Report No: AUS0002913 . Uganda Uganda Poverty Assessment Strengthening Resilience to Accelerate Poverty Reduction . June 27 2022 . POV . Uganda Poverty Assessment: Strengthening Resilience to Accelerate Poverty Reduction 27 June 2022 ii © 2022 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org 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, completeness, or currency of the data included in this work and does not assume responsibility for any errors, omissions, or discrepancies in the information, or liability with respect to the use of or failure to use the information, methods, processes, or conclusions set forth. 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 construed or 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 The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Attribution—Please cite the work as follows: “World Bank. 2022. Uganda Poverty Assessment: Strengthening Resilience to Accelerate Poverty Reduction. © World Bank.” Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522- 2625; e-mail: pubrights@worldbank.org. iii Table of Content Acknowledgment .......................................................................................................................................... i Acronyms and Abbreviations ....................................................................................................................... ii Overview...................................................................................................................................................... iii I. Poverty and Inequality: Facts ................................................................................................................. v Multiple shocks in 2012–2020 led to volatile poverty rates................................................................. v Every third individual experienced transient poverty in the last five years, while every fifth remained chronically poor ................................................................................................................... vi The Northern and Eastern regions continued having the highest poverty rates, as well as the biggest gaps in human capital outcomes and access to basic infrastructure .................................................. vi Working in agriculture and lack of education are the strongest predictors of high poverty ............ viii Households with more children, larger household size, and with a single female income earner were more likely to have higher poverty rates .................................................................................. viii II. Why has there been limited progress in poverty reduction? .............................................................. ix Rural households and the poor experienced more frequent shocks and have limited safety nets .... ix Half of the population was vulnerable to poverty in 2019/20 – education, asset ownership, and working status being key determinants.............................................................................................. xii The change from subsistence agriculture to paid employment was at a higher pace among the wealthier and more educated ............................................................................................................xiv Internal migration was not widely accessible for the poorest households living in lagging areas .....xvi COVID-19 slowed down the pace of structural change and increased vulnerability as many were pushed into agriculture ...................................................................................................................... xvii Agricultural development, which is the key for poverty reduction, has not experienced substantial changes in production practices in recent years .............................................................................. xviii Inequality of opportunities in access to basic services persisted and was exacerbated by the COVID- 19 pandemic........................................................................................................................................ xix Increasing competition in the telecommunications sector could increase affordability and access, and reduce poverty ............................................................................................................................. xxi III. Way forward: A policy agenda ......................................................................................................... xxiii Raise productivity and income earning opportunities...................................................................... xxiii Strengthen household resilience ...................................................................................................... xxiv Chapter 1. Most recent poverty and inequality trends ............................................................................ 27 I. Background .......................................................................................................................................... 27 II. Trends in poverty, inequality, and shared prosperity since 2016/17 ................................................. 31 iv Poverty indicators ............................................................................................................................... 31 Inequality and shared prosperity ........................................................................................................ 34 IV. Sociodemographic characteristics of the poor in 2019/20 ............................................................... 38 Household and individual level characteristics ................................................................................... 38 Community characteristics ................................................................................................................. 45 V. Geographic disparities in poverty and their correlates ...................................................................... 49 Sources of welfare disparities within and across regions ................................................................... 49 Geographic factors behind subregional welfare disparities ............................................................... 53 VI. Policy implications ............................................................................................................................. 57 References .............................................................................................................................................. 59 Annex 1 ................................................................................................................................................... 60 Chapter 2. Most recent trends in the labor market and structural change ............................................. 67 I. Economic growth and structural change ............................................................................................. 67 II. The distributional aspects of structural change.................................................................................. 70 III. Impacts of COVID-19 on the labor market ......................................................................................... 76 IV. Internal migration .............................................................................................................................. 80 Magnitude and direction of migration flows ...................................................................................... 81 Characteristics of internal migration .................................................................................................. 84 Structural change, internal migration and monetary welfare ............................................................ 87 IV. Agriculture ......................................................................................................................................... 88 V. Policy implications .............................................................................................................................. 94 References .............................................................................................................................................. 96 Chapter 3. Vulnerability in Uganda ........................................................................................................... 97 I. Shocks and coping strategies ............................................................................................................... 97 Exposure to shocks before COVID-19 ................................................................................................. 97 Coping strategies before COVID-19 .................................................................................................. 102 Shocks and coping strategies during the COVID-19 pandemic ......................................................... 106 II. Quantifying vulnerability to poverty ................................................................................................. 110 The concept of vulnerability to poverty............................................................................................ 110 Estimating vulnerability to poverty in Uganda in 2019/20 ............................................................... 111 III. Policy implications............................................................................................................................ 116 References ............................................................................................................................................ 118 Annex 2 ................................................................................................................................................. 119 v Chapter 4. Inequality of opportunities .................................................................................................... 124 I. Background ........................................................................................................................................ 124 II. Human Opportunity Index for Ugandan children ............................................................................. 125 Uganda’s most recent performance in human capital ..................................................................... 125 Constructing Human Opportunity Index for Uganda ........................................................................ 127 Progress towards opportunities for all children in 2019/20 ............................................................. 130 Explaining changes in HOI between 2012/13 and 2019/20.............................................................. 134 III. Impact of COVID-19 on human capital ............................................................................................ 136 IV. Policy implications ........................................................................................................................... 139 References ............................................................................................................................................ 141 Annex 3 ................................................................................................................................................. 142 Chapter 5. The role of the telecommunications sector for poverty reduction ...................................... 144 I. Development of the telecommunications sector .............................................................................. 144 Trends in the telecommunications sector during the last decade ................................................... 144 Distribution of telecommunication services ..................................................................................... 146 II. Welfare impacts of competition in the mobile telecommunications market in Uganda ................. 150 III. Policy implications............................................................................................................................ 153 References ............................................................................................................................................ 155 Annex 4 ................................................................................................................................................. 156 vi List of figures Figure O1. Headcount poverty rates across rural and urban areas during 2012/13–2019/20 (before and during COVID-19), % ..................................................................................................................................... v Figure O2. Gini coefficient across rural and urban areas during 2012/13–2019/20 (before and during COVID-19) ..................................................................................................................................................... v Figure O3. Poverty transitions between 2015/16 and 2019/20 by head of household characteristics in 2015/16 year, % ........................................................................................................................................... vi Figure O4. Headcount poverty rates in 2019/20 across areas, regions and subregions, % ....................... vii Figure O5. Poverty rate in 2016/17 and Human Capital Index circa 2016/17 by subregions .................... vii Figure O6. Poverty rate in 2016/17 and access to electricity by subregions*............................................ vii Figure O7. Population density in Uganda in 2020** .................................................................................. vii Figure O8. Night lights in Uganda in 2020*** ............................................................................................ vii Figure O9. Headcount poverty rates in 2019/20, depending on head of household sector of employment, %................................................................................................................................................................. viii Figure O10. Headcount poverty rates depending on head of household education level in 2019/20, % viii Figure O11. Headcount poverty rates in households with different number of children in 2019/20, % .... ix Figure O12. Headcount poverty rates depending on head of household marital and gender status in 2019/20, % ................................................................................................................................................... ix Figure O13. Share of households that experienced at least one shock across survey years and rural and urban consumption quintiles, % ................................................................................................................... x Figure O14. Shocks and decline in different dimensions of wellbeing in 2019/20, % of households reporting decline ........................................................................................................................................... x Figure O15. Distribution of coping strategies during the last 12 months in 2019/20 across residence, head of household gender and rural/urban consumption quintiles, % ....................................................... x Figure O16. Share of population with at least one shock during last 12 month across poverty transition status during 2015/16–2019/20, % .............................................................................................................. x Figure O17. Incidence of shocks among Ugandans during March–June 2020, % ....................................... xi Figure O18. Type of coping strategies used by households in Uganda during March–June 2020 across different dimensions, % of all strategies ..................................................................................................... xi Figure O19. Incidence of beneficiaries of SCG in 2019/20 among individuals 60 years+ by consumption quintiles, % .................................................................................................................................................. xii Figure O20. Incidence of beneficiaries of NUSAF in 2019/20 among individuals 15 years+ by consumption quintiles, % .................................................................................................................................................. xii Figure O21. Poverty and vulnerability rates in Uganda in 2019/20 across rural and urban areas, and regions, % ....................................................................................................................................................xiii Figure O22. Vulnerability decomposition into poverty- and risk-induced components across rural and urban areas, and regions, %........................................................................................................................xiii Figure O23. Changes in probability of being vulnerable to poverty with regards to non-vulnerable by different household characteristics ............................................................................................................xiv Figure O24. Annualized growth rates of employment share in agriculture and GDP per capita, % ..........xiv Figure O25. Value added in agriculture in per capita terms in UGX constant prices and the share of dry months based on SPEI index ....................................................................................................................... xv Figure O26. Share of paid employment across different groups of population using LFS 14–64, %..........xvi Figure O27. Share of working age individuals employed in agriculture by consumption quintiles using UNPS 14–64, % ............................................................................................................................................xvi Figure O28. Individual consumption and its annual growth rate across household heads’ sector of employment in 2015/16 and 2019/20, UGX and % ................................................................................... xvii Figure O29. Individual consumption and its annualized growth rates across household heads’ migration status in 2015/16 and 2019/20, UGX and % .............................................................................................. xvii Figure O30. Working respondents using different rounds of the HFPS, % of respondents ...................... xvii Figure O31. Work stoppages among those who worked in the previous round across rural/urban areas and economic sectors, % ........................................................................................................................... xvii Figure O32. Share of working respondents among refugees and Ugandans before COVID-19 and during the first year of the pandemic, %.............................................................................................................. xviii Figure O33. Share of households in planting activities who used fertilizers at least once on any land plot during any of the agricultural seasons in 2013/14 and 2019/20, % ........................................................... xix Figure O34. Share of households in planting activities who used improved seeds on any land plot during either of the two agricultural seasons and who received any advice from extension services in 2013/14 and 2019/20, % ........................................................................................................................................... xix Figure O35. Coverage rate, HOI and inequality for access to education and health among children in 2019/20, % ...................................................................................................................................................xx Figure O36. Coverage rate, HOI and inequality for access to basic services among children in 2019/20, % .....................................................................................................................................................................xx Figure O37. Decomposing trends in the HOI between 2012/13 and 2019/20, percentage points .............xx Figure O38. School attendance and participation in any type of schooling for children ages 3–18 years in March 2020 (before lockdown) and in March/April 2021 by area, region and pre–COVID-19 consumption quintiles, % .................................................................................................................................................. xxi Figure O39. Access to mobile phones among individuals 16 years+ in 2019/20, % .................................. xxii Figure O40. Effect of improved competition on Uganda’s poverty rate, change in percentage points.... xxii Figure 41. Annual percentage growth rates of GDP at market prices based on constant local currency and population growth rates in Uganda and comparators in 2010–2019, %............................................. 27 ii Figure 42. Contributions to change in real GDP per capita growth, by value added from key economic sectors, changes in percentage points........................................................................................................ 28 Figure 43. Contribution to changes in real GDP per capita growth, by demand side, changes in percentage points ....................................................................................................................................... 28 Figure 44. Working status among respondents to Uganda High-Frequency Phone survey, % .................. 29 Figure 45. Households with income below average monthly income during 12-month period prior to the first lockdown, % of receiving income ........................................................................................................ 29 Figure 46. Headcount poverty rates across rural and urban areas during 2012/13–2019/20, % .............. 32 Figure 47. Headcount poverty rates across rural and urban areas during 2016/17–2019/20 (before and during COVID-19), % ................................................................................................................................... 32 Figure 48. Depth of poverty across rural and urban areas during 2016/17–2019/20 (before and during COVID-19), % ............................................................................................................................................... 33 Figure 49. Severity of poverty across rural and urban areas during 2016/17–2019/20 (before and during COVID-19), % ............................................................................................................................................... 33 Figure 50. Headcount poverty rates across regions during 2016/17–2019/20, % ..................................... 33 Figure 51. Headcount poverty rates across regions during 2016/17–2019/20 (before and during COVID- 19), % .......................................................................................................................................................... 33 Figure 52. International headcount poverty rates in Uganda and comparators in 2019 circa years based on 1.90 USD 2011 PPP daily poverty line .................................................................................................... 34 Figure 53. Consumption per adult equivalent Gini coefficient across rural and urban areas during 2016/17–2019/20 using UNHS ................................................................................................................... 34 Figure 54. Consumption per adult equivalent Gini coefficient across rural and urban areas during 2016/17–2019/20 (before and during COVID-19) using UNHS .................................................................. 34 Figure 55. Consumption per adult equivalent Gini coefficient across countries circa 2019 year .............. 35 Figure 56. Annualized consumption per adult equivalent growth rates by deciles during 2016/17– 2019/20 in rural and urban areas using UNHS, %....................................................................................... 35 Figure 57. Annualized consumption per adult equivalent growth rates by deciles across rural and urban areas during 2016/17–2019/20 (pre-COVID) using UNHS, % ..................................................................... 36 Figure 58. Annualized consumption per adult equivalent growth rates by deciles across rural and urban areas in 2019/20 (before and during COVID-19) using UNHS, %................................................................ 36 Figure 59. Growth-redistribution decomposition of poverty changes in 2016/17–2019/20 (before and during COVID-19) using UNHS in rural and urban areas, percentage points ............................................. 36 Figure 60. Annualized consumption per adult equivalent growth rates across rural and urban areas, regions during 2015/16 and 2019/20 using UNPS data, % ......................................................................... 37 Figure 61. Headcount poverty rates in 2019/20 across areas, regions and subregions, % ........................ 38 Figure 62. Share of the poor population in 2019/20 across areas and regions, % ..................................... 38 Figure 63. Headcount poverty rates depending on income earner composition 2019/20, % ................... 40 iii Figure 64. Population shares depending on income earner composition 2019/20, % .............................. 40 Figure 65. Headcount poverty rates in households with different number of children in 2019/20, % ..... 40 Figure 66. Headcount poverty rates depending on head of household marital and gender status in 2019/20, % .................................................................................................................................................. 40 Figure 67. Selected gender and marital status coefficients from regressions explaining logarithm of consumption per adult equivalent in rural and urban areas in 2019/20, % ............................................... 41 Figure 68. Headcount poverty rates depending on head of household education level in 2019/20, % .... 41 Figure 69. Consumption per capita returns to head of household education after controlling for other factors in 2019/20, % .................................................................................................................................. 41 Figure 70. Headcount poverty rates depending on head of household working status 2019/20, % ......... 42 Figure 71. Head of household working status across poverty status in 2019/20, % .................................. 42 Figure 72. Headcount poverty rates in 2019/20 depending on head of household sector of employment, %.................................................................................................................................................................. 43 Figure 73. Ownership of selected assets in 2016/17 and 2019/20 in Uganda and among the poor households, % ............................................................................................................................................. 44 Figure 74. Ownership of selected assets in 2019/20 in rural and urban areas across poor and non-poor households, % ............................................................................................................................................. 44 Figure 75. Availability of land and livestock in 2019/20 in rural and urban areas across poor and non- poor households, %..................................................................................................................................... 44 Figure 76. Access to improved water and sanitation in 2019/20 in rural and urban areas across poor and non-poor households, % ............................................................................................................................. 