.8 1.2 Harari 15.3 3.4 8 3.2 0 NA Addis Ababa 11.8 4 7.8 4 1.1 3.6 Dire Dawa 20.6 4.7 13.5 1.6 7.7 2 National 31.4 3.9 21.6 3.3 10.2 3.3 Note: Food gap is only calculated for those households that report a food shortage. Source: WMS, 2005; 2011; 2016. The next two sections will look at standard indicators nature of this report (a Poverty Assessment), the ultimate of coverage, targeting and beneficiary incidence of aim of social protection programming (to reduce poverty and Ethiopia’s main social protection programs. Due to data vulnerability), and the way in which the Government of Ethio- availability, the focus will be on the PSNP and HFA (humani- pia measures and monitors poverty, the emphasis will be on tarian food aid). For both programs, the chapter will first look targeting as it relates to consumption poverty.56 Throughout at the national picture, before disaggregating and decom- the analysis, we will use an estimate of pre-benefit household posing some of the indicators at lower administrative levels, consumption expenditures in order not to conflate targeting mainly regional and woreda-level. Targeting of the programs accuracy and program benefits. Annex 5 describes in detail will be assessed on a variety of indicators of well-being (food the technical approach to the analysis. security, assets, household consumption, etc.). Given the 56 To assess targeting accuracy, we use a measure of pre-transfer total consumption expenditures. Given the focus of both PSNP and HFA on food security, one might ask why the chapter focuses on total consumption and not food consumption. While this is a valid question, the rank correlation coefficient between quintiles of total consumption expenditure and food consumption expenditure is higher than 0.8, meaning that focusing on one or the other will not have any substantive impacts on the analysis. 162 ETHIOPIA POVERTY ASSESSMENT Box 21 I  mpacts of the PSNP Over the years PSNP demonstrated notable impacts, though their scope varies by region and period. In the highland regions the impact evaluations of the PSNP carried out on panel surveys every two years between 2006 and 2014 reported that food security has improved: The impact estimates showed that the number of months a household cannot satisfy its food needs was reduced on average by more than 40 percent, from 3.1 to 1.7 months for public works beneficiaries. During the same period positive impact estimates were also recorded on consumption, dietary diversity, housing conditions, and, to some extent, livestock holdings: Both food and total consumption per capita doubled (in real terms), dietary diversity improved by about 25 percent, and positive impacts on livestock assets where observed for the poorest 20 percent beneficiary households but not for the rest of the relatively better-off households. In the lowlands however, where public works where introduced in 2010, evaluations found no significant impact until 2014 on most relevant outcomes except a decrease in food insecurity of almost one month for the poorest 50 percent of beneficiaries. The evaluation reports also found that overall PSNP participants markedly reduced their use of distress asset sales. In ad- dition, some impact evaluation rounds suggest that PSNP has led to an increase in girls’ grade attainment and has improved schooling efficiency by 10–20 percent. Similarly there is some evidence that participation in PSNP lowers fertility and leads to a delay in marriage of adolescent girls. However, these impacts were not sustained over the years. The impact evaluation sample and methodology were revised in 2016. During 2016-2018, despite several challenges including the El Nino and La Nina induced droughts, no adjustments of benefits value despite two digits inflation, and worsening of implementation performance with respect to timely payments delivery, the program sustained its impacts: food security (+0.1 months), household consumption (+10%), and dietary diver- sity (+0.1 food groups) continued to improve in highlands. In the lowlands the impact estimates on food security are positive and higher than in the past - a reduction of about half month in the food gap for the overall PSNP beneficiaries sample compared to the control group. Source: Berhane, Hirvonen, and Hoddinott (2015a); Berhane, Hirvonen, and Hoddinott (2015b); Berhane et al. (2018a); Ber- hane et al. (2018b); Hoddinott and Mekasha (2017). CHAPTER VI. POVERTY AND SOCIAL PROTECTION 163 3. THE RURAL PRODUCTIVE SAFETY NET PROJECT The PSNP is overall well-targeted to the poor and compares favorably with Public Works programs in other countries. First stage woreda-selection does not add much to the targeting performance of PSNP, reflecting the difficulty of geographical targeting in a country with relatively small spatial welfare disparities in rural areas. Pro-poor selection of households within woredas however compensates for relatively weak geo- graphical targeting. PSNP is better targeted towards the poor in the highlands, yet even in the lowlands the PSNP is progressive. Contrary to conventional wisdom, Afar outperforms the other regions on targeting, thanks to the fact that all woredas in Afar are included in PSNP and hence there is no first-stage exclusion of the poor due to woreda selection. Within an overall positive picture, three main issues emerge: First, the number of beneficiaries by region bears little relation to poverty or food security needs by region, with case- loads exceeding poverty or food security indicators in certain regions and substantially falling short in other regions. Second, selection of woredas adds little to targeting performance, suggesting a rethink on whether geographical targeting can actually work in a country such as Ethiopia. And third, under-coverage remains an important issue, with only 13 percent of the poor population in Ethiopia covered by PSNP in 2016. 3.1 The national picture 2015, when PSNP covered less than six percent of the pop- ulation, but then sharply increased in 2016 as a result of the COVERAGE AND BENEFICIARY INCIDENCE EL Nino drought. At the regional level, coverage decreased most in Tigray and increased sharply in Somali (Table 28). While absolute coverage of PSNP increased between In 2016, coverage of the PSNP was highest in Afar (32 per- 2011 and 2016, relative coverage slightly decreased. cent), Somali (30 percent) and Tigray (19 percent). Harari and While slightly over nine percent of the total population was Oromia had the lowest coverage at about five percent of the covered by the PSNP in 2011, this decreased by a half a per- population (with the exception of Addis Ababa, Gambella centage point by 2016. Coverage declined substantially until and Benishangul-Gumuz that are not covered by the PSNP). Table 28 SUBSTANTIAL  CHANGES IN REGIONAL COVERAGE BETWEEN 2011 AND 2016 Share of the population covered by PSNP, by region 2011 2016 DIFFERENCE Tigray 30.9% 19.6% -11.3% Afar 31.0% 31.8% 0.8% Amhara 12.4% 9.1% -3.3% Oromia 4.4% 5.0% 0.6% Somali 8.5% 29.9% 21.4% Benishangul-Gumuz 0.0% 0.0% 0.0% SNNPR 8.7% 5.6% -3.1% Gambella 0.0% 0.0% 0.0% Harari 7.9% 4.5% -3.4% Addis Ababa 0.0% 0.0% 0.0% Dire Dawa 13.7% 14.3% 0.6% National 9.2% 8.7% -0.5% Note: Coverage is the share of the population covered by PSNP. Source: PSNP administrative data; CSA population projections. World Bank staff calculations. 164 ETHIOPIA POVERTY ASSESSMENT In 2016, two-thirds of all PSNP beneficiaries were clus- had a relatively low coverage in 2016 (Table 28), its beneficia- tered in three regions. Amhara region accounted for 24 ry incidence was nevertheless high due to its large popula- percent of all PSNP beneficiaries, Oromia for 22 percent and tion. Somali, despite having a relatively small population, still Somali region for 21 percent (Figure 90). While Oromia region accounted for a large share of overall beneficiaries. Figure 90 AMHARA,  OROMIA, AND SOMALI REGIONS ACCOUNT FOR TWO-THIRDS OF PSNP BENEFICIARIES Distribution of PSNP beneficiaries by region, 2016 25.0% 23.7% 21.7% 20.9% 20.0% 15.0% 13.0% 12.7% 10.0% 7.0% 5.0% 0.8% 0.1% 0.0% Amhara Oromia Somali SNNPR Tigray Afar Dire Dawa Harari Source: WMS, 2016. World Bank staff calculations. The slight decrease in coverage between 2011 and Overall, 13 percent of the poor population in Ethiopia was 2016 applied to all income groups and to food-secure covered by the PSNP in 2016 (Figure 91). and insecure households alike. In 2011, 17 percent of the poorest quintile of the Ethiopian population was covered by A DETAILED TARGETING ASSESSMENT PSNP, decreasing to 13 percent in 2016 (Figure 91). Cov- Beneficiary targeting for the PSNP happens in different erage of higher-income groups decreased too: Close to six stages. The procedure to target households can be broken percent of the wealthiest 20 percent of the Ethiopian popu- down into three steps. In the first step, the Federal Ministry of lation was covered by the PSNP in 2011, decreasing to four Agriculture, following consultations with the regions, defines percent in 2016. In both years, PSNP coverage was higher the caseload per woreda based on the history of food aid for low-income than higher-income quintiles, showing the in the woreda. When the PSNP started in 2005, woredas progressive nature of PSNP targeting. Looking at self-report- included were the ones that had received three consecutive ed food insecurity rather than consumption poverty, cover- years of food aid prior to 2005. In the second step, kebele age of both the food-insecure and secure slightly decreased selection and caseloads are determined by the woreda. Fi- but remains progressive: In 2016, about 17 percent of the nally, households are selected at kebele level though com- nationwide food-insecure people were covered by PSNP, munity-based targeting. This subsection will assess overall compared to seven percent of the food-secure population.57 targeting of the PSNP but will also break down the targeting 57 The results on food security need to be interpreted with caution. In contrast to consumption expenditures, where it is possible to construct a pre-PSNP consumption estimate, we cannot get to a pre-PSNP food security situation based on the WMS. For instance, part of the food-secure population that is covered by PSNP may in fact have been food-insecure if it were not for PSNP. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 165 Figure 91 COVERAGE  DECREASED ACROSS THE DISTRIBUTION Share of population covered by PSNP, by pre-transfer consumption quintile, pre-transfer poverty status, and self-reported food security, 2016 2011 2016 20 18 17.4 17.5 17.1 16.5 16 14 13.3 13.4 13.3 12 11.3 10.4 9.5 10 8.3 8.3 8 7.6 7.3 6.7 6 5.6 4.9 4.1 4 2 0 Q1 Q2 Q3 Q4 Q5 Food No food Below PL Above PL shortages shortages Quintiles of pre-transfer consumption expenditures Self-reported food Pre-transfer poverty shortages rate Source: WMS, 2011; 2016. World Bank staff calculations. performance by the different selection steps. Given data lim- somewhat better in 2016.58 Despite the progressive nature itations, step 1 (selection of woredas) and step 2 (selection of of targeting, there was considerable inclusion of the upper kebeles) will be combined in the analysis. quintiles: Close to ten percent of beneficiaries were in the top consumption quintile in 2016, and 21 percent were in the At the national level, PSNP is well targeted towards upper 40 percent of the consumption distribution. In terms the poor. About one-third of PSNP beneficiaries were in the of binary poverty status, 39 percent of PSNP beneficiaries poorest quintile in terms of consumption in 2016, and 60 were below the national poverty line in 2016. The beneficia- percent of beneficiaries were in the bottom 40 percent (Fig- ry incidence for PSNP compares favorably in cross-country ure 92). Targeting performance was largely similar in 2011 comparisons (Box 22). and 2016, though inclusion of the bottom 40 percent was Figure 92 MOST  OF PSNP BENEFICIARIES ARE IN THE LOWER CONSUMPTION QUINTILES Share of beneficiaries by quintile, 2011 and 2016 40 33 33.8 2011 2016 35 30 27 25 23.6 20 18.8 18.1 14.6 15 11.6 10 9.5 10 5 0 Q1 Q2 Q3 Q4 Q5 Quintiles of pre-transfer consumption expenditures Source: WMS, 2011; 2016. World Bank staff calculations. 58 Higher inclusion of the bottom 40 percent in 2016 is largely due to the broader sample in the 2016 HCES (the inclusion in the sam- ple of the previously uncovered pastoral zones of Somali and Afar). 166 ETHIOPIA POVERTY ASSESSMENT Box 22 T  argeting of PSNP Public Works Looks Good in Cross-Country Comparison How does the coverage and beneficiary incidence of PSNP compare to programs in other countries? To answer that question, we use harmonized data from the World Bank’s Atlas of Social Protection – Indicators of Resilience and Equity (ASPIRE). ASPIRE is a compilation of Social Protection and Labor (SPL) indicators gath- ered from officially-recognized international household surveys in 123 countries.59 Given that ASPIRE indicators are mainly used for cross-country comparison, trade-offs need to be made to make the data as comparable as possible. For this current comparison, the trade-offs are the following: We focus only on Public Works programs (which accounts for the bulk of PSNP), we use post-transfer final consumption aggregates, and we use the US$1.9 international poverty line. Querying the ASPIRE database for countries with Public Work programs and data between 2010 and 2016 resulted in 11 hits. Three hits were removed from the comparison as the countries and/or programs were deemed quite incomparable to Ethiopia (The Philippines, Mexico and Argentina).60 Overall, Ethiopia compares favorably to other countries when it comes to coverage and beneficiary inci- dence. In 2016, slightly over seven percent of the Ethiopian population was covered by the PSNP Public Works, which is higher than most of the other countries in the comparison, except Malawi, India and Afghanistan. Targeting was also relatively good in Ethiopia, with the share of Public Works beneficiaries that is in the bottom quintile being higher than in any of the other African countries, though lower than in India and Nepal (Figure 93). In contrast, the share of PW beneficiaries that is below the international poverty line is lower in Ethiopia (35 percent) than in Malawi (74 percent), Rwanda (81 percent), and Niger (65 percent). This is however largely explained by Ethiopia’s relatively low poverty rate compared to those countries. Figure 93 ETHIOPIA’S  BENEFICIARY INCIDENCE COMPARES FAVORABLY Share of beneficiaries by quintile, Public Works programs Q1 Q2 Q3 Q4 Q5 45 40 35 30 25 20 15 10 5 0 Post-transfer consumption quintiles Source: ASPIRE, 2019. WMS, 2016. World Bank staff calculations. 59  Accessible at http://datatopics.worldbank.org/aspire/about_aspire 60 The eight comparator countries are Rwanda (2013), Malawi (2013), Niger (2014), Tanzania (2012), Nepal (2010), India (2011), and Afghanistan (2011). CHAPTER VI. POVERTY AND SOCIAL PROTECTION 167 PSNP is also well targeted on other non-monetary in- main difference between beneficiaries and non-beneficia- dicators of living standards. Figure 94 shows standard- ries however, as seen by the largest bar in Figure 94, is the ized differences in a number of indicators of living standards “greenness” of the area where they live: PSNP beneficiaries between PSNP and non-PSNP beneficiaries. Differences are live in areas that have far lower scores on the normalized dif- standardized so as to easily interpret the length of the bars ference vegetation index (NDVI), a measure of the greenness as the magnitude or importance of the differences. Overall, of the vegetation. In simple terms, PSNP beneficiaries live in PSNP beneficiaries are more likely to be poor, more likely dryer places with less vegetation and less suitable for rainfed to report food shortages, and are more remote in terms of agriculture -likely a result of path-dependency as woredas distance from key infrastructure assets. They also have low- included in PSNP are the ones that have a history of receiving er scores on standardized asset and livestock indices. The emergency food aid. Figure 94 PSNP  BENEFICIARIES ARE WORSE-OFF ON A VARIETY OF WELFARE INDICATORS Standardized mean differences between PSNP and non-PSNP beneficiaries, 2016 Rural remoteness index Poorest 15% Poorest 20% Poorest 10% Widow HH head Self-reported food shortage Disability in HH HH engages in agriculture Age HH head HH size Livestock index HH has wage income Durable assets index NDVI (vegetation index) -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Note: Bars in green are statistically significant at the 10% level or lower. Source: WMS, HCES; 2016. World Bank staff calculations. Though PSNP is well targeted overall, regional benefi- people. Somali for instance had 1.7 million PSNP beneficia- ciary numbers are not strongly related to regional pat- ries in 2016, even though the WMS survey of the same year terns of poverty and food insecurity. Table 29 compares estimated a far lower number of people in poverty or food-in- the number of PSNP beneficiaries by region with the number security. And third, there seems to be no objective reason to of poor and food-insecure people in that region, estimated not include the western lowlands regions of Gambella and from the 2016 WMS. Though findings depend on the specif- Benishangul in the PSNP, as their needs (again proxied in ic indicator considered, three broad patterns are clear: First, terms of poverty or food insecurity) exceed the ones of Harari more populated regions such as Oromia and SNNPR have low and rural Dire Dawa.61 Looking at chronic poverty (as estimat- beneficiary numbers compared to needs, with needs proxied ed in Chapter 4) rather than poverty or food insecurity largely by the number of poor or food-insecure people. Second, in tells a similar story: Regions’ shares in national chronic pover- the lowlands regions of Afar and Somali the PSNP benefi- ty seems quite disconnected from regions’ shares in the total ciary numbers exceed the number of poor or food-insecure PSNP beneficiary caseloads (Figure 95).62 61 Doing the same calculations for the 2011 survey results in qualitatively similar results. The patterns presented in Table 4 are not influenced too much by the 2015/16 drought. 62 The ESS are only representative at regional level for Tigray, Oromia, Amhara and SNNPR. The estimates of chronic poverty for Somali and Afar need to be interpreted with care. 168 ETHIOPIA POVERTY ASSESSMENT Table 29 THE  NUMBER OF PSNP BENEFICIARIES BY REGION Food shortage in the last 12 months -% yes- and number of months of shortage, 2005-2011-2016 NUMBER OF POOR NUMBER OF FOOD- NUMBER OF POOR PEOPLE PRE-PSNP INSECURE PEOPLE NUMBER OF PSNP PEOPLE TRANSFERS (SELF-REPORTED) BENEFICIARIES Tigray 1,368,769 1,415,335 610,640 1,010,750 Afar 391,037 488,569 151,221 562,082 Amhara 5,280,434 5,451,790 2,119,424 1,890,985 Oromia 7,941,991 7,993,015 3,495,489 1,733,622 Somali 1,141,383 1,144,429 304,177 1,673,009 Benishangul-Gumuz 264,158 264,158 84,535 0 SNNPR 3,597,101 3,635,469 2,210,394 1,039,959 Gambella 81,350 81,350 14,145 0 Harari 10,310 10,310 0 10,723 Dire Dawa 37,519 38,543 15,140 64702 Note: Number of poor people and number of food-insecure people are calculated from the HCES/WMS using survey sampling weights. The number of poor people pre-PSNP transfers deducts PSNP transfers from the consumption aggregate to get an estimate of pre-PSNP poverty levels. Source: HCES, WMS, 2016. World Bank staff calculations. Figure 95 SUBSTANTIAL  DISPARITIES BETWEEN REGIONS’ CONTRIBUTION TO OVERALL CHRONIC POVERTY AND REGIONS’ SHARES IN OVERALL PSNP CASELOAD Regions’ contribution to national chronic poverty and national PSNP caseload, 2016 Oromia Amhara SNNPR Tigray Somali Afar 0.4% 2.2% 100.0% 7.0% 2.9% 90.0% 80.0% 20.9% 70.0% 39.7% 12.7% 60.0% 50.0% 13.0% 40.0% 33.6% 23.7% 30.0% 20.0% 10.0% 18.2% 21.7% 0.0% Share in national chronic poverty Share in national PSNP bene ciaries Source: WMS, 2016. World Bank staff calculations. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 169 STEPS 1 & 2: SELECTION OF WOREDAS AND Despite a fairly good woreda selection, poverty rates KEBELES are relatively high even in non-PSNP woredas. In 2016, PSNP woredas had a poverty rate of 28 percent, compared The federal Ministry of Agriculture, in consultation with to 23 percent in non-selected woredas. This relatively small the regions, select woredas to be included in the PSNP difference highlights the difficulty of geographical targeting in and their caseload based on historical receipt of food Ethiopia: Poverty in Ethiopia is widely spread across regions assistance. Once a woreda is selected for the PSNP, there and woredas and there are few specific spatially-defined is no set time limit for the woreda to stay in PSNP.63 Over- pockets of poverty. In such a situation, a first-stage spatial all, woredas selected for the PSNP are on average worse-off targeting will inevitably exclude a large share of the poor. To than not-selected woredas. PSNP woredas have higher pov- assess to what extent the geographical targeting in the PSNP erty rates, report more food shortages and are more remote contributes to its overall targeting performance, we need to (Figure 96). Asset and livestock holdings are lower in PSNP use a targeting indicator that is decomposable in a geograph- woredas and, most importantly, they are less green (lower ical component (the choice of woredas) and a within-unit score on the normalized difference vegetation index). Given component (choice of households within woredas). The “tar- that woredas are selected into the PSNP based on the history geting differential” provides such a measure (Box 23). of food aid, it is no surprise that woreda greenness is the main correlate of a woreda being in the PSNP. Figure 96 THE  GREENNESS OF WOREDA VEGETATION IS THE STRONGEST CORRELATE OF WOREDA SELECTION FOR PSNP Standardized mean differences between PSNP and non-PSNP woredas, 2016 Self-reported food shortage Rural remoteness index Poorest 15% Poorest 20% Poorest 10% Age HH head HH size Disability in HH HH has wage income Widow HH head HH engages in agriculture Durable assets index Livestock index NDVI (vegetation index) -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 Note: Bars in green are statistically significant at the 10% level or lower. Source: WMS, HCES; 2016. World Bank staff calculations 63 No woreda, once selected for the PSNP, has ever exited the program. This is normal and reflects the chronic nature of social needs and the need for a long-term social protection system. 170 ETHIOPIA POVERTY ASSESSMENT Box 23 D  ecomposing targeting performance The targeting differential To decompose the overall targeting performance of the PSNP into an “inter-woreda” (or “inter-kebele”) and an “in- tra-woreda” component, we use the targeting differential. The targeting differential (TD), introduced by Ravallion (2000), is defined as a program’s coverage of the poor minus its coverage of the non-poor. A TD of zero means that coverage of the poor is equal that of the non-poor, indicating a random allocation of the program. Theoretically, the TD can range between -100 (zero coverage of the poor and complete coverage of the non-poor) and +100 (complete coverage of the poor and zero leakage to the non-poor). In practice the possible range of the TD is determined by the size of the program and the prevailing poverty rate. If the size of the program is insufficient to cover all the poor, the maximum attainable TD is given by G/H, where G is the coverage of the program and H the poverty rate. The lower bound of the TD is then given by -G/(1-H). An appealing feature of the TD is that it is decomposable into a “geographical selection” and a “household selection” com- ponent, measuring, respectively, the higher-level administration’s performance in correctly identifying the poorest lower-lev- el administrative units and the lower-level administrative units’ performance in correctly identifying the poorest households. Given that the targeting differential is a function of program coverage and prevailing poverty rates, it cannot be used to compare targeting performance across administrative divisions. To compare targeting performance across regions, we define a relative targeting differential. The relative targeting differential measures how close the actual target- ing differential approaches the maximum possible targeting differential. The maximum possible targeting differential is determined by program coverage and prevailing poverty rates. If program coverage is either insufficient or just sufficient to cover all the poor, the maximum possible targeting differential is given by G/H. If program coverage exceeds the pov- erty rate, the maximum possible TD is equal to 1-((G-H/(1-H)). The relative targeting differential then is simply the actual TD divided by the maximum possible. In 2016, PSNP targeting achieved about 20 percent of the maximum possible – the relative targeting differential was 20 percent. Keeping into account the size of the PSNP in 2015/16, the maximum attainable TD was 33.8 percent -assuming PSNP only included households below the national poverty line.64 The actual TD was 6.6 percent: Coverage of the poor was 6.6 percentage points higher than coverage of the non-poor.65 Of this difference in poverty rates, selec- tion of woredas accounted for 1.7 percentage points (one-fourth of the TD), while selection of households within wore- das accounted for close to five percentage points (three-fourths of the TD) - Figure 97. Woredas did however manage to select the poorer kebeles. Doing the decomposition at kebele level shows that selection of kebeles and selection of households within kebeles each account for about half of the targeting differential. Figure 97 WOREDA  SELECTION DOES NOT ADD MUCH TO PSNP’S TARGETING PERFORMANCE Decomposition of the targeting differential Geographical selection Household selection 7 Targeting differential 6 5 3.4 4 4.9 (pp) 3 2 3.2 1 1.7 0 Woreda Kebele Note: Targeting differential is the difference between coverage of the poor and that of the non-poor. Source: WMS, 2016. World Bank staff calculations. 64 This is obtained by dividing PSNP’s coverage in 2016 (8.3 percent) by the prevailing pre-PSNP poverty rate (24.4 percent) 65 In terms of self-reported food insecurity, the relative targeting differential in 2016 was 11 percent (the actual TD of 9.2 divided by the maximum possible of 81 percent). CHAPTER VI. POVERTY AND SOCIAL PROTECTION 171 Selection of woredas, though progressive, does not the 20 best-off PSNP woredas with the 20 worst-off non- contribute much to the PSNP’s targeting performance. PSNP woredas. The numbers need to be treated with In 2016, the PSNP covered 13.3 percent of the poor and caution, as the HCES is not representative at the woreda 6.7 percent of the non-poor, resulting in a targeting differ- level. However, unless sampling in these woredas were ential of 6.6 percentage points. About one-fourth of the TD systematically geared towards wealthier kebeles and is explained by selection of woredas while three-fourths is households (in PSNP woredas) and towards poorer kebe- explained by section of households within woredas (Figure les and households (in non-PSNP woredas), they will still 97). In terms of self-reported food security, the pattern is indicate a signal. The 20 best-off PSNP woredas in the largely similar : PSNP coverage of the food-insecure is nine 66 HCES have low poverty rates (5 percent), low food inse- percentage points higher than coverage of the food-secure curity (3 percent), more durable assets and livestock, and (TD of 9.2 percent), of which a mere 0.8 percentage points are less remote (lower score on a rural remoteness index). is explained by woreda selection67. Selection of kebeles The 20 worst-off non-PSNP woredas on the other hand within woredas however is strongly correlated with kebele have far worse indicators. The PSNP woredas are how- food insecurity: Selection of kebeles accounted for close ever less green (lower NDVI score than the poor woredas to 60 percent of the TD for food insecurity, while household outside PSNP). This underscores a basic point on the geo- selection within kebeles accounted for 40 percent. graphical targeting of the PSNP: Greenness of a woreda is the single main predictor of whether or not a woreda is in Though the selection of woredas contributes to the PSNP. And while greenness is correlated with poverty, the PSNP’s targeting performance, there is room to in- correlation is far from perfect. crease this contribution. To illustrate, Table 30 compares Table 30 THERE  IS ROOM TO IMPROVE WOREDA SELECTION Selected household indicators in the 20 best-off PSNP woredas and the 20 worst-off non-PSNP woredas, 2016 20 BEST-OFF PSNP 20 WORST-OFF WOREDAS NON-PSNP WOREDAS 413 OTHER WOREDAS Poverty rate (pre PSNP and HFA) 4.6% 61.1% 25.8% Food shortages (% yes) 3.3% 13.9% 11.6% NDVI 4,674 5,159 4,731 Durable assets index 0.08 -0.303 -0.219 Rural remoteness index -0.342 0.225 0.202 Livestock index 0.186 -0.008 0.256 Total HH obs in woredas 336 384 8,492 Note: Coverage is the share of the population covered by PSNP. Source: WMS, 2011; 2016. World Bank staff calculations. 66 Households are considered food-insecure if they reported having experienced food shortages in the past 12 months preceding the survey. 67 In terms of food insecurity, the maximum attainable TD in 2016 was 81 percent (calculated as the PSNP coverage of 8.3 percent divided by the prevalence of self-reported food insecurity of 10.3 percent). The actually observed TD of 9.2 percent indicates that PSNP achieved about 11 percent of its theoretical best targeting performance. However, as food insecurity is only observed after assistance, these numbers will be underestimated. 172 ETHIOPIA POVERTY ASSESSMENT Figure 98 GREENNESS  AND POVERTY ARE RELATED, BUT ONLY WEAKLY Incidence of poverty and food security by quintile of the vegetation index Poverty (%) Food shortages (%) 40 35.7 35 30 27.9 26 24.1 25 18.9 20 16.7 15.7 15 12.4 10 7.1 6.5 5 0 1 2 3 4 5 Quintiles of rural NDVI Source: WMS, HCES, 2016. World Bank staff calculations. STEP 3: SELECTION OF HOUSEHOLDS member. Selected households were on average smaller and had less livestock and fewer durable assets. As was also Selection of households within PSNP woredas is gen- the case with woreda selection, the main difference between erally good. Considering only the woredas that were in PSNP and non-PSNP households within PSNP woredas is PSNP (in 2016), selected households were more remote, the greenness of their surroundings: Within PSNP woredas, more likely to be below the poverty line, and more likely to PSNP households live in places that have less vegetation, report food shortages (Figure 99). They were also more likely and are presumable less suitable for agriculture, compared to be headed by a widow and include a disabled household to non-PSNP households. Figure 99 SELECTION  OF HOUSEHOLDS WITHIN PSNP WOREDAS IS GENERALLY GOOD Standardized mean differences between PSNP and non-PSNP households within PSNP woredas, 2016 Rural remoteness index Poorest 15% Poorest 20% Poorest 10% Widow HH head Self-reported food shortage Disability in HH HH engages in agriculture Age of HH head HH size Livestock Index HH has wage income Durable assets index NDVI (vegetation index) -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Note: Bars in green are statistically significant at the 10% level or lower. Source: WMS, HCES; 2016. World Bank staff calculations. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 173 There is nevertheless substantial inclusion of the top (Table 31). However, they have fewer assets, less livestock, consumption quintiles in PSNP woredas. In 2016, the are more remote, and are headed by a less educated head. PSNP covered, in PSNP woredas, 13 percent of households PSNP households in the upper quintiles tend to be smaller in the fourth consumption quintile and 10 percent of house- and more likely to be female-headed. They are more likely to holds in the top consumption quintile (Figure 100). A clos- be in the direct support component of PSNP (20 percent of er look at the characteristics of these households shows a PSNP households in the upper two quintiles are direct sup- picture that is more nuanced than that of blatant mistarget- port households, compared to 13 percent in the bottom two ing. Relative to non-PSNP households, PSNP households in quintiles). While these households are not consumption-poor the upper quintiles have higher consumption expenditures today, one could make a case that these households, given and calorie intake on a per adult equivalent basis, are less their low asset base, are vulnerable to future poverty. likely to report food shortages and have a lower food gap Figure 100 WITHIN  PSNP WOREDAS, COVERAGE IS PROGRESSIVE BUT STILL INCLUDES THE UPPER QUINTILES PSNP coverage by pre-transfer quintile, PSNP woredas only 2011 2016 35 32.5 30 27.5 25.5 25 23.4 23.7 20 16.8 17.9 14.3 15 12.6 10.2 10 5 0 Q1 Q2 Q3 Q4 Q5 Pre-transfer consumption quintile Source: WMS, 2016. World Bank staff calculations. 174 ETHIOPIA POVERTY ASSESSMENT Table 31 HOUSEHOLD  INCLUSION ERRORS SEEM LESS SALIENT THAN THEY APPEAR Selected indicators for PSNP households in the upper quintiles and non-PSNP households in PSNP woredas, 2016 PSNP HOUSEHOLDS NON-PSNP HOUSEHOLDS IN Q4 AND Q5 IN PSNP WOREDAS Female head of household (% yes) 34.1 16.6 HH completed primary (% yes) 6.3 11.8 Age HH 45.7 45 Widow HH (% yes) 14.9 8.2 Household size 4.2 5.9 Durable asset index -0.174 -0.015 Livestock index 0.002 0.118 Consumption expenditures per AE 16,706 11,842 Rural remoteness index 0.234 0.106 Food shortage (% yes) 11.5 14.8 Food gap (# of months) 0.38 0.5 Calories per day per AE 3,702 2,973 Note: The rural remoteness index is higher in more remote places. The food gap is calculated on all households, including those house- holds that did not report a food shortage. Source: WMS, 2016. World Bank staff calculations. To summarize, PSNP is overall well-targeted to poor 3.2 Highlands vs Lowlands households. While woreda selection does not contribute much to the targeting performance, with woredas included In line with evidence from the PSNP impact evaluations, in PSNP being not much worse-off compared to non-se- PSNP is well-targeted in the highlands. 36 percent of lected woredas, this is compensated by a good targeting beneficiaries in the highlands are in the poorest quintile, and performance within woredas. Despite this overall positive 60 percent of beneficiaries are in the poorest 40 percent. In picture, three main issues emerge: First, there is substantial the lowlands too, PSNP is reasonably well-targeted, with 65 under-coverage, with PSNP covering only 13 percent of the percent of beneficiaries in the poorest 40 percent. Relative to poor in 2016. Second, woreda selection can be improved the highlands though, fewer beneficiaries in the lowlands are to strengthen overall targeting performance, with the caveat in the poorest quintile while more are in the second quintile that geographical targeting is difficult in Ethiopia. Third, re- (37 percent). On the other hand, leakage to the upper quin- gional distribution of PSNP beneficiaries does not seem to tiles is higher in the highlands: 24 percent of beneficiaries in align to any indicator of monetary welfare or food security. the highlands are in the two upper consumption quintiles, The next section takes a more disaggregated look at high- compared to 15 percent in Somali and Afar. lands and lowlands. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 175 Figure 101 BENEFICIARY  INCIDENCE IS PROGRESSIVE BOTH IN HIGHLANDS AND LOWLANDS Distribution of PSNP beneficiaries across pre-transfer quintiles, 2016 Highlands Lowlands 40 36.1 36.5 35 30 28.3 25 23.1 20.2 20 17.2 15 12.7 10.9 9 10 6.1 5 0 Q1 Q2 Q3 Q4 Q5 Pre- transfer consumption quintile Source: WMS, 2016. World Bank staff calculations. PSNP is well-targeted on a range of other indicators as 2016 was 92 percent. The actual TD of 14.1 percent means well, particularly in the highlands. Relative to non-bene- that targeting in the lowlands achieved 15 percent of the ficiaries, PSNP beneficiaries in both highlands and lowlands maximum possible. Note that geographical targeting contrib- are poorer, have less durable assets, are headed by an old- utes substantially to overall targeting performance in the low- er head, and live in places that have less vegetation (Fig- lands but not in the highlands. This is explained by the fact ure 102 and Figure 103). In the lowlands however, there are that PSNP covers a large share of woredas in the lowlands, no discernable differences between PSNP and non-PSNP which implies there is less scope for geographical exclusion households when it comes to remoteness, self-reported food of the poor in the lowlands as compared to the highlands.68 shortages, and livestock holdings. To summarize, PSNP targeting is progressive in both Relative to what is theoretically possible, the high- the highlands and the lowlands. The poorest 20 per- lands achieve a better targeting differential. Taking into cent of households are better covered in the highlands, but account PSNP coverage in the highlands and the pre-trans- leakage to the upper quintiles (the better-off households) fer poverty rate, the maximum attainable TD in the highlands is higher in the highlands too. In 2016, the lowlands were amounted to slightly over 27 percent. The actual TD of 5.8 over-covered given the prevailing poverty and food insecurity means that targeting in the highlands achieved 21 percent rates. This over-coverage led to a higher targeting differen- of the maximum possible. In the lowlands, given that PSNP tial in absolute terms, but a lower targeting performance in coverage in 2016 was higher than the poverty rate, it would terms what could theoretically be achieved (a lower relative theoretically have been possible to cover all the poor. A TD of TD). First stage woreda selection contributes more to overall 100 percent would however not have been attainable, as the targeting performance in the lowlands, not because woreda over-coverage of the PSNP in the lowlands meant that non- selection is better per se, but because PSNP covers a higher poor households needed to be included as well (to reach share of woredas in the lowlands and therefore has lower the allocation). The maximum possible TD in the lowlands in exclusion as compared to the highlands. 68 In the 2016 HCES/WMS sample, 80 percent of woredas in the lowlands were included in the PSNP. In the highlands, 40 percent of woredas were included. 176 ETHIOPIA POVERTY ASSESSMENT Figure 102 PSNP  IS WELL TARGETED IN Figure 103 AND  ALSO IN THE LOWLANDS, THE HIGHLANDS… BUT TO A LESSER EXTENT Standardized mean differences between Standardized mean differences between PSNP and non-PSNP households in the PSNP and non-PSNP households in the highlands, 2016 lowlands, 2016 Self-reported Self-reported shortage food food shortage Age of AgeHH of HH head head remoteness RuralRural remotenessindex index Poorest Poorest 15% 15% Poorest Poorest 15% 15% Poorest Poorest 20% 20% Poorest Poorest 20% 20% Poorest Poorest 10% 10% Poorest Poorest 10% 10% HH engages HH engages in agriculture in agriculture Widow HH head Widow HH head Widow HH head Widow HH head Disability Disability in HH in HH Livestock Livestock Index Index HH engages HH engages in agriculture in agriculture Disability Disability in HH in HH Age Age of HH of HH head head HH sizeHH size HH sizeHH size Self-reported Self-reported shortage food food shortage HH has has wage HHwage incomeincome RuralRural remoteness remotenessindex index Livestock Livestock Index Index HH has has wage HHwage incomeincome Durable Durable assets index assets index Durable Durable assets index assets index (vegetation NDVINDVI index) (vegetation index) NDVINDVI (vegetation index) (vegetation index) -1 -1 -0.5 -0.5 0 0 0.5 0.5 1 1 -0.5 -0.4 -0.5-0.3 -0.4-0.2 -0.3-0.1 0 0.1 -0.2 -0.1 0.1 0.3 0 0.2 0.2 0.3 Note: Bars in green are statistically significant at the 10% Note: Bars in green are statistically significant at the 10% level or lower. level or lower. Source: WMS, HCES; 2016. World Bank staff calculations. Source: WMS, HCES; 2016. World Bank staff calculations. Figure 104 THE  HIGHLANDS ACHIEVE A HIGHER RELATIVE TARGETING DIFFERENTIAL Decomposition of absolute TD and TD as share of maximum attainable, 2016 Geographical selection Household selection Relative TD 25 21.4 TD, absolute and relative 20 15.3 15 10 7.4 5 4.7 6.7 0 1.1 Highlands Lowlands Note: Targeting differential is the difference between coverage of the poor and that of the non-poor. Relative TD expresses the TD as a percentage of what is possible under perfect targeting. Source: WMS, 2016. World Bank staff calculations. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 177 3.3 A Regional Perspective Afar. Coverage of the poorest quintile is highest in Amha- ra, with 43 of beneficiaries in the poorest 20 percent (Figure A regional analysis of PSNP incidence using the HCES/ 105). In Tigray, the single largest share of beneficiaries are WMS is constrained by issues of sample size. While the also in the bottom quintile (34 percent), while in Afar the sec- surveys are representative at regional levels and at urban and ond quintile accounts for most of the beneficiaries (mirroring rural levels within regions, the number of PSNP households the lowlands pattern presented in Figure 101). A common within regions is not always sufficiently high to have confi- pattern in the three regions is that the top consumption quin- dence in the patterns. This subsection briefly looks at benefi- tile accounts for a higher share of beneficiaries than either the ciary incidence for those regions where the sample contains third or fourth quintile. Although hypothetical, the over-cover- a sufficient number of PSNP beneficiaries. We put this “suf- age of the top quintile could be due to traditional perceptions ficient number” at an arbitrary 300 households. Using this of vulnerability, which are not necessarily related to consump- cut-off point, the region of Tigray, Afar, and Amhara qualify tion poverty: PSNP households in the fifth quintile are more for a regional analysis. likely to be small, female-headed (by a widow), and tend to be in the direct support component. While they are not con- The distribution of PSNP beneficiaries is progressive sumption-poor, they conform with traditional perceptions of in the three considered regions of Tigray, Amhara and vulnerability of households headed by an older woman. Figure 105 PSNP  IS PROGRESSIVE IN THE THREE CONSIDERED REGIONS Distribution of PSNP beneficiaries across pre-transfer consumption quintiles, by region, 2016 50 A. Tigray 50 B. Afar 39.9 50 40 50 50 40 50 35.5 33.6 39.9 39.9 40 30 4035.5 35.5 40 30 40 33.6 33.6 21.7 30 20 30 16.5 30 20 30 21.7 21.713.4 12.9 13 20 20 16.5 16.5 10 20 20 8.7 10 13.4 13.4 12.9 12.9 4.8 13 13 8.7 8.7 10 0 10 10 0 10 4.8 4.8 Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 0 0 0 0 Pre-transfer consumption quintiles Pre-transfer consumption quintile Q1 Q1 Q2 Q2 Q3 Q3 Q4 Q4 Q5 Q5 Q1 Q1 Q2 Q2 Q3 Q3 Q4 Q4 Q5 Q5 Pre-transfer Pre-transfer consumption consumption quintiles quintiles Pre-transfer Pre-transfer consumption consumption quintile quintile C. Amhara 50 43.1 50 40 50 43.1 43.1 40 30 40 22.5 30 20 30 22.5 22.5 12.3 13.6 20 20 8.6 10 12.3 12.313.6 13.6 8.6 8.6 10 0 10 Q1 Q2 Q3 Q4 Q5 0 0 Pre-transfer consumption quintile Q1 Q1 Q2 Q2 Q3 Q3 Q4 Q4 Q5 Q5 Pre-transfer Pre-transfer consumption consumption quintile quintile Source: HCES, WMS, 2016; World bank staff calculations. 178 ETHIOPIA POVERTY ASSESSMENT In all three regions, household selection within wore- The finding that Afar achieved the highest targeting das is the main contributor to targeting performance. differential in 2015/16 may seem surprising. The PSNP In Tigray and especially in Amhara, the first-stage selection evaluations conducted by IFRPI routinely find evidence of of woredas does not add significantly to the targeting differ- poor targeting in the lowland regions of Afar and Somali.71 ential, meaning that PSNP woredas in those regions have It appears that this discrepancy is partly driven by differenc- poverty rates similar to non-PSNP woredas.69 In absolute es in outcome variables. Evaluating PSNP targeting in Afar terms, Afar has the highest targeting differential (coverage of based on household consumption expenditures and house- the poor is 24 percentage points higher than coverage of the hold durable assets results in a good targeting performance. non-poor). Even accounting for differences in PSNP cover- Evaluating it based on a livestock index however changes age and poverty rates across the regions, Afar comes closest this result, making PSNP targeting regressive (Table 32). to the maximum possible TD (relative TD of 27 percent), fol- Statements on the quality of targeting will therefore crucially lowed by Amhara (24 percent) and Tigray (15 percent). The depend on the yardstick against which targeting is evaluated. relatively good performance of Afar is due to the fact that the While targeting in Afar is good when evaluated against dura- PSNP in Afar covers all woredas: There is no ex-ante exclu- ble assets and consumption expenditures, the same cannot sion of the poor because of a first-stage woreda selection. 70 be said when evaluated against livestock. Figure 106 AMONG  THE THREE REGIONS, AFAR HAS THE HIGHEST ABSOLUTE AND RELATIVE TARGETING DIFFERENTIAL Decomposition of absolute targeting differential and share of maximum possible, 2016 Geographical selection Household selection Relative TD 30 27.4 25 23.5 20 Actual TD 15.4 15 17 10 5 8.8 8.6 7 1.4 0.7 0 Tigray Afar Amhara Note: Targeting differential is the difference between coverage of the poor and that of the non-poor. Relative TD expresses the TD as a percentage of what is possible under perfect targeting. Source: WMS, 2016. World Bank staff calculations 69 In Afar, all woredas are included in the PSNP. The relatively high contribution of woreda selection in Afar can be explained by the fact that the poorer woredas tend to have the most beneficiaries. 70 The finding that Afar has a relatively high TD while the lowlands as a whole (Afar and Somali) have a relatively low TD must mean that Somali region has a low TD. Indeed, based on the 2016 HCES/WMS, the PSNP in Somali achieved a mere four percent of the maximum possible. This is however based on a small sample (231 PSNP beneficiaries in the HCES sample in Somali region). 71 The 2018 PSNP-4 midline evaluation reports following PSNP coverage by livestock quintile for Afar: 54 percent for the lowest live- stock quintile (Q1), 60 percent in Q2, 61 percent in Q3, 57 percent in Q4, and 47 percent in the highest livestock quintile. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 179 Table 32 TARGETING  PERFORMANCE IN AFAR DEPENDS ON THE OUTCOME VARIABLE CONSIDERED Share of PSNP beneficiaries by quintiles of consumption, durable assets, and livestock; Afar 2016 CONSUMPTION DURABLE HOUSEHOLD PSNP (PRE-BENEFIT) ASSETS LIVESTOCK INDEX Q1 33.6 33.5 8.8 Q2 39.9 34.5 35.1 Q3 4.8 20.5 18.6 Q4 8.7 6.3 11.3 Q5 13 5.2 26.3 Note: Table shows the share of PSNP beneficiaries by quintile of the relevant variable. Source: HCES; WMS, 2016. World Bank staff calculations. 4. HUMANITARIAN FOOD AID While PSNP aims to cover the chronically poor and food-insecure, HFA aims to cover people with acute food needs, mainly related to shocks. Overall, HFA was reasonably well targeted in 2016, with HFA benefi- ciaries being poorer, more remote, more likely to report food shortages, and having fewer assets and living in less green places. Similar to the PSNP, first-stage geographical targeting does not significantly add to HFA targeting performance, reflecting the fact that food insecurity is not geographically clustered in Ethiopia. Despite an overall good targeting, inclusion errors exist, with a substantial share of HFA beneficiaries being significantly better-off on a wide range of indicators. These inclusion errors are mainly due to HFA targeting in woredas where PSNP is not active: HFA is well targeted in woredas where PSNP is active, though poorly targeted where PSNP is not active. About 2 percent of the population in Ethiopia were covered by both PSNP and HFA in 2016, and this overlap was well-targeted to the poorest. Results suggest HFA targeting could be improved by harmonizing PSNP and HFA and revisiting, in tandem with PSNP, the procedures for first-stage geographical targeting. While the PSNP seeks to address chronic and predict- dependent on the resources available. Woreda staff are able food insecurity in the lean season, humanitarian responsible for selecting kebeles and defining the kebele food aid (HFA) addresses acute food insecurity caused caseload. Household targeting is undertaken by community by shocks, mainly drought. HFA needs are typically as- structures. sessed twice yearly during seasonal assessments with the Given that HFA aims to address acute food insecurity results released in a joint Government-Humanitarian Partners following an adverse shock, it is not supposed to be appeal document. The multi-agency seasonal assessment targeted explicitly towards poverty. To assess HFA tar- produces woreda level estimates of numbers of people in geting, this section considers a whole range of indicators, need of assistance and duration of assistance. It does so similar to the PSNP analysis presented earlier. Given that by reviewing woreda level assessments and depends heav- there is nevertheless a consensus that the poor should be ily on subjective data on crop production and rainfall. While prioritized in case of a shock, we also look at targeting of HFA each round of support of HFA is provided to the full approved as it relates to consumption poverty. beneficiary caseload, the duration of support provided is 180 ETHIOPIA POVERTY ASSESSMENT 4.1 The national picture While not supposed to be explicitly targeted towards poverty, poorer households were more likely to receive The latest HCES/WMS surveys were implemented HFA. In 2016, 31 percent of HFA beneficiaries were in the during a bad drought year. The 2015/16 El Nino drought bottom consumption quintiles and another 28 percent in caused the HFA caseload to swell from about 2.5 million in the second quintile (Figure 108). Beyond the second quin- 2014/15 to 10 million in 2015/16.72 Oromia accounted for the tile, beneficiaries were equally likely to come from the third, bulk of the HFA beneficiaries in 2015/16 (36 percent), followed fourth, of top quintile. Almost 30 percent of HFA beneficiaries by Amhara (22 percent) and Somali (15 percent). Coverage in 2016 were in the top two quintiles in terms of pre-food-aid of HFA was largest in Somali (27 percent of the population consumption, indicating a fair degree of leakage (for PSNP, covered), Afar (25 percent) and Tigray (23 percent). Compar- the corresponding share was 21 percent – see Figure 92). ing the numbers in Figure 107 with the numbers reported in HFA targeting was progressive in 2011 as well, though less Table 28 above reveals substantial disparities between needs so than in 2016. In terms of binary poverty status, 37 percent and allocations, with the number people deemed in need of of HFA beneficiaries were below the national poverty line in HFA exceeding the number of self-reported food-insecure 2016 (using a pre-food aid estimate of consumption expen- people in several regions (mainly Afar, Somali and Tigray). ditures), compared to 39 percent for PSNP. Figure 107 PEOPLE WHO NEED HUMANITARIAN FOOD ASSISTANCE OROMIA ACCOUNTED FOR THE BULK OF HFA NEEDS  People in need of food aid according to HRD, 2016 10.2 M 1.2M TIGRAY AMHARA AFAR 2.2M 439,218 BENESHANGUL GUMUZ 79,357 7 79 357 DIRE DAWA 56,771 14, 14,500 14,5 ,500 , HARERI Addis Ababa GAMBELA 39,800 3.7M SOMALI SNNPR OROMIA 1.5M 756,483 Number of relief food ## bene ciaries per region Source: Joint GoE and DP assessment, 2016 72 Perhaps due to the timing of the survey (July 2015 to July 2016), the HCES does not reflect this big caseload: In the survey, five percent of people reported receiving food aid-about half the share compared to administrative data. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 181 Figure 108 THE  BOTTOM QUINTILES ARE OVERREPRESENTED AMONG HFA BENEFICIARIES Distribution of HFA beneficiaries across pre-transfer quintiles, 2011 and 2016 2011 2016 35 31 30 27.4 28 24 25 20 18.2 17.6 15.3 15 12.4 12.8 13.3 10 5 0 Q1 Q2 Q3 Q4 Q5 -food aid consumption Quintile of pre Source: WMS, 2011, 2016. World Bank staff calculations. Other indicators also suggest a decent targeting per- of productive assets (livestock). Similar to PSNP, the main formance of HFA. Relative to non-beneficiaries, beneficia- correlate of a household getting HFA is the greenness of ries lived in more remote places, were poorer, and more likely its surroundings: Relative to non-beneficiaries, beneficiary to report food shortages in the 12 months prior to the survey. households lived in places that, in 2016, had far lower values HFA beneficiaries scored lower on a durable assets index, on the normalized difference vegetation index. though there was no discernible difference in ownership Figure 109 TARGETING  OF HFA BENEFICIARIES IS PROGRESSIVE Standardized mean differences between HFA and non-HFA households, 2016 Rural remoteness index Poorest 15% Poorest 20% HH size Poorest 10% Self-reported food shortage HH engages in agriculture Age of HH head HH has wage income Livestock Index Widow HH head HH has some with a disability Durable assets index NDVI (vegetation index) -2 -1.5 -1 -0.5 0 0.5 1 1.5 Note: Bars in green are statistically significant at the 10% level or lower. Source: WMS, HCES; 2016. World Bank staff calculations. 182 ETHIOPIA POVERTY ASSESSMENT An important question is whether HFA effectively re- of values observed in the same period in previous years. As sponded to the intensity of the drought in 2016. The such, high levels of VCI indicate that vegetation is better than NDVI shows that HFA on average went to households who normal, while low levels indicate that vegetation is worse than lived in places that are less green than other places (a spa- normal. Exceptionally low levels indicate a drought shock. As tial comparison). A more important question is whether HFA shown in Figure 110, households in places that had excep- was more likely to cover households in places that had, in tionally poor vegetation during the 2015 Meher harvest were 2015/16, an exceptionally low greenness compared to a far more likely to receive food aid compared to households long-term average (a temporal comparison). The Vegetation in places with normal or good vegetation, indicating that HFA Condition Index (VCI) compares current NDVI to the range responded to the spatial intensity of the drought. Figure 110 HFA  WENT TO HOUSEHOLDS WHO LIVED IN PLACES WITH EXCEPTIONALLY POOR VEGETATION Local polynomial smoothing between receiving HFA and vegetation anomalies .8 .7 .6 HHs started food aid .3 .4.2 .1 0 .5 0 20 40 60 80 100 Vegetation anomalies, harvest season Source: Sohnesen, 2018. Though HFA targeting is progressive, there are never- HFA beneficiaries from the top two quintiles lived in smaller theless substantial inclusion errors: Close to 30 percent households, had a higher calorie intake, and far higher con- of beneficiaries were in the top two consumption quintiles sumption expenditures, both pre- and post HFA. They were in 2016. As HFA is not targeted on poverty per se, such a not more likely to have experienced food shortages nor did pattern would not be a cause for concern if other indicators they have fewer assets or less livestock (quite to the con- would show that these households were in need of acute trary). While these households lived in more remote and drier aid following a shock. This narrative is not supported by the places, they were actually better-off than households in less data: Relative to non-beneficiaries in the same woredas, remote, slightly greener places (Table 33). CHAPTER VI. POVERTY AND SOCIAL PROTECTION 183 Table 33 HFA  INCLUSION ERRORS ARE FAIRLY SALIENT Selected indicators for HFA households in the upper quintiles and non-HFA households in HFA woredas, 2016 NON-HFA BENE- HFA BENEFICIA- FICIARIES IN HFA STATISTICALLY RIES IN Q4/Q5 WOREDAS SIGNIFICANT HH size 5.10 5.62 ** Durable asset index 0.12 0.13 Rural remoteness index 0.29 0.04 ** Livestock index 0.22 0.03 HH head widow (% yes) 8.8% 10.0% Age of HH head 42.40 45.00 Female HH head (% yes) 20.0% 20.7% HH has a disabled member (% yes) 7.2% 7.2% HH does agriculture (% yes) 90.5% 86.9% HH Head has completed primary education (% yes) 15.7% 12.5% NDVI (vegetation) 3,795 4,205 *** Consumption per AE (post food aid) – 19,629 12,847 *** December 2010 prices Consumption per AE (pre-food aid) - 18,588 12,847 *** December 2010 prices Calorie intake per day per adult 3,476 2,799 *** Food shortage past 12 months (% yes) 9.6% 12.2% Food gap past 12 months (# months) 0.27 0.36 Number of observations 332 7,837 Note: The rural remoteness index is higher in more remote places. The food gap is calculated on all households, including those house- holds that did not report a food shortage. NDVI is the normalized differenced vegetation index, with higher values denoting more vegeta- tion. ***: Statistically significant at 1%; **: Statistically significant at 5%. Source: WMS, 2016. World Bank staff calculations. 4.2 Woreda vs household targeting woredas - Figure 111. Looking at food shortages rather than poverty, the pattern is largely similar: Coverage of the food Like the PSNP, selection of woredas and their case- insecure was 3.2 percentage points higher than coverage loads does not add much to overall targeting per- of the food secure, and woreda selection did not contribute formance of the HFA. In 2016, the poverty targeting to the TD (Figure 111). In other words, HFA beneficiaries in differential for HFA was 3.3 percent, meaning that HFA cov- 2016 were not clustered in woredas with higher levels of erage of the poor was 3.3 percentage points higher than self-reported food shortages.73 This finding again reflects the HFA coverage of the non-poor. Over 80 percent of the TD fact that food insecurity, much like poverty, is not geograph- is explained by selection of kebeles and households within ically clustered in Ethiopia. woredas, and less than 20 percent by first-stage selection of 73 Of course, this result needs to be interpreted with caution, as receiving food aid may lead to fewer self-reported food shortages. 184 ETHIOPIA POVERTY ASSESSMENT Figure 111 WOREDA  SELECTION DOES NOT ADD MUCH TO HFA TARGETING PERFORMANCE Decomposition of absolute TD for HFA, 2016 Geographical selection Household selection 4 3.5 Targeting differential 3 2.5 2 2.7 3.4 1.5 1 0.5 0.6 0 -0.2 -0.5 Poverty Food shortage Source: WMS, 2016. World Bank staff calculations To summarize, targeting of HFA is progressive in the this section, we use the HCES and WMS data to look at sense that poorer and more food-insecure households overlap and complementarity between these two social pro- were more likely to be covered. There are however sub- tection and food security interventions. An important caveat stantial inclusion errors with a fair share of genuinely bet- is in order: Though participation in PSNP and/or HFA was ter-off households benefiting from HFA in 2016. While HFA measured by two separate questions in the survey ques- targeting is progressive both in the highlands and lowlands, tionnaire, it is not entirely clear to what extent respondent the poorest households (the bottom quintile) tend to be un- households correctly made that distinction. For instance, in der-covered in the lowlands. The difficulty of geographical 2015/16 the duration of PSNP support was extended from targeting in Ethiopia is also borne out of the HFA analysis: its normal term of six months to nine months to respond First stage woreda selection adds little to the HFA targeting to the ongoing drought. It is unclear whether beneficiary performance. Despite its overall pro-poor nature, HFA seems households labeled this as the PSNP or as HFA. to suffer from a regional mismatch also observed for PSNP, According to the survey, about two percent of the with certain regions being substantially overcovered relative Ethiopian population benefited from both PSNP and to needs while others substantially under-covered. There is HFA in 2015/16. HCES data suggest that of all PSNP ben- ample room to further improve HFA targeting by revisiting the eficiaries in 2016, one in four also received food aid. Of all regional caseloads and the first-stage selection of woredas. food aid beneficiaries, one in three were also PSNP clients. Overlap was highest in the lowland regions of Afar (18 per- 4.3 Overlap and complementarity cent of the population) and Somali (8 percent of the popula- between HFA and PSNP tion). There is a large overlap between PSNP and HFA at the woreda-level: More than 90 percent of PSNP woredas also In the absence of a comprehensive social registry in received HFA, and over 70 percent of HFA woredas also Ethiopia, the extent of overlap between the PSNP and benefit from PSNP.74 HFA cannot be assessed with administrative data. In 74  Hirvonen et al., 2019. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 185 Complementarity between PSNP and HFA is often Beneficiary characteristics suggest that both pro- conceptualized as HFA covering non-PSNP house- grams largely reach who they are supposed to reach holds in case a woreda is affected by a shock. 75 If and complementarity is, to some extent, achieved. that is the case, we would expect PSNP to mainly cover PSNP beneficiaries are clearly worse off than the average the bottom quintiles (the poor) while HFA would over-cover rural population and have the “typical” characteristics of the households that are not in the PSNP and that are presum- poor, as presented in Chapter 2 (Table 34). In contrast, HFA ably in higher quintiles. This is partly supported by the data: beneficiaries are largely comparable to the average rural In PSNP woredas, coverage of PSNP decreases monotoni- population, with the exception that they have significantly cally as one moves to higher quintiles. HFA coverage follows lower calorie intake. The descriptives presented in Table 34 a weak U-shaped pattern, with both poorer and better-off are consistent with HFA covering average households who households (in terms of consumption) more likely to be cov- were faced with an acute shock which temporarily made ered (Figure 112). Overall, coverage of either PSNP or HFA their calorie intake decrease, while PSNP covers house- (or both) in PSNP woredas is progressive, with coverage de- holds who would be poor in most states of the world. creasing as we approach higher quintiles. The issue of un- der-coverage is obvious from Figure 112: Even within PSNP woredas, two-thirds of the poorest (the bottom quintile) re- main un-covered by either PSP or HFA. Figure 112 HFA  COMPLEMENTS THE PSNP IN PSNP WOREDAS Coverage by program and quintile in PSNP woredas, 2016 PSNP only HFA only Both 40 35 30 8.9 25 6.6 6.4 20 5.7 2.9 2.8 2 15 1.9 4.8 10 4.9 18.8 17.2 14 5 10.9 9 0 Q1 Q2 Q3 Q4 Q5 Pre -bene t consumption quintiles Source: WMS, 2016. World Bank staff calculations. 75 Indeed, a recent assessment by IFPRI found that in qualitative interviews woreda and community respondents indicated that non- PSNP households are prioritized in the allocation of HFA, though PSNP households are also eligible. 186 ETHIOPIA POVERTY ASSESSMENT Table 34 PSNP  AND HFA REACH DIFFERENT BENEFICIARY GROUPS Beneficiary characteristics by program, 2016 PSNP AND RURAL PSNP ONLY HFA ONLY HFA POPULATION HH size 5.56 6.04 6.22 5.82 Durable asset index -0.30 -0.10 -0.29 -0.01 Rural remoteness index 0.41 0.59 1.14 0.06 Livestock index -0.04 0.23 -0.01 0.16 HH head widow (% yes) 14.3% 6.4% 9.4% 9.0% Age of HH head 45.81 44.07 44.81 44.50 Female HH head (% yes) 29.6% 17.6% 24.9% 18.5% HH has a disabled member (% yes) 9.1% 3.9% 6.5% 6.3% HH does agriculture (% yes) 92.2% 93.0% 95.1% 91.4% HH Head has completed primary education (% yes) 4.2% 10.3% 5.8% 11.2% NDVI (vegetation) 3,581 3,598 3,310 4,811 Calorie intake per day per adult 2,770 2,696 2,585 3,039 Food shortage past 12 months (% yes) 20.0% 11.7% 22.0% 10.9% Food gap past 12 months (# months) 0.62 0.38 0.72 0.35 Number of observations 1,290 668 338 18,320 Note: The rural remoteness index is higher in more remote places. The food gap is calculated on all households, including those house- holds that did not report a food shortage. NDVI is the normalized differenced vegetation index, with higher values denoting more vegeta- tion. ***: Statistically significant at 1%; **: Statistically significant at 5%. Source: WMS, 2016. World Bank staff calculations. Another aspect of complementarity is that HFA ap- Overlap between PSNP and HFA benefits the poorest. pears to be better targeted in woredas that also have 43 percent of people who received both PSNP and HFA in PSNP. 37 percent of HFA beneficiaries in PSNP woredas 2016 were in the bottom consumption quintile, and 74 per- were in the bottom consumption quintile in 2016 and an- cent were in the bottom 40 percent of welfare (Figure 114). other 27 percent came from the second quintile (left-hand Looking at the pre-benefit poverty rates (right-hand side pan- side graph of Figure 113). In contrast, the distribution of HFA el of Figure 114), it is clear that the combined HFA-PSNP beneficiaries in non-PSNP woredas seemed fairly random, beneficiaries are the poorest: Over half of them would be with no better inclusion of the bottom consumption quintiles below the national poverty line if PSNP/HFA benefits were (right-hand side graph of Figure 113). The HFA inclusion er- removed from their consumption. Looking at post-aid pov- rors summarized in Table 33 are mainly due to targeting in erty rates (based on the official consumption aggregate), 35 non-PSNP woredas. In PSNP woredas, HFA can draw on percent of combined beneficiaries were poor in 2016, com- pre-existing PSNP structures and experiences, which posi- pared to the national poverty rate of 24 percent. The figures tively affects targeting performance of HFA. in Table 34 confirm that combined PSNP+HFA beneficiaries are by and large the poorest and most vulnerable. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 187 Figure 113 HFA  IS BETTER TARGETED IN PSNP WOREDAS HFA beneficiaries by quintile, PSNP woredas HFA beneficiaries by quintile, non-PSNP woredas 40 36.7 40 30 27.4 30 26.7 22.5 19 20 20 16.8 15 12 12 12 10 10 0 0 Q1 Q2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 Q5 Pre-bene t consumption quintiles Pre-bene t consumption quintiles Note: Pre-benefit consumption quintiles are based on a consumption aggregate that is filtered of PSNP and HFA benefits. Source: WMS, HCES, 2016. World Bank staff calculations Figure 114 OVERLAP  IS WELL TARGETEDS Share of PSNP + HFA beneficiaries by pre-aid pre- and post-aid poverty rates for HFA, PSNP, consumption quintile, 2016 and HFA+PSNP beneficiaries, 2016 6060 6060 51.4 51.4 5050 5050 43.5 43.5 4040 4040 3535 35.1 35.1 31.9 31.9 30.1 30.1 29.4 29.4 3030 3030 24.7 24.7 2020 2020 12.5 12.5 1010 7.3 7.3 6.7 6.7 1010 00 00 Q1Q1 Q2Q2 Q3Q3 Q4 Q4 Q5 Q5 PSNP PSNP only only HFA HFA only only PSNP PSNP and and HFA HFA Pre PSNP Pre and PSNP andHFA HFAconsumption quintiles consumption quintiles Poverty Poverty(pre (prebene bene ts) ts) Poverty Poverty (of (of cial-after cial-after bene bene ts) ts) Note: Pre-benefit consumption quintiles are based on a consumption aggregate that is filtered of PSNP and HFA benefits. Source: WMS, HCES, 2016. World Bank staff calculations 188 ETHIOPIA POVERTY ASSESSMENT Conclusions Overall, this Chapter painted a positive picture of tar- either Tigray or Amhara. This was not because targeting com- geting of Ethiopia two largest social protection and food mittees in Afar were superior in selecting poor households. It security programs. Both in 2011 and 2016, both the PSNP was because Afar includes all woredas in PSNP and hence and the HFA were progressive in the sense that households does not have a first-stage exclusion of the poor. and people from the lower income quintiles were more likely to Given that targeting of households within selected be covered. Looking beyond monetary living standards, ben- woredas is generally good, improving PSNP’s ability to eficiaries of both programs tended to be worse-off on a wide reach the poor will likely require a shift in the approach range of indicators of well-being. As such, both programs pos- to geographical targeting. A possible scenario is to ex- itively contribute to Ethiopia’s pro-poor development strategy. pand the coverage of PSNP to all rural woredas, but select a Despite this overall positive assessment, there is room far smaller number of beneficiaries per woreda to manage the for improvement. While targeting differentials for both pro- fiscal implications. Including all rural woredas has the obvious grams are positive, indicating their progressiveness, they are advantage to remove the “exclusion by design” feature of relatively small compared to what could be obtained under the current woreda selection approach. It would also reflect hypothetical (and unattainable) perfect targeting. The main the reality that geographical targeting in Ethiopia is wicked avenues to further improve targeting performance of the pro- hard. Having a small beneficiary caseload within each wore- grams is through revisiting the process to determine the re- da could also help in reducing the inclusion errors, as the gional allocation of beneficiaries and the first-stage selection small caseload would only be sufficient to cover the poorest of woredas within regions. of the poor in the woreda. An extension of the PSNP to all rural woredas but with smaller woreda allocations should of The regional beneficiary numbers seem not to be well course be properly costed to assess the fiscal implications. connected to actual needs. Both for PSNP and HFA, the more populated regions appear under-covered relative to For the lowland regions, a smaller but more flexible needs (with the exception of Amhara), while the smaller regions PSNP may be envisaged. In 2016, the PSNP beneficiary are substantially over-covered. This under- and over-coverage numbers in Afar and Somali exceeded the number of poor introduces significant inclusion and exclusion errors: In the people in these regions and far exceeded the food-insecure. under-covered regions, the bulk of the poor remain excluded Arguably, a better use of scarce resources would be to reduce from any form of assistance; in the over-covered regions, rela- the core caseload in these regions but design a flexible system tively better-off households need to be included as the benefi- that can be scaled up rapidly and efficiently in times of crises. ciary allocation for these regions exceeds the number of poor Finally, the data presented in this Chapter suggest that and food-insecure in these regions (most notably the case in issues of food insecurity have become progressively Somali and Afar). Designing a way to bring regional caseloads less salient in Ethiopia. This largely reflects Ethiopia’s de- more in line with needs will further improve targeting perfor- velopment success over the past decades. Poverty, while it mance and the programs’ impacts on poverty. has also decreased substantially, has remained stickier, going Within regions, first-stage targeting of woredas adds lit- from 39 percent in 2005 to 24 percent in 2016. Overall, while tle to overall targeting performance. This is not so much a only one in ten people in Ethiopia suffer from self-reported result of outright mistargeting than it is of the fact that pover- food shortages, one in four still below the poverty line. Given ty is not geographically concentrated in Ethiopia. In a context the evidence on the impact of PSNP on household consump- such as Ethiopia’s where poverty is spread across all areas tion the Government could consider reorienting the focus of of the country, geographical targeting will inevitable lead to safety nets to poverty in general and target benefits to the exclusion of a large share of the country’s poor. To illustrate, poorest rather than the narrowly-defined food insecure. This of all Ethiopia’s poor in 2016, half lived in woredas covered shift would lead to a more inclusive social protection policy. At by PSNP.76 The other half lived in woredas not in PSNP and the same time, reducing fragmentation of the national safety are, by design, excluded. The role of woreda selection in over- net by consolidating the PSNP and emergency drought relief all targeting performance is nicely illustrated by Afar. For the (HFA) under one scalable rural safety net would improve the PSNP, Afar had, in 2016, a better targeting performance than efficiency and effectiveness of the social protection system. 76  Looking at self-reported food insecurity, 66 percent of the self-reported food insecure lived in PSNP woredas. CHAPTER VI. POVERTY AND SOCIAL PROTECTION 189 190 ETHIOPIA POVERTY ASSESSMENT CHAPTER VII Inequality of Opportunity in Ethiopia While inequality in household consumption expenditures in Ethiopia is low, there are substantial disparities in access to key services and opportunities. This chapter uses the Human Opportunity Index to assess access to opportunities for children in Ethiopia, how access to these opportunities is determined by circumstances outside one’ control, and how coverage of opportunities has evolved between 2011 and 2016. It also investigates the strength of the intergenerational relationship of educational attainment, how that relationship differs across characteristics, and how it has changed over time. Overall, access to opportunities has improved between 2011 and 2016 and disparities in access have narrowed, leading to an increase in the Human Opportunity Index. The location of the household and household wealth, circumstances that are largely outside children’s control, are the main factors determining access to key opportunities: 40 percent of children 15-18 in urban areas were enrolled in secondary school in 2016, compared to 10 percent of rural children. Half of children (15-18) in households from the top consumption quintile had completed primary school, compared to less than 20 percent of children in the bottom consumption quintile. The extent to which parental education influences children’s education has weakened between 2011 and 2016, reflecting an increase in intergenerational mobility. The probability of a child being enrolled in primary school has become less dependent on parental education, while the opposite happened for enrolment in secondary school. Improvements in access to education happened for children with poorly educated parents in urban areas and children with relatively higher educated parents in rural areas. Household consumption levels have a large influence on whether the household’s children go to school. The effect of household consumption on access to primary school did not change between 2011 and 2016, while its effect on enrolment in secondary school increased. The effect of household welfare levels on children’s schooling is significantly stronger in rural areas. The implication of these results is that children of poor households and poorly-educated parents in rural areas are in danger of being left behind. Breaking the intergenerational transmission of poverty will require that children of extremely poor households in rural areas accumulate more schooling, which may require the introduction of additional policy instruments. CHAPTER VII. INEQUALITY OF OPPORTUNITY IN ETHIOPIA 191 Introduction Over the last two decades, Ethiopia has experienced Gaps in access to basic opportunities may be predictive significant economic growth. Although economic growth of future gaps in outcomes, and widening disparities in started from a relatively low base, the pattern of growth saw outcomes may become problematic for widespread eco- a major shift from negative GDP growth in 1998 to double nomic growth for a country (World Bank (2006), Bour- digit growth by 2004. Today Ethiopia is the second most guignon, Ferreira et al. (2007); Ferreira and Walton (2006)). populous country in Africa, the fastest growing economy in If inequality of opportunity is pervasive and persistent, the the region, and one of the fastest in the world. GDP growth effects have the potential to entrench the existing gaps be- has remained high at close to eight percent in 2018. De- tween different socioeconomic groups and reproduce living spite this growth, there has been a relatively weak transmis- standards across generations. The link between inequality sion between economic growth and poverty reduction in the of opportunity and inequality of outcomes is important, as country over the last decade (see Chapter 1) and social and policies to address gaps now are aimed at reducing welfare economic outcomes for the poor remain precarious. gaps in the future. Though there has been an increase in recent years, in- An important indicator of access to opportunity is the equality of outcomes in Ethiopia remains relatively low. extent of intergenerational mobility in a country – how The Gini Coefficient of 0.33 in 2016 was slightly higher than much a child’s outcomes depends on those of his or the 0.30 of 2011, but remains low in regional comparison. her his parents. In an ideal society, a child’s access to op- There were however increasing disparities between urban portunities is independent of his or her parents’ outcomes. and rural parts of the country. These changes naturally lead Societies with equal access to opportunities have high social to a focus on better understanding what is generating these mobility since children are not constrained by their circum- disparities - not only in outcomes, but in access to opportu- stances at birth (Chetty, Hendren et al. (2014), Solon (2014)). nities and its effects on socioeconomic mobility. Data constraints do not allow us to match future economic outcomes of children with those of their parents’. As a result, Evidence in this and the previous poverty assessment this chapter will focus on intergenerational mobility in edu- has shown divergent trends of growth based on geo- cation along two dimensions: differences in average educa- graphic location. The urban-rural differences highlighted in tional attainment, and the relationship between a child and the last poverty assessment (World Bank (2015)) have been parents’ level of education. consolidated. The last two decades have seen heavy invest- ment in some of the most substantial safety net programs This chapter investigates inequality of opportunity and in the region. While this has had a poverty-reducing effect, intergenerational mobility in Ethiopia between 2011 the consumption gaps between rural and urban areas have and 2016, and addresses two key questions: (i) To what increased. Of particular concern is the contraction of the in- extent does the gap in access to key services in Ethiopia de- come accruing to the bottom 10 percent of households. pend on circumstance variables? How has this evolved over time, and what are the biggest drivers of unequal access? Generating a better understanding of inequality of op- And (ii) What are the patterns of intergenerational educational portunity (IOp) is key to thinking about how poverty attainment in Ethiopia? Has his changed over time, and are and inequality in a society may develop in future years. there differences across different parts of the population? Broadly speaking, the focus of inequality research has ex- panded to place greater emphasis on access to opportuni- This chapter is divided into several sections. Section 2 ties and its relationship to birth circumstances such as race, investigates the Human Opportunity Index (HOI) for Ethiopia, gender, and location of birth, or even the wealth of a house- shows how this has changed since 2011, and decomposes hold a child is born into. These circumstances are generally it into key components. Section 3 analyses the intergener- thought of as being outside of an individual’s control. If ac- ational transmission of educational attainment, and shows cess to opportunities is influenced by circumstances that are how the strength of the relationship differs by various parts of out of the control of individuals, then any resulting inequal- the population. The final section presents some concluding ity is fundamentally unfair. If, in contrast, all individuals in a remarks and potential policy avenues that can be explored. society have similar access to opportunities, any remaining inequality in outcomes will mainly be the result of differences in talent, hard work, or choices, which can be considered fair. 192 ETHIOPIA POVERTY ASSESSMENT 2. THE HUMAN OPPORTUNITY INDEX IN ETHIOPIA 2.1 Background and methodology The extent of inequality of opportunity is measured us- ing the D-index. This index calculates how much access to The Human Opportunity Index (HOI) is widely used to services varies by birth characteristics, such as socio-eco- measure inequality of opportunity. The HOI captures nomic status of a households and the location of the house- both (i) the overall access to basic services, such as edu- hold. A D-index of zero indicates perfect equality (no gaps in cation, water and electricity, (the coverage rate); and (ii) in- access to services across circumstance groups), whereas a equality in access (Barros, Ferreira et al. (2008)). If access to D-index of one indicates perfect inequality. More information a basic service is perfectly equal, then the HOI is the same on the construction of the HOI and the D-index is provided as the coverage rate. As access becomes more and more in Box 24. unequal, so the HOI becomes lower and lower. Box 24 C  onstructing the HOI and the D-Index The central question behind the HOI is to what extent circumstances beyond one’s control influence the one’s access to a set of important basic services. Simply put, the HOI takes the coverage level of a basic service or “opportunity” (for example whether a child is enrolled in primary education) and combines this with the extent to which that opportunity is determined to be beyond the control of the child (for example being born in a rural rather than urban area or being a girl rather than a boy). Ideally, random circumstances should play no role in determining access to opportunities. The D-index measures dissimilar access rates to a given basic opportunity for groups of children where groups are defined by circumstance characteristics (for example, area of residence, or gender) compared to the average access rate to the same service for the population as a whole. To formulate groups the sample is stratified into groups or “cells,” so that all individuals in any given cell have the same combination of circumstances. The result- ing subgroups are known in the literature as “types” (Barros, Ferreira et al. (2008)). These cells are then compared to one another. The difference in outcomes between cells can be attributed to inequality of opportunity, while the differences within cells can be considered the result of effort or luck. The D-index summarizes all the gaps into a single measure by weighting them according to the population share in each circumstance group. The D-index generates a value between 0 and 1. In a society in which there is no inequality of access the D-index is be zero. If average access is denoted by p, the specific access rate of group i is pi, and the share of group i in the population is given by ßi then the D-index is: n ∑ 1 D= ßi | pi – p| 2p i =1 The HOI can then be calculated as: HOI = p (1 – D) The measure is also decomposable so that the extent to which specific opportunity sets contribute to the dis- similarity can be assessed. This means that the contribution of different circumstances to overall inequality of opportunity can be determined. CHAPTER VII. INEQUALITY OF OPPORTUNITY IN ETHIOPIA 193 To assess access to opportunities in Ethiopia, we fo- for good health and nutrition which also affects outcomes cus on the six key outcomes: (i) Primary school enrol- in adulthood, while access to electricity can produce many ment, (ii) primary school completion, (iii) secondary school potential benefits, including increasing the time available for enrolment, (iv) access to electricity; (v) access to an improved study. Circumstance variables include rural/urban location, water source; and (vi) access to a health post within 5km gender, religion, agro-ecological zone, consumption quintile (Table 35). The importance of these outcome variables is ob- of the child’s household, region, and the household’s score vious. Education is a key opportunity with large effects on on the rural accessibility index (RAI).77 earnings and welfare later in life. Access to safe water is key Table 35 KEY  OUTCOME VARIABLES FOR THE ANALYSIS Variable DEFINITION Age 7-14 Enrolment in to primary school A child aged 7-14 is registered to attend primary school. Age 15-18 Completed primary school A child aged 15-18 has completed primary school. Enrolment in secondary education A child aged 15-18 is registered to attend secondary school. Age 7– 18 Access to electricity A child lives in a household with an electrical connection. Access to improved water A child has access to an improved water source (piped water, a protected water source). Distance to Health Post A child lives in a household that is under 5km from a health post Sources: Staff calculations based on LFS 2013. 77  Accessibility was also tested using a market accessibility variable, and the results were broadly consistent. 194 ETHIOPIA POVERTY ASSESSMENT 2.2 Descriptive statistics and for secondary school (15 to 18) were enrolled in secondary school in 2016. This indicates that the education opportunity results set is quite different for the older cohort of children (15-18) IMPROVEMENTS IN ACCESS TO KEY SERVICES than the younger cohort (7-14). The proportion of children AND OPPORTUNITIES with access to electricity improved marginally by about three percentage points but remained low at 20 percent in 2016. There were across the board improvements in cover- 30 percent of children 15-18 had completed primary school age of the key outcome variables between 2011 and in 2016, which nevertheless represented a large increase 2016. Figure 115 presents these measures across six out- from 20 percent in 2011. come variables over this five-year period. The first panel of the figure shows coverage rates. The proportion of age-eli- The HOI improved for all outcome variables, while gible children (7 to 14) enrolled in primary school increased the D-index decreased, indicating that circumstances from 61 percent in 2011 to 71 percent in 2016. There were played less of a role in determining outcomes in 2016 also large improvements in access to an improved water than they did in 2011. Proportional increases in the HOI source, and in the share of children that lived within 5km of were particularly strong for primary school completion, sec- a health post. ondary school enrollment and proximity to a health post. The HOI was lowest, and showed the smallest improvement, for Despite these improvements, access to electricity, proximity to a hospital, and it was also very low (under 10) for health care, primary school completion and secondary access to electricity. enrollment remain low and unequal. The improvement in primary school enrollment rates did not translate fully into The largest improvements in the D-index were seen for either primary school completion or secondary school enroll- the primary completion and secondary enrollment out- ment. Fewer than one in five children who were age-eligible comes. Although the coverage rates of these variables were Figure 115 EQUAL  ACCESS TO OPPORTUNITIES INCREASED BETWEEN 2011 AND 2016 Coverage, HOI, and D-Index for different outcomes, 2011 and 2016 80 Coverage HOI D-index 70 60 Percent/Score 50 40 30 20 10 0 ol ol ol of d w s th r st co ry s ol l l ss al ter st co ry s ol l ol of d w s th r st El dar hoo o El dar hoo e e s s ho ho ho ho ho ci cho ho po po po at at 5k ro cce 5k ro cce 5k pro cce a of d w c pr y sc c tri sc sc sc sc pr y sc th s a a a al al y y y y y m ve m ve m ve Im ty y ty ar ar ar r r he he he t En lete ima a En lete ima a ci ci im im im nd im tri tri n n in Imp in Imp pr pr pr pr co ec ec ec se se se d ed om led d d d El lle e et l d d d l l ro ro ro in lle lle lle pl p p En En En ith ith ith om om ro ro ro W W W En C C C 2011 2016 Source: Calculations from WMS 2015/16 and HCES 2015/16. CHAPTER VII. INEQUALITY OF OPPORTUNITY IN ETHIOPIA 195 low in both 2011 and 2016 (see panel 1 of the Figure 115), household’s consumption quartile. Figure 116 shows the role of circumstances in determining access decreased that the rural-urban electricity access gap is also extreme- by around 16 points for both over the time period. The D-in- ly large, with an access rate of 90 percent in urban areas dex was lowest for enrollment in primary education, driven but just under 10 percent in rural areas. Ethiopian Orthodox by the fact that enrollment rates for age-eligible children was households are just above the mean access rate of about 23 already relatively high. percent, while non-orthodox households are slightly below. Households in the top quartile are about five times as like- However, unequal coverage of infrastructure access ly to have access to electricity than the poorest quarter of variables remains a challenge. The D-index was highest households. for proximity to a hospital and for access to electricity, exem- plifying the rural-urban divide in access to these services that Differences in primary school enrollment rates are not will be explored in the coming decomposition. An example so stark between circumstances, as the mean access serves to highlight some of these differences. A rural, female rate is fairly high. As shown in Figure 117, which is restrict- child living in the drought prone highlands is 53 percent less ed to children aged between 7 and 14, the rural-urban gap likely to have access to electricity and 64 percent less likely is about 16 percentage points – far smaller than it was for to have access to a hospital compared to a male child living electricity access. As with all the outcome variables consid- in an urban area. ered, the difference between female and male children was extremely small, though the female mean was slightly higher. LOCATION AND HOUSEHOLD WEALTH ARE THE The biggest differences in access are at the regional level. Al- MAIN DETERMINANTS OF ACCESS most 90 percent of age-eligible children living in Addis Ababa Location and household wealth are the most signifi- are enrolled in primary school, while the figure for Somali is cant contributors to inequity in access to basic oppor- just over half. In fact, on this measure Somali lags far behind tunities. Though access to opportunities increased across the next lowest region, which is Afar. The gap between the the board between 2011 and 2016, the gains were unevenly richest quintile of households and the poorest is approxi- spread. As a result, gaps between rural and urban areas and mately as large as the gap between urban and rural loca- between richer and poorer households remained large. For tions. The patterns for enrollment in secondary education (for example, roughly half of urban children between the ages those aged between 15 and 18) are essentially replicated, 15-18 were enrolled in secondary school, but only 10 of chil- but at a much lower level. The mean age-eligible secondary dren in rural households are. Similarly, about half of children enrollment rate for age-eligible children was under 20 per- between 15 and 18 in the top consumption quintile had com- cent. It was as high as 40 percent in urban areas, and as low pleted primary school in 2016, compared to less than 19 as 11 percent in the poorest quintile of households. Figures percent in the bottom quintile (Annex Figure 12) for this and for the access rates for the other outcome vari- ables are shown in Annex Figures 12 to 15. Access gaps by gender are relatively small. This is con- sistent both when measuring at an individual level for chil- About two thirds of households live within 5km of one dren, and when using the gender of the household head for of the country’s approximately 17 000 health posts, with variables such as electricity access. However, there are some the biggest differences occurring between regions. Al- access differentials when comparing religious groups. Chil- most all urban households are located within the 5km thresh- dren belonging to households identifying as Ethiopia Ortho- old, as shown in Figure 118. The access differential between dox tend to have higher access rates than children who live in male- and female-headed households is extremely small, other Christian (Catholic or Protestant) or Muslim households. while the access differences across religious groups are also relatively small. Regional differences are stark, however, and Unsurprisingly, the biggest electricity access differ- range from almost full access in Addis Ababa and almost 90 entials are based on rural-urban location and on the percent in Dire Dawa to around 50 percent in Somali and Afar. 196 ETHIOPIA POVERTY ASSESSMENT Figure 116 ACCESS  TO ELECTRICITY IS LARGELY DETERMINED BY LOCATION AND HOUSEHOLD WEALTH Coverage of access to electricity by circumstance variables, 2016 100% Urban 90% 80% 70% 60% Richest 50% Highest RAI 40% Orthodox 30% Female Mean 20% Male Non−orthodox 10% Poorest Lowest RAI 0% Rural n r n e tle de tio io til in in ig en ca qu qu el G Lo R y n ilit io ib pt ss m su ce on ac C al ur Source: Calculations from WMS 2015/16 and HCES 2015/16. R Figure 117 PRIMARY  SCHOOL ENROLMENT CRUCIALLY DEPENDS ON THE REGION WHERE ONE IS BORN Coverage of access to primary education by circumstance variables, 2016 100% 90% Urban Addis Ababa Gambella Richest Orthodox Amhara Harari Highest RAI 80% Female B−G Tigray Other Christian Dire Dawa Mean 70% SNNPR Afar Oromiya Rural Male Poorest 60% Muslim Lowest RAI 50% Somali 40% 30% 20% 10% 0% n r n e tle de tio io til in in ig en ca qu qu el G Lo R y n ilit io ib pt ss m su ce on ac C al ur R Source: Calculations from WMS 2015/16 and HCES 2015/16. CHAPTER VII. INEQUALITY OF OPPORTUNITY IN ETHIOPIA 197 Figure 118 LOCATION  DETERMINES PROXIMITY TO A HEALTH POST Coverage of having a health post within 5km by circumstance variables, 2016 Addis Ababa 100% Urban 90% Dire Dawa Harari B−G Highest RAI 80% Richest Other Christian SNNPR 70% Female Orthodox Gambella Amhara Mean 60% Male Tigray Oromiya Rural Muslim Poorest 2nd lowest RAI Somali 50% Afar 40% 30% 20% 10% 0% n r n n e tle de tio io io til in in ig eg en ca qu qu el R G Lo R y n ilit io ib pt ss m su ce on ac C al ur R Source: Calculations from WMS 2015/16 and HCES 2015/16. Box 25 Gender differences and the HOI  In the HOI analysis, gender matters only for education outcome variables. Estimating the HOI and D-index sepa- rately for male and female children, and for male- and female-headed households yields some interesting results. Primary school enrollment rates are relatively high, and although the female HOI for this outcome variable is slightly higher than the male HOI, the difference between the two is not statistically significant. Females between the ages of 15 and 18 are more likely to have completed primary school than males in the same age cohort, and they have a higher HOI as well. There are no significant differences in the HOI or D-index between male and female children based on secondary school enrollment, access to basic infrastructure or healthcare facilities. The differences between children living in male- versus female-headed households are more prominent. Once again there are no differences for the primary enrollment outcome variable, but children living in female-headed house- holds are about 10 percentage points more likely to have completed primary school than those in male-headed households. This result is robust to controls for urban residence and household consumption. The primary comple- tion HOI is also higher for children in female-headed households. However, the D-index for this outcome variable is almost identical for male- and female-headed households for this variable, indicating that the role of differential ac- cess by circumstances actually matters less. A similar qualitative result holds for the secondary enrolment outcome variable, though the difference in the HOI is not quite as large. Finally, there are no significant differences in the HOI or D-index for the infrastructure or health access variables. 198 ETHIOPIA POVERTY ASSESSMENT Decomposing inequality of opportunity into its circum- enrolment is mostly explained by differences between stance components shows that location is the biggest rural and urban areas and difference in household driver of unequal access. The application of the Shapley wealth. The same is true for the infrastructure access vari- Value approach allows for the estimation of the contribution ables – two thirds of unequal access to electricity is explained of each circumstance to the D-index. 78 As shown in Figure by a household’s rural location, as is about half of unequal 119 the only outcome for which the rural-urban variable is not access to an improved water source (Figure 119). the main explanatory factor is for age-eligible primary school Household poverty is usually the second most import- enrollment. For this variable the religion of the household ex- ant factor in explaining access to services. One fifth of plains just over 40 percent of the inequality of opportunity. Un- differences in primary school enrollment is explained by the derlying this finding is the fact that children who live in Muslim household’s consumption quartile. Children in poorer house- households are less likely to be enrolled in primary school. holds are also less likely to complete primary school, even This is strongly correlated with populations living in pastoral- after controlling for location and accessibility factors. The ist areas and in drought-prone lowlands, where household household’s consumption quartile also explains 16 percent movements in relation to rainfall and the availability of grazing and 18 percent of inequality in secondary school enrollment lands and water are more common. and access to an improved water source, respectively. Dif- Inequality of opportunity as measured by differences ferences by gender play a relatively minor role in explaining in primary school completion and secondary school inequalities across all seven outcome variables (Box 25). Figure 119 LOCATION  AND HOUSEHOLD WEALTH ARE THE MAIN SOURCES OF INEQUITY IN ACCESS TO OPPORTUNITIES Decomposition of the D-Index by circumstance variable, 2016 70 60 circumstance to IOp Contribution of each 50 40 30 20 10 0 Enrolled primary Completed Enrolled Electricity access Improved water Within 5km of school primary school secondary school health post Rural Female Religion Eco. zone Region Consumption quartile RAI Source: Calculations from WMS 2015/16 and HCES 2015/16. 78  For further details on theory and application see Hoyos and Narayan (2011) and Abras, Cuesta et al. (2012). CHAPTER VII. INEQUALITY OF OPPORTUNITY IN ETHIOPIA 199 3. INTERGENERATIONAL EDUCATIONAL MOBILITY The chapter now turns to another question that is cen- intergenerational mobility is lower in urban areas. We do this tral to understanding inequality of opportunity, that of analysis separately for 2011 and 2016, to measure how these intergenerational educational mobility. If children’s edu- relationships have changed over time. Depending on the cational attainment is largely influenced by the educational specification, the child’s education outcome variable either attainment of their parents, then there will be a high transmis- reflects the years of educational attainment or is binary and sion of poverty or prosperity across generations. This section reflects primary or secondary enrollment. The age range for begins by outlining the methodology used in the estimation, children used in the estimation is generally 7 to 18, but is before presenting results on the relationship between paren- adjusted to 7 to 14 when looking at primary enrollment, and tal education and the probability that an age-eligible child is 15 to 18 when looking at primary completion or secondary enrolled in primary school. The strength of the link between enrollment. household consumption and age-eligible enrollment is then While the gap between rural-urban primary enrollment explored. Finally, the analysis investigates the correlation be- rates decreased between 2011 and 2016, the gap in tween years of parental education and years of education secondary enrolment increased. Table 36 presents sum- attained by the child. Box 26 summarizes the extent of inter- mary statistics for some of the key variables that are used in generational mobility in a sample of African countries.79 the analysis in this section. The data used are for fathers and There are three central questions in this section: (i) To children (aged 7 to 18) who are co-resident in the 2011 and what extent does a child’s educational outcomes depend on 2016 samples. The average number of years of education the education of his or her parents? (ii) How does this differ for a rural father living with a primary-school-aged child in a by different groups in society? (iii) How has this changed in rural area was 1.8 in 2011 and 2.2 in 2016. This is in contrast Ethiopia between 2011 and 2016? For the main results, the to the urban comparison which is 6.6 years and 7.3 years, relationship is estimated using an OLS regression with the respectively. There was considerable improvement in primary following specification: school enrollment rates for children aged 7 to 14 between 2011 and 2016.80 In rural areas the enrollment rate increased Childedu = ß0 + ß1Fatheredu + ß2Childage + ß3Urban + by more than 10 percentage points, while in urban areas it ß4Fatheredu * Urban + ϵ increased by three percentage points to 90 percent. Second- where Urban is a dummy variable indicating whether child ary enrollment rates were low overall, and the gap between was born in an urban area. A low estimated value of the co- rural and urban areas widened. In 2016 only about 6 percent efficient ß1 means that relative intergenerational mobility is of children aged 15 to 18 were enrolled in secondary school, high in rural areas, because education attainment does not compared to just over one third of age-eligible children in depend on one’s parents. Similarly, if the coefficient, ß4 , is urban areas. The average number of years of education for positive and statistically significant, this means that relative a 15-year-old child increased in both rural and urban areas over the period. 