44 Figure 77. Availability of government health centers, primary and secondary schools in the original or neighboring communities in 2019/20 for poor and non-poor population, % ............................................ 45 Figure 78. Distance to government health centers, primary and secondary schools from a village center in the original or neighboring communities in 2019/20 for poor and non-poor population, kilometers .. 45 Figure 79. Access to other essential services and infrastructure in the original or neighboring communities in 2019/20 for poor and non-poor population, % ................................................................ 46 Figure 80. Distance to other essential services and infrastructure from a village center in the original or neighboring communities in 2019/20 for poor and non-poor population, kilometers.............................. 46 Figure 81. Share of population living in communities with services ranked as poor, % ............................ 46 Figure 82. Poverty transitions between 2015/16 and 2019/20 by head of household characteristics in 2015/16, % .................................................................................................................................................. 47 Figure 83. Household asset ownership in 2015/16 across poverty transitions status between 2015/16 and 2019/20, % ........................................................................................................................................... 48 Figure 84. Headcount poverty rates across regions and rural and urban areas within regions in 2019/20, %.................................................................................................................................................................. 50 iv Figure 85. Urban versus rural and Central versus other regions: endowments and rewards explaining log difference in consumption per adult equivalent in 2019/20, percentage points ...................................... 51 Figure 86. Urban versus rural and Central versus other regions: type of endowments explaining log difference in consumption per adult equivalent in 2019/20, percentage points ...................................... 51 Figure 87. Kampala versus other regions: endowments and rewards explaining log difference in consumption per adult equivalent in 2019/20, percentage points ............................................................ 52 Figure 88. Kampala versus other regions: type of endowments explaining log difference in consumption per adult equivalent in 2019/20, percentage points .................................................................................. 52 Figure 89. Headcount poverty rates by subregions in 2016/17, % ............................................................. 53 Figure 90. Number of poor by subregions in 2016/17, people, people ..................................................... 53 Figure 91. Subregional Human Capital Index by subregions circa 2016/17 ............................................... 54 Figure 92. Correlation between poverty rate in 2016/17 and Human Capital Index circa 2016/17 by subregions ................................................................................................................................................... 54 Figure 93. Share of population with access to improved sanitation by subregions in 2016/17, % ............ 55 Figure 94. Share of population with access to electricity by subregions in 2016/17, % ............................ 55 Figure 95. Correlation between poverty rate and access to improved sanitation by subregions.............. 55 Figure 96. Correlation between poverty rate in 2016/17 and access to electricity by subregions ............ 55 Figure 97. Market accessibility index .......................................................................................................... 56 Figure 98. Correlation between poverty rate in 2016/17 and logarithm of market accessibility index .... 56 Figure 99. Population density in Uganda in 2020 ....................................................................................... 56 Figure 100. Night-time light in Uganda in 2020 .......................................................................................... 56 Figure 101. Drought risk index .................................................................................................................... 57 Figure 102. Share of population exposed to pluvial flood risk, %............................................................... 57 Figure 103. Annualized growth rates of employment share in agriculture and GDP per capita, % ........... 68 Figure 104. Value added per working person by economic sectors, million UGX in constant 2016/17 prices ........................................................................................................................................................... 68 Figure 105. Access to electricity in the initial period and GDP per capita growth rates in Uganda and comparators, % ........................................................................................................................................... 69 Figure 106. Completion of lower secondary school in the initial period and GDP per capita growth rates in Uganda and comparators, % ................................................................................................................... 69 Figure 107. Value added in agriculture in per capita terms in UGX constant prices and the share of dry months in Uganda based on SPEI index in % .............................................................................................. 70 Figure 108. Fertility rate in the initial period and GDP per capita growth rates in Uganda and comparators, births per woman and % ...................................................................................................... 70 Figure 109. Working age population and working in paid employment and subsistence agriculture in 2012/12 and 2018/19, millions of people .................................................................................................. 71 v Figure 110. Growth in working age population and working in paid employment and subsistence agriculture in 2018/19 using 2012/12 as a base ......................................................................................... 71 Figure 111. Structure of the working age population by type of working status across areas and time, % .................................................................................................................................................................... 72 Figure 112. Structure of the working age population by type of working status across education level and time, % ........................................................................................................................................................ 72 Figure 113. Structure of the working age population by type of working status across age groups and time, % ........................................................................................................................................................ 72 Figure 114. Structure of the working age population by type of working status across gender and time, % .................................................................................................................................................................... 72 Figure 115. Median total monthly earnings among employees in 2011/12 and 2019/20, UGX in 2011/12 prices ........................................................................................................................................................... 73 Figure 116. Median total earnings per hour worked among employees in 2011/12 and 2019/20, UGX in 2011/12 prices ............................................................................................................................................ 74 Figure 117. Annualized real growth rates of labor income per capita by sources and across population groups between 2013/14 and 2019/20, % ................................................................................................. 75 Figure 118. Share of agriculture in employment by consumption per adult equivalent quintiles, % ........ 76 Figure 119. Consumption per adult equivalent and its annual growth rate in 2015/15 and 2019/20 depending on head of household sector of employment, % ..................................................................... 76 Figure 120. Stringency index and cumulative number of COVID-19 cases in Uganda from January 2020 to December 2021 ........................................................................................................................................... 77 Figure 121. Working respondents across different rounds, % ................................................................... 78 Figure 122. Work stoppages among those who worked in previous rounds across rural/urban areas and economic sectors, % ................................................................................................................................... 78 Figure 123. Structure of work stoppages by economic sector, %............................................................... 78 Figure 124. Reasons for work stoppages across rounds 1 and 7, % of respondents who stopped working .................................................................................................................................................................... 78 Figure 125. Economic sector of current employment across rounds, % .................................................... 79 Figure 126. Share of working respondents among refugees and Ugandans before COVID-19 and during the first year of pandemic, % ...................................................................................................................... 79 Figure 127. Share of households unable to cover basic needs and access services among refugees and Ugandans, %................................................................................................................................................ 80 Figure 128. Share of urban population in total population in Uganda and comparators, % ..................... 81 Figure 129. Main reasons for migrating across different groups as of 2016/17, % ................................... 82 Figure 130. Structure of internal migrants by area of origin and destination as of 2016/17, % ................ 83 Figure 131. Employment sector of household head using UNHS 2016/17, %............................................ 87 vi Figure 132. Monthly consumption per adult equivalent and consumption annualized growth rates across household heads’ migration status from the UNPS 2015/16 and 2019/20, UGX and % ............................ 88 Figure 133. Subregional poverty rates versus different agricultural indicators ......................................... 89 Figure 134. Growth rate in value added in agriculture versus growth rates in rural population during 2010 to 2020, % .......................................................................................................................................... 90 Figure 135. Real annualized growth rates of median crop production per hectare and median agricultural income per capita by consumption per adult equivalent quintiles during 2013/14 and 2019/20 among the population engaged in agricultural activities, %................................................................................... 91 Figure 136. Share of households engaged in crop and livestock production activities by consumption per adult equivalent quintiles, % ...................................................................................................................... 92 Figure 137. Share of agricultural income in total income by consumption per adult equivalent quintiles, %.................................................................................................................................................................. 92 Figure 138. Share of households engaged in planting activities who used fertilizers at least once on any land plot during any of the agricultural seasons in 2013/14 and 2019/20, % ............................................ 92 Figure 139. Share of households engaged in planting activities who used improved seeds on any land plot during either of the two agricultural seasons and who received any advice from extension services during the last 12 months in 2013/14 and 2019/20, % .............................................................................. 93 Figure 140. Share of households engaged in planting activities who used mechanized equipment during the last 12 months in 2013/14 and 2019/20, % ......................................................................................... 94 Figure 141. Total number of natural disasters and population affected* in Uganda during 1999–2020 .. 97 Figure 142. Experience with climate conditions for agricultural production in Uganda and selected comparators compared to 10 years ago in circa 2016–2018, %................................................................. 97 Figure 143. Share of households with at least one shock during last the 12 months across rounds and by rural and urban areas, % ............................................................................................................................. 99 Figure 144. Share of households that experienced at least one shock during the last 12 months across years and regions, % ................................................................................................................................... 99 Figure 145. Share of households that experienced at least one shock during the last 12 months across years and rural and urban consumption per adult equivalent quintiles, % ............................................... 99 Figure 146. Incidence of different shocks in 2019/20 across regions, % of households .......................... 100 Figure 147. Shocks and decline in different dimensions of wellbeing in 2019/20, % .............................. 100 Figure 148. Characteristics of shocks between 2013/14 and 2019/20 .................................................... 101 Figure 149. Types of coping strategies among households experienced shocks during the last 12 months in 2019/20, % ............................................................................................................................................ 103 Figure 150. Distribution of coping strategies during the last 12 months in 2019/20 across residence, head of household gender and rural/urban consumption quintiles. % ............................................................ 103 Figure 151. Incidence of beneficiaries of SCG during the last 12 months in 2019/20 among individuals 60 years+ by consumption quintiles, % ......................................................................................................... 104 vii Figure 152. Incidence of beneficiaries of NUSAF during the last 12 months in 2019/20 among individuals 15 years+ by consumption quintiles, % .................................................................................................... 104 Figure 153. Getting any type of social assistance based on HFPS, % of respondents .............................. 105 Figure 154. Incidence of social assistance by types of aid based on HFPS, % of respondents ................. 105 Figure 155. Main coping strategy used across key shocks experienced by households during the last 12 months in 2019/20, % ............................................................................................................................... 106 Figure 156. Share of population with at least one shock during last 12 month across poverty transition status during 2015/16-2019/20, % ........................................................................................................... 106 Figure 157. Incidence of shocks among Ugandans during March–June 2020, % ..................................... 107 Figure 158. Types of shocks faced by Ugandans during March–June 2020, %......................................... 107 Figure 159. Type of coping strategies used by households in Uganda during March–June 2020 across different dimensions, % of all strategies .................................................................................................. 108 Figure 160. Reasons for borrowing to face COVID-19 emergency during March–July/August 2020, % .. 108 Figure 161. Level of worry that the household will not be able to repay all the money borrowed within the repayment period, % of those who borrowed ................................................................................... 108 Figure 162. Incidence of shocks among Ugandans during March-June 2020 and refugees during March- Oct/Nov 2020, % of households ............................................................................................................... 109 Figure 163. Vulnerability to poverty is characterized by the mean and variance of household consumption ............................................................................................................................................. 110 Figure 164. Poverty and vulnerability rates in Uganda in 2019/20 across rural and urban areas and regions, % .................................................................................................................................................. 112 Figure 165. Poverty and vulnerability rates in Uganda in 2019/20 across subregions, % ........................ 112 Figure 166. Vulnerability decomposition into poverty- and risk-induced components across rural and urban areas in 2019/20, and regions, % ................................................................................................... 112 Figure 167. Vulnerability decomposition into poverty- and risk-induced components across subregions in 2019/20, % ................................................................................................................................................ 112 Figure 168. Ratio of idiosyncratic to covariate shocks across rural and urban areas and regions in 2019/20, % ................................................................................................................................................ 113 Figure 169. Ratio of idiosyncratic to covariate shocks across subregions in 2019/20 ............................. 113 Figure 170. Poverty and vulnerability nexus by head of household’s sector of employment in 2019/20, % .................................................................................................................................................................. 114 Figure 171. Poverty and vulnerability nexus by head of household’s education level in 2019/20, % ..... 114 Figure 172. Probability of being vulnerable to poverty with regards to non-vulnerable in 2019/20, marginal effects ........................................................................................................................................ 115 Figure 173. Human capital index in Uganda and selected comparators in 2020 ..................................... 126 Figure 174. Selected components of HCI in Uganda and other countries in 2020 versus logarithm of GDP per capita 2017 PPPs................................................................................................................................. 127 viii Figure 175. The coverage, HOI and D-index for access to education and health among children in 2019/20, % ................................................................................................................................................ 131 Figure 176. Contribution of each circumstance to inequality of opportunities in education and health in 2019/20, % ................................................................................................................................................ 132 Figure 177. The reasons for not attending the school among children 6–16 years old by different circumstances in 2019/20, % .................................................................................................................... 133 Figure 178. Coverage, HOI and D-index for access to basic services among children in 2019/20, %....... 133 Figure 179. Contribution of each circumstance to inequality of opportunities in basic services in 2019/20, %................................................................................................................................................................ 134 Figure 180. Trends in the HOI between 2012/13, 2016/17 and 2019/20, %............................................ 135 Figure 181. Decomposing trends in the HOI between 2012/13 and 2019/20, by scale and distribution effects, change in percentage points ........................................................................................................ 136 Figure 182. School attendance and participation in any type of schooling among children ages 3–18 years in March 2020 (before lockdown) and in March/April 2021 by area, region and pre-COVID-19 consumption quintiles, %.......................................................................................................................... 138 Figure 183. Access to medicine when needed in the HFPS, % of households .......................................... 138 Figure 184. Access to medical treatment when needed in the HFPS, % of households........................... 138 Figure 185. Main reasons for inability to access medical treatment in June 2020 according to the HFPS, % of households without access ................................................................................................................... 139 Figure 186. Mobile-cellular subscriptions per 100 inhabitants based on ITU .......................................... 144 Figure 187. Active mobile-broadband subscriptions per 100 inhabitants based on ITU.......................... 144 Figure 188. Price of ICT baskets in percent of GNI per capita in 2020...................................................... 145 Figure 189. Usage of computers, internet, and ownership of mobile phones in 2016/2017 and 2019/2020, % ............................................................................................................................................ 146 Figure 190. Usage of computers, internet, and ownership of mobile phones in 2019/2020, % .............. 147 Figure 191. Reasons for not using internet among individuals 16 years+ in 2019/2020, % ..................... 148 Figure 192. Access to mobile phones among individuals 16 years+: owned or shared in 2019/2020, % 149 Figure 193. Distribution of phones by types among individuals 16 years+ in 2019/2020, % ................... 149 Figure 194. Access to mobile money registered accounts among individuals 16 years+ in 2019/20, %.. 150 Figure 195. Expenditure share on ICT services by consumption quintiles in 2019/20 ............................. 152 Figure 196. Relative welfare gains after improved competition, percent ................................................ 152 Figure 197. Effect of improved competition on national poverty (change in percentage points) ........... 153 Figure 198. Effect of improved competition on inequality as measured by the Gini index (change in Gini coefficient points) ..................................................................................................................................... 153 ix Figure A1. Community mobility trends in Uganda using Google mobility report and the 5-week period Jan 3–Feb 6, 2020 as a baseline .................................................................................................................. 60 Figure A2. Population structure of areas and regions in the UNHS before and during COVID-19 using original weights, % ...................................................................................................................................... 60 Figure A3. Population structure of areas and regions in the UNHS before and during COVID-19 using adjusted weights, % .................................................................................................................................... 60 Figure A4. Share of the population with decreased living standards during the last 12 months across rural and urban areas during 2016/17–2019/20, % ................................................................................... 62 Figure A5. Share of the population with decreased living standards during the last 12 months across regions in 2019/20 during 2016/17–2019/20, % ........................................................................................ 62 Figure A6. Share of the population with income being very unstable during the last 12 months across rural and urban areas during 2016/17–2019/20, % ................................................................................... 63 Figure A7. Share of the population with income being very unstable during the last 12 months across regions in 2019/20 during 2016/17–2019/20, % ........................................................................................ 63 Figure A8. Share of dry months using SPEI across regions during 2016–2020, % ...................................... 64 Figure A9. Growth rates in the intensity of night lights across regions during 2016–2020, %................... 64 Figure A10. Predicted consumption distribution among Ugandans ......................................................... 123 List of tables Table 1. Sociodemographic characteristics of Ugandan households, 2019/20 .......................................... 39 Table 2. Marginal effects after probit regressions...................................................................................... 48 Table 3. Key household characteristics across different areas and regions in 2019/20............................. 50 Table 4. Migration by region of origin among households whose household head was an internal migrant at some point in the five years prior up to 2016/17, % .............................................................................. 83 Table 5. Migration by destination among households whose household head was an internal migrant at some point in the five years prior up to 2016/17, % .................................................................................. 84 Table 6. Characteristics of households depending on internal migration status during the five-year period up to 2016/17 ............................................................................................................................................. 84 Table 7. Multilevel mixed-effects logistic regression for a probability of having a head of household who migrated internally at some point in the five-year period up to 2016/17, marginal effects ..................... 86 Table 8. List of opportunities .................................................................................................................... 128 Table 9. List of circumstances ................................................................................................................... 130 Table A.1 Descriptive statistics for the variables selected for the consumption model by areas and across consumption per adult equivalent quintiles ............................................................................................. 121 x Table A.2 Regression results: estimates of consumption equation (log consumption per capita) .......... 121 Table A.3 Coverage of human opportunities across areas, regions, and consumption quintiles in 2012/13 .................................................................................................................................................................. 