79 Father’s education rather than mother’s education is used in this analysis. The mother’s education variable for school-aged children piles up at zero years of formal education – in fact it is zero until the 75th percentile. The father’s education variable contains zeros up to the 25th percentile, and is 5 years at the 75th percentile. This tends to make results using mother’s education a lot more imprecise than when using father’s education. Regressions using a different specification in which “highest education either mother or father” yield very similar results compared to a regression using father’s education only. 80 The full cumulative distribution functions of educational attainment for children aged 7 to 18 in 2011 and in 2016 is provided in Annex Figure 16. 200 ETHIOPIA POVERTY ASSESSMENT Table 36 THE  RURAL-URBAN GAP IN EDUCATION SLIGHTLY DECREASED BUT REMAINS HIGH Education indicators of father and children, 2011-2016 RURAL URBAN DIFFERENCE 2011 Father years edu. (with child aged 7-14) 1.8 6.6 *** Father years edu. (with child aged 15-18) 1.3 6.1 *** Age-eligible enrolled in primary 60% 87% *** Age-eligible enrolled in secondary 2% 27% *** Years of education 15 year old 3.1 6.2 *** 2016 Father years edu. (with child aged 7-14) 2.2 7.3 *** Father years edu. (with child aged 15-18) 2.1 7.2 *** Age-eligible enrolled in primary 71% 90% *** Age-eligible enrolled in secondary 6% 35% *** Years of education 15 year old 3.9 6.5 *** Source: WMS, 2011; 2016. World Bank staff calculations. CHAPTER VII. INEQUALITY OF OPPORTUNITY IN ETHIOPIA 201 Box 26 Intergenerational mobility in African countries  A study of 23 African countries by Alesina, Hohmann et al. (2019) shows that since independence intergen- erational mobility has increased, largely reflecting the rise in educational attainment. Countries in which the parental generation has lower than average years of education tend to have lower mobility. Conversely, a more high- ly educated parental generation is associated with a higher level of intergenerational mobility. South Africa, Nigeria, and to a lesser extent Botswana, Kenya, Ghana, and Tanzania, have higher mean education of the parental genera- tion, and also higher levels of mobility. On the other hand, Ethiopia, Senegal, Sudan and Mali have both relatively low average education levels, and lower levels of intergenerational mobility. Research by Azomahou and Yitbarek (2016) shows that Ethiopia, Burkina Faso, Sudan, South Sudan and Morocco have very high parental generation illiteracy rates at around 90%. Children of parents without formal schooling in these countries have a very low probability of completing primary school – around 20 percent. This is in contrast to countries with a better educated parental generation, in which primary school completion rates are generally at or above 70 percent. Figure 120 UPWARD  INTERGENERATIONAL SOCIAL MOBILITY IN DIFFERENT AFRICAN COUNTRIES Source: Reproduction of Figure 3 panel (a) in Alesina, Hohmann et al. (2019). The figure above, from Alesina, Hohmann et al. (2019), shows estimates of intergenerational mobility for individuals aged 14 to 18. The figure shows upward mobility – the brighter the color the higher the upward mobility. Upward mobility in this case is defined as “the likelihood that children whose parents have not completed primary school will themselves go on to complete at least primary education”. The results show that Ethiopia has among the lowest rates of upward mobility for countries for which data are available, reflecting the relatively poor performance in education. 202 ETHIOPIA POVERTY ASSESSMENT The relationship between children’s education and The educational effect of living in an urban area in- their parents’ education weakened slightly between creased between 2011 and 2016. The urban coefficient 2011 and 2016, indicative of a small increase in inter- in Table 37 rose from 1.3 years to 1.44 years, on average: generational mobility. The regression output in Table 37 Relative to a rural child of the same age, an urban child had shows the results of estimating years of education of the completed on average 1.44 more years of education. The child on the years of education of his or her father, with con- interaction term between urban location and father’s years trols included for the child’s age, urban location, and an inter- of education was not statistically significant in 2016. Taken action term between father’s education and urban location. together this means that the educational effect of happening On average, a one year increase in parental education was to live in an urban area increased over the period, but the associated with a 0.10 year increase in the child’s education intergenerational relationship in urban versus rural areas did in 2016, while controlling for age and location. not change. There were no statistically significant differenc- es between male and female children (regression output not shown) in either 2011 of 2016. Table 37 INTERGENERATIONAL  MOBILITY HAS IMPROVED BETWEEN 2011 AND 2016 Intergenerational educational attainment: 7-18 Y=CHILD'S YEARS EDU. 2011 2016 Father's education 0.12*** 0.10*** (0.01) (0.01) Urban 1.30*** 1.44*** (0.06) (0.06) Urban * Father's education -0.02* -0.001 (0.01) (0.01) Child's age 0.51*** 0.42*** (0.01) (0.01) Constant -3.95*** -3.27*** (0.06) (0.06) Observations 23 074 22 711 R2 0.49 0.42 Note: *** p<0.01, ** p<0.05, * p<0.1. Source: Calculations from WMS 2010/11 and 2015/16, and HCES 2010/11 and 2015/16. Another way of understanding the intergeneration- and children with better-educated fathers in rural ar- al relationship between education is to investigate eas. Figure 121 shows that the relationship between father’s the probability that an age-eligible child is enrolled in education and the probability of being enrolled in primary school conditional on his or her parents’ level of edu- school became flatter over time in urban areas, reflecting a cation. This is achieved by running a logistic regression of more equal access to primary school in urban areas. The whether the age-eligible child is enrolled or not on parental relative improvement of rural children with well-educated par- education while controlling for other variables such as loca- ents can be seen by the closing of the gap at the top of the tion. Figure 121 shows this relationship in 2011 and in 2016 distribution between 2011 and 2016. However, the differenc- for both rural and urban areas. es at the lower end of the distribution between rural and ur- ban children remain stark: 70 percent of urban children with The biggest improvements were concentrated among an uneducated father are enrolled in primary school, com- children with poorly-educated fathers in urban areas, pared to 50 percent of rural children. CHAPTER VII. INEQUALITY OF OPPORTUNITY IN ETHIOPIA 203 Figure 121 ENROLMENT  IN PRIMARY SCHOOL BECAME LESS DEPENDENT ON FATHER’S EDUCATION IN URBAN AREAS Probability of a child being enrolled in primary school, by father’s education, 2011 and 2016 2011 2016 .9 .9 Probability of primary enrollment Probability of primary enrollment .8 .8 .7 .7 .6 .6 .5 .5 .4 .4 .3 .3 .2 .2 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Father’s years of education Father’s years of education Rural Urban Rural Urban Note: Analysis restricted to children aged 7 to 14. Figures show the predicted probability of enrollment for age-eligible children following a logistic regression. Sample sizes for rural were 9,248 and 9,612 in 2011 and 2016 respectively. Sample sizes for urban were 7,515 and 7,611 in 2011 and 2016 respectively. Source: Calculations from WMS 2010/11 and 2015/16, and HCES 2010/11 and 2015/16. The dynamic is amplified when considering secondary The improvement in urban areas was driven by small school enrollment. The two panels in Figure 122 show the gains in the probability of secondary enrollment for chil- predicted probabilities of secondary enrollment for individu- dren of poorly-educated fathers. The overall age-eligible als aged 15 to 18 over the range of their father’s education. secondary enrollment rate in urban areas increased by eight The relationship for children in rural areas was completely flat percentage points (from 27 percent to 35 percent), and much in 2011, which is not surprising considering the overall sec- of that was concentrated in gains accruing to children of fa- ondary enrollment rate of about two percent. The enrollment thers with between zero and five years of education. In other rate in 2016 was still only six percent, but it is clear that most words, there was an intercept change for the urban line, but of that change came from rural households with better-edu- it did not flatten out to the same extent as did the primary cated fathers. enrollment urban line presented in the previous figure.81 All else equal, children of poorly-education parents in urban areas were more likely to go to secondary school than the children of well-educated parents in rural areas. This resulted in an in- crease in this measure of secondary education intergenera- tional mobility in urban areas, but a decrease in rural areas. 81 Similar analysis was undertaken to investigate differences between religious groups and the agro-ecological zones in which households were situated. For the former, the main differences occurred in the probability of secondary school enrollment, in which children in Ethiopian Orthodox households were more likely to be enrolled for any level of parental education than were children in non-Orthodox Christian and Muslim households. Patterns between the agro-ecological zones show very few differences at the top of the distribution. The children of poorly-educated parents in the drought-prone lowlands had lower primary enrollment probabili- ties than the other four agro-ecological zones in both 2011 and 2016. 204 ETHIOPIA POVERTY ASSESSMENT Figure 122 ENROLMENT  IN SECONDARY SCHOOL BECAME MORE DEPENDENT ON FATHER’S EDUCATION IN RURAL AREAS Probability of a child being enrolled in secondary school, by father’s education, 2011 and 2016 2011 2016 .8 .8 Probability of secondary enrollment Probability of secondary enrollment .6 .6 .4 .4 .2 .2 0 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 Father’s years of education Father’s years of education Rural Urban Rural Urban Note: Analysis restricted to children aged 15 to 18. Figures show the predicted probability of enrollment for age-eligible children following a logistic regression. Sample sizes for rural were 3,022 and 2,913 in 2011 and 2016 respectively. Sample sizes for urban were 2,967 and 3,071 in 2011 and 2016 respectively. Source: Calculations from WMS 2010/11 and 2015/16, and HCES 2010/11 and 2015/16. CHAPTER VII. INEQUALITY OF OPPORTUNITY IN ETHIOPIA 205 THE ROLE OF HOUSEHOLD WELFARE IN ACCESSING EDUCATION IS INCREASING, ESPECIALLY IN RURAL AREAS While access to education has increased between 2011 The increased disparity in access to education and and 2016, the increase was larger for children from quantity of education attained between poor and bet- wealthier households. As a result, household welfare has ter-off children is driven by rural areas. While there is a become more important in determining access to education. 82 positive association between household consumption levels The probability of the poorest children being enrolled in primary and schooling of the household’s children in both rural and school did not significantly increase between 2011 and 2016, urban areas, the association is significantly stronger in rural while this increase was significant at the top of the distribution areas (a positive and statistically significant interaction effect in (left-hand side graph of Figure 123). Changes in the predicted Annex Table 6). This result suggests that the poorest house- probability of secondary school enrollment were concentrated holds in rural areas face substantial financial constraints in in the top quartile of households, while the probability of being getting and, especially, keeping their children in school. Specif- enrolled in secondary school for children in the bottom half ically, children from poor rural households tend to start school of the distribution was not different in 2011 and 2016 (right- at a later age, affecting their ability to progress successfully hand side graph of Figure 123). This means that the gains in through school. This effect is seen predominantly at the pri- secondary enrollment between 2011 and 2016 went mainly mary school level. This pattern holds when also controlling for to children in relatively richer households. For children in poor distance to actual school facilities. Demand-side interventions households, and households in the middle of the consumption that provide financial support or incentives to extremely poor distribution, the probability of attending secondary school did rural households may be necessary to build the human capital not change significantly over the period. of their children and break the intergenerational poverty trap. Figure 123 ENROLMENT  IN SCHOOL BECAME MORE DEPENDENT ON HOUSEHOLD WELFARE BETWEEN 2011 AND 2016 School enrollment probabilities by household consumption, 2011 and 2016 Primary school enrollment Secondary school enrollment Probability of secondary enrollment .8 .8 Probability of primary enrollment .6 .6 .4 .4 .2 .2 0 0 6 7 8 9 10 11 6 7 8 9 10 11 Log of real consumption per adult equivalent Log of real consumption per adult equivalent 2011 2016 2011 2016 Source: Calculations from WMS 2010/11 and 2015/16, and HCES 2010/11 and 2015/16. 82 The effect of household welfare on accessing primary school did not significantly change between 2011 and 2016. However, the influence of household wealth on the likelihood of enrolment in secondary school significantly increased. 206 ETHIOPIA POVERTY ASSESSMENT Conclusion This chapter presented evidence on how inequality of Intergenerational mobility in education also improved opportunity and intergenerational mobility in Ethiopia between 2011 and 2016. Overall, the effect of parental ed- have changed over the last five years. Understanding ucation on children’s education has weakened, though is still these two concepts better is a crucial part of thinking about strong and significant. The overall dynamic masks different how poverty and inequality may develop and be tackled in fu- dynamics at different level of education: While enrolment in ture years. This is important, as although Ethiopia has made primary school has become less dependent on parents’ ed- significant gains in poverty reduction, and inequality of out- ucation levels, enrolment in secondary school has become comes remains relatively low, the economic growth of the last more dependent on parent’s education. decade has not been equally shared within the country. En- Education gains between 2011 and 2016 were concen- suring that there is equal opportunity of access to services is trated among the children of better-educated parents in linked to more sustainable and inclusive economic growth as rural areas and less-educated parents in urban areas. investing in the productivity and welfare of all citizens through The implication is that the children of poorly-educated rural access to opportunities equates to improved economic out- parents are in danger of being left behind. This is confirmed comes in the future. by the finding that the impact of household welfare levels on There has been an overall improvement in equal access child schooling is substantially higher in rural than urban ar- to basic services but there are still large disparities be- eas. The poorest rural households may need supplementary tween circumstance groups. Rural vs urban location and support to get and keep their children in school for longer. household welfare are the main sources of inequity in ac- Overall, the results tell a consistent story that crucial cess to key opportunities. Urban children and children from investments need to be made in providing access to households in the top consumption quintile have far better services in rural areas, so that children born in rural access to key services compared to rural or poorer children. households are afforded the same opportunities as those born in urban areas. CHAPTER VII. INEQUALITY OF OPPORTUNITY IN ETHIOPIA 207 References Aguero, J., Carter, M. & May, J. (2007). “Poverty and Blinder, A. (1973). “Wage discrimination: Reduced form and inequality in the first decade of South Africa’s democracy: structural estimates.” The Journal of Human Resources What can be learnt from panel data from KwaZulu-Natal?”. 8(4): 436-455. Journal of African Economies, 16(5): 782-812. Central Statistical Agency of Ethiopia (2017). LSMS - Alderman, H. & Garcia, M. (1993). Poverty, household food Integrated surveys on agriculture: Ethiopia Socioeconomic security and nutrition in rural Pakistan. Research paper 96. Survey (ESS) report. Addis Ababa, Central Statistical Washington DC: IFPRI. Agency of Ethiopia. Bachewe, F., & Headey, D. (2017). Urban wage behavior Chauvin, J. P., Glaeser, E., Ma, Y., & Tobio, K. (2017). What and food price inflation in Ethiopia. The Journal of is different about urbanization in rich and poor countries? Development Studies, 53 (8), 1207-1222. cities in Brazil, China, India and the United States. Journal of Urban Economics, 98, 17–49. Bachewe F, F. Yimer, B. Minten, and P. Dorosh. (2016). Agricultural prices during drought in Ethiopia: An Christaensen, L., De Weerdt, J., & Kanbur, R. (2016). assessment using national producer data (January 2014 Urbanization and poverty reduction: The role of secondary to January 2016) ESSP Working Paper 88 Washingto DC: towns in Tanzania. Analysis and Policy Brief No 18. Institute International Food Policy Research Institute. of Development Policy and Management. Baulch, B. & Hoddinott, J. (2000). “Economic mobility Christiaensen, L., & Kanbur, R. (2017). Secondary towns and poverty dynamics in developing countries.” Journal of and poverty reduction: Refocusing the urbanization agenda. Development Studies 36(6): 1-24. Annual Review of Resource Economics, 9, 405-419. Beegle, K., de Weerdt, J., & Dercon, S. (2011). Migration Christiaensen, L., & Todo, Y. (2014). Poverty reduction and economic mobility in Tanzania: Evidence from a during the rural-urban transformation: The role of missing tracking survey. Review of Economics and Statistics, 93 (3), middle. World Development, 63, 43-58. 1010-1033. Dang, H.-A. H., P. F. Lanjouw and R. Swinkels (2017). Who Berhane, G., Hirvonen, K. and J. Hoddinott. (2015a). The remained in poverty, who moved up, and who fell down? Implementation of the Productive Safety Nets Programme, Poverty Reduction in the Course of African Development. 2014: Highlands Outcomes Report. M. Nissanke and M. Ndulo. Oxford, Oxford University Press. Kumar, N. and J. Hoddinott. (2015b). The Implementation de Brauw, A., Mueller, V., & Woldehanna, T. (2017). Does of the Productive Safety Nets Programme, 2014: Lowlands internal migration improve overall well-being in Ethiopia? Outcomes Report. Journal of African Economics, 27 (3), 1-19. Berhane, G., Gilligan, D., Hirvonen, K., Hoddinott, Dercon, S., & Krishnan, P. (2000). Vulnerability, seasonality J., Kumar, N., and A. Seyoum-Tafesse. (2018a). The and poverty in Ethiopia. Journal of Development Studies, Productive Safety Net Programme 4 Midline Survey 2018. 36(6), 25-53. Highland Outcomes Report. Dercon, S. (2004). Growth and shocks: Evidence from Berhane, G., Gilligan, D., Hirvonen, K., Hoddinott, rural Ethiopia. Journal of Development Economics, 74(2), J., Kumar, N., and A. Seyoum-Tafesse. (2018a). The 309-329. Productive Safety Net Programme 4 Midline Survey 2018. Dercon, S. (2006). Economic reform, growth and the poor: Lowland Outcomes Report. Evidence from rural Ethiopia. Journal of Development Economics. 81(1). 1-24. 208 ETHIOPIA POVERTY ASSESSMENT Dercon, S., Gilligan, D., Hoddinott, J., & Woldehanna, T. Glewwe, P. & Gibson, J. (2005). “Analysis of poverty (2009). The impact of roads and agricultural extension dynamics”. Handbook on Poverty Statistics: Concepts, on consumption growth and poverty in fifteen Ethiopian Methods and Policy Use. Geneva: United Nations. villages. American Journal of Agricultural Economics, 91(4), Glewwe, P. (2012). “How much of observed economic 1007-1021. mobility is measurement error? IV methods to reduce Dercon, S., Hoddinott, J., & Woldehanna, T. (2012). Growth measurement error bias, with an application to Vietnam”. and chronic poverty: Evidence from rural communities in The World Bank Economic Review, 26(2): 236-264. Ethiopia. Journal of Development Studies, 48(2), 238-253. Gollin, D., Jedwab, R., & Vollrath, D. (2016). Urbanization Devereux, S., R. Sabates-Wheeler and R. Slater (2008), with and without industrialization. Journal of Economic Ethiopia’s Productive Safety Net Programme (PSNP): 2008 Growth, 21(1), 35–70. Assessment Report Study Team. Henderson, J. V., & Kriticos, S. (2018). The development of Dorosh, P., Thurlow, J., Kebede, F., Ferede, T. & Taffesse, A. the African system of cities. Annual Review of Economics, (2018). Structural change and poverty reduction in Ethiopia: 10, 287-314. Economy-wide analysis of the evolving role of agriculture. Henderson, J. V., Storeygard, A., & Deichmann, U. (2017). Strategy Support Program Working Paper 123. IFPRI. Has climate change driven urbanization in Africa? Journal of Duranton, G. (2014). Growing through cities in developing Development Economics, 124, 60–82. countries. The World Bank Research Observer, 30, 39-73. Hill, R. and E. Tsehaye (2018). Growth, safety nets and Duranton, G. (2016). Agglomeration effects in Colombia. poverty: Assessing progress in Ethiopia from 1996 to 2011. Journal of Regional Science, 56(2), 210–238. Policy Working Research Paper 8380. Washington DC, World Bank. Fafchamps, M. (2012). Reprint of development, agglomeration, and the organization of work. Regional Hirvonen, K., Mascagni, G. & Roelen, K. (2016). Linking Science and Urban Economics, 42(5), 765–778. taxation and social protection: Evidence on redistribution and poverty reduction in Ethiopia. Working Paper 2016/111. FDRE (2016). Urban Food Security Strategy. Federal UNU-WIDER. Democratic Republic of Ethiopia: Ministry of Urban Development and Housing. Hirvonen K., Sohnesen, T. and Bundervoet, T. (2018). “Impact of Ethiopia’s 2015 drought on child undernutrition”. FDRE (2017). Ethiopia’s Progress Towards Eradicating International Food Policy Research Institute (IFPRI) ESSP Poverty: An Interim Report on 2015/16 Poverty Analysis Working Paper 114. Study. Ethiopia: National Planning Commission. Hirvonen, K., Hoddinott, J., Lind, J., Sabates-Wheeler, R., Firpo, S., N. Fortin and T. Lemieux (2009). “Unconditional Tefera, M., and Seyoum-Tafesse, A. (2019). Humanitarian quantile regressions.” Econometrica 77(3): 953-973. Food Assistance 2018 – Process Evaluation Report. PPT Franklin, S. (2019). Slums to failed estates? Evidence from presentation shared with WB team. a lottery for low-cost housing in Ethiopia. Unpublished Hirvonen, K., T. P. Sohnesen and T. Bundervoet (2018). manuscript. Impact of Ethiopia’s 2015 drought on child undernutrition. Fuje, H. (2018). Welfare dynamics and drought in Ethiopia. Ethiopia Strategy Support Program. Washington DC, CSAE Conference 2018: Economic Development in Africa. International Food Policy Research Institute. Oxford. Hoddinott, J. and T. Mekasha. (2017). Social protection, Gindling, T. H., & Newhouse, D. (2014). Self-employment in household size and its determinants: Evidence from the developing world. World Development, 56, 313–331. Ethiopia. ESSP Working Paper 107. HARNESSING CONTINUED GROWTH FOR ACCELERATED POVERTY REDUCTION 209 Hulme, D. and A. Shepherd (2003). “Conceptualizing Minten, B., Dereje, M., Bachewe, F. and T. Seneshaw. chronic poverty.” World Development 31(3): 403-423. (2018). “Evolving Food Systems in Ethiopia: Past, Present and Future.” ESSP Working Paper. Washington DC: IFPRI. Jalan, J. and M. Ravallion (1998). “Transient poverty in post- reform rural China.” Journal of Comparative Economics Moreno-Monroy, A. (2012). Critical commentary. informality 26(2): 338-357. in space: Understanding agglomeration economies during economic development. Urban Studies, 49(10), 2019–2030. Jedwab, R., Christiaensen, L., & Gindelsky, M. (2015). Demography, urbanization and development: Rural push, Mukim, M. (2016). How do cities in Ethiopia create jobs? urban pull and... urban push? Journal of Urban Economics, World Bank Policy Research Working Paper 7866. 98, 6-16. Nakamura, S., Bundervoet, T., & Nuru, M. (2019). Rural Kafle, K., K. McGee, A. Ambel and I. Seff (2016). “Once roads, poverty, and resilience: Evidence from Ethiopia. poor always poor? Exploring consumption- and asset- World Bank Policy Research Working Paper No. 8800. based poverty dynamics in Ethiopia.” Ethiopian Journal of Oaxaca, R. (1973). “Male-female wage differentials in urban Economics 25(2): 37-76. labor markets.” International Economic Review 14(3): Kamei, A., & Nakamura, S. (2019). Urban agglomerations 693-709. and employment transitions in Ethiopia. Unpublished Puga, D. (2010). The magnitude and causes of manuscript. agglomeration economies. Journal of Regional Science, Kefyalew Endale, Pick, A. and Tassew Woldehanna, 2019, 50(1), 203–219. Financing Social Protection in Ethiopia: A Long-term Quintero, L. E., & Roberts, M. (2018). Explaining spatial Perspective, OECD Development Policy Papers No. 15, variations in productivity: evidence from Latin America and OECD. the Caribbean. World Bank Policy Research Working Paper Kosec., K., Ghebru, H., Holtemeyer, B., Mueller, V. and E. 8560. Schmidth. (2018). “The Effect of Land Access on Youth Ravallion, M. and M. Huppi (1991). “Measuring changes in Employment and Migration Decisions: Evidence from Rural poverty: A methodological case study of Indonesia during Ethiopia.” The American Journal of Agricultural Economics an adjustment period.” World Bank Economic Review 5(1): 100(3): 931-954. 57-82. Lipton, M. and M. Ravallion (1995). Poverty and policy. Ravallion, M. and S. Chen (2003). “Measuring pro-poor Handbook of Development Economics. J. Behrman and T. growth.” Economics Letters 78(1): 93-99. N. Srinivasan. Amsterdam, North Holland. III. Ravallion, M., 2000. Monitoring targeting performance when McCulloch, N. and B. Baulch (2000). “Simulating the decentralized allocations to the poor are unobserved. World impact of policy upon chronic and transitory poverty in rural Bank Economic Review 14 (2), 331– 345. Pakistan.” Journal of Development Studies 36(6): 100-130. Rosenthal, S. S., & Strange, W. C. (2004). Evidence on McKay, A. and D. Lawson (2003). “Assessing the extent the nature and sources of agglomeration economies. In and nature of chronic poverty in low income countries: Handbook of regional and urban economics (Vol. 4, pp. Issues and evidence.” World Development 31(3): 425-439. 2119–2171). Elsevier. Melo, P. C., Graham, D. J., & Noland, R. B. (2009). A meta- Schmidt, E. and M. Kedir (2009). Urbanization and spatial analysis of estimates of urban agglomeration economies. connectivity in Ethiopia: Urban growth analysis using GIS. Regional science and urban Economics, 39(3), 332–342. ESSP 2 Discussion Paper 003. Addis Ababa, IFPRI. 210 ETHIOPIA POVERTY ASSESSMENT Schmidt, E., Dorosh, P., Jemal, M.K., & Smart, J. (2018). World Bank (2015). Ethiopia Poverty Assesment 2014. Ethiopia’s spatial and structural transformation: Public Washington DC: The World Bank. policy and drivers of change. IFPRI/EDRI Strategy Support World Bank (2019a). Ethiopia Urban Land Supply Program Working Paper 119. and Affordable Housing: Housing Deep Dive Report. Seff, I. and D. Jolliffe (2016). “Multidimensional poverty Unpublished manuscript. dynamics in Ethiopia: How do they differ from consumption- World Bank (2019b). Ethiopia Gender Diagnostic Report. based poverty dynamics?” Ethiopian Journal of Economics Washington DC: The World Bank. 25(2): 1-35. World Bank. (2015a). Ethiopia Poverty Assessment. Sohnesen, T. (2018). “Measurement and economic impact Washington, D.C.: World Bank. of drought: The case of Ethiopia’s El-Nino drought”. Internal working paper. World Bank. (2015b). Ethiopia Urbanization Review: Urban institutions for a middle-income Ethiopia. Washington, D.C.: Sohnesen, T. P. (2018). Droughts, measurement and World Bank. impact: The case of Ethiopia in 2015. CSAE Conference 2018: Economic Development in Africa, Oxford. World Bank. (2016). 5th Ethiopia Economic Update: Why so idle? Wages and employment in a crowded labor market. UN-Habitat. (2007). Situation Analysis of Informal Washington, D.C.: World Bank. Settlements in Addis Ababa. Nairobi, Kenya: UN-Habitat. World Bank. (2016). Ethiopia’s Great Run. The growth Valderrama, D. and M. Viveros (2014). Sectoral gains acceleration and how to pace it. Washington DC: The World and population shifts. Poverty and inequality measures Bank. in practice: A basic reference guide with Stata examples, World Bank Group, Poverty and Equity Global Practice. World Bank. (2018). Review of the Ethiopian Rural Roads Program: Study Report. Washington, D.C.: World Bank. Vandercasteelen, J., Beyene, S.T., Minten, B., & Swinnen, J. (2018). Big cities, small towns, and poor farmers: Evidence from Ethiopia. World Development, 106, 393-406. HARNESSING CONTINUED GROWTH FOR ACCELERATED POVERTY REDUCTION 211 212 ETHIOPIA POVERTY ASSESSMENT ANNEX 1 Additional material for Chapter 1 ETHIOPIA’S WELFARE MONITORING SYSTEM As briefly mentioned at the beginning of Chapter I, 1996 poverty line of ETB 1,075 to estimate the poverty rates. Ethiopia monitors the evolution of poverty, inequality In 2011, the method changed: The original food basket un- and non-monetary dimensions of welfare through the derpinning the 1996 poverty line was re-costed at 2011 pric- HCES and WMS. The HCES and WMS have been conduct- es and an allowance for non-food consumption was added, ed roughly every five years between 1995/96 and 2015/16. resulting in a poverty line of ETB 3,781 (2011 prices). Still With five welfare and poverty surveys over a period of 20 another approach was used in 2016, when the 2011 pover- years, Ethiopia compares favorably to other IDA countries. ty line was updated to 2016 prices using the GDP deflator. There is however room to further strengthen the welfare These changes in methods jeopardize the consistency of the monitoring system to better fit the context of a fast-grow- reported poverty trends. Using a consistent way to construct ing and fast-changing country vulnerable to frequent shocks. price deflators or formulating scientific arguments as to why This relates particularly to transparency and documentation a change in deflator is warranted (and presenting the trend at the one hand and frequency at the other. using the original deflator as well) would improve the integrity of the poverty statistics. Transparency and documentation Frequency While the household consumption data collected by HCES is considered good quality, no public docu- The HCES and WMS are currently implemented every mentation is provided on the construction of the con- five years. One might argue that this is not frequent enough sumption aggregate. Given that different assumptions and in a fast-growing and changing country with frequent weath- approaches can lead to different aggregates and different er shocks to confidently inform effective policy-making. To il- poverty rates, this lack of documentation is important. A lustrate, the results of the latest 2015/16 survey may suggest good practice is to ensure replicability, whereby the con- that progress in rural Ethiopia has slowed substantially, and struction of the aggregate is explained in detail in the data that renewed efforts are needed to uplift rural areas. This sur- documentation, together with the code used in constructing vey was however implemented amidst the El Nino drought, the aggregate. This practice is used, for instance, by the Liv- so the rural numbers may reflect the short-term impact of a ing Standards Measurement Study. severe shock and as such mask a positive long-term trend. Having more frequent poverty measurements would allow Another element of transparency and consistency distinguishing longer-term trends from short-term fluctua- relates to the method of adjusting the consumption tions and closer monitoring of living standards. aggregate or the poverty line for differences in prices across space and over time. As surveys are implemented There are broadly two options to increase the frequen- in different years with different prices, prices need to be ex- cy of poverty monitoring: Increase the frequency of the pressed in a constant period to allow for a valid comparison. HCES/WMS or use the ESS to provide interim poverty es- There are several ways of constructing price deflators, and timates. One option, and likely the preferred one, would be each way will typically lead to slight or non-so-slight differ- to increase the frequency of the WMS/HCES to every three ences in poverty rates. Between 1995/6 and 2015/16, NPC years, which would need to be financed. A second option has used different ways of deflating the consumption aggre- would be to use the ESS to estimate interim poverty num- gate or inflating the poverty line. In 2000 and 2005, the con- bers in between successive HCES surveys. For instance, the sumption aggregate was expressed in 1996 prices using the ESS round that is being implemented in 2019 can be used CPI, and the deflated aggregates were compared with the to provide an interim poverty estimate until the next HCES ANNEX 1: ADDITIONAL MATERIAL FOR CHAPTER 1 213 has been implemented, completed and analyzed. Though fixed in 1996. That is, it reflects the food consumption pat- the design and implementation modalities of HCES and ESS terns of the poor more than 20 years ago. Given the rapid are different, statistical techniques can be used to maxi- changes in Ethiopia over this period, it is likely that the con- mize comparability.83 This option has the advantage of being sumption patterns of the poor have changed substantially lower-cost. and that the original food basket no longer reflects the cur- rent situation. As a good practice, the food basket and pov- A final reflection on frequency concerns the poverty erty line should be re-estimated periodically to make sure it line. The poverty line that is currently being used to assess still reflects actual consumption patterns of the poor. poverty in Ethiopia is still based on a food basket that was Annex Figure 1 THE  FIVE ETHIOPIA’S The five main agro-ecological zones of Ethiopia Drought prone, highland Drought prone, lowland Humid moisture reliable, lowland Moisture reliable, highland Pastoralist NA N 0 100 200 300 400 500 km Source: HCES, 2011; 2016. World Bank staff calculations. 83 Based on a technique called “survey to survey imputation”, poverty numbers estimated from the ESS-3, implemented in 2016, are statistically indistinguishable from the 2016 official HCES poverty numbers. The point estimates were however higher in the ESS. 214 ETHIOPIA POVERTY ASSESSMENT Annex Table 1 MEDIAN  HOUSEHOLD CONSUMPTION INCREASED IN ALL REGIONS EXCEPT AFAR AND AMHARA Regional median annual consumption per adult equivalent in December 2015 prices TOTAL URBAN RURAL % % % 2011 2016 CHANGE 2011 2016 CHANGE 2011 2016 CHANGE Tigray 9,308 10,749 15.5 13,786 15,665 13.6 8,604 9,705 12.8 Afar 9,031 8,503 -5.9 11,317 15,339 35.5 8,277 7,892 -4.7 Amhara 9,395 9,219 -1.9 10,289 16,095 56.4 9,301 8,758 -5.8 Oromia 9,748 10,993 12.8 10,758 14,090 31.0 9,615 10,894 13.3 Somali 9,197 10,195 10.9 11,052 12,143 9.9 8,868 10,128 14.2 Benishangul-Gumuz 9,671 10,641 10.0 11,640 14,659 25.9 9,506 9,971 4.9 SNNPR 9,278 9,972 7.5 10,308 14,089 36.7 9,169 9,692 5.7 Gambella 9,134 11,382 24.6 10,304 13,862 34.5 8,837 10,210 15.5 Harari 11,255 16,739 48.7 12,448 18,392 47.8 10,638 15,607 46.7 Addis Ababa 10,377 12,718 22.6 10,377 12,718 22.6 - - - Dire Dawa 9,610 12,203 27.0 9,540 15,876 66.4 9,733 11,280 15.9 Source: HCES, 2011; 2016. World Bank staff calculations. Annex Figure 2 SENSITIVITY  OF POVERTY MEASURES TO DIFFERENT POVERTY LINE DEFLATORS: REGIONAL Sensitivity of poverty measures to different poverty line deflators: regional 2011 2016 40.0 Percentage of poor 30.0 20.0 10.0 0.0 SNNPR SNNPR Amhara Harari Amhara Harari B. Gumuz Gambella B. Gumuz Gambella Somali Somali Tigray Afar Tigray Oromia Afar Oromia Addis Ababa Addis Ababa Dire Dawa Dire Dawa Case 1 (2011 PL - of cial, 2016 PL - 2011 PL adjusted Case 2 (1996 PL brought to 2011 and 2016 using CPI ) uisng CPI) Source: HCES, 2011; 2016. World Bank staff calculations. ANNEX 1: ADDITIONAL MATERIAL FOR CHAPTER 1 215 Annex Figure 3 MAP  OF QUARTILES OF RAIN FALL SHOCKS IN 2015 Based on average z-score for the months of June, July, August and September 2015 Note: 1st quartile indicates the most severe rain shortfall, while the 4th quartile indicates more than average rainfall. Source: CHIIRPS; World Bank staff calculations. 216 ETHIOPIA POVERTY ASSESSMENT ANNEX 2 Additional material for Chapter 2 Annex Figure 4 CORRELATES  OF RURAL CONSUMPTION IN 2016, ALL NOMADIC ZONES INCLUDED Never married (reference: status of married) Marital head Divorced Household head characteristics Widowed Adult education Education of head (reference: no Primary incomplete education) Primary complete Secondary incomplete Secondary complete Age of head Female head Employed head Others household (reference: Household level: socio-economic variables Main livelihood of Salary_employment Casual_labor others) Self -livestock production Self - crop & livestock Self - service/manufacturing Household size Dependency ratio # non-head employed members # non-head illiterate members Owns land Owns livestock Safetynet participation Drought prone lowland group: ence: drought prone (reference: Geography/location related Ecological highland) Moisture reliable lowland zones Moisture reliable highland Pastoralist Far (between 2 to 3kms) Distance Distance town wether ence (refer- close) close) to all (refer- road Very far (more than 3kms) Far (b/n 1 & 2hrs) to Very far (more than 2hrs) Border zone -40 -20 0 20 40 60 Source: HCES, WMS; 2016. World Bank staff calculations. ANNEX 2: ADDITIONAL MATERIAL FOR CHAPTER 2 217 Annex Table 2 POVERTY  INDICATORS BY SUBGROUPS, 2016 CHARACTERISTICS POVERTY SHARE POVERTY HEAD CONSUMPTION POVERTY GAP POVERTY GAP POPULATION HOUSEHOLD SUBGROUPS (SEVERITY) SQUARED (DEPTH) PER AE COUNT SHARE MEAN Rural 80.9 88.0 10,946 25.6 7.4 3.1 Big cities 7.6 4.4 17,947 13.8 3.2 1.1 Urban/rural Medium cities 5.9 3.0 19,782 11.9 2.8 1.0 Small cities 5.7 4.7 17,671 19.4 5.4 2.3 All urban 19.1 12.0 18,519 14.8 3.7 1.4 Tigray 5.8 6.6 14,108 27.0 7.1 2.7 Afar 1.9 1.9 12,282 23.6 4.2 1.2 Amhara 23.0 25.5 12,340 26.1 6.2 2.2 Oromia 37.8 38.3 12,022 23.9 6.8 3.0 Somalia 5.8 5.5 10,408 22.4 8.4 3.8 Region Benishangul-Gumuz 1.1 1.3 13,373 26.5 5.6 1.8 SNNPR 19.9 17.5 12,204 20.7 7.5 3.1 Gambella 0.4 0.4 13,855 23.0 5.8 2.1 Harari 0.3 0.1 21,059 7.1 3.0 1.9 Addis Ababa 3.6 2.6 16,237 16.8 4.1 1.4 Dire Dawa 0.5 0.3 17,428 15.4 3.0 0.9 Drought prone highland 22.5 19.9 13,422 20.8 5.1 2.0 Drought prone lowland 7.4 10.0 11,129 31.7 10.9 5.0 Agro-ecological Moisture reliable lowland 58.4 58.5 12,136 23.6 6.8 2.8 groups Moisture reliable highland 4.3 4.7 14,705 25.4 5.9 2.0 Pastoralists 7.4 6.9 11,211 21.9 7.3 3.1 Less than or equal to 30 12.8 4.9 17,197 9.0 2.2 0.9 30 to 44 41.9 43.7 11,958 24.6 7.0 2.9 Age of head 50 to 59 28.5 34.9 10,985 28.9 8.5 3.5 60 and above 16.9 16.5 12,201 22.9 6.4 2.6 No education 48.0 58.0 10,498 28.4 8.8 3.7 Primary incomplete 29.7 28.0 11,920 22.1 5.8 2.3 Primary complete 3.2 1.8 15,513 13.0 2.8 0.8 Education of Secondary incomplete 5.1 2.8 18,128 12.8 2.8 1.0 head Secondary complete 1.7 0.5 19,440 6.1 1.3 0.4 Post-secondary 4.5 0.7 25,569 3.5 0.6 0.2 Adult education 7.7 8.3 11,692 25.2 6.1 2.3 Table continue on next page. 218 ETHIOPIA POVERTY ASSESSMENT Table continued from previous page. CHARACTERISTICS POVERTY SHARE POVERTY HEAD CONSUMPTION POVERTY GAP POVERTY GAP POPULATION HOUSEHOLD SUBGROUPS (SEVERITY) SQUARED (DEPTH) PER AE COUNT SHARE MEAN Never married 2.2 1.0 23,407 10.2 2.8 1.1 Marital status Married 84.1 87.1 11,964 24.3 6.9 2.8 of head Divorced 4.3 3.5 14,277 19.1 5.6 2.3 Widowed 9.4 8.4 12,790 21.0 6.0 2.5 Male 19.9 83.8 12,079 24.6 7.0 2.9 Sex of head Female 80.1 16.2 13,676 19.1 5.3 2.1 Agriculture 72.0 80.3 10,756 26.2 7.5 3.1 Sector Industry 4.6 3.6 15,029 18.1 4.6 1.8 of head’s Service 15.1 8.9 18,734 13.9 3.9 1.6 employment No sector (head has no 8.2 7.2 13,574 20.7 5.7 2.4 employment) Less than 0.5 28.7 19.1 15,700 15.7 4.1 1.6 Dependency Between 0.5 and 1 28.6 27.3 12,298 22.5 6.5 2.7 ratio Between 1 and 2 28.6 34.6 10,443 28.5 8.2 3.4 More than 2 14.1 18.9 9,794 31.5 9.5 4.1 1 1.8 0.2 31,544 2.2 0.4 0.2 2 4.9 1.2 20,117 5.7 1.0 0.3 3 10.0 3.9 16,296 9.3 2.1 0.8 Household size 4 14.3 7.7 14,159 12.7 3.0 1.1 5 16.8 13.9 12,174 19.5 5.1 1.9 6 17.0 19.4 11,012 26.9 7.1 2.7 7 and above 35.4 53.7 9,318 35.7 11.2 4.9 Salary employment 8.0 3.9 19,976 11.3 3.0 1.1 Casual labor 2.9 3.6 11,304 28.6 8.8 3.8 Self-crop production 51.5 56.4 10,874 25.7 7.1 2.8 Main livelihood Self-livestock 4.7 5.3 9,814 26.2 8.5 3.5 of household Self-crop & livestock 19.8 21.8 10,908 25.7 7.8 3.5 Self-manufacturing 1.3 1.1 15,412 20.5 4.8 1.9 Self-service 7.3 4.3 18,651 13.9 3.7 1.4 Other livelihoods 4.4 3.7 15,313 19.6 5.7 2.4 Table continue on next page. ANNEX 2: ADDITIONAL MATERIAL FOR CHAPTER 2 219 Table continued from previous page. CHARACTERISTICS POVERTY SHARE POVERTY HEAD CONSUMPTION POVERTY GAP POVERTY GAP POPULATION HOUSEHOLD SUBGROUPS (SEVERITY) SQUARED (DEPTH) PER AE COUNT SHARE MEAN Close (less than 2kms) 54.2 46.2 13,908 20.0 5.5 2.2 Distance to Far (between 2 and 3 16.4 16.1 10,994 23.1 6.4 2.6 health facility kms) Very far (more than 3kms) 29.4 37.6 10,396 30.1 9.1 3.9 Close (less than 2kms) 51.2 41.0 14,353 18.8 5.0 2.0 Distance to all Far (between 2 and 3 9.0 9.1 10,865 23.8 6.6 2.7 weather roads kms) Very far (more than 3kms) 39.8 49.8 10,226 29.4 8.8 3.8 Close (less than 1hr 55.6 42.6 14,235 18.1 4.8 1.9 walking) Distance to town Far (between 1 and 2hrs) 21.6 25.7 10,845 28.2 7.4 2.9 Very far (more than 2hrs) 22.9 31.7 10,060 32.7 10.3 4.4 Border/no Border zones 20.0 25.0 11,678 29.4 9.3 4.0 border zones Non-border zones 80.0 75.0 12,570 22.1 6.1 2.5 Source: Own calculations from HCES, WMS, 2016. 