142 Table A.4 Coverage of human opportunities across areas, regions, and consumption quintiles in 2016/17 .................................................................................................................................................................. 142 Table A.5 Coverage of human opportunities across areas, regions, and consumption quintiles in 2019/20 .................................................................................................................................................................. 143 List of boxes Box 1. Previous poverty assessments and strategic documents ................................................................ 30 Box 2. Welfare trends according to the Uganda National Panel Survey in 2015/16 and 2019/20 ............ 37 Box 3. Determinants of poverty transitions based on UNPS 2015/16 and UNPS 2019/20 ........................ 47 Box 4. Can Uganda’s oil resources contribute to poverty reduction? ........................................................ 61 Box 5. Main data, concepts and definitions used in the Poverty Assessment ........................................... 65 Box 6. Economic impacts of COVID-19 on refugees ................................................................................... 79 Box 7. Impact of weather on the economic wellbeing of households in Uganda .................................... 101 Box 8. Social assistance in Uganda............................................................................................................ 104 Box 9. Shocks among refugees in Uganda ................................................................................................ 109 Box 10. Explaining the intuition of HOI using a simple example .............................................................. 129 xi Acknowledgment This study was prepared by a team led by Aziz Atamanov (Senior Economist) and Nistha Sinha (Senior Economist). The contributing authors are listed in alphabetical order: Aziz Atamanov, Eduardo Alonso Malasquez Carbonel, Takaaki Masaki, Cara Ann Myers, Rogelio Granguillhome Ochoa and Nistha Sinha. The team is grateful for guidance and support from Keith E. Hansen (Country Director), Rosemary Mukami Kariuki (Country Manager), Pierella Paci (Practice Manager), Allen Dennis (Program Leader) and Nobuo Yoshida (Lead Economist). The team wants to acknowledge constant support and collaboration from Uganda government officials, in particular, from the Uganda Bureau of Statistics. The team has benefitted from excellent comments provided by peer-reviewers Carlos Rodriguez Castelan (Lead Economist, EAWPV), Ruth Hill (Lead Economist, EPVGE), Carolina Mejia-Mantilla (Senior Economist, ELCPV) and Dhiraj Sharma (Senior Economist, EAEPV). Useful comments and suggestions were also provided by Paul Corral (Senior Economist), Talip Kilic (Senior Economist), Eva Liselotte Lescrauwaet (Senior Operations Officer), Franklin Mutahakana (Senior Operations Officer), Fatima Naqvi (Senior Social Protection Specialist), Giulia Ponzini (Economist), Benjamin Christopher Reese (Senior Operations Officer), Rachel K. Sebudde (Senior Economist), and Richard Walker (Senior Economist). We are very thankful to Rural Livelihoods Information System team from FAO for providing harmonized income data for the Uganda National Panel Survey. Excellent administrative and operational support was provided by Esther Ampumuza (Team Assistant), Martin Buchara (Program Assistant), Clare Busingye (Senior Executive Assistant), Santosh Kumar Sahoo (Program Assistant), and Tsehaynesh H. Michael Seltan (Program Assistant). Finally, the team thanks Virginia Larby for proof-reading and editing the report. Acronyms and Abbreviations CIAT International Center for Tropical Agriculture CIDR CUNY Institute for Demographic Research CIESIN Center For International Earth Science Information Network CUNY City University Of New York DCRM Displacement Crisis Response Mechanism ECA Europe And Central Asia EM-DAT Emergency Events Database FAO Food And Agriculture Organization GDP Gross Domestic Product GNI Gross National Income GRUMP Global Rural-Urban Mapping Project GSMA Groupe Special Mobile Association HCI Human Capital Index HFPS High-Frequency Phone Survey HOI Human Opportunity Index IBRD International Bank for Reconstruction and Development ICT Information Communications Technology IDA International Development Association IFPRI International Food Policy Research Institute ILO The International Labor Organization ITU International Telecommunications Union LDC Least Developed Countries LFS Labor Force Survey LIC Low-Income Countries LIPW Labor-Intensive Public Works MAI Market Accessibility Index ND-GAIN Notre Dame Global Adaptation Initiative NDP National Development Plan NUSAF The Northern Uganda Social Action Fund PCO Partial Collusive Oligopoly Structure PDM Parish Development Model RHFPS Refugee High-Frequency Phone Survey RULIS Rural Livelihoods Information System SCG Senior Citizens Grant SPEI Standardized Precipitation Evapotranspiration Index UBOS Uganda Bureau of Statistics UGX Ugandan Shilling UNESCO United Nations Educational, Scientific, and Cultural Organization UNHCR United Nations High Commissioner for Refugees UNHS Uganda National Household Survey UNPS Uganda National Panel Survey VIIRS Visible Infrared Imaging Radiometer Suite VNL VIIRS Nighttime Lights WB World Bank WDI World Development Indicators WELCOM Welfare And Competition Microsimulation ii Overview The share of Uganda’s population that lives below the poverty line has fluctuated over the last seven years, greatly influenced by shocks that have tested the resilience of the people. About 30 percent of the country’s population was poor in 2019/20, which is comparable to the poverty rate of 30.7 percent in 2012/13.1 The pattern of fluctuating poverty rates is largely driven by the experience of rural households. There was a surge in the poverty rate between 2012/13 and 2016/17 – linked to the drought in 2016/17 – followed by improvement in 2019/20 prior to the pandemic, when favorable weather conditions helped lift rural incomes. The COVID-19 pandemic pushed both urban and rural residents into poverty. Inequality, which reflects the extent to which different population groups benefit from Gross Domestic Product (GDP) growth, and affects the transmission of growth into poverty reduction, remained largely unchanged over this period and may even have worsened in urban areas. The findings of this report show that previously identified patterns and drivers of Uganda’s pover ty changes persisted well into 2020 – shaped by low productivity and high vulnerability. Since 2012/13 there were emerging signs of structural change with workers moving from low to high productivity activities. Workers moved out of subsistence agriculture into paid work, and the share of workers engaged in the services sector has increased, which augurs well for income growth. Nevertheless, agricultural productivity was falling until recent years, and its increase after 2017 was mostly associated with good weather rather than improvement in production practices. The progress in structural change was negatively affected by COVID-19 pandemic when many people returned to agriculture following the job losses and closure of small businesses. Identified inequality of economic opportunities and unequal accumulation of the human capital could hold back structural change in employment. The ability to change one’s economic sector of work appears to be unequal with the shift out of agriculture mainly taking place amongst men, older individuals, those with at least some level of formal education, and those from more well-off households. Unequal access to opportunities and unequal access to basic public services among young population identified in this report may therefore hold back the productivity and income gains that could come from structural change in employment sectors. Income generation strategies of households are also impacted by their resilience capabilities as the reported frequency of extreme weather shocks has increased in recent years. The pandemic further tested these capabilities. Accelerating poverty reduction in such a setting requires a two-pronged strategy. While at the macroeconomic level, policies addressing growth fundamentals are important for reducing poverty, from a microeconomic perspective, the report’s analysis shows that two strategies will be crucial. The first strategy is to lift the productivity and incomes of poor households in both rural and urban areas. While tackling agricultural productivity and job creation are at the top of the agenda here, making mobile phone services more widely accessible and affordable is a potential opportunity. The second strategy is to strengthen people’s resilience to shocks, particularly in rural areas. To have an impact, policies in both these areas will have to address the inequality in opportunities analyzed in the report. 1 Household budget surveys in Uganda span across two years. Detailed information about data used is provided in Annex 1. iii The rest of this overview presents key findings of the report. The next section synthesizes key facts about Uganda’s poverty reduction experience up to 2020. These facts set the stage for the section that follows examining reasons behind limited progress in poverty reduction. The final section reviews the key policy points for action. The report’s analysis is based on new analysis of available data sources as well as published analytical reports such as the Systematic Country Diagnostic Update (World Bank; International Finance Corporation; Multilateral Investment Guarantee Agency 2021), the Country Economic Memorandum (World Bank 2022), and the previous Poverty Assessment (World Bank 2016). iv I. Poverty and Inequality: Facts2 Multiple shocks in 2012–2020 led to volatile poverty rates Most recently, the COVID-19 shock wiped out previous reductions in rural poverty rates and accelerated the increase in urban poverty rates. Disaggregation of the 2019/20 household survey between the pre- COVID-19 and COVID-19 periods, sheds some light on trends right before the pandemic, revealing a drop in rural poverty of six percentage points and very little change in urban poverty between 2016/17 and the pre-COVID period in 2019/20. However, poverty rates soared in the COVID-19 period of 2020, increasing by 7.2 percentage points in rural areas (wiping out previous gains), and by 3.1 percentage points in urban areas (Figure O1). In addition, the COVID-19 shock aggravated both the depth and severity of poverty, meaning that more money will be needed to lift the poor out of poverty and that inequality among the poor increased dramatically as well. Figure O1. Headcount poverty rates across rural and Figure O2. Gini coefficient across rural and urban areas urban areas during 2012/13–2019/20 (before and during 2012/13–2019/20 (before and during COVID-19) during COVID-19), % Source: UNHS, WB staff calculations. Note: A detailed explanation of poverty and inequality concepts and the data used to measure them is provided in the Annex 1. Adjusted weights were employed for the analysis using the pre-COVID and COVID sub-samples as explained in Chapter 1. At the national level, inequality in consumption did not change much between 2016/17 and 2019/20, however there were substantial changes within urban and rural areas. Inequality in consumption, measured by the Gini coefficient, has barely changed since 2016/17 in Uganda (Figure O2). However, inequality in urban areas has increased, mainly in the COVID-19 period, while rural inequality decreased significantly in the same period. The narrowing of rural inequality may be a reflection of a contraction in consumption amongst the better off rural households. Most of the change in poverty rates was due to average growth in household consumption rather than favorable shifts in the distribution of this growth. In rural areas between 2016/17 and the pre-COVID-19 period in 2019/20, the increase in the mean average household consumption (growth effect) accounted for more than 90 percent of all poverty reduction during the period. Meanwhile, the contribution of reduction in inequality (the distribution effect) was limited. This was reversed in 2019/20 between the pre-COVID and the COVID-19 period, when poverty increased in rural areas solely due to lower average household consumption (negative growth effect), and the reduction in inequality curbed the increase in 2 For more detailed analysis check the Chapter 1. Most recent poverty and inequality trends. v poverty. The agricultural sector – where the majority of the poor were working – was affected by COVID- 19-related mobility restrictions to a lesser extent after the first lockdown, which can explain the poverty reducing distribution impact in rural areas. In urban areas, both growth and inequality effects contributed to the increase in poverty observed in 2019/20 during both the pre-COVID-19 and COVID-19 periods. Every third individual experienced transient poverty in the last five years, while every fifth remained chronically poor Poverty is a mix of persistent and transient – another reflection of the limited resilience of households. Figure O3. Poverty transitions between 2015/16 and 2019/20 by head Overall, about 20 percent of the of household characteristics in 2015/16 year, % population was persistently poor over the 2015/16 to 2019/20 period and about half of the population was never poor. The remaining 30 percent of the population experienced transient poverty: 18 percent of those who were poor in 2015/16 had moved out of poverty in 2019/20, while the remaining 12 percent moved from being non-poor in 2015/16 to being poor in 2019/20. The largest shares of persistent poor were observed among heads of household without formal Source: UNPS, WB staff calculations. education and who were either unemployed or out of the labor force in 2015/16 (Figure O3). The Northern and Eastern regions continued having the highest poverty rates, as well as the biggest gaps in human capital outcomes and access to basic infrastructure Location plays a key role in poverty status in Uganda, with rural areas and the Northern and Eastern regions exhibiting the highest poverty rates. In 2019/20, the poverty rate in rural areas was 33.8 percent – much higher than the urban rate of 19.8 percent (Figure O4). Indeed, more than 80 percent of all the poor in Uganda live in rural areas. Poverty rates also varied substantially across regions. Poverty in the Central region was about 15 percent, or almost three times lower than the poverty rate in the Eastern and Northern regions, which reached 42 percent and 40 percent, respectively. In particular, within the Eastern and Central regions, rural households were also poorer compared to urban households. Subregions also had stark differences in poverty rates; the Kampala subregion had the lowest poverty rate of just four percent, which was 20 times lower than that of the Karamoja subregion. vi Figure O4. Headcount poverty rates in 2019/20 across areas, Subregional spatial disparities in poverty are regions and subregions, % also reflected in gaps in human capital outcomes and access to basic infrastructure, which translates into low growth potential. Subregions with higher poverty rates also have lower Human Capital indices (HCI),3 (see Figure O5) and lower levels of infrastructure access, including access to electricity (Figure O6). In contrast, the combination of high population density, better access to infrastructure and markets, and higher human capital indices seem to contribute to subregional economic development, measured by night-time light (NTL). Thus, Kampala and its surrounding areas stand out in terms of night-time light and economic development (Figure O7 and Figure Source: UNHS, WB staff calculations. O8). Figure O5. Poverty rate in 2016/17 and Human Capital Figure O6. Poverty rate in 2016/17 and access to Index circa 2016/17 by subregions electricity by subregions* Figure O7. Population density in Uganda in 2020** Figure O8. Night lights in Uganda in 2020*** Kampala Kampala Sources: *World Bank 2020a; UNHS 2016/17, World Bank staff calculations. ** WorldPop. *** Elvidge et al. 2021). 3 Human Capital Index encompasses stunting, mortality rates, years of schooling, and test scores in one indicator. vii Working in agriculture and lack of education are the strongest predictors of high poverty Households working in agriculture Figure O9. Headcount poverty rates in 2019/20, depending on were the most likely to be poor in head of household sector of employment, % Uganda. According to Figure O9, households whose head worked in agriculture had the highest headcount poverty rate (35.5 percent), which was seven percentage points higher than the average poverty rate among population from households with employed heads (28.7 percent). Put another way, 77 percent of poor households worked in agriculture. Source: UNHS, WB staff calculations. Figure O10. Headcount poverty rates depending on The education level of the head of household was head of household education level in 2019/20, % another determinant of poverty and consumption levels in 2019/20. Across both rural and urban areas, there was a strong negative association between the level of education and poverty rates (Figure O10) – with much lower poverty rates for households whose head had a higher education level. For example, the poverty rate among households with uneducated heads reached about 48 percent in 2019/20 (this group accounted for 17 percent of all heads). Source: UNHS 2019/20, WB staff calculations. This was almost twice as high as the poverty rate of households in which the household head finished primary education – 25.7 percent (also 17 percent of all heads). Even having primary incomplete education, after controlling for other household characteristics potentially affecting wellbeing, increases consumption by 18 percent compared to households where the head does not have any education at all. Importantly, spatial decomposition analysis also found that differences in education levels among heads of household were one of the biggest endowment factors accounting for the urban-rural consumption gap. Households with more children, larger household size, and with a single female income earner were more likely to have higher poverty rates Demography also plays an important role in poverty status, which is especially relevant given Uganda’s high population growth rates and burgeoning young population. Poor households were significantly larger with more children. Whereas the average household size among non-poor households in 2019/20 was 4.3 members, poor households averaged 5.5 members. In 2019/20, the headcount poverty rate in viii households with two or three children in 2019/20 was about 34 percent – 10 percentage points higher than among households with one or two children (Figure O11). Once controlled for other factors, having one additional child (ages 0–13 years) was associated with a reduction in consumption per individual by three percent. Gender is also an important correlate of poverty. Female-headed households in rural areas, female- headed divorced and married households had higher poverty rates than their male counterparts (Figure O12). Furthermore, analysis of poverty rates by income earner composition showed a significantly higher poverty rate among households with one female earner in paid employment compared to households with one male earner in paid employment, potentially pointing to gender inequalities in the labor market. Figure O11. Headcount poverty rates in households Figure O12. Headcount poverty rates depending on head with different number of children in 2019/20, % of household marital and gender status in 2019/20, % Source: UNHS 2019/20, WB staff calculations. Note: *The difference between female and male heads of household is statistically significant. II. Why has there been limited progress in poverty reduction? Rural households and the poor experienced more frequent shocks and have limited safety nets4 Experiencing shocks more frequently need not push vulnerable households into poverty; however, access to safety nets and other means to smooth incomes is limited. According to the UNPS data, the share of households that experienced shocks5 during the last decade before COVID-19 ranged from 30 percent to 40 percent in Uganda, with rural and the poorest households affected the most (Figure O13). These shocks frequently resulted in a decline in income and assets (Figure O14). To cope with the impact of shocks, households might employ a variety of strategies, including using savings, soliciting help from relatives and neighbors, and reducing consumption. The poor in Uganda were much more likely to use coping mechanisms such as an involuntary decline in food consumption (Figure O15). Furthermore, the 4 For more detailed analysis check the section I. Shocks and coping strategies in Chapter 3. 5 Respondents were asked if they experienced any shock during last 12 months from the wide list of distress events such as droughts, irregular rains, flooding, death of income earner(s) and so forth with a possibility to select shock not in the list. ix UNPS data showed correlation between households who experienced shocks and the likelihood of being chronically poor or falling into poverty (Figure O16). Figure O13. Share of households that experienced at Figure O14. Shocks and decline in different least one shock across survey years and rural and urban dimensions of wellbeing in 2019/20, % of households consumption quintiles, % reporting decline Source: UNPS 2019/20, WB staff calculations. Source: UNPS 2019/20, WB staff calculations. Figure O15. Distribution of coping strategies during the Figure O16. Share of population with at least one last 12 months in 2019/20 across residence, head of shock during last 12 month across poverty household gender and rural/urban consumption transition status during 2015/16–2019/20, % quintiles, % Source: UNPS 2019/20, World Bank staff calculations. Source: UNPS 2015/16 and 2019/20, WB staff Note: All coping strategies are taken into account regardless of calculations. their rank. Note: The same households were considered during two rounds. The majority of Ugandan households have experienced at least one shock during the COVID-19 pandemic, especially the poor and those living in the Northern region, who also used inferior coping x mechanisms. The Uganda High-Frequency Phone Survey (HFPS) found that almost 60 percent of households in the first round experienced at least one shock after March 2020 (Figure O17).6 Figure O17. Incidence of shocks among Ugandans The incidence of shocks was higher among the during March–June 2020, % poorest households from the first pre-COVID-19 consumption per adult equivalent quintile (65 percent) and those living in the Northern region (69 percent).7 Similar to the pre-COVID-19 period, the poorest households were less likely to rely on savings as a coping strategy and more likely to reduce food consumption when they experienced a shock (Figure O18).8 Shocks such as increased food prices, business failure and falling output prices had the highest incidence. Source: HFPS round 1, World Bank staff calculations. Figure O18. Type of coping strategies used by households in Uganda during March –June 2020 across different dimensions, % of all strategies Source: HFPS round 1, WB staff calculations. Note: All coping strategies used are included. Despite reasonable accuracy in targeting and effectiveness for the program beneficiaries, the current levels of expenditure on social protection in Uganda are low. According to the World Bank (2020a), the allocation to social development – which includes social protection expenditures – was only 0.7 percent 6 The incidence of shocks is not comparable across pre-COVID-19 UNPS data and HFPS. Both surveys used a different list of shocks. 7 According to the Uganda Refugee High-Frequency Phone Survey (RHFPS), almost 90 percent of refugee households experienced at least one shock between March and October/November 2020. In contrast to Ugandans, refugees did not have access to savings to cope with shocks but relied mostly on aid and help from friends and relatives. 8 One important caveat here is that the recall period in the UHFPS is much shorter than in the UNPS and this may affect coping strategies as well. xi of the overall government budget in FY19/20. Expenditure on the two largest direct income support programs, such as the Senior Citizens Grant (SCG) and The Northern Uganda Social Action Fund (NUSAF 3), was just 0.14 percent of GDP in FY17/18, which is lower than in neighboring countries like Kenya and Rwanda who spent 0.4 percent and 0.3 percent of GDP respectively, on direct income support. The most recent data confirms very low coverage of SCG and NUSAF programs, but also demonstrates their pro-poor incidence. As shown in Figure O19, about nine percent of all individuals aged 60 years and above benefited from SCG, and the coverage rate was twice as high among the poorest individuals from the bottom consumption quintiles. About 0.6 percent of individuals aged 15 years and above benefited from NUSAF and, similar to SCG, the coverage rate was higher among the poorest individuals (Figure O20). Figure O19. Incidence of beneficiaries of SCG in Figure O20. Incidence of beneficiaries of NUSAF in 2019/20 among individuals 60 years+ by consumption 2019/20 among individuals 15 years+ by consumption quintiles, % quintiles, % Source: UNHS 2019/20, WB staff calculations. Half of the population was vulnerable to poverty in 2019/20 – education, asset ownership, and working status being key determinants9 The high frequency of shocks and their correlation with poverty is significant because half of the population in Uganda was vulnerable to poverty in 2019/20. The vulnerability to poverty rate in Uganda was estimated to be 50 percent as of 2019/20 (Figure O21). This is much higher than the observed poverty rate of 30 percent in the same year. Rural areas are, as expected, characterized by both higher poverty and vulnerability rates (34 percent and 59 percent, respectively) compared to urban areas (20 percent and 26 percent, respectively). Among regions, the highest vulnerability rates were found in the poorest Northern and Eastern regions (66 percent and 64 percent, respectively). In relative terms, however, the largest difference between poverty and vulnerability rates was in the Western region, where the vulnerability rate was twice as high as the poverty rate. Risk-induced vulnerability prevailed in urban areas, while poverty-induced vulnerability prevailed in the poorest Eastern and Northern regions.10 Figure O22 shows that, at the national level, risk-induced 9 For more detailed analysis check the section II. Quantifying vulnerability to poverty in Chapter 3. 10 Vulnerability can be driven either by permanent low consumption (poverty-induced) or high volatility of consumption (risk- induced). Poverty-induced vulnerability happens when the expected mean of consumption already lies below the poverty line. Households face risk-induced vulnerability when their expected consumption is higher than the poverty line, but high estimated variance leads to a probability higher than the established threshold of 29 percent for a given year. xii vulnerability slightly prevails over poverty-induced vulnerability. The difference widens in urban areas, where high consumption volatility accounts for almost 70 percent of all vulnerability. By contrast, in rural areas vulnerability is equally driven by low consumption and its high volatility. At the regional level, vulnerability in the poorest Eastern and Northern regions is poverty-induced to a larger extent, while in the Central and Western regions vulnerability was more associated with high consumption fluctuations. Figure O21. Poverty and vulnerability rates in Uganda Figure O22. Vulnerability decomposition into poverty- in 2019/20 across rural and urban areas, and regions, and risk-induced components across rural and urban % areas, and regions, % Source: UNHS 2019/20, authors’ calculations. While idiosyncratic shocks are more prevalent on the national level, covariate shocks, such as weather and price shocks, are more important for rural residents.11 The ratios between the percentage of households that would fall below the poverty line from an idiosyncratic shock versus a covariate shock are higher than one at the country level, rural and urban areas and different regions. This shows that the impact of idiosyncratic shocks was consistently higher than the impact of covariate shocks. In relative terms, however, the role of idiosyncratic shocks is much more prevalent in urban areas and the Central and Western regions, compared to rural areas and the Northern and Eastern regions. This suggests that covariate shocks such as weather, locust, and price shocks play a more important role for rural residents. Education, asset ownership, and working status were found to be the main determinants of vulnerability in Uganda. For households where the household head has an incomplete primary education the probability of being vulnerable drops by 15 percent relative to households where the head has no education. Ownership of selected assets such as TVs and motorcycles, used as a proxy of monetary wellbeing, were also associated with significantly lower chances of being vulnerable. Having the head of household out of the labor force or in subsistence farming increases the probability of being vulnerable compared to having the head of household in paid employment (Figure O23). 11Idiosyncratic shocks include household specific shocks such as health issues, job losses and so forth, which have a weak correlation across households living in the same community. In contrast, covariate shocks are correlated across households within communities or, in other words, households from the same community experience similar shocks. These may include price and weather shocks, political crises, and so forth. xiii Figure O23. Changes in probability of being vulnerable to poverty with regards to non-vulnerable by different household characteristics Source: UNHS 2019/20, authors’ calculations. Note: Base category is in parentheses. The change from subsistence agriculture to paid employment was at a higher pace among the wealthier and more educated12 Figure O24. Annualized growth rates of employment There was a noteworthy decline in the share of share in agriculture and GDP per capita, % agricultural employment in Uganda between 2011/12 and 2018/19; however, this well- known marker of structural change was not accompanied by a robust GDP per capita growth as in most countries with similar economic development.13 The rate of decline of the share of agricultural employment in Uganda was faster than in most comparators – about two percent annually between 2011/12 and 2018/19 (Figure O24). Nevertheless, Uganda’s economic growth lagged behind, with the annualized growth rate of GDP per capita close to only one percent. Source: LFS and WDI (modelled estimates). World Bank staff calculations. 12 For more detailed analysis check Chapter 2. Most recent trends in the labor market and structural change. 13 In order to compare the performance of Uganda in regard to structural change, we have selected countries with GDP per capita measured in constant international dollars in 2017 purchasing power parity similar to Uganda in 2011. xiv Figure O25. Value added in agriculture in per capita terms in Agricultural productivity was falling before UGX constant prices and the share of dry months based on 2017 when it started to grow due to SPEI index resumed economic growth and labor moving to services and industry sector; with performance impacted by weather shocks. At the same time, productivity in services and industry sectors, which absorbed excessive labor in agriculture, continued to decline across all years. In addition, given the importance of the agricultural sector in the economy, weather shocks affected Uganda’s economic performance in this period. There was a significant correlation in the country between the value added in agriculture in per capita terms and the precipitation Source: Global SPEI database accessed in Oct. 2021 and UBOS. during the last ten years (Figure O25). Thus, Note: Value added was taken for fiscal year to account for lagged impact of weather. For example, the observation for the year of 2006 higher value added in agriculture happened was based on value added from 2006/2007, while SPEI index was in recent years with lower shares of dry based on 2006 year. months. A notable feature of structural change is that at this early stage it is also unequal – sectoral change in employment is mainly observed among men, older individuals, and those with at least some level of formal education. Overall, rural residents, females, individuals of 14–24 years of age, and those without education were less likely to have paid employment as their primary job in 2011/12 (Figure O26). The structural change during the 2011/12 to 2018/19 period – measured by the falling share of subsistence agriculture and the increasing share of paid employment – was also pronounced among groups that already had higher levels of paid employment, with one notable exception of rural areas. Wealthier individuals experienced faster rates of sectoral change in employment during the period from 2013/14 to 2019/20. As shown in Figure O27, the share of employment in agriculture among the working population was declining across all quintiles from 2013/14 to 2019/20, but the rate of decline was higher among individuals from the wealthier quintiles. This may be related to constraints that the poorest face trying to enter the non-agriculture sectors, such as low levels of human capital, limited access to infrastructure and markets, and the up-front costs of transitioning from one sector to another. xv Figure O26. Share of paid employment across different Figure O27. Share of working age individuals groups of population using LFS 14–64, % employed in agriculture by consumption quintiles using UNPS 14–64, % Source: LFS 2011/12 and 2018/19, World Bank staff Source: UNPS, World Bank staff calculations. calculations. Note: Cross-section weights are used. Internal migration was not widely accessible for the poorest households living in lagging areas14 Internal migration may not have been widely accessible for the poorest households living in lagging agricultural regions, constraining their abilities to access better opportunities. About nine percent of household heads in Uganda in 2016/17 reported migrating at some point during the last five years. Rural to urban migration accounted for the smallest share in overall internal migration flows in 2016/17 (16 percent). Results from regression analysis revealed that, after controlling for other factors, households with less educated heads, households with more dependents, and households living in areas with predominantly subsistence agriculture and inadequate infrastructure were less likely to migrate. On average, those who managed to move out of agriculture into other sectors – as well as those who migrated from rural to urban areas – benefited from gains in individual consumption between 2015/16 and 2019/20.15 Households in which the household head switched from the agricultural sector in 2015/16 to the non-agricultural sector in 2019/20 demonstrated the highest annualized consumption growth rate (Figure O28). With regards to internal migration, consumption grew in all groups during 2015/16 and 2019/20, but the annual growth rate was the same among those households who remained in rural areas and those who moved from rural to urban areas (Figure O29). 14 For more detailed analysis check the section IV. Internal migration in Chapter 2. 15 A panel component from the UNPS 2015/16 and 2019/20 rounds was used to check consumption per adult equivalent changes in households depending on the household head’s sector of employment and internal migration status. xvi Figure O28. Individual consumption and its annual Figure O29. Individual consumption and its annualized growth rate across household heads’ sector of growth rates across household heads’ migration status employment in 2015/16 and 2019/20, UGX and % in 2015/16 and 2019/20, UGX and % Source: UNPS 2015/16 and 2019/20, World Bank staff calculations. Note: The same households across years are used. Note: N=2,673 households. COVID-19 slowed down the pace of structural change and increased vulnerability as many were pushed into agriculture COVID-19 may have slowed down the pace of structural change in Uganda, reversing momentum and pushing some individuals into the agricultural sector where they continue to be vulnerable to weather shocks. The share of agriculture in employment increased after the first lockdown when some individuals stopped working and some shifted to working in agriculture (Figure O30). This negatively affected the pace of structural change. The latest round of the HFPS – conducted after the second lockdown – showed a different pattern. Work stoppages were more universally distributed across areas and sectors and less related to COVID-19 restrictions (Figure O31). This may be related to droughts in most parts of the country in the first agricultural season of 2021. Figure O30. Working respondents using different Figure O31. Work stoppages among those who rounds of the HFPS, % of respondents worked in the previous round across rural/urban areas and economic sectors, % Source: HFPS, World Bank staff calculations. Note: Only the same respondents across all seven rounds are kept. xvii Figure O32. Share of working respondents among refugees Work stoppages after COVID-19 were more and Ugandans before COVID-19 and during the first year of pronounced among refugees than Ugandans, the pandemic, % with slower income and employment recovery. For refugees, the share of working respondents dropped from 56 percent before COVID-19 to 36 percent in October/November 2020, and further to 32 percent in February/March 2021 (Figure O32). While work stoppages were also pronounced among Ugandans, employment fully returned to the pre-pandemic level after June 2020. In addition, lower shares of refugee households reported income levels in February/March 2021 being the same or higher than pre-COVID compared Source: RHFPS and HFPS, WB staff calculations. to numbers reported by Ugandans in February 2021. Agricultural development, which is the key for poverty reduction, has not experienced substantial changes in production practices in recent years16 Agricultural growth has largely been dependent on favorable weather as opposed to substantial changes in production practices. Poverty reduction depends on improved roved agricultural performance as the majority of poor households engage in subsistence agriculture and do not necessarily have access to non-farm opportunities. The median value of crop production per hectare among households engaged in growing crops increasing by eleven percent annually between 2013/14 and 2019/20.17 However, this agricultural growth does not seem to be accompanied by commensurate changes in production practices and the use of improved inputs. For example, the share of households who used organic fertilizers increased from 11 to 14 percent, pesticides from 14 to 17 percent, and the use of inorganic fertilizers remained at about seven percent (Figure O33). Use of fertilizers remained unequal. Getting advice from extension services and using improved seeds fell between 2013/14 and 2019/20 (Figure O34). These findings suggest that positive external conditions, most notably favorable weather patterns, may have played the chief role in contributing to the sector’s robust growth. In particular, the years of 2018 and 2019 were considered to have very favorable weather conditions for agriculture. On the flip side, this also signals the sector’s vulnerability to poor weather conditions, such as the 2016/17 drought that caused a decline in agricultural incomes and an increase in the poverty rate. 16 For more detailed analysis check the section IV. Agriculture in Chapter 2. 17 To check the latest performance of the agricultural sector and the underlying factors behind agricultural production, we have used Rural Livelihood Information System (RULIS) harmonized datasets constructed from the UNPS for 2013/14 and 2019/20 years. xviii Figure O33. Share of households in planting Figure O34. Share of households in planting activities who activities who used fertilizers at least once on any used improved seeds on any land plot during either of the land plot during any of the agricultural seasons in two agricultural seasons and who received any advice 2013/14 and 2019/20, % from extension services in 2013/14 and 2019/20, % Source: UNPS 2013/14 and 2019/20, World Bank staff calculations. Inequality of opportunities in access to basic services persisted and was exacerbated by the COVID- 19 pandemic18 The Human Opportunity Index (HOI) was used to measure inequality of opportunities among children in Uganda. The HOI explores how the personal “circumstances” for which a child cannot be held accountable – like location, gender, household composition, or parental wealth – can affect the child’s probability of accessing basic services that are necessary to succeed in life, like timely education, clean water, electricity, and decent housing. In other words, the HOI measures access rates (coverage) to basic services (opportunities) adjusted by inequality and allows to measure progress toward universal equitable access to opportunities. The HOI also indicates which socio-demographic characteristics influence a child’s likelihood of access to a particular opportunity and identifies those with the largest contributions to inequality. The report focused the analysis on opportunities in education, access to basic services, and health. For education, inequality was found to be higher in opportunities that captured the quality of services and was largely explained by differences in location and monetary wellbeing. Figure O35 shows the coverage rates (dots) and HOI (bars) with the gap between the two reflecting the relative inequality of opportunity among children of different circumstances. None of the education opportunities had universal 100 percent coverage, which can be viewed as an aspirational goal for a society. Opportunities related to the quality of education such as starting and finishing school on time had the lowest coverage and highest inequality. If inequality is split by circumstances, location explained more than 40 percent of all inequality in school enrollment for primary age children and finishing school in time, with monetary wellbeing being the second largest contributor. Inequality in starting school on time was driven by monetary wellbeing and head of household education levels. Visits to health facilities and access to medication when needed were accessible for only 67 percent of children, with significant inequality across groups, especially regarding monetary wellbeing. As shown 18 For more detailed analysis check Chapter 4. Inequality of opportunities. xix in Figure O35, the HOI in access to health services was close to 61 percent, which was 6 percentage points lower than the average coverage rate of 67 percent. The main contributing factor for inequality in access to health services was monetary wellbeing, measured by consumption per capita quintiles. It accounted for more than 55 percent of the inequality observed, compared to only 17 percent related to location. Only 48 percent of the poorest children from the bottom quintile were able to visit health facilities and access medication when ill, compared to 91 percent of children from the richest quintile. Figure O35. Coverage rate, HOI and inequality for Figure O36. Coverage rate, HOI and inequality for access to education and health among children in access to basic services among children in 2019/20, % 2019/20, % Source: UNHS 2019/20, World Bank staff calculations. Note: Age of children is shown in brackets. Access to all infrastructure services in Uganda except drinking water was very limited and very unequal. For all services, except drinking water, the coverage rate in 2019/20 was very low (Figure O36). Access to electricity improves substantially if electricity off grid (mostly solar) is counted in, but the coverage was still below 60 percent. Access to sanitation and hand washing facilities was the lowest. The most unequal access was observed for the opportunity to receive electricity from the national grid. Getting access to electricity off-grid reduces inequality substantially. Figure O37. Decomposing trends in the HOI between Location, especially regional disparities, was 2012/13 and 2019/20, percentage points responsible for the largest chunk of inequality in access to drinking water (78 percent), access to sanitation (41 percent), access to electricity from the national grid (55 percent), and access to electricity on and off-grid (44 percent). Inequality in access to a hand washing facility was an exception, with monetary wellbeing being the most important contributor to inequality (45 percent) and the head of household’s education level and location accounting for equal, but much smaller shares Source: UNHS 2019/20, World Bank staff calculations. in inequality (25 percent each). xx Almost all of the improvements in the HOI indicators between 2012/13 and 2019/20 were due primarily to higher coverage rates for entire population without improving existing inequality, with access to off- grid electricity being the one noticeable exception where equitable distribution played a more important role. Figure O37 shows contribution of scale19 and distribution effects20 for the changes in 11 opportunities between 2012/13 and 2019/20. For most improved opportunities, distribution effect played a marginal role and any improvement observed was mostly due to a scale effect. The increase in the opportunity to access electricity (on- and off-grid) was the only one coming from both scale and distribution effects. Figure O38. School attendance and participation in any type of The COVID-19 pandemic has stalled the schooling for children ages 3–18 years in March 2020 (before progress Uganda had been making in lockdown) and in March/April 2021 by area, region and pre– improving human capital accumulation, COVID-19 consumption quintiles, % in particular education. During the period March 2020–October 2021, schools in Uganda were fully or partially closed for 83 weeks – the longest closure in the world (UNESCO). Participation in schooling declined dramatically (Figure O38): about 90 percent of children aged 6–18 years participated in schooling before the lockdown was introduced in March 2020, but by March/April 2021, this had dropped to only 46 percent of children participating in any learning activities. Source: HFPS, World Bank staff calculations. Note: Age range in the HFPS data was from 3–18 years and differs from the age range used for the HOI. Increasing competition in the telecommunications sector could increase affordability and access, and reduce poverty21 The adoption of mobile telecommunication services has the potential to improve individual welfare. This may happen through multiple channels, including better labor market outcomes, more variety and better quality of goods and services, lower prices for buyers, improved student learning, expanded financial inclusion, and increased access to information and new markets through mobile broadband internet (Aker & Mbiti 2010, Aker, Ksoll & Lybbert 2012). Despite these benefits, there is a gap between service availability (i.e., coverage) and usage of mobile telecommunication services, especially regarding the adoption of mobile internet (Maudi & Dubus 2020, Granguillhome Ochoa et al. 2022). In Uganda, affordability remains one of the main constraints for the country to realize the full potential of digital transformation and its associated welfare benefits. The Information Communication 19 A change in the overall coverage for the entire population without any changes in inequality. 20 A change in the equality of access to the opportunity between the circumstance groups. 21 For more detailed analysis check Chapter 5. The role of the telecommunications sector for poverty reduction. xxi Technology (ITC) sector is highly concentrated, which affects the level of competition for telecom operators and subsequently the prices they charge customers. According to the International Telecommunications Unit (ITU), which monitors affordability of selected standardized baskets of ICT measured in prices expressed as a percentage of Gross National Income (GNI) per capita, for four out of five standardized ICT baskets, the price in Uganda was higher than the median for African and least developed countries. While Uganda has relatively high levels of ICT penetration, access to mobile phones varies across groups, with the poorest individuals having lower ownership rates. Overall, 59 percent of individuals ages 16 years and above had access to mobile phones in 2019/20. However, access to mobile phones among the poorest consumption quintile was half the level of the top quintile: 36 percent versus 79 percent, accordingly (Figure O39). About 70 percent of all owned mobile phones were basic ones with text and calling functions only. The poorest individuals from the bottom consumption quintile were more likely to own basic phones compared to the richest individuals from the top quintile: 83 percent versus 58 percent, accordingly. As a direct result of increased competition in the telecommunications sector, the reduction in average ICT prices, and the entrance of new users of telecommunication services, poverty is expected to decline by about 0.8 percentage points, without significant changes in inequality. Simulation shows that improvement in competition is associated with an average total increase in individual consumption by 1.7 percent, with 0.9 percent coming from current users and 0.8 percent coming from new users. Among current users, the positive impact increases by consumption quintile, given their larger expenditures on ICT services and higher coverage rates. In contrast, the positive impact of improving competition for new users is pro-poor, with larger consumption growth observed among the population from the bottom wealth quintiles. Consistent with increased consumption, poverty is expected to fall by about 0.8 percentage points (Figure O40). Figure O39. Access to mobile phones among Figure O40. Effect of improved competition on Uganda’s individuals 16 years+ in 2019/20, % poverty rate, change in percentage points Source: UNHS, World Bank staff calculations. Source: UNHS 2019/20, World Bank staff calculations. Note: Poverty is measured using the updated poverty line, with the poverty rate of about 30 percent in 2019/20. xxii III. Way forward: A policy agenda Raise productivity and income earning opportunities Investing in human capital development is a critical pathway for poverty reduction in Uganda. In 2019/20, the education level of the head of household was found to be the key determinant of poverty and consumption levels, highlighting the imperative to invest in educating young population. Furthermore, policies and investments focusing on expanding access to basic services like health and education should also focus on factors that improve the quality of those services to deliver more significant gains in outcomes and ultimately reduce poverty. This is particularly important in the wake of the COVID-19 related school closures, during which 60 percent of children were not engaged in any learning activities. Ensuring that these students return to school and are able to recover the lost learning will be a critical part of the human capital recovery process. As access to public services gradually expands, targeting lagging regions can improve opportunities for the poorest. Location and monetary wellbeing are the most significant factors explaining the inequality in most human-capital related indicators for Ugandan children. Given the strong correlations between education level, poverty status and economic mobility, increasing opportunities for children in poorer and more vulnerable regions could be an important pathway for breaking the cycle of poverty. In addition, more attention will need to be placed on distribution to ensure that expansion is equitable and continues to increase opportunities for those who need it the most, including refugees. Economic transformation will require a fundamental shift in the nature of production – from low investment, informal activities – to higher-capital, more productive employment. In Uganda, the agricultural sector – in which the majority of the population currently works – has the lowest productivity level compared to services and industry. Structural change is integral to economic growth and poverty reduction in Uganda, involving the reallocation of labor from less productive (especially subsistence agriculture) to more productive sectors. In particular, accelerating wage job creation is key for faster economic transformation and poverty reduction. Households that transition out of agriculture into other sectors and those who migrate from rural to urban areas tend to benefit from increases in consumption, demonstrating the poverty reducing effects of structural change. Therefore, reducing the barriers and costs to migration could enable more vulnerable households and individuals in Uganda to take advantage of opportunities in other sectors and in more economically vibrant places. COVID-19 may have slowed down the pace of structural change in Uganda, reversing the momentum of economic transformation, and calling for policies to revitalize the industry and services sectors. The lockdowns and mobility restrictions imposed to limit the spread of COVID-19 had a disproportionally higher impact on urban areas and those working in the services sector. Therefore, restoring growth in the industry and service sectors will be an important part of the COVID-19 recovery strategy. In addition, providing targeted support to refugee and host community businesses and the self-employed can contribute to the mitigation of shocks while fostering job creation in refugee hosting districts. Finally, Uganda has identified ‘digital transformation’ as one of the key drivers of growth and increasing access to ICT can have poverty reducing benefits as well. The digital sector represents one of the fastest growing sectors in Uganda, with positive spillover effects on other sectors of the economy, the combination of which can play a key role in the post-COVID-19 recovery (World Bank 2020c). Providing xxiii affordable and ubiquitous telecommunication services is particularly important to promote technology- based empowerment, which is especially relevant for Uganda’s fast growing, young population. Currently, affordability remains one of the main constraints for Uganda to realize the full potential of digital transformation. However, simulations show that increased competition in the telecommunications sector – and the associated reduction in average ICT prices – as well as the entrance of new users can generate welfare gains with a limited impact on equality. Strengthen household resilience A key component of sustainable poverty and vulnerability reduction in Uganda is addressing the country’s high susceptibility to climate, health, and forced displacement-related shocks and the exposure of the poor and vulnerable