220 ETHIOPIA POVERTY ASSESSMENT ANNEX 3 Additional material for Chapter 3 Annex Figure 5 MOST  OF THE REDUCTION IN POVERTY HAPPENED AMONG THE SELF-EMPLOYED Employment type decomposition of poverty changes, 2011-2016 2 Change in poverty headcount 0 -2 -4 -6 -8 -10 2000-2005 2005-2011 2011-2016 Not self-employed Self-employed Population shift Interaction Source: HCES, 2011; 2016. World Bank staff calculations. Annex Figure 6 CHANGES  IN AGRICULTURAL EMPLOYMENT, MANUFACTURING EMPLOYMENT AND SERVICES EMPLOYMENT 2000 TO 2016 Changes in agriculture and manufacturing employment Changes in agriculture and services employment 5 5 % change in agricultural employment % change in agricultural employment 0 0 −5 −5 −10 −10 −6 −4 −2 0 2 4 −4 −2 0 2 4 % change in manufacturing employment % change in services employment Source: HCES, 2011; 2016. World Bank staff calculations. ANNEX 3: ADDITIONAL MATERIAL FOR CHAPTER 3 221 Annex Figure 7 MAPS  OF TRAVEL TIME TO THE NEAREST CITY OF 50 000 PEOPLE 1984, 1994, 2007 AND 2015 Source: Schmidt, Dorosh et al (2018). 222 ETHIOPIA POVERTY ASSESSMENT ANNEX 4 Additional material for Chapter 4 This chapter uses the first three rounds of ESS data, to account for variation in the cost of living across regions. corresponding to 2012, 2014, and 2016. The ESS is a In line with Fuje (2018) (who uses a poverty line at the 40th nationally representative panel survey of households that percentile) this relative poverty line is different to the official comprised rural households and small towns in all rounds, poverty line, and it was chosen as a 2016 benchmark against but urban (large town) households in rounds 2 and 3 only. which progress from 2012 can be measured, rather than as a Much of the analysis of dynamics is therefore focused on a national poverty line to be used in wider studies. balanced panel of rural and small town residents. The ESS is Rates of attrition for these households were low over representative at the rural and small town level in all rounds, the three rounds of the ESS, with little evidence to sug- and additionally at the urban level in rounds 2 and 3. It is also gest that the probability of attrition differed over the representative for the regions of Amhara, Oromia, Tigray, and distribution of consumption. Figure 15 shows that the SNNPR, which cover approximately 90 percent of the popu- overall retention rate for all rural and small town households lation. In this report, the other regions are generally displayed between 2012 and 2016 was over 90 percent. 94 percent of as the regional group “Other”. rural households that were interviewed in ESS 2012 were suc- The poverty line used in the analysis is an annual adult cessfully re-interviewed in ESS 2016. For households in small equivalent level of consumption of ETB 3 299. This is towns, the corresponding rate was about 85 percent. Attrition the consumption level of the 25th percentile of the overall in urban households was higher than in other areas, and was distribution in 2016.84 All monetary values are converted to close to 16 percent between ESS 2014 and ESS 2016. their 2016 equivalents, and values are also spatially adjusted 84  This is approximately the poverty rate that is estimated using the national poverty line in the 2015/16 HICES. ANNEX 4: DATA AND ADDITIONAL MATERIALS FOR CHAPTER 4 223 Annex Figure 8 PROBABILITY  OF BEING RETAINED BETWEEN 2012 AND 2016 – RURAL AND SMALL TOWN HOUSEHOLDS ONLY 100% Probability of being retained 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% Poorest 2 3 4 5 6 7 8 9 Richest Consumption quintiles Source: Own calculations from ESS 2012, 2014 and 2016. Annex Figure 9 KERNEL  DENSITY DISTRIBUTIONS OF REAL CONSUMPTION EXPENDITURE FOR RURAL AND SMALL TOWN HOUSEHOLDS Rural and small town households .8 .6 Density .4 .2 0 6 8 10 12 Log of real adult equivalent consumption expenditure 2012 2014 2016 Source: Own calculations from ESS 2012, 2014 and 2016. 224 ETHIOPIA POVERTY ASSESSMENT Annex Figure 10 NON-ANONYMOUS  GIC FOR RURAL AND SMALL TOWN HOUSEHOLDS, 2012 TO 2016 Rural and small town households Wave 1 to wave 3 50 Annual percentage change 37.5 25 12.5 0 −12.5 −25 0 .1 .2 .3 .4 .5 .6 .7 .8 .9 1 Consumption expenditure percentiles Confidence interval (95 %) Estimated difference Source: Own calculations from ESS 2012, 2014 and 2016. Annex Figure 11 PROPORTION  OF RURAL AND SMALL TOWN HOUSEHOLDS WITH POSITIVE CONSUMPTION CHANGE BASED ON 2012 DECILES 100% 90% 83% 80% 69% 70% Percentage 60% 54% 50% 46% 43% 37% 40% 30% 26% 25% 20% 16% 7% 10% 0% Poorest 2 3 4 5 6 7 8 9 Richest 2012 consumption deciles Source: Own calculations from ESS 2012, 2014 and 2016. ANNEX 4: DATA AND ADDITIONAL MATERIALS FOR CHAPTER 4 225 Annex Table 3 MARGINAL  EFFECTS OF PROBIT FOR POVERTY EXIT TRANSITIONS OUT OF TRANSITIONS OUT OF POVERTY POVERTY WAVE 1 VARIABLES W1 TO W2 W1 TO W3 WAVE 1 VARIABLES W1 TO W2 W1 TO W3 Household Head 0.146* -0.046 PSNP in kebele 0.002 0.001 (0.086) (0.081) Age of household head (0.002) (0.002) 0.001 0.009 Number of cows -0.024 -0.056 (0.010) (0.010) Male household head (0.083) (0.080) -0.106 0.074 Access to bank/finance 0.123* -0.061 (0.066) (0.065) Primary education (0.072) (0.071) 0.044 0.086 Small town 0.504** 0.369 (0.165) (0.151) Secondary education (0.222) (0.235) -0.135 0.073 Oromia Household (0.101) (0.103) -0.039* 0.039* -0.129 0.025 Size SNNPR (0.021) (0.022) (0.107) (0.103) 0.029 -0.107*** -0.044 0.249** Number of children Tigray (0.044) (0.040) (0.132) (0.101) -0.059 0.062 -0.193 -0.013 Dependency ratio Other (0.053) (0.051) (0.181) (0.190) -0.101 -0.008 0.165 -0.135 Land per adult (hectares) Drought-prone highland (0.216) (0.260) (0.105) (0.112) 0.191 0.251 0.012 -0.220 Land per adult squared Moisture-reliable highland (0.155) (0.221) (0.175) (0.179) -0.001 0.000 0.367*** -0.264* Distance to road Drought-prone lowland (0.002) (0.002) (0.093) (0.145) 0.001 0.003* 0.397** 0.299** Distance to pop. Center Lowland pastoralist (0.002) (0.001) (0.167) (0.131) -0.000 -0.002* Distance to market Observations 469 464 (0.001) (0.001) Note: Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 0.080 0.068 Extension program Source: Own calculations from ESS 2012, 2014 and 2016. (0.063) (0.070) 0.015 0.037 Ever in PSNP (0.081) (0.074) 226 ETHIOPIA POVERTY ASSESSMENT Annex Table 4 MARGINAL  EFFECTS OF PROBIT FOR POVERTY ENTRY TRANSITIONS INTO TRANSITIONS INTO POVERTY POVERTY WAVE 1 VARIABLES W1 TO W2 W1 TO W3 WAVE 1 VARIABLES W1 TO W2 W1 TO W3 Household Head 0.037 PSNP in kebele -0.001* -0.001 (0.035) Age of household head (0.001) (0.001) -0.009*** -0.011*** Number of cows 0.077** 0.081** (0.004) (0.004) Male household head (0.032) (0.034) -0.029 -0.047 Access to bank/finance -0.076*** -0.062** (0.028) (0.030) Primary education (0.026) (0.030) -0.048 -0.152* Small town -0.060 -0.152** (0.068) (0.081) Secondary education (0.064) (0.075) -0.221*** -0.172*** Oromia Household (0.036) (0.037) 0.013 0.018* -0.126*** -0.062 Size SNNPR (0.009) (0.010) (0.044) (0.046) 0.008 0.004 -0.188*** -0.091 Number of children Tigray (0.017) (0.020) (0.052) (0.058) 0.070*** 0.074*** -0.138 0.054 Dependency ratio Other (0.023) (0.024) (0.087) (0.106) -0.099*** -0.078** -0.078*** -0.058 Land per adult (hectares) Drought-prone highland (0.030) (0.032) (0.030) (0.038) 0.005*** 0.004** -0.011 -0.093 Land per adult squared Moisture-reliable highland (0.001) (0.002) (0.084) (0.072) 0.000 0.001 0.015 0.203*** Distance to road Drought-prone lowland (0.000) (0.001) (0.043) (0.061) -0.000 -0.000 -0.064 -0.225*** Distance to pop. Center Lowland pastoralist (0.000) (0.001) (0.075) (0.040) 0.001** 0.000 Distance to market Observations 2,095 2,070 (0.000) (0.000) Note: Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 -0.003 -0.040 Extension program Source: Own calculations from ESS 2012, 2014 and 2016. (0.024) (0.029) -0.010 -0.046 Ever in PSNP (0.030) (0.037) ANNEX 4: DATA AND ADDITIONAL MATERIALS FOR CHAPTER 4 227 ANNEX 5 Additional material for Chapter 6 The PSNP analysis presented in Chapter 6 is based on the Support (DS) or Public Works (PW) beneficiaries. This mat- 2011 and 2016 Household Consumption Expenditure Sur- ters for the analysis, given that the magnitude of transfers veys. The survey includes a variable on whether a household and duration of support is different between the two modal- is, or has ever been, a PSNP beneficiary. This variable has ities. To assign PSNP households to the DS or PW modal- four response options: (i) graduated, (ii) currently participat- ity, we constructed a statistical model based on the 2016 ing, (iii) dropped out, (iv) never participated. In the analysis Ethiopian Socio-Economic Survey (ESS). The ESS contains we mainly focus on response option (ii), the households that information on whether a PSNP household is in DS or PW. were PSNP beneficiary during the year of the survey. A logistical regression model was applied to estimate the correlates of being in DS. This model was then used to the Additional data cleaning was necessary both for the 2011 same independent variables in the HCES surveys to estimate and 2016 survey. In both surveys, a number of households the probability of a PSNP household being in DS. Project in Gambella and Benishangul-Gumuz regions reported to administrative data was used to assign PSNP households be PSNP beneficiaries, even though the PSNP is not active to DS or PW, separately for each region (so that the share there. Similarly, a number of households in large cities report- of households assigned to DS in each region is the same as ed to be part of PSNP. For these households, the decision the share of households in that region in DS according to ad- was made to recode them to non-beneficiaries. Overall, the ministrative data). Overall, this imputation exercise seems to final analysis sample consisted of 2,171 PSNP households in have worked well, with PSNP households estimated to be in 2011 and 1,668 in 2016. DS being smaller, more likely to headed by an older, widowed Though the HCES surveys identify PSNP beneficiary house- woman, more likely to have a disabled household member, holds, they do not allow to distinguish between Direct and less likely to be involved in agriculture (Annex Table 4). Annex Table 5 DIFFERENCE  IN CHARACTERISTICS BETWEEN PW AND DS HOUSEHOLDS DS status estimated by ESS prediction model DS HHS PW HHS NON PSNP HHS HH size 3.83 6.09 5.97 Female head (share yes) 0.63 0.21 0.16 Disability in HH (share yes) 0.19 0.06 0.06 Age head 63.19 42.69 44.94 Widow head (share yes) 0.53 0.06 0.08 Poor (share yes) 0.33 0.33 0.27 Food share 0.58 0.55 0.55 Asset index -0.5 -0.35 0.07 Remoteness index 0.83 1.66 -0.07 HH engaged in agriculture 0.88 0.95 0.98 Note: Asset index is a standardized score on a composite asset indicator. Lower values mean fewer assets. Remoteness index is based on variables measuring the distance to key infrastructure assets (roads,…). Higher scores mean more remote. Source: HCES, 2016; ESS, 2016. 228 ETHIOPIA POVERTY ASSESSMENT Total real household consumption expenditures per adult Given that PSNP is supposed to target the poorest house- equivalent is among the welfare metrics used in this chap- hold, we assume a marginal propensity to consume of 1; ter. The household expenditures calculated from the survey that is, we assume that all PSNP payments are consumed data and used throughout this report are, by construction, and hence show up in consumption expenditures.86 The total inclusive of all aid and donations. However, to study ben- value of PSNP payments is subtracted from total household efit incidence and coverage of social protection projects in consumption expenditures. The resulting pre-transfer con- general and PSNP in particular, we need to try to obtain a sumption variable is then spatially and temporally deflated counterfactual: What households’ welfare levels would have and divided by household size expressed in adult equiva- been in absence of the PSNP. To remove PSNP benefits from lents. All analyses were also conducted with a marginal household expenditures, we calculate total PSNP payments propensity to consume of 0.5 and with the final unadjusted based on the prevailing wage rate (obtain from project data), consumption aggregates (the official ones). As expected, re- household size (which determines household transfer value), sults remain qualitatively similar. and type of support (PW vs DS). For the 2016 HCES, the Correcting the aggregate for HFA benefits is more compli- value of PSNP transfers was estimated as follows: cated as there is no comprehensive data on the volume of PWi,j = Daily wage ratej * Number of days worked per food aid provided nor the number of rounds (months) during monthj * 6 which aid was provided. Based on administrative data from humanitarian partners, it appears seven rounds of food aid DSi,j = Daily wage ratej * Eligible days per month * 12 were provided in 2015/16. It is estimated that the value of Where PWi,j is the annual value of benefits received by a PW HFA benefits are on average 15 percent higher than PSNP. household i in woreda j. The daily wage rate differs across Monthly HFA benefits at household level were calculated us- woredas and the number of days worked per month is a ing the same household size rule as in PSNP, and with a function of household size. PW beneficiaries work for six benefit rate 15 percent higher than PSNP. This is multiplied months per year (hence the monthly payment is multiplied by 7 to account for the seven rounds of food aid in 2015/16. by 6). DSi,j is the annual value of benefits received by a DS The HFA analysis was also conducted using the unadjusted household i in woreda j. Eligible days per month for DS aggregate. Results were qualitatively similar. households is also a function of household size.85 Month- ly payments are multiplied by 12 as DS households receive support all-year-round. For 2011, similar formulas were used, but with the 2011 wage rate. 85 Following rule was used to determine the number of days per month per household: Up until four household members, the house- hold is eligible to work five days per month per household member (20 days for a household of four). For a household of between five and eight members, the total number of days worked per month is capped at 20. For households of more than eight members, the total number of days worked per month is capped at 25. Though according to the project manual larger households should get more working days (more than 25), the observation on the field is that large households work fewer days than planned for in the project documents, likely related to budget constraints. Applying this rule to the 2016 HCES results in an average of 3.8 beneficia- ries per PSNP household, which is exactly the same as the number obtained from project administrative data. 86 Given that XX PSNP payments are made in-kind (food), this would not appear to be an unreasonable assumption. ANNEX 5: DATA AND ADDITIONAL MATERIALS FOR CHAPTER 6 229 ANNEX 6 Additional material for Chapter 7 Annex Figure 12 LOCATION  AND HOUSEHOLD WEALTH DETERMINES PRIMARY SCHOOL COMPLETION Completion of primary school by circumstance variables, children 15-18, 2016 100% 90% 80% 70% Urban 60% Addis Ababa 50% Richest Dire Dawa Tigray Highest RAI Harari 40% Female Orthodox B−G Gambella 30% Other Christian Amhara Afar SNNPR Mean Oromiya Male Somali 20% Muslim Lowest RAI Rural Poorest 10% 0% n r n n e tle de tio io io til in in ig eg en ca qu qu el R G Lo R y n ilit io ib pt ss m ce su ac on C al ur R Source: Calculations from WMS 2015/16 and HCES 2015/16. 230 ETHIOPIA POVERTY ASSESSMENT Annex Figure 13 LOCATION  AND HOUSEHOLD WEALTH DETERMINE ENROLMENT IN SECONDARY SCHOOL Completion of primary school by circumstance variables, children 15-18, 2016 100% 90% 80% 70% 60% 50% Urban 40% Addis Ababa Dire Daw 30% Gambella Harari Richest Highest RAI Female Orthodox B−G Tigray 20% SNNPR Mean Other Christian Afar Amhara Oromiya Male 10% Rural Muslim Somali Poorest Lowest RAI 0% n r n n e tle de tio io io til in in ig eg en ca qu qu el R G Lo R y n ilit io ib pt ss m ce su ac on C al ur R Source: Calculations from WMS 2015/16 and HCES 2015/16. Annex Figure 14 LOCATION  LARGELY DETERMINES ACCESS TO IMPROVED WATER Access to an improved water source by circumstance variables, children 7-18, 2016 Addis Ababa 100% Urban Dire Dawa 90% Gambella Richest 80% Harari B−G Highest RAI Tigray 70% Orthodox Afar Female 60% Muslim Amhara SNNPR Mean Male Oromiya Somali 50% Rural Other Christian Poorest Lowest RAI 40% 30% 20% 10% 0% n r n n e tle de tio io io til in in ig eg en ca qu qu el R G Lo R y n lit io bi pt i ss m ce su ac on C al ur R Source: Calculations from WMS 2015/16 and HCES 2015/16. ANNEX 6: ADDITIONAL TABLES AND FIGURES FOR CHAPTER 7 231 Annex Figure 15 LOCATION  DETERMINES PROXIMITY TO A HOSPITAL Access to a hospital within 5km, children 7-18, 2016 100% Addis Ababa 90% 80% 70% Harari 60% Urban Dire Dawa 50% 40% 30% Richest Highest RAI 20% Orthodox Gambella Female 10% Muslim SNNPR Tigray Oromiya Mean Male Other Christian Somali Amhara B−G Poorest 0% Rural Afar Lowest RAI n r n n e tle de tio io io til in in ig eg en ca qu qu el R G Lo R y n ilit io ib pt ss m ce su ac on C al ur R Source: Calculations from WMS 2015/16 and HCES 2015/16. Annex Figure 16 EDUCATIONAL  ATTAINMENT IMPROVED FOR SCHOOL-AGED CHILDREN BETWEEN 2011 AND 2016 Years of educational attainment, children 7-18, 2011-2016 1 .8 Cumulative probability .6 .4 .2 0 0 2 4 6 8 10 12 14 16 Years of education children aged 7 to 18 2011 2016 Source: Calculations from HCES 2010/11 and HCES 2015/16. 232 ETHIOPIA POVERTY ASSESSMENT Annex Table 6 RELATIONSHIP  BETWEEN HOUSEHOLD CONSUMPTION LEVEL AND CHILDREN’S SCHOOLING, 2016 VARIABLES YEARS OF EDUCATION Age 8 0.884*** (0.001) Age 9 1.538*** (0.001) Age 10 1.948*** (0.001) Age 11 2.307*** (0.001) Age 12 2.478*** (0.001) Age 13 2.686*** (0.001) Age 14 2.850*** (0.001) Age 15 2.911*** (0.001) Age 16 3.077*** (0.001) Age 17 3.141*** (0.001) Age 18 3.153*** (0.001) Female 0.024*** (0.000) Log consumption 0.128*** (0.000) Rural -1.377*** (0.002) Rural*Log consumption 0.099*** (0.000) Constant -2.278*** (0.002) Observations 36 041 Note: *** p<0.01, ** p<0.05, * p<0.1. Results show coefficients from a negative binomial regression with years of education as the dependent variable. Sample is restricted to individuals aged between 7 and 18. Source: Calculations from WMS 2015/16, and HCES 2015/16. ANNEX 6: ADDITIONAL TABLES AND FIGURES FOR CHAPTER 7 233 234 ETHIOPIA POVERTY ASSESSMENT