population to these shocks. The significant fluctuations in Uganda’s poverty story over the last decade highlights the extent to which ongoing shocks have limited the sustained reduction of poverty in the country. A large portion of Ugandans are clustered around the poverty line and are therefore sensitive to sudden changes in external conditions. As such, mitigating the severity of shocks, increasing the coverage of social protection programs for Ugandans and refugees, and developing better coping mechanisms are essential for buffering against the negative impact of shocks. Expanding safety nets for both Ugandans and refugees is an important strategy to help the vulnerable face shocks without falling (deeper) into poverty. Efforts should be made to ensure that households do not have to resort to coping mechanisms that jeopardize their long-term wellbeing and future prospects, such as reducing food consumption and forgoing investments in the health and education of children. Given the limited fiscal space, it is critical to ensure that the sector’s resources, including the potential expansion of some of the existing programs, are channeled to the poor and vulnerable in the neediest geographical areas. This allocation should consider the potential for social protection programs to both address deficiencies in human capital and help mitigate the vulnerability of households to shocks. For example, the Disaster Risk Financing (DRF) pilot – under the Uganda Social Action Fund (NUSAF) 3 project – is an example of a successful adaptive social protection system that could be expanded under this framework. Furthermore, policies and programs should aim to foster inclusion and cohesion between refugees and host communities by maintaining public service provision and investments for both groups, such as the Displacement Crisis Response Mechanism (DCRM).22 Introducing the monitoring refugees in the national statistical systems is an important way to facilitate their inclusion. While social protection programs can play an important role, policies to promote insurance and savings schemes can also contribute to mitigating the negative impacts of shocks. Relying on savings is considered one of the more preferable coping mechanisms for dealing with a shock (as opposed to involuntary reduction in food, for example). Given that the vast majority of Uganda’s labor force (especially the poor and vulnerable) work in the informal sector, traditional social insurance schemes may not be viable and alternative schemes must be designed. For example, policies can provide fiscal incentives to improve the take-up of voluntary savings schemes by informal sector workers. Financing 22 The DCRM is a new framework to handle displacement-related shocks. It is a pre-planned and pre-financed mechanism for shock response. The process of disbursing resources for public service provision from the DCRM is agreed upon in advance. The government then selects indicators regarding public service provision and monitors them over time. If the indicators drop below threshold levels, the DCRM rapidly and automatically disburses resources. The public service investments financed by DCRM resources are pre-agreed and include the development of schools, water supplies and health care facilities. xxiv options for farmers should also be more accessible, including the development of insurance schemes for farmers to protect against climate change. That said, limited access and the lack of an enabling environment are key reasons why agriculture finance and insurance remain at sub-optimal levels in Uganda. Increasing the resilience of the agricultural sector is essential to address the higher poverty and vulnerability rates in rural areas. Almost 60 percent of rural Ugandans are vulnerable to poverty compared to just 26 percent of urban Ugandans. Furthermore, the Northern and Eastern regions, which depend mainly on agriculture, have particularly high vulnerability rates at almost two-thirds of the population. Importantly, while idiosyncratic shocks are more prevalent on the national level, covariate shocks such as weather and price shocks, are more important for rural residents who depend on agriculture. Therefore, increasing the resiliency of the agricultural sector could go a long way to reducing the overall levels of vulnerability for rural Ugandans. More broadly speaking, diversifying household incomes and increasing access to non-farm jobs are fundamental to improving resilience and reducing vulnerability to shocks. In order to achieve this, investing more in human capital accumulation, particularly education, is essential. Households with higher levels of education are less likely to work in agriculture, are less vulnerable to shocks, and have better coping mechanisms for dealing with shocks. Investments in education take time to pay off, and thus more immediate strategies should also be employed to help households better cope with shocks in the near term. Nevertheless, for Uganda’s long-term future, economic transformation and increasing education levels are vital for the country to successfully reduce poverty and vulnerability. xxv References Aker, J. C.; & Mbiti, I. M. (2010). "Mobile Phones and Economic Development in Africa." Journal of Economic Perspectives 24 (3): 207-232. doi:10.1257/jep.24.3.207. Aker, J. C.; Ksoll, C. & Lybbert, T. J. (2012). "Can Mobile Phones Improve Learning? Evidence from a Field Experiment in Niger." American Economic Journal: Applied Economics 4 (4): 94-120. doi:10.1257/app.4.4.94. Elvidge, C.D.; Zhizhin, M.; Ghosh T.; Hsu F.C.; & Taneja J. (2021). Annual time series of global VIIRS nighttime lights derived from monthly averages: 2012 to 2019. Remote Sensing 13(5), p.922, doi:10.3390/rs13050922. Granguillhome Ochoa, R.; Lach, S.; Masaki, T.; & Rodríguez-Castelán, C. (2022). "Mobile internet adoption in West Africa." Technology in Society 68: 101845. doi:doi.org/10.1016/j.techsoc.2021.101845. Hasbi, M. & Dubus, A. (2020). "Determinants of mobile broadband use in developing economies: Evidence from Sub-Saharan Africa." Telecommunications Policy 44 (5). doi:doi.org/10.1016/j.telpol.2020.101944. International Telecommunications Unit (2022). https://www.itu.int/itu-d/sites/statistics/. UNESCO (United Nations Educational, Scientific and Cultural Organisation). Dashboards on the Global Monitoring of School Closures Caused by the COVID-19 Pandemic. World Bank (2016). The Uganda Poverty Assessment Report 2016: Farms, Cities and Good Fortune - Assessing Poverty Reduction in Uganda from 2006 to 2013. World Bank, Washington, DC. World Bank (2020). Uganda Economic Update, 14th Edition, February 2020: Strengthening Social Protection to Reduce Vulnerability and Promote Inclusive Growth. World Bank, Washington, DC. World Bank (2022). Growth, trade, and transformation. Country Economic Memorandum for the Republic of Uganda. World Bank, Washington, DC. World Bank; International Finance Corporation; Multilateral Investment Guarantee Agency (2021). Uganda Systematic Country Diagnostic Update. World Bank, Washington, DC. World Bank and UNHCR. Uganda Refugee High-Frequency Phone Survey on COVID-19 (RHFPS). Uganda Bureau of Statistics. Uganda National Panel Survey (UNPS). Uganda Bureau of Statistics. Uganda National Household Survey (UNHS). Uganda Bureau of Statistics. Uganda High-Frequency Phone Survey on COVID-19 (HFPS). xxvi Chapter 1. Most recent poverty and inequality trends I. Background Uganda’s economic development lagged comparators during the last decade before the COVID-19 crisis. Uganda’s GDP was growing by about 5 percent on average during the last decade before the eruption of the COVID-19 pandemic in 2020. This rate of growth was the lowest compared to a selected set of countries in the region (Figure 41). At the same time, population growth accelerated from 3.2 percent during 2010–2014 to 3.6 percent during 2015–2019, while population growth rates among comparators were lower during the same period. As a result, the GDP per capita growth rate in Uganda during the last decade – and particularly during the 2015–2019 period – was lower compared to countries with similar economic backgrounds. It is projected that the population in Uganda will reach 59 million by 2030, with about one million youth entering the labor market annually (World Bank 2020a). To absorb new entrants to the labor market, Uganda will need to achieve and sustain a high level of economic growth.23 Figure 41. Annual percentage growth rates of GDP at market prices based on constant local currency and population growth rates in Uganda and comparators in 2010–2019, % Source: World Development Indicators (WDI), accessed on 9/27/2021. 23 Opportunities and challenges associated with potential usage of oil reserves are discussed in Box 4. 27 Note: Comparators were selected using Find my Friends tool and were also used in the Systematic Country Diagnostic and Systematic Country Diagnostics update (World Bank 2015; World Bank, International Finance Corporation, Multilateral Investment Guarantee Agency 2021) Economic growth has been particularly volatile since 2015, affected by weather and COVID-19 shocks. GDP per capita growth was negative in 2016/17 and 2019/20. Figure 42 and Figure 43 show contributors to economic growth from the supply and demand side. On the supply side, the growth trajectory was affected by the drought in 2016/17, with the recovery driven primarily by the industry and services sectors. On the demand side, the reversal in economic growth in 2016/17 led to a decline in private consumption, with a strong counterbalancing impact from net exports. The recovery in the next two years was driven by private and government consumption and investment. Economic activity stalled again during the first half of 2020 due to a COVID-19 related domestic lockdown that lasted for over four months, border closures, and the spillover effects of a slowed down global economy. This hit the industrial and service sectors the most and resulted in a reduction of private consumption and investment. Reduced revenues coupled with higher spending measures to cope with the COVID-19 crisis led to a significant widening of the fiscal deficit, which reached 7.2 percent of GDP in FY20. This was the highest fiscal deficit during the last ten years – almost double the average fiscal deficit of 3.6 percent of GDP in 2010-2020 (World Bank, International Finance Corporation, Multilateral Investment Guarantee Agency 2021). Figure 42. Contributions to change in real GDP per capita Figure 43. Contribution to changes in real GDP per growth, by value added from key economic sectors, capita growth, by demand side, changes in percentage changes in percentage points points Source: Uganda Bureau of Statistics, accessed on 9/27/2021. The COVID-19 pandemic and associated containment measures affected the population through the labor market and income channels. Following the mobility restrictions that were put in place in March through June 2020, about 16 percent of respondents had stopped working by June,24 according to the COVID-19 Uganda High Frequency Phone Survey (UHFPS). The rate of work stoppage was highest in urban areas and in the service sector. Female respondents were more likely to stop working then male respondents and the gap was particularly pronounced in urban areas and among young respondents ages 15–30 years. The employment rate returned almost fully to the pre-March levels by August 2020, but income levels still did not reach the average levels observed before the pandemic (Figure 44 and Figure 45). In addition, many people shifted from working in industry and services to agriculture, somewhat 24 This is equivalent to 19 percent of those who worked before the lockdown. 28 reversing (at least temporarily) the structural change trend observed before the COVID-19 pandemic. The second lockdown, during June–July 2021, had a smaller negative impact on employment. As reported in October/November 2021, about 11 percent of respondents stopped working compared to the previous round, and work stoppages were more universally distributed compared to June 2020. More than 50 percent of work stoppages in October/November 2021 after the second lockdown occurred in the agriculture sector and were to a less extent associated with COVID-19 related restrictions compared to stoppages in June 2020. Observed differences in the extent, nature and distribution of work stoppages may be related to strictness of the first lockdown which was shorter, but tighter than the second one. In addition, higher work stoppages in the agricultural sector can be related to seasonality and prolonged dry spells observed in different parts of the country in 2021. Figure 44. Working status among respondents to Figure 45. Households with income below average Uganda High-Frequency Phone survey, % monthly income during 12-month period prior to the first lockdown, % of receiving income Source: Uganda High-Frequency Phone survey, different rounds. World Bank staff calculations. Note: The same households with the same respondents across all rounds were used to compared working status. Besides negative economic impacts, COVID-19 widened inequality in access to education during the pandemic, which remains the key threat to human capital development in Uganda. During the period March 2020–October 2021, schools in Uganda were fully or partially closed for 83 weeks – the longest in the world (UNESCO). This was longer than the average of 91 days for Eastern and Southern Africa and resulted in substantial learning losses. Access to distant learning activities was quite low and unequal. According to the HFPS, about 84 percent of children aged 3–18 years attended school before the lockdown and school closure in March 2020. Overall, only 41 percent of children either returned to school or were engaged in any learning/education activities from home in March/April 2021. This is less than half the school attendance before the lockdown. Moreover, participation in education and learning activities was very unequal, with children in urban areas and from the wealthiest pre-COVID consumption quintiles having much higher chances to study compared to children in rural areas and those from the poorest quintiles. This poverty assessment has two objectives. The first one is to explore the most recent trends in socio- economic wellbeing since 2016/17, distinguishing between pre- and COVID periods. The second objective is to fill the knowledge gap by measuring vulnerability to poverty, assessing inequality of opportunities, 29 simulating distributional impact of increased competition in the telecommunication sector and analyzing the patterns and determinants of internal migration. To answer the question and fill the knowledge gaps, a wide variety of surveys are used. The report uses three rounds of Uganda National Household Survey, three rounds of Uganda Panel Survey, seven rounds of Uganda High-Frequency Phone Survey, Uganda Refugee High-Frequency Phone Survey and AfroBarometer data (more detailed information about the surveys used can be found in Box 5). In addition, the report uses geospatial data measuring night lights, precipitation, population density, distance to roads and so forth. Box 1. Previous poverty assessments and strategic documents The most recent poverty assessments covered the period from 2005/2006 to 2016/17. The 2016 Poverty Assessment covered the period from 2005/06 to 2012/13 (World Bank 2016). It concluded that poverty reduction during this period was 80 percent driven by agricultural growth. Agricultural growth, however, was not driven by technology adoption or change from subsistence to more commercial agriculture, rather by favorable weather and market conditions. The report also highlighted the high vulnerability and limited role of structural transformation and migration in poverty reduction. The latest Poverty Diagnostic focused on the period covering 2012/13 and 2016/17 (World Bank 2019). It demonstrated that poverty increased in 2016/17 compared to 2012/13, mainly in rural areas and due to an extreme weather shock. The rural population remains vulnerable and increasing their participation in non- agricultural economic activities will be fundamental to increasing their living standards. The Uganda Systematic Country Diagnostic Update (World Bank, International Finance Corporation, Multilateral Investment Guarantee Agency 2021) selected four key areas to focus on: improving agricultural productivity, creating jobs, enhancing human capital and women empowerment, and strengthening household and community resilience. Addressing most of the development issues above is embedded into the third National Development Plan (NDP) for 2020/21–2024/25 (National Planning Authority 2020). The third NDP’s objectives include enhancing value addition in key growth opportunities, creating jobs through the private sector, increasing quality of productive infrastructure, enhancing productivity and social wellbeing of the population and strengthening the role of the state in facilitating development. The plan identified 18 programs to achieve these objectives including agro-industrialization of agriculture, digital transformation, human capital development program, regional development program, and sustainable urbanization. Most recently, the Parish Development Model (PDM) was initiated and adopted by the Government, with a parish as the key administrative unit for implementation of Government Programmes and as the best suited delivery mechanism for the attainment of NDPIII development goals at the grassroots level. The PDM will focus on four sectors for wealth and job creation, including: commercial agriculture, industry, services and Information, and communication technology. 30 II. Trends in poverty, inequality, and shared prosperity since 2016/17 Interruption in data collection of the Uganda National Household Survey (UNHS) 2019/20 related to COVID-19 can potentially affect comparability of poverty rates across years and within the survey. Field data collection for the UNHS usually takes 12 months to account for the seasonality of agricultural production. The sampled enumeration areas are spread out equally across the country for each quarter of the year. The field work in the UNHS 2019/20 was affected by the COVID-19 lockdown introduced in March 2020. It started in September 2019 but was interrupted in March 2020. Data collection resumed in July 2020 and ended in November 2020. As a result, the survey lasted 12 months, but fully omitted the second quarter (April–June 2020). This creates potential comparability issues between previous rounds of the UNHS, which covered all quarters, and the UNHS 2019/20. Separating pre-COVID and COVID periods in UNHS 2019/20 can provide interesting information on the impact of COVID-19. However, comparing poverty rates across these periods and comparing them with annual poverty rates from the previous rounds is prone to the same seasonality issue, particularly if such time-sensitive indicators as employment are considered. In addition, even though the UNHS 2019/20 is nationally representative, two sub-samples of the UNHS 2019/20 have different population structure with larger urban population in the COVID sub- sample. To address some of these challenges, the report made adjustments to the data and conducted robustness checks. The report applied post-stratified population weights in the UNHS 2019/20 to ensure subregional and rural/urban population shares in two subsamples (pre and COVID-19) have the same distribution as a nationally representative yearly survey (Figure A2 and Figure A3). These adjusted weights were employed for the analysis using the pre-COVID and COVID sub-samples of the UNHS 2019/20. Additional robustness checks include comparing poverty rates from similar quarters in 2016/17 and 2019/20. Overall, results qualitatively stayed the same whether comparable quarters were used or not, therefore the report uses UNHS 2019/20 as it is for the annual analysis. However, it is also shown below that before COVID-19 the regional poverty trends using the UNHS and consumption changes using UNPS were not fully aligned. Poverty trends from the UNHS were also not well aligned with subjective wellbeing indicators from the same survey and selected geospatial indicators. This calls for caution in interpreting the most recent regional trends, particularly in using the pre-COVID and COVID sub-samples. Poverty indicators Since 2012, robust poverty reduction has stalled, with rural poverty fluctuating up and down and urban poverty increasing slightly. Figure 46 shows the evolution of poverty rates in Uganda across rural and urban areas since 2012. Relatively stable poverty rates at the national level mask changes in rural and urban poverty. Due to the episode of drought in 2016/17 (World Bank 2019), rural poverty increased by 1.5 percentage points from 35 percent in 2012/13 to 36.5 percent in 2016/17 and then declined to 33.8 percent in 2019/20 after the drought crisis subsided. Urban poverty was relatively stagnant between 2012/13 and 2016/17 but increased afterwards from 17.2 percent to 19.8 percent in 2019/20. It seems that poverty was falling in rural areas before the pandemic, but COVID-19 wiped out the gains and also accelerated the increase in poverty in urban areas. Given that 2019/20 covers both pre-COVID and COVID periods, showing the full year poverty rate may hide important variations across them. Even though not strictly comparable, this disaggregation may be indicative of whether poverty would have fallen if COVID-19 had not happened. Indeed, Figure 47 demonstrates that both national and rural poverty 31 fell much faster between 2016/17 and the pre-COVID period in 2019/20, but the progress has been completely reversed during the COVID-19 period of 2020. Thus, rural poverty fell by six percentage points between 2016/17 and the pre-COVID period in 2019/20 but increased by 7.2 percentage points. The trend in urban poverty did not change, but disaggregation shows that the vast majority of the increase in urban poverty happened during COVID-19. To check the robustness of these findings, a simple regression analysis was carried using the UNHS 2019/20 where one of the explanatory variables was a dummy for COVID-19 period. Even after controlling for quarter or month of the survey (seasonality), results show that households in the COVID-19 period experienced large and statistically significant decline in consumption level ranging from 11 to 17 percent depending on the control variables used. Figure 46. Headcount poverty rates across rural and Figure 47. Headcount poverty rates across rural and urban areas during 2012/13–2019/20, % urban areas during 2016/17–2019/20 (before and during COVID-19), % Source: UNHS, WB staff calculations. The depth and severity of poverty demonstrated similar patterns to poverty headcount rates being negatively affected by COVID-19. The depth of poverty (or poverty gap) indicates how far, on average, poor households are from the poverty line. The severity of poverty captures both how far the poor are from the poverty line and consumption inequality among the poor. Both indicators fell sharply between 2016/17 and the pre-COVID period of 2019/20 at the national level and rural areas (Figure 48 and Figure 49). No positive trend was observed in urban areas during the same period. After COVID-19, both the depth and severity of poverty indicators deteriorated significantly, signaling that more money would be needed to lift the poor out of poverty and that inequality among the poor increased dramatically as well. The depth and severity of poverty have increased in rural and urban areas, but the increase in the COVID- 19 period was less sharp compared to rural areas. 32 Figure 48. Depth of poverty across rural and urban areas Figure 49. Severity of poverty across rural and urban during 2016/17–2019/20 (before and during COVID-19), areas during 2016/17–2019/20 (before and during % COVID-19), % Source: UNHS, WB staff calculations. The largest changes in poverty across 2016/17 and 2019/20 were observed in the Eastern and Western regions. The poorest Eastern region demonstrated a very strong reduction in poverty between 2016/17 and 2019/20, with poverty falling by about 12 percentage points even despite the COVID-19 shock (Figure 50). Looking separately at the pre-COVID and COVID periods reveals that, without COVID-19, poverty in the Eastern region could have fallen even further, but unfortunately the pandemic has eroded much of the gains (Figure 51). The second region with significant poverty change was the Western region, where poverty increased by five percentage points, with the largest increase observed during the COVID-19 period. Poverty increased slightly by about three percentage points in the Northern region, but it was not statistically significant. Poverty declined slightly before COVID-19 and increased afterwards in the Central region, but these changes were also not statistically significant. Figure 50. Headcount poverty rates across regions Figure 51. Headcount poverty rates across regions during 2016/17–2019/20, % during 2016/17–2019/20 (before and during COVID- 19), % Source: UNHS, WB staff calculations. 33 Figure 52. International headcount poverty rates in Based on the international poverty line of US$1.90 Uganda and comparators in 2019 circa years based on PPP, about 40 percent of Uganda’s population 1.90 USD 2011 PPP daily poverty line were poor in 2019/20. This is similar to the poverty rate in Zimbabwe in 2019 (40 percent) and Kenya in 2015 (37 percent). At the same time, international poverty rates were significantly higher in Tanzania in 2018 (49 percent), Rwanda in 2016 (57 percent) and Malawi in 2019 (74 percent). Ghana had the lowest international poverty rate in 2016 (13 percent) followed by Ethiopia with a poverty rate of 31 percent in 2015 (Figure 52). Source: World Bank. Inequality and shared prosperity Inequality did not grow between 2016/17 and 2019/20 at the country level – in contrast to substantial changes within urban and rural areas. Inequality in consumption, measured by the Gini coefficient, has barely changed since 2016/17 at the national level (Figure 53). However, this picture changes dramatically for rural and urban inequality. Inequality in urban areas has increased, mainly in the COVID-19 period, while rural inequality dropped significantly in the same period (Figure 54). This widened the gap in Gini coefficients across two areas even further, making the distribution of consumption in urban areas more unequal. In comparison to regional peers, the Gini in Uganda was similar to the Gini in Ghana and Rwanda and was slightly lower than the Gini in Tanzania and Kenya, although with slightly different reference years (Figure 55). Figure 53. Consumption per adult equivalent Gini Figure 54. Consumption per adult equivalent Gini coefficient across rural and urban areas during coefficient across rural and urban areas during 2016/17–2019/20 using UNHS 2016/17–2019/20 (before and during COVID-19) using UNHS Source: UNHS 2016/17 and 2019/20, WB staff calculations. Note: Inequality indicators and growth incidence curves were based on consumption per adult equivalent spatially adjusted by poverty lines. 34 The decline in consumption between 2016/17 and 2019/20 was much more pronounced among urban residents. Figure 56 demonstrates how consumption per adult equivalent grew/declined across deciles in urban and rural areas using the UNHS data. In rural areas consumption grew on average by 0.3 percent annually, mainly due to positive growth among the bottom 20–70 percent of the consumption distribution. In contrast, nobody experienced positive consumption growth in urban areas, but those from the bottom 20-70 percent of the distribution were affected the least by the negative downturn. Economic growth before the COVID-19 pandemic mainly benefited rural residents, but they were also affected much more severely by the COVID-19 shock compared to urban residents. Figure 57 compares consumption growth during 2016/17 and the pre-COVID period of 2019/20. Consumption of urban residents was negative, while consumption of rural residents grew by 2.5 percent annually, without significant differences across the distribution except for the poorest decile, who did not seem to benefit much from the economic growth. The positive gains in the consumption of rural residents were fully wiped out during the COVID-19 period when consumption declined by about 14 percent, on average (Figure 58). However, the bottom 40 percent of rural population was affected much less than the top 60 percent, which explains the sharp reduction in inequality observed during this period. Urban residents lost, on average, about 6 percent of their consumption, with the bottom 40 percent affected less than the top 60 percent of the urban population. Figure 55. Consumption per adult equivalent Gini Figure 56. Annualized consumption per adult equivalent coefficient across countries circa 2019 year growth rates by deciles during 2016/17–2019/20 in rural and urban areas using UNHS, % Source: World Bank. Source: UNHS 2016/17 and 2019/20, WB staff calculations. 35 Figure 57. Annualized consumption per adult equivalent Figure 58. Annualized consumption per adult growth rates by deciles across rural and urban areas equivalent growth rates by deciles across rural and during 2016/17–2019/20 (pre-COVID) using UNHS, % urban areas in 2019/20 (before and during COVID-19) using UNHS, % Source: UNHS 2016/17 and 2019/20, WB staff calculations. In the COVID-19 period, reducing inequality curbed the increase in rural poverty but contributed to the increase in poverty in urban areas. Changes in poverty in Uganda can be broken down into “pure growth” and “distribution” components to determine how each factor contributed to the change in the poverty rate observed. In rural areas between 2016/17 and the pre-COVID-19 period in 2019/20, the increase in the mean average household consumption (growth effect) was a dominant poverty reduction force, accounting for more than 90 percent of all poverty reduction during the period, while the contribution from the reduction in inequality (distribution effect) was limited. This was reversed in 2019/20 between the pre-COVID-19 and during COVID-19 periods, when poverty increased in rural areas solely due to lower average household consumption (negative growth effect), with the reduction in inequality diminishing the total increase in poverty. As shown later in the text, the agricultural sector – where the majority of the poor work – was less affected by the COVID-19 related measures after the first lockdown, which can explain the poverty reducing distribution impact in rural areas. In urban areas, both growth and inequality effects contributed to the increase in poverty observed in 2019/20 during pre-COVID-19 and COVID-19 periods. Figure 59. Growth-redistribution decomposition of poverty changes in 2016/17 –2019/20 (before and during COVID-19) using UNHS in rural and urban areas, percentage points Source: UNHS, WB staff calculations. 36 Box 2. Welfare trends according to the Uganda National Panel Survey in 2015/16 and 2019/20 Due to seasonality and potential comparability issues in the latest UNHS 2019/20 round, which capture pre-COVID- 19 and COVID-19 periods, the Uganda National Panel Survey (UNPS) data are analyzed for the years of 2015/16 and 2019/20 to see whether poverty was falling before COVID-19, as is suggested by the UNHS data. The annual UNPS data collection period starts and ends in February of each year (the data collection in 2019/20 was finished before the lockdown in March 2020). The welfare aggregate is constructed using the same methodology as for the UNHS data. Figure 60. Annualized consumption per adult equivalent growth rates across rural and urban areas, regions during 2015/16 and 2019/20 using UNPS data, % Source: UNPS 2015/16 and 2019/20, WB staff calculations. Note: Quintiles are constructed using official welfare aggregate spatially adjusted by regional poverty lines. Figure 60 shows consumption per adult equivalent growth rates by quintiles at the national, rural/urban and regional levels. Consumption growth was positive everywhere, including urban areas and the Central and Northern regions. This is not fully aligned with the UNHS data, which showed that poverty fell significantly only in the Eastern region and was rather stagnant everywhere else. When comparing consumption growth between the UNHS and UNPS data, it is important to recognize that the period of the UNPS data was longer (2015/16–2019/20), while the UNHS starting point was in 2016/17 (which also coincided with an economic downturn in that year). Indicators about households' subjective wellbeing demonstrated universal improvement before COVID- 19 and universal worsening during COVID-19. To triangulate and reconcile monetary poverty trends since 2016/17, the report checked the evolution of subjective wellbeing indicators and geospatial data. According to the analysis of subjective wellbeing indicators from the UNHS, such as share of population who reported a decline in living standards or the share of population with unstable income, there was universal improvement in these indicators before COVID-19 and universal worsening during COVID-19. Weather conditions were more favorable for agricultural production in all regions in 2019/20 compared to the 2016/17 survey period, while nighttime lights also indicated positive economic development in Uganda since 2016 and before COVID-19. The more detailed triangulating exercise can be found in the Annex 1. 37 IV. Sociodemographic characteristics of the poor in 2019/20 This section compares sociodemographic characteristics of poor and non-poor households, focusing on household/individual and community level indicators. Where possible, simple cross-tabulations are supplemented by multivariate analysis based on regression techniques that identify the correlates of poverty or household consumption controlling for the influence of other factors.25 Household and individual level characteristics Location plays a key role in monetary welfare status, with poverty being overwhelmingly rural in Uganda. The poverty rate in Uganda was much higher in rural areas compared to urban in 2019/20: 33.8 versus 19.8 percent, respectively (Figure 61). More than 80 percent of the poor population in Uganda resides in rural areas (Figure 62). Regional and subregional disparities in headcount poverty rates were also stark, with poverty being much higher in the Eastern and Northern regions and being twenty times lower in Kampala compared to the Karamoja subregion. Figure 61. Headcount poverty rates in 2019/20 across Figure 62. Share of the poor population in 2019/20 areas, regions and subregions, % across areas and regions, % Source: UNHS 2019/20, WB staff calculations. Larger household size and more children are strongly associated with higher poverty rates. Poor households have significantly larger households with more children. Thus, average household size among the poor households was 5.5 members in 2019/20 compared to 4.3 members among non-poor households (Table 1). The number of children is also significantly correlated with poverty. The headcount poverty rate in households with two or three children in 2019/20 was about 34 percent – which is 10 percentage points higher than among households with one or two children (Figure 65). Once controlled for other factors, having one additional child (ages 0–13 years) was associated with a reduction in consumption per adult equivalent by three percent. Households in rural areas were particularly vulnerable to these demographic factors as they had more children and, as a result, a higher dependency ratio. 25 This simple analysis does not aim to find causal relationships, as doing so requires more sophisticated techniques and panel data that tracks the same households over time. Consumption per adult equivalent was adjusted for spatial variation in prices using location specific poverty lines. It was also normalized to preserve the mean of nominal consumption. 38 Female-headed households in rural areas, female-headed divorced and married households had higher poverty rates. Simple comparison of poverty rates finds significant difference in poverty rates among female- and male-headed households in rural areas only (36 versus 33 percent, respectively). As the term ‘female-headed household’ includes married, unmarried, divorced, and widowed women, poverty rates for separate categories were also reported in Figure 66. Significant gender gaps in poverty rates in favor of men were found for divorced and married heads. Once controlled for other factors, the relationship between gender and monetary wellbeing becomes more complex. Figure 67 shows coefficients from a log consumption regression, which compares consumption of households with different gender and marital status keeping other factors constant. Male-headed single households are used as the base category, so all reported results are compared to it. Regardless of gender and marital status, consumption levels of all groups were lower than those of male- headed single households. For some categories in rural areas, results were significantly different across men and women. Thus, both divorced male and female-headed households in urban areas had much lower consumption than single male-headed households, but for female-headed households the negative relationship was much higher than for male-headed (24 versus 10 percent respectively). In contrast, married male-headed households have lower negative relationship with consumption than married female-headed households (24 versus 17 percent). There were gender gaps for other categories as well, but they were not statistically significant, probably because of the small number of observations for these groups. Table 1. Sociodemographic characteristics of Ugandan households, 2019/20 Area Poverty status Uganda Rural Urban Non-poor Poor Rural poor Urban poor Household size 4.6 4.8 4.1 4.3 5.5 5.6 5.2 Children 0–13 2.0 2.2 1.6 1.8 2.7 2.8 2.4 Adults 14–59 2.3 2.3 2.3 2.2 2.6 2.5 2.6 Elderly 60+ 0.2 0.3 0.2 0.2 0.3 0.3 0.2 Dependency ratio, % 45 48 38 43 52 53 48 Head of household age 44.4 45.5 41.8 44.1 45.5 45.4 45.7 Gender of household head, % Men 68 69 66 68 67 68 63 Women 32 31 34 32 33 32 37 Highest level of education completed by household head, % No education 17 19 11 13 28 28 29 Primary incomplete 36 42 23 34 44 45 39 Primary complete 17 17 17 18 15 15 15 Secondary incomplete 20 17 28 23 12 12 14 Secondary complete 2 1 5 3 0 0 1 Post-secondary but not university 4 3 7 5 1 1 2 University 3 1 8 4 0 0 0 Source: UNHS 2019/20, WB staff calculations. Analysis of poverty rates by income earner composition showed a significantly higher poverty rate among households with one female earner in paid employment compared to households with one male earner in paid employment, potentially pointing to gender inequalities in labor markets. While paid employment and households with single income earners did not represent the largest share of the Ugandan population, it was nevertheless striking that 15 percent of households with just one male earner in paid employment were below the poverty line compared to 27 percent of households with just one female earner in paid employment (Figure 63 and Figure 64). In contrast, poverty rates among male and 39 female earners in subsistence agriculture were quite similar, at around 32 percent to 34 percent. In addition, when the type of work was not specified, the overall poverty rate was roughly similar for more than one female earner or more than one male earner – around 28 percent to 32 percent. Paid employment is typically viewed as an important pathway out of poverty and the overall lower poverty rates were observed for both female and male earners in paid employment. However, the substantially higher poverty rates observed among households with one female earner in paid employment compared to one male are notable and merit further research into the driving causes that are preventing these households from realizing the poverty-reducing benefits of paid employment. Figure 63. Headcount poverty rates depending on Figure 64. Population shares depending on income income earner composition 2019/20, % earner composition 2019/20, % Source: UNHS 2019/20, WB staff calculations. Note: *Significant difference in mean poverty rate at 1 percent, ** at 5 percent and *** at 10 percent. Figure 65. Headcount poverty rates in households Figure 66. Headcount poverty rates depending on head with different number of children in 2019/20, % of household marital and gender status in 2019/20, % Source: UNHS 2019/20, WB staff calculations. Note: *significant difference in mean poverty rate at 1 percent, ** at 5 percent and *** at 10 percent. The education level of the head of household was the key determinant of poverty and consumption levels both in rural and urban areas. Figure 68 shows headcount poverty rates for households depending on head of household education status. There is a strong negative association between the level of education and poverty rates. Households with heads having higher education levels demonstrated lower 40 poverty rates. The difference is striking. For example, poverty in households with uneducated heads reached about 48 percent in 2019/20 (this group accounted for 17 percent of all heads). This was six times higher than the poverty rate of households where the household head finished secondary education (which is only 3.3 percent of all heads). Figure 69 shows the relationships between head of household education level and consumption per adult equivalent. Even having primary incomplete education increases consumption by 18 percent compared to households where the head does not have any education at all. The highest returns are expectedly observed for those households where the head had a university degree. Consumption among these households was more than 95 percent higher than consumption of households with an uneducated head. Poor households – and those living in rural areas – tend to have less educated heads. Almost one-third of poor household heads do not have education, compared to only ten percent of non-poor households. Figure 67. Selected gender and marital status coefficients from regressions explaining logarithm of consumption per adult equivalent in rural and urban areas in 2019/20, % a) Urban areas b) Rural areas Source: UNHS 2019/20, WB staff calculations. Note: * significant difference in coefficients at 1 percent, ** at 5 percent and *** at 10 percent. Figure 68. Headcount poverty rates depending on head Figure 69. Consumption per capita returns to head of of household education level in 2019/20, % household education after controlling for other factors in 2019/20, % Source: UNHS 2019/20, WB staff calculations. 41 Head of household engagement in subsistence agriculture is associated with high poverty rate. The working status of heads of household was correlated with poverty rates in 2019/20 (Figure 70). Thus, households with heads in any type of paid employment had a poverty rate about 23 percent, which was almost twice as low as poverty rates among households with heads either in subsistence agriculture (40 percent) or unemployed or out of the labor force (39 percent). A striking difference was observed across rural and urban areas. Thus, in rural areas, engagement in subsistence agriculture results in lower poverty compared to being out of the labor force or unemployed. By contrast, in urban areas the poverty rate of those households with heads in subsistence agriculture was almost double the poverty rate of households whose heads were not working at all. This indicates that there were households in urban areas who had sources of income other than work. Consistent with welfare impacts of employment, heads in paid employment were more likely to be found among non-poor households, while heads in subsistence agriculture among the poor ones (Figure 71). Figure 70. Headcount poverty rates depending on head Figure 71. Head of household working status across of household working status 2019/20, % poverty status in 2019/20, % Source: UNHS 2019/20, WB staff calculations. The sector of employment of the head of household was highly correlated with poverty status, with those engaged in agriculture more likely to be poor. Almost 60 percent of household heads in Uganda were working in agriculture in 2019/20. The second largest sector, commerce, food, and accommodation absorbed about 15 percent of household heads. The labor market profile of heads from poor households was heavily skewed towards agriculture – about 77 percent of heads worked in this sector. According to Figure 72, households whose head worked in agriculture had the highest headcount poverty rate (35.5 percent). This was about seven percentage points higher than the average poverty rate of households with an employed head (28.7 percent) and twice as high as the poverty rate of households whose head worked in commerce or the accommodation sector (15.9 percent). 42 Figure 72. Headcount poverty rates in 2019/20 depending on head of household sector of employment, % Source: UNHS 2019/20, WB staff calculations. Poor households cannot afford such important assets as mobile phones, TVs, radios, solar panels, and motorcycles, but ownership rates had improved in the 12 months to the 2019/20 round of the UNHS. Households were asked if any member owned a particular asset during the interview and if they owned it 12 months ago. Overall, average rates of ownership for all selected assets increased during the last 12 months among all Ugandans and especially for poor households (Figure 73). Mobile phones demonstrated the fastest growth and remained the most widespread asset, with about 68 percent of rural and 86 percent of urban households owning them. The gap between the poor and non-poor households was persistently high, particularly in urban areas, where 90 percent of non-poor households had mobile phones compared to only 65 percent among the poor households (Figure 74). Ownership of solar panels increased as well during the period considered. A striking gap in ownership of a TV was observed in urban areas, where 49 percent of non-poor households owned this asset compared to only five percent of the poor households. Such a drastic difference in access to information technologies will affect the capacity of poor households to adapt to the COVID-19 pandemic. Ownership of cattle and land was associated with higher consumption in rural areas, while in urban areas land ownership was negatively related to consumption. Ownership of agricultural land was about 74 percent among households in rural areas and 39 percent in urban areas (Figure 75). Poor households in urban areas were much more likely to own agricultural land compared to non-poor households (62 percent versus 39 percent) – likely indicating that agriculture is a relatively less profitable sector of work for an urban resident. The same pattern was observed in urban areas with regards to ownership of cattle. In urban areas, the ownership of land is associated with lower consumption by five percent, while access to livestock was associated with higher consumption by five percent. In contrast, the rural poor were less likely to own cattle, signaling that they may not have capital to obtain the livestock and it plays a different role for them compared to urban poor. Once controlled for other factors, the ownership of land and cattle is associated with higher consumption in rural areas. 43 Figure 73. Ownership of selected assets in 2016/17 and Figure 74. Ownership of selected assets in 2019/20 2019/20 in Uganda and among the poor households, % in rural and urban areas across poor and non-poor households, % Source: UNHS 2016/17 and 2019/20, WB staff calculations. Figure 75. Availability of land and livestock in 2019/20 Figure 76. Access to improved water and sanitation in in rural and urban areas across poor and non-poor 2019/20 in rural and urban areas across poor and non- households, % poor households, % Source: UNHS 2019/20, WB staff calculations. Access to improved sanitation was very low in Uganda in 2019/20, especially for rural residents and the poor. Access to improved drinking water was much better in Uganda in 2019/20 compared to improved sanitation. Thus, close to 74 percent of the rural population and 87 percent of the urban population had access to improved drinking water (Figure 76). There was no difference in access to improved water within rural areas between the poor and non-poor population, suggesting that access may be more related to infrastructure investments on a community level. The urban non-poor had very high access to improved drinking water – 90 percent, which was 14 percentage points higher than access among the poor urban residents. In contrast to drinking water, access to improved sanitation was very low in both urban and rural areas (39 percent and 25 percent, respectively). The poor versus non-poor gap in access to improved sanitation in urban areas was particularly large – 22 percent versus 42 percent, respectively. 44 Community characteristics Access to and quality of community services such as health, education, and infrastructure play as important a role for monetary wellbeing as well as household-specific characteristics. Living in communities with better access to essential public services is expected to increase household consumption and reduce poverty by providing better economic opportunities and increasing monetary returns to household characteristics. Overall, government health centers and schools were available to the majority of the population, regardless of poverty status, but the non-poor population lived in communities with shorter distances to these services. Community officials were asked about the availability of different government services to the members of the community. Figure 77 shows that government health centers and government schools were available for most of the population either within their own community or in a neighboring community. The difference between poor and non-poor people was observed with regards to the distance to these services (Figure 78). Thus, poor people lived in the communities with longer distances from the village center to the services. For example, the distance from the village center to the government secondary school was, on average, about 6.3 kilometers for the poor population compared to 4.9 kilometers for the non-poor population. Figure 77. Availability of government health centers, Figure 78. Distance to government health centers, primary and secondary schools in the original or primary and secondary schools from a village center in neighboring communities in 2019/20 for poor and non- the original or neighboring communities in 2019/20 for poor population, % poor and non-poor population, kilometers Source: UNHS 2019/20, WB staff calculations. Note: Community level indicators are calculated for population. Access to the national paved road, agricultural markets, banks and mobile service points was lower compared to government schools and health centers, with larger gaps between poor and non-poor population. Access to other essential services and infrastructure in 2019/20 is shown in Figure 79. Overall, access to paved national roads, markets selling agricultural products, banks and mobile service points was lower compared to access to the government health centers and schools. Moreover, the gap between the poor and non-poor population was more pronounced. For example, 72 percent of non-poor people live in communities with accessible banks either in the same or neighboring communities compared to only 58 percent of poor people. The poor also face longer distances to these essential services compared to the 45 non-poor (Figure 80). Overall, banks were located the furthest from a village center – about 19 kilometers for the poor and 15 kilometers for the non-poor. Figure 79. Access to other essential services and Figure 80. Distance to other essential services and infrastructure in the original or neighboring infrastructure from a village center in the original or communities in 2019/20 for poor and non-poor neighboring communities in 2019/20 for poor and non- population, % poor population, kilometers Source: UNHS 2019/20, WB staff calculations. Note: Community level indicators are calculated for population. In contrast, while government health centers and primary schools are more accessible in terms of distance, the perceived quality of these services lags behind the quality of agricultural markets, roads, banks and mobile service points. Figure 81. Share of population living in communities with Community officials were asked to rank the services ranked as poor, % quality of public services offered in their community. Almost 25 percent of the population lived in communities with health centers and primary schools that were perceived to be of poor quality. The perceived the quality of government secondary schools was higher (Figure 81). The views on the quality of markets selling agricultural outputs, national paved roads and mobile service points were mostly positive. The poor population was slightly more likely to live in the communities where primary, secondary schools, agricultural markets and mobile service points were ranked as poor. Source: UNHS 2019/20, WB staff calculations. 46 Box 3. Determinants of poverty transitions based on UNPS 2015/16 and UNPS 2019/20 In order to understand the correlates of changes in poverty, the report used two rounds from the UNPS, which track the same households forming a panel dataset. The UNPS is not used to measure and report poverty but contains a welfare aggregate constructed using official UBOS methodology to measure poverty in the UNHS and regional poverty lines. To achieve a poverty rate similar to the official one reported using UNHS, the analysis used the maximum value of UNPS consumption per adult equivalent, spatially adjusted by poverty lines for the 30th percentile (close to the poverty rate in UNHS 2019/20). This value was used as the poverty line in UNPS 2015/16 and 2019/20 and resulted in poverty rates of about 38 percent for 2015/16 and 30 percent for 2019/20 after applying cross-sectional weights. After forming a balanced panel and using panel weights, poverty rates were about 39 percent for 2015/16 and 31 percent for 2019/20. Using panel data allows us to classify all households into four mutually exclusive groups: poor in both periods (chronic poor), non-poor in the first period and poor in the second period, poor in the first period and non-poor in the second period, and non-poor in both periods. Overall, about 20 percent of the population was chronic poor and remained poor in 2015/16 and 2019/20, about half of the population was out of poverty in both periods, about 18 percent moved from being poor in 2015/16 to being non-poor in 2019/20, and finally about 12 percent moved from being non-poor in 2015/16 to being poor in 2019/20. Figure 82 shows the distribution of all groups across head of household characteristics. The largest shares of chronic poor were observed among heads of household who did not have formal education and were either unemployed or out of the labor force. Better education, working in services and public administration were also closely related to being non-poor in the two periods. The largest shares of those who moved out of poverty in 2019/20 while being poor in 2015/16 were observed among households where heads worked in agriculture, were unemployed or out of the labor force and had some primary education in 2015/16. Table 2 shows marginal effects after the probit regressions estimated for two subsamples. The first regression was run for the subsample of poor households in 2015/16 and explored correlates of probability for moving out of poverty in 2019/20. The second regression was run for the subsample of non-poor households in 2015/16 and explored correlates of probability of moving into poverty in 2019/20. Figure 82. Poverty transitions between 2015/16 and 2019/20 by head of household characteristics in 2015/16, % Source: UNPS, WB staff calculations. Note: Panel population weight for 2015/16 was used. Poor households in 2015/16 were more likely to move out of poverty in 2019/20 if their heads had better education, lived in the Central region, had a smaller household size, and were older. Heads of households working 47 in other sectors were less likely to move out of poverty compared to heads working in agriculture. Regarding non- poor households in 2015/16, they were more likely to become poor in 2019/20 if they lived in rural areas, had a larger household size and younger, less educated heads. Households with heads working in industry and construction sectors were more likely to move into poverty in 2019/20 compared to households working in agriculture. Table 2. Marginal effects after probit regressions moved out of poverty in became poor in 2019/20 variables in 2018/19 2019/20 among poor in 2015/16 among non-poor in 2015/16 household size -0.0188** 0.0193*** head is female 0.06 -0.01 head age 0.00544*** -0.00197** Household head without education is base some primary 0.232*** -0.04 completed primary 0.245*** -0.139*** some secondary 0.315*** -0.153*** completed secondary 0.489*** -0.217*** post-secondary -0.289*** Household head working in agriculture is base industry and construction 0.140* -0.06 services and public administration 0.15 -0.03 other sectors -0.01 0.06 unemployed or out of labor force 0.07 -0.02 Central region is base Eastern -0.285*** 0.189*** Northern -0.406*** 0.219*** Western -0.132* 0 urban 0.03 -0.125*** Observations 862 1,707 Source: UNPS, WB staff calculations. Note: Significant at 1 percent, ** at 5 percent and *** at 10 percent. Figure 83. Household asset ownership in 2015/16 across poverty transitions status between 2015/16 and 2019/20, % Source: UNPS and RULIS, WB staff calculations. Note: Panel household weight for 2015/16 was used. Ownership and usage of mechanized equipment is shown only for households engaged in agricultural activities in both periods. The chronic poor were not able to afford or were less likely to own and use important assets. Figure 83 shows ownership of different assets among households grouped by their poverty status in 2015/16 and 2019/20. For this figure, those who were poor at least once in 2015/2016 or 2019/20 were labeled as transient poor. Being 48 chronically poor is associated with a lower likelihood of owning such important assets as a TV, radio, mobile phones, motorcycles and so forth. Among households engaged in agricultural activities in both periods, the chronic poor were also least likely to use or own mechanized equipment. 26 Overall, households in transient poverty were better off than the chronically poor, but the highest asset ownership was observed among households that were never poor in both periods, as would be expected. V. Geographic disparities in poverty and their correlates Sources of welfare disparities within and across regions Uganda’s development is very spatially uneven, with Kampala standing out as a place with the lowest poverty rate. Poverty rates varied a lot in Uganda in 2019/20 across the regions (Figure 84). Poverty in the Central region was about 15 percent or almost three times lower than poverty in the Eastern and Northern regions, which reached 42 percent and 40 percent, accordingly. Poverty rates varied within some regions as well, with poverty rates being the highest in the rural areas and lowest in urban areas. Thus, the urban headcount rate in the Eastern region was 30 percent compared to 44 percent in rural areas of the same region. Large statistically significant differences in the poverty rates were observed in the Central region as well, where rural poverty was double urban poverty (22 percent versus 10 percent, respectively). The Kampala subregion stands out with a very low poverty rate of just four percent. Spatial variation in monetary wellbeing can be driven by the concentration of people with certain favorable characteristics in some places and by conditions affecting returns to these characteristics. Sociodemographic characteristics of households were different across rural and urban areas, Kampala and different regions (Table 3). Urban areas had a much higher concentration of smaller households, with lower dependency ratios, and with better educated heads engaged in paid employment. In contrast, rural areas had a much higher concentration of people with less favorable characteristics. Substantial differences exist across regions as well. The households in the Central region and Kampala stand out as having better educated heads, working in paid employment, and having lower household size. These differences in characteristics could explain lower poverty rates in urban areas, Kampala and the Central region. In addition, these areas can also have better infrastructure, lower distances to markets, higher concentration of economic activities and rates of public investment, which can affect the returns to household characteristics such as education, occupations and so forth (called agglomeration effects). In order to understand spatial welfare disparities, it is important to disentangle these two factors. This, in turn, will provide more guidance for the design of poverty alleviation policies in different areas. 26 Mechanized equipment is measured as usage of ploughs, tractors, chain/band saws, shellers, harrows/cultivators, weeders, planters and sprayers during the last 12 months. 49 Figure 84. Headcount poverty rates across regions and rural and urban areas within regions in 2019/20, % Source: UNHS 2019/20, WB staff calculations. Table 3. Key household characteristics across different areas and regions in 2019/20 Areas Regions Uganda Rural Urban Kampala Central Eastern Northern Western Head of household working status, percent Paid employment 59 54 72 84 77 53 46 53 Out of labor force or unemployed 15 14 15 15 10 14 22 14 Subsistence farming 26 32 13 1 13 32 32 33 Head of household education, percent No education 17 19 11 6 12 16 22 20 Primary incomplete 36 42 23 15 28 41 39 40 Primary complete 17 17 17 16 17 15 18 18 Secondary incomplete 20 17 28 33 26 21 15 17 Secondary complete 2 1 5 8 4 2 1 1 Post-secondary, but not university 4 3 7 9 5 4 4 3 University 3 1 8 14 7 2 2 2 Demographics Household size 4.6 4.8 4.1 3.4 4.0 5.2 4.5 4.8 Dependency ratio, percent 45 48 38 27 39 50 48 46 Source: UNHS 2019/20, WB staff calculations. Spatially adjusted consumption per adult equivalent is used to analyze the determinants of spatial welfare variation. In order to compare consumption across different areas, it should be adjusted for spatial variation in prices. There are eight poverty lines constructed for rural and urban areas of each of Uganda’s four regions. These region-specific poverty lines are assumed to incorporate the cost of living differences faced by the poor in different regions and areas. Consumption per adult equivalent was adjusted using region-specific poverty lines and normalized to preserve the mean. To explore the basic factors behind the spatial disparities, regional differences in logarithm of consumption per adult equivalent were decomposed using the Oaxaca-Blinder method (Jann 2008) into the parts associated with (1) differences in household characteristics or endowments, such as demographic composition, education of the head of the household, and access to employment opportunities; and (2) differences in rewards or 50 returns to these characteristics or endowments. The analysis estimates the relative importance of the role of household characteristics, henceforth referred to as the endowments effect, and the role of the associated marginal welfare gains, the coefficient effects, both at the mean and each decile of the consumption distribution. The consumption gap between urban and rural areas was driven largely by better endowments in urban areas. Figure 85 shows the contributions of endowments and rewards to the gap between urban and rural areas. Better endowments play a more important role than rewards in explaining the urban-rural gap. Among endowments, the head of household education explains the largest chunk of differences in consumption, while labor market and demographics contribute to the urban-rural gap to a much lower extent. Figure 85. Urban versus rural and Central versus Figure 86. Urban versus rural and Central versus other other regions: endowments and rewards explaining regions: type of endowments explaining log difference in log difference in consumption per adult equivalent consumption per adult equivalent in 2019/20, percentage in 2019/20, percentage points points Source: UNHS 2019/20, WB staff calculations. The positive gap between the Central region and other regions is driven slightly more by gaps in rewards than gaps in endowments. Figure 85 shows the contributions of endowments and rewards to the gap in consumption between the Central and other regions. Figure 86 looks into the contributions of three broad types of endowments in explaining the consumption gap. Differences in head of household education explained about half of the consumption gap between the Central and other regions. Differences in demographics (household size and dependency ratio) explained a relatively larger share of the welfare gap between the Central and Eastern regions and did not play any role in the gap between the Central and Western regions. Differences in rewards played a more important role in explaining the gap between Kampala and the Eastern and Western regions, while differences in endowments were more important for the gap between Kampala, Central and Northern regions. Figure 87 shows the contributions of endowments and rewards to the positive consumption gap between Kampala and other regions. In all regions, except the Central and Northern regions, the role of rewards prevailed. In other words, the consumption gap between Kampala and the Western and Eastern regions can – to a larger extent – be attributed to rewards rather than better endowments per se. This may signal the existence of agglomeration effects in Kampala, 51 when high density of economic activity, higher concentration of educated population and better infrastructure contribute to higher rewards to endowments. This effect is less pronounced in explaining the gap between the Central region (excluding Kampala) and Kampala. One potential explanation may be related to migration, which equalizes welfare gains within the Central region. Figure 87. Kampala versus other regions: endowments Figure 88. Kampala versus other regions: type of and rewards explaining log difference in consumption endowments explaining log difference in consumption per adult equivalent in 2019/20, percentage points per adult equivalent in 2019/20, percentage points Source: UNHS 2019/20, WB staff calculations. Since the role of household characteristics and associated marginal welfare gains can vary across distributions, the analysis looked beyond averages and explored differences across the entire spectrum of welfare distributions.27 The rural-urban consumption gap was increasing sharply across the distribution, signaling that the poorest in rural and urban areas were not so different. The role of endowments and returns did not change across the distribution though. The consumption gaps between the Central and the Eastern and Western regions were increasing across the distribution, but endowments and returns played a different role. The gap in endowments between households in the Central and Eastern regions was increasing at the top of the distribution, while the dominance of rewards effect in explaining the gap between the Central and the Western regions was mainly due to wealthier households from the top of the distribution. The consumption gap between Kampala and the Central and the Northern regions was the largest for poorer households, while the gap between Kampala and the Eastern and Western regions was relatively flat across the distribution. In each regional gap, the contribution of rewards was higher for the poor, while the contribution of endowments was higher for wealthier households. This means that the poorest in Kampala get higher rewards for their characteristics compared to similar households in other regions, while the richest in Kampala have much better endowments than the richest in other regions, and this plays a more important role in explaining the consumption gap. Finally, education continues to be the main explanator of welfare gaps between Kampala and other regions as well as across the distribution. 27 Results are based on a method proposed by Firpo, Fortin, and Lemieux (2009). 52 Geographic factors behind subregional welfare disparities In this subsection the report explores some of the determinants of poverty at the subregional level in Uganda. Poverty rates from the 2016/17 round of the UNHS instead of 2019/20 round are used for several reasons. First, as discussed earlier, there were gaps in data collection in 2019/20 and if respondents were not properly stratified across time in subregions, that could affect accuracy of subregional poverty. Second, the data from 2019/20 captured the impact of COVID-19, and this shock could affect the underlying relationships between poverty and the determinants of poverty. Third, the sample size of the 2016/17 data is larger than the sample size of the 2019/20 data and can provide more accurate estimates of subregional poverty. Finally, in this exercise the analysis focuses on ranking subregions and the ranking is not expected to change drastically across years unless particular shocks happen in specific areas or major investments were made to bring lagging regions up to par. Figure 89. Headcount poverty rates by subregions in Figure 90. Number of poor by subregions in 2016/17, 2016/17, % people, people Source: UNHS 2016/17, World Bank staff calculations. Source: UNHS 2016/17, World Bank staff calculations. Spatial disparities in poverty are closely linked to spatial gaps in human capital outcomes. Among Uganda’s subregions, there are significant variations in poverty rates, ranging from a low in the Kampala subregion of six percent to a high of 65 percent in Karamoja in the Northern region and 62 percent in Bukedi in the Eastern region, as of 2016/2017 (Figure 89). The largest number of poor people were located in Busoga subregion, followed by Bukedi and West Nile (Figure 90). Just as poverty rates vary substantially among subregions, so do other indicators – many of which are correlated with poverty. Within Uganda, lower access to public services, such as health and education, often results in poorer human development outcomes. In fact, subregions with lower Human Capital Index (HCI), which encompasses stunting, 53 mortality rates, years of schooling, and test scores all in one indicator,28 also tend to have higher poverty rates, as can be seen in the correlation in Figure 92. The difference in the HCI between Kampala and Karamoja is as large as the difference in poverty rates. Just for illustration purposes, the HCI of 0.49 in Kampala is close to the HCI observed in Egypt or Cambodia, while the HCI of 0.29 in Karamoja is close to the HCI observed in Chad or Central African Republic. Figure 91. Subregional Human Capital Index by Figure 92. Correlation between poverty rate in 2016/17 subregions circa 2016/17 and Human Capital Index circa 2016/17 by subregions Source: World Bank 2020b Sources: World Bank 2020b; UNHS 2016/17, World Bank staff Note: calculations. Similarly, subregions with lower levels of basic infrastructure access also exhibit higher poverty rates. The same subregions that lagged in terms of human capital also show lower levels of infrastructure access, including access to improved sanitation and electricity (Figure 93 and Figure 94). These are subregions that also have lower population density, which may make the provision of these services more costly on a per capita basis. In contrast, urban areas like the Kampala subregion benefit from a high population density, allowing public service provision to take advantage of compact population (Figure 99). However, this contributes, inevitably, to the ongoing gap between more urban, richer subregions and more rural, poorer subregions. This suggests that non-monetary poverty and monetary poverty have a significant positive correlation in Uganda, as spatial differences in access both reflect higher poverty rates and simultaneously cause higher poverty rates (Figure 95 and Figure 96). A large swath of the population is disconnected from the main road network and markets, which is correlated with higher poverty rates. Access to markets is not universally spread across Uganda, with the Northern subregions being largely disconnected from other major cities (Figure 97). In fact, in Karamoja 28 The HCI measures the amount of human capital that a child born today can expect to attain by the age of 18. It is constructed based on five constituent indicators: (i) probability of survival to age 5; (ii) expected years of schooling for children; (iii) quality of learning; (iv) adult survival rate; and (v) the proportion of children who are stunted. 54 subregion, about half of the rural population lives farther than two km away from all-season roads.29 Not only does this constrain access to basic public services like schools and health clinics, but it also raises the cost of transportation and limits mobility. For agricultural households, this can reduce their ability to commercialize crops, leaving them to focus on lower-value subsistence crops. It can also limit short-term migration opportunities that would enable households to diversify their income sources. Figure 93. Share of population with access to improved Figure 94. Share of population with access to electricity sanitation by subregions in 2016/17, % by subregions in 2016/17, % Figure 95. Correlation between poverty rate and Figure 96. Correlation between poverty rate in 2016/17 access to improved sanitation by subregions and access to electricity by subregions Source: UNHS 2016/17, World Bank staff calculations. 29 Based on WB staff calculations. See more details on how the share of rural population in proximity to all-season roads is calculated based on publicly available geospatial datasets in Annex 1. 55 Figure 97. Market accessibility index Figure 98. Correlation between poverty rate in 2016/17 and logarithm of market accessibility index Source: World Bank staff calculations (See Annex 1 for Source: UNHS 2016/17 and the Market Access Index based on further details on how the Market Access Index (MAL) World Bank staff calculations (see Annex 1 for further details on is constructed). how the MAI is constructed). Note: Note: Higher values of log of market accesability index means better access. Regions with low access to infrastructure, markets, and low density of population demonstrate low economic development. One way to approximate subregional economic development is by measuring the change in night-time light (NTL) observed over time. As one can see, Kampala and surrounding areas stand out in terms of night-time light and economic development (Figure 100). This is a result of a combination of high population density, better access to infrastructure and markets, and high human capital indicators, which were shown in previous figures, and which contributed to vibrant urban agglomeration effects, economic development, and lower poverty rates. Figure 99. Population density in Uganda in 2020 Figure 100. Night-time light in Uganda in 2020 Kampala Kampala Source: WorldPop. Source: VIIRS annual VNL V2 Elvidge et al. 2021. Many people in Uganda are vulnerable to climatic risks with some particularly pronounced in areas with high poverty rates. Many areas in Uganda are subject to climatic risks, making the population living there vulnerable to poverty. Even though there is no direct relationship between climatic risks and observed poverty rates, some of the poorest subregions in Uganda are particularly vulnerable to weather shocks, namely droughts and flooding (Figure 101 and Figure 102). Indeed, the drought risk is very high in the 56 poorest Karamoja subregion. Similarly, the population living in Karamoja subregion, and its neighboring Elgon subregion, are also prone to flash floods. Figure 101. Drought risk index Figure 102. Share of population exposed to pluvial flood risk, % Source: Index of drought risks. Carrão et al. 2016. Source: Flood risk is measured based on the share of Note: The drought risk index from Carrão et al. 2016 is based people who are exposed to pluvial flood with a water on the assessment of data from the period 2000–2014 and is depth of at least 10cm for a return period of 20 years. a combination of drought hazard, drought exposure, and drought vulnerability. A value of 1 indicates the highest level of drought risk and zero the lowest. VI. Policy implications A key component of sustainable poverty and vulnerability reduction in Uganda is addressing the country’s high susceptibility to climate, health, and forced displacement-related shocks and the exposure of the poor and vulnerable population to these shocks. The significant fluctuations in Uganda’s poverty story over the last decade, highlights the extent to which ongoing shocks have limited sustained poverty reduction in the country on top of slowed down economic growth in per capita terms. A large portion of Ugandans are clustered around the poverty line and are therefore sensitive to sudden changes in external conditions. As such, mitigating the severity of shocks, expanding safety nets for Ugandans and refugees, and developing better coping mechanisms are essential for buffering against the negative impact of shocks. With location playing a major role in determining poverty status in Uganda, targeted policies and investments in lagging regions could help decrease the gap. Poverty in the Eastern and Northern regions, which reached 42 percent and 40 percent in 2019/20 respectively, is almost three times higher than in the Central region, which was about 15 percent. Furthermore, within regions, rural households are also poorer compared to urban households. As such, targeted safety nets and poverty reduction schemes for households in these regions, such as NUSAF3, could be an efficient way to reduce poverty. Furthermore, 57 policies and programs should aim to foster inclusion and cohesion between refugees and host communities by maintaining public service provision and investments for both groups, such as the Displacement Crisis Response Mechanism (DCRM).30 In addition, increasing mobility could enable households from these regions to migrate for jobs in more economically robust regions and help them to contribute to development of lagging regions through remittances. Households working in subsistence agriculture are the most likely to be poor in Uganda, suggesting that addressing the constraints to agricultural growth and enabling workers to transition out of subsistence agriculture could bring about substantial gains in poverty reduction. Among poor households, 77 percent of their household heads worked in subsistence agriculture (compared to 60 percent of household heads overall). Because subsistence agriculture is so susceptible to external shocks (particularly weather), households working in this sector are often poor or vulnerable to falling into poverty. Therefore, increasing the productivity and climate resiliency of agriculture could help mitigate the impact of external shocks on the sector. In addition, creating more economic opportunities outside of subsistence agriculture could enable households to diversify their income sources and transition into more stable employment. The education level of the head of household is found to be the key determinant of poverty and consumption levels in 2019/20, highlighting the imperative to invest in human capital development as a critical pathway for poverty reduction. Across both rural and urban areas, there is a strong negative association between the level of education and poverty rates, with much lower poverty rates for households whose household head had a higher education level. Even having primary incomplete education increases consumption by 18 percent compared to households where the head does not have any education at all. Importantly, spatial decomposition of consumption gaps finds that differences in education levels among the head of household is one of the biggest factors accounting for the urban-rural consumption gap. Therefore, while education investments may take longer to pay off, they are essential to poverty reduction in Uganda. With the country’s rapidly growing young population, these investments are especially pertinent now to shape the future generation’s prospects. Addressing demographic pressures can also contribute to poverty reduction. With high population growth rates, Uganda’s economic growth is still not translating into substantial per capita GDP gains. Furthermore, poor households have significantly larger households with more children. Households in rural areas are particularly vulnerable to these demographic factors as they have more children and, as a result, a higher dependency ratio. Large investments in human capital development, such as the provision of health and education services, are necessary to meet the demands of this burgeoning young population, and it is also an effective strategy to empower women and girls and contribute to lower fertility, thereby reducing poverty as well. 30 The DCRM is a new framework to handle displacement-related shocks. It is a pre-planned and pre-financed mechanism for shock response. The process of disbursing resources for public service provision from the DCRM is agreed upon in advance. Government then selects indicators regarding public service provision and monitors them over time. If the indicators drop below threshold levels, the DCRM rapidly and automatically disburses resources. The public service investments financed by DCRM resources are pre-agreed and include the development of schools, water supplies and health care facilities. 58 References Elvidge, C.D.; Baugh, K.E; Zhizhin, M. & Hsu, F.C. (2013). “Why VIIRS data are superior to DMSP for mapping nighttime lights,” Asia-Pacific Advanced Network 35, vol. 35, p. 62. Elvidge, C.D.; Zhizhin, M.; Ghosh T.; Hsu F.C.; Taneja J. (2021). Annual time series of global VIIRS nighttime lights derived from monthly averages: 2012 to 2019. Remote Sensing, 13(5), p.922, doi:10.3390/rs13050922. Firpo, S.; Fortin, N. M.; & Lemieux, T. (2009). "Unconditional Quantile Regressions," Econometrica, Econometric Society, vol. 77(3), pages 953-973, May. Goodman, S.; BenYishay, A.; Lv, Z.; & Runfola, D. (2019). GeoQuery: Integrating HPC systems and public web-based geospatial data tools. Computers & Geosciences, 122, 103-112. Jann, B. (2008). "The Blinder–Oaxaca decomposition for linear regression models," Stata Journal, StataCorp LP, vol. 8(4), pages 453-479, December Jedwab, R. & Storeygard, A. (2022). “The Average and Heterogenous Effects of Transportation Investments: Evidence from sub-Saharan Africa 1960-2010.” Journal of the European Economic Association, 20(1): 1–38, https://doi.org/10.1093/jeea/jvab027. Weiss D.J.; Nelson A.; Gibson H.S.; Temperley W.H.; Peedell S.; Lieber A.; Hancher M.; Poyart E.; Belchior S.; Fullman N.; Mappin B.; Dalrymple U.; Rozier J.; Lucas T.C.D.; Howes R.E.; Tusting L.S.; Kang S.Y.; Cameron E.; Bisanzio D.; Battle K.E.; Bhatt S.; & Gething P.W. (2018). A global map of travel time to cities to assess inequalities in accessibility in 2015. Nature. January 2018 553: 333–336. World Bank Group. (2015). Uganda Systematic Country Diagnostic: Boosting Inclusive Growth and Accelerating Poverty Reduction. World Bank, Washington, DC. World Bank. (2016). The Uganda Poverty Assessment Report 2016: Farms, Cities and Good Fortune – Assessing Poverty Reduction in Uganda from 2006 to 2013. World Bank, Washington, DC World Bank. (2019). Detour from the Poverty Reduction Path: Uganda Poverty Update Note – Uganda National Household Survey 2016-17. World Bank, Washington, DC. World Bank. (2020a). Tackling the Demographic Challenge in Uganda. World Bank, Washington, DC. World Bank. (2020b). Uganda Economic Update, 14th Edition, February 2020: Strengthening Social Protection to Reduce Vulnerability and Promote Inclusive Growth. World Bank, Washington, DC. World Bank. World Bank; International Finance Corporation; Multilateral Investment Guarantee Agency. (2021). Uganda Systematic Country Diagnostic Update. World Bank, Washington, DC. World Bank (2022). Growth, trade, and transformation. Country Economic Memorandum for the Republic of Uganda. World Bank, Washington, DC. 59 Annex 1 Figure A1. Community mobility trends in Uganda using Google mobility report and the 5-week period Jan 3–Feb 6, 2020 as a baseline Source: Google LLC "Google COVID-19 Community Mobility Reports." https://www.google.com/covid19/mobility/ Accessed: February, 2022. Figure A2. Population structure of areas and regions Figure A3. Population structure of areas and regions in in the UNHS before and during COVID-19 using the UNHS before and during COVID-19 using adjusted original weights, % weights, % Source: UNHS 2019/20, WB staff calculations. 60 Box 4. Can Uganda’s oil resources contribute to poverty reduction?31 In the medium term, fiscal revenues from oil production can boost Uganda’s fiscal space significantly. Assuming no further delays, the first production of oil in Uganda is expected in 2025, with all three fields becoming operational. Oil production is projected to quickly reach its peak at 230,000 barrels per day, and then, in the absence of further discoveries, gradually decline after 2030. Assuming an oil price of USD 70 per barrel throughout in real terms, analysis suggests that fiscal revenues from oil production can reach about USD 3.3 billion per year during peak production (2030). This would likely correspond to about 4.9 percent of GDP and about a quarter of fiscal revenues from other sources at that time. Today, that revenue would cover a large share of the deficit (net borrowing) in the fiscal budget. Therefore, although Uganda’s hydrocarbon reserves are by no means large, the fiscal revenues they may generate can help accelerate the country’s economic and social developments. If used well, the additional revenues from hydrocarbons can help close Uganda’s human development and infrastructure gaps. Better provision of infrastructure services, like transportation and electricity, will be a crucial determinant of productivity growth and structural transformation. While the authorities have already begun to support oil production and its commercialization by undertaking investments in critical infrastructure such as roads, investments in human development are still yet to be seen once oil revenues start to flow. Another important fiscal challenge is the efficient spending of oil revenues, considering the economy’s absorptive capacity constraints. In the long term, oil revenue inflows can also pose structural challenges to the Ugandan economy. While the fiscal revenues from oil can help build much needed physical and human capital in Uganda, the economic gains from this process are neither guaranteed nor automatically generated. Most importantly, adverse economic impacts that may be generated by windfall revenues can limit the competitiveness of Uganda’s tradable sectors (a phenomenon known as Dutch disease) and lower the country’s growth potential (a “resource curse”). Furthermore, the actual development of Uganda’s oil reserves carries a number of risks, which could j eopardize the potential poverty-reducing effect. A delay in production, dissolution of prospects, or a sharp decrease in oil prices can all distort the fiscal outlook drastically. Even if oil production remains economically viable in the long term, with a narrow fiscal space and frontloaded capex requirements at the outset, the government can face financial bottlenecks. Once the oil starts flowing, one of the most important fiscal challenges becomes the volatility of oil revenues. Such extreme volatility and uncertainty can be very harmful if transmitted to the non- oil economy through sudden adjustments in public expenditures. Finally, the remoteness and environmental sensitivity of the Lake Albert region – where the oil reserves are located – heightens the probability of negative environmental impact. In addition to the likely ethical concerns regarding biodiversity, these impacts can also directly reduce the country’s tourism potential. Oil revenues provide an excellent opportunity to extricate Uganda from a low equilibrium by enhancing infrastructure, building human capital, and promoting private economic activity. However, capitalizing on this opportunity is not easy. First, the authorities will need to get the fiscal management right to achieve stabilization and efficiency objectives. Second, policies will need to target conditions that can reinforce Dutch disease dynamics in the economy. These include trade barriers (i.e., difficulties in exporting final products or importing intermediary inputs), non-inclusive prospects for economic participation, and institutional weaknesses that can facilitate rent-seeking, among others. Third, policies should respond effectively to concerns about graft, environmental impact, lagging regions, and disadvantaged groups to preempt grievances and social unrest. In the final analysis, there is only one way to make the most out of oil, and that is to get these policies right. 31 This box draws heavily from World Bank (2022). 61 Triangulating monetary poverty trends since 2016/17 Subjective wellbeing Subjective wellbeing indicators showed much more volatility than poverty indicators in all geographic areas. Given that the developments of monetary wellbeing before the COVID-19 shock are not fully consistent between the UNHS and UNPS data, results for selected subjective indicators from the UNHS data are reported as well. The first indicator demonstrates the share of population who experienced a decline in living standards during the last 12 months by rural and urban areas in Figure A4, while Figure A5 shows the same indicator for different regions. Indicators are reported for 2016/17, the pre-COVID-19 and during COVID-19 periods of 2019/20. Regardless of the geographic location used (except for the Western region), the share of the population with decreased living standards had a pronounced U-shape. It fell before the COVID-19 pandemic but increased sharply during the COVID-19 period. Thus, in 2016/17 about half of the population in Uganda reported worsening living standards and this share declined to 31 percent in 2019/20 before the COVID-19 period but then increased again to 50 percent during the COVID- 19 period. The reduction of the share of the population with worse living standards in urban areas and the Northern and Central regions after 2016/17 and before COVID-19, is different from the stagnant poverty rates observed in the UNHS data and is more consistent with growing GDP per capita in that period. The second indicator shows the share of the population who reported household income being very unstable during the last 12 months (Figure A6 and Figure A7). It also has a U-shape but is less pronounced compared to the first indicator on the changes in living standards. In particular, the share of the population with very unstable household income dropped before COVID-19 by about five percentage points but increased rapidly afterwards by ten percentage points. This trend was particularly distinct in urban areas, which did not show any change in monetary poverty using the UNHS data. Figure A4. Share of the population with decreased Figure A5. Share of the population with decreased living standards during the last 12 months across rural living standards during the last 12 months across and urban areas during 2016/17–2019/20, % regions in 2019/20 during 2016/17–2019/20, % Source: UNHS, WB staff calculations. Differences between UNHS monetary poverty trends and the trends in subjective wellbeing indicators may be related to conceptual differences between monetary and subjective wellbeing indicators. They 62 can also be related to the ability of households to smooth consumption during the time of shocks, with subjective wellbeing indicators being more sensitive to the short-term impact on wellbeing. The inconsistency may also be related to comparability issues – driven by differences in the fieldwork in the UNHS 2016/17 and UNHS 2019/20 data collection. This factor might have also affected the analysis of poverty before and during the COVID-19 period. Figure A6. Share of the population with income being Figure A7. Share of the population with income being very unstable during the last 12 months across rural very unstable during the last 12 months across regions and urban areas during 2016/17–2019/20, % in 2019/20 during 2016/17–2019/20, % Source: UNHS, WB staff calculations. Geospatial data Compared to the 2016/17 survey period, weather conditions were more favorable for agricultural production in all regions in 2019/20. Weather is an important determinant of poverty in Uganda, given the strong link between weather patterns, agricultural production, and rural household incomes. Thus, the drought was one of the main factors behind the increase in poverty in Uganda in 2016/17 (World Bank 2019). Mean monthly values of the Standardised Precipitation-Evapotranspiration Index (SPEI) were used to check regional weather conditions. Figure A8 shows the distribution of months with dry weather defined as when the SPEI index was below -0.5. In 2016, dry weather was prevalent in more than 75 percent of all months (the Northern region was an exception), and this share dropped substantially afterwards. The decline in the episodes of dry weather from 2017–2019 is expected to have a poverty reducing impact, particularly in the poorest Northern and Eastern regions, which rely heavily on agricultural activities – especially among the poor population. Nighttime lights also indicated positive economic development in Uganda since 2016 and before COVID- 19. The second indicator used to triangulate the trends in monetary wellbeing was based on monthly Visible Infrared Imaging Radiometer Suit (VIIRS) Nighttime Lights satellite images. There are many studies showing that the data on nightlight intensity is correlated with economic activity. Therefore, this indicator is used as a proxy of economic activities after 2016 in four regions. Figure A9 shows yearly growth rates in the average nightlights indicator across regions. There was a reduction in nightlight intensity in 2016 in all regions except the Northern region. This was consistent with lower GDP per capita in 2016/17 and the increase in poverty everywhere, except the Northern region. After 2016, there seems to be a rebound in 63 economic activity in all regions as reflected by the higher intensity of night lights. Intensity of night lights continued to grow after 2017 as well, but at a much slower pace, and halted in 2020 during the pandemic. Geospatial data points out to a rather universal growth in economic activities after 2016. It is important to mention though that improved economic activities do not necessarily have a welfare enhancing effect on the poor. Still this indirect measure, along with improved weather conditions, provide some support to the view that poverty most likely fell in all regions in Uganda before COVID-19. Figure A8. Share of dry months using SPEI across Figure A9. Growth rates in the intensity of night lights regions during 2016–2020, % across regions during 2016–2020, % Source: SPEI global database accessed through Goodman et al. Source: Elvidge, Zhizhin, Hsu. 2013; Goodman et al. 2019. 2019, WB staff calculations. Note: Month is dry is SPEI mean value <-0.5 Market Access Index In this report, market access is defined as a measure of accessibility from one origin to all destinations based on travel distance (or travel time). More formally, market access for a given location (or origin) i can be expressed as follows: = ∑ − where refers to the population of a location (or destination) d, is travel time from cell o to destination d, and is a trade elasticity or decay parameter measuring how trade volumes fall as travel times increase. Following Jedwab and Storeygard (2022), is set at 3.8. The origins are defined as the centroids of 0.0833 x 0.0833 degrees (roughly 10km) grid cells defined over the territory of Uganda. Destinations that are considered in the construction of this index include all cities with a population greater than 100,000.32 In other words, market access is the weighted sum of population in all the destinations, which are weighted by travel time/distance. Travel time is computed based on the friction map generated by Weiss et al. (2018). Finally, to aggregate this index to the sub-region level, the weighted mean of the index value by subregion is computed using grid-level population data from WorldPop as weights. 32 We used the projected city population data in 2020 from https://www.citypopulation.de/. 64 Rural Access Index The Rural Access Index (RAI) measures the share of rural population living within 2km away from all- season roads. To construct this index, three sources of data are used: OpenStreetMap, WorldPop 2015 population estimates, and GRUMP Global Rural-Urban Mapping Project, Version 1 (GRUMPv1) (CIESIN, Columbia University, CUNY, CIDR, IFPRI, and CIAT 2017). The methodology as laid out in https://rai.azavea.com/ is used: • Select commonly used tags from OpenStreetMap (Trunk, Primary, Secondary, Tertiary) that serve as an approximation for all-season roads • Create a mask based on 2 km buffer on these roads • Create a mask based on urban areas as defined by GRUMP urban extents polygons • Summarize the population remaining on the 100-meter raster dataset from WorldPop 2020. Box 5. Main data, concepts and definitions used in the Poverty Assessment Main data The Uganda National Household Survey (UNHS) is the main source of data for the calculation of official poverty rates in the country. The UNHS was conducted in 1999/00, 2002/03, 2005/06, 2009/10, 2012/13, 2016/17 and 2019/20. The most recent UNHS samples were designed to allow for the generation of representative estimates at the national level, for urban and rural areas and for the 15 sub-regions of Uganda. The poverty assessment relies on the UNHS to report official poverty and inequality numbers, analyze internal migration, measure vulnerability to poverty, construct the human opportunity index, and simulate the impact of competition in the telecommunications sector on poverty. The latest round of the UNHS in 2019/20 was affected by the COVID-19 pandemic. The field work was interrupted during the lockdown, which effectively fully omitted the second quarter (April–June 2020). The implications of this are discussed in the main text of the report. The Uganda National Panel Survey (UNPS) is another annual multi-topic household survey that commenced in 2009/10. The UNPS 2009/10 was followed by additional rounds of data collection in 2010/11, 2011/12 and 2013/14, 2015/16, 2017/18, 2018/19 and 2019/20 rounds. Similar to the UNHS, the UNPS collects detailed consumption data. In addition, the UNPS collects detailed and high-quality information on income and agricultural activities. It provides representative estimates at the national, rural-urban and regional levels. The poverty assessment relies on this survey to measure chronic and transient poverty, explore the impact of shocks and the distribution of coping strategies, and finally to analyze the determinants of agricultural development in the country by measuring utilization of fertilizers, improved seeds, and extensions services. The annual Labor Force Survey (LFS) is specifically designed to construct representative labor market indicators. Stand-alone labor force surveys were conducted in 2012, 2016/17, 2017/18 and 2018/19. The sample was nationally representative with three main reporting domains: national, rural/urban, and disaggregated by sex. Where particular indicators had sufficient data, further disaggregation was made by age groups and other characteristics. The poverty assessment relies on the LFS to analyze employment trends and structural changes in the economy. In June 2020, the Uganda Bureau of Statistics (UBOS), with the support from the World Bank, officially launched the High Frequency Phone Survey (HFPS) to track the impacts of the COVID-19 pandemic monthly for a period of 12 months. The survey aimed to recontact the entire sample of households that had been interviewed during the Uganda National Panel Survey (UNPS) 2019/20 round and that had phone numbers for at least one household member or reference individual. The first round (baseline) of the survey was conducted in June 2020 and interviewed 2,227 households. A subsequent six rounds attempted to reach the same households with the sample 65 size in the latest seventh round of 1,950 observations. The poverty assessment relies on the HFPS to measure impact of COVID-19 on Ugandan population. The World Bank (WB) in collaboration with the Uganda Bureau of Statistics (UBOS) and the United Nations High Commissioner for Refugees (UNHCR) launched and conducted the Refugee High Frequency Phone Survey (RHFPS). The RHFPS tracked the impacts of the pandemic between October 2020 and March 2021. The survey sample includes respondents with active phone numbers that were selected randomly from the Profile Global Registration System (ProGres) of UNHCR, and the refugee household survey carried out by UBOS and the World Bank in 2018. The sample was representative at seven strata constructed as a combination of regions and different countries of origin. In order to reduce the bias related to only interviewing households with phone numbers and non-response, the data from the 2018 representative refugee household survey was used to produce and calibrate the weights for all three rounds of the phone survey. Main concepts and definitions Poverty is measured by comparing welfare measure with some defined threshold called poverty line below which an individual is considered poor. The headcount ratio is the proportion of the population that is classified as poor. The poverty line indicates the minimum level of welfare required for healthy living. Healthy living is usually defined as being able to afford the food covering the minimum required caloric intake and additional non-food items. Updated poverty line used in this poverty assessment is based on the UNHS 2016/17. Depth of poverty or poverty gap indicates how far off an average poor household is from the poverty line. It captures the mean consumption shortfall relative to the poverty line across the whole population. Poverty severity or squared poverty gap considers the distance separating the poor from the poverty line (the poverty gap) as well as the inequality among the poor. Conceptually, poverty severity puts a higher weight on households/individuals, who are further below the poverty line. The Gini coefficient is a measure of inequality. A Gini coefficient of 0 indicates perfect equality while 1 signifies complete inequality. The welfare measure or welfare aggerate in this study is measured by consumption per adult equivalent as calculated by the Uganda Bureau of Statistics unless stated otherwise. For inequality measures and growth rates, consumption and income aggregates were spatially deflated by regional poverty lines to account for differences in prices across different areas unless stated otherwise. Quintiles are based on consumption per adult equivalent, which was spatially deflated by regional poverty lines unless stated otherwise. Improved sanitation includes access to a flush toilet, a VIP latrine, a covered pit latrine with a slab, an uncovered pit latrine with a slab, and a compost toilet. All access should be unshared with other households. Improved drinking water includes access to piped water, borehole, protected well/spring, gravity flow scheme, rain and bottled water. 66 Chapter 2. Most recent trends in the labor market and structural change I. Economic growth and structural change Overall, the share of the working population employed in the agricultural sector in Uganda declined over the last decade. Uganda has multiple surveys that collect labor market information, including the Labor Force Survey (LFS), the Uganda National Household Survey (UNHS), and the Uganda National Panel Survey (UNPS). In addition, there are modelled estimates provided by the International Labor Organization (ILO) and accessible in the World Development Indicators (WDI). These different sources provide varying results on the degree and pace of structural change. Indeed, there is a substantial mismatch in the levels and trends for indicators, such as the share of the working population employed in the agricultural sector. According to the LFS and the UNPS data, the share of the working age population that is employed in agriculture fell by almost 10 percentage points during the period from 2011/12 to 2019/20, while the WDI numbers show a flat trend after 2013. The UNHS data for 2019/20 is difficult to interpret in terms of structural change because it covered the COVID-19 period. That said, between the 2012/13 and the 2016/17 rounds of the UNHS, the share of employment in agriculture fell as well. Despite the substantial decline in the share of agricultural employment, GDP per capita growth during this period was very modest. In order to benchmark Uganda’s performance and better situate the structural change in employment in the country, the report selected countries with GDP per capita in 2017 PPP in constant prices similar to Uganda’s in 2011. The analysis compares the growth rate of the share of agriculture in total employment as well as the real GDP per capita growth rate for the same period. Both indicators are shown in Figure 103. It is important to note that for Uganda, the report uses the growth in the share of employment from the Labor Force Survey (LFS) microdata rather than the WDI for comparison since LFS data are designed to provide a more accurate representation of labor market performance.33 Clearly, the rate of decline of the share of agricultural employment in Uganda was faster than in most comparators – 2.3 percent annually between 2011/12 and 2018/19. However, Uganda’s performance in economic growth lagged behind, with the annualized growth rate of GDP per capita close to only one percent. 33 The same concern could be applicable to other countries as well but calculating shares of agricultural employment from LFS for all comparators is beyond the scope of this work. 67 Figure 103. Annualized growth rates of employment share in agriculture and GDP per capita, % Source: LFS and WDI (modelled estimates). World Bank staff calculations. Agriculture was the only sector showing slight Figure 104. Value added per working person by signs of increasing productivity after 2016/17 due economic sectors, million UGX in constant 2016/17 to resumed economic growth and labor moving prices to services and industry sectors. Figure 104 shows the value added per worker in the three sectors (agriculture, industry, and services) in constant prices in millions of UGX over four periods of time. The agricultural sector had the lowest productivity level compared to services and industry. However, it was the only sector in which productivity increased after 2016/17 and which potentially continued to grow in 2019/20 when value added in agriculture increased in real terms by about five percent (lack of employment figures for this period prevents us from checking this). At the same time, productivity in services and industry Source: UBOS for value added and LFS for the number of sectors, which absorbed excessive labor in working people in each sector. Note: Number of workers is based on age 14–64 and includes agriculture, continued to decline across all years. those in subsistence agriculture. Lagging initial conditions in human capital and lower access to infrastructure might have contributed to lower GDP per capita growth rate in Uganda during the last decade. To understand the correlates of economic performance, report analyzes GDP per capita growth rates versus several socio-economic indicators in the initial period: completion of lower secondary school and access to electricity. The initial period is country specific and selected for each country when it had GDP per capita in PPP terms close to the GDP per capita in PPP terms of Uganda in 2011. GDP per capita growth rates are higher in countries with higher initial access to electricity and higher initial completion rates of lower secondary school (Figure 105 and Figure 106). Low electricity access and lower secondary school completions rates observed in Uganda in 2011/12 might be correlated with the modest economic performance. Furthermore, access to 68 electricity increased significantly during 2011/12 and 2018/19, but no change was observed in the lower secondary school completion rate, and this lack of human capital accumulation may have also contributed to slower economic growth rates. Given Uganda’s agro-climatic conditions and the importance of the agricultural sector in the economy, weather shocks also affected Uganda’s economic performance in this period. As shown in Figure 107, there was a significant negative correlation in Uganda between the value added in agriculture in per capita terms and the share of months with dry weather from 2011 to 2020.34 Indeed, the lowest value added in agriculture per capita was observed in 2016 – a year in which about 75 percent of all months were dry. In contrast, in 2020 the value added in agriculture was the highest and it also coincided with very favorable weather conditions as shown by the fewest number of dry months. The correlation becomes even stronger if the value added from fishing is subtracted from the aggregate for agriculture value added overall, since the value added in fishing is positively correlated with dry weather. Modest GDP per capita growth in Uganda was also affected by high fertility rates and population growth. High fertility rates require higher economic growth to sustain positive economic growth in per capita terms. However, the average annual GDP growth rate in Uganda during the period from 2011 to 2018 was about two percent, which was lower than in most comparators. In contrast, the fertility rate in Uganda in 2011 (Figure 108) was one of the highest among selected countries and the average population growth from 2010 to 2019 was high and increasing, dampening the impacts of GDP growth in per capita terms. In addition, fertility rates in Uganda are strongly correlated with the socio-economic status of a household, with almost two times higher fertility rates for women in the lowest wealth quintile compared to those in the highest wealth quintile – 7.1 children versus 3.8 children respectively (UBOS and ICF 2018). This limits the impact of economic growth on poverty reduction yet further. Figure 105. Access to electricity in the initial period and Figure 106. Completion of lower secondary school in GDP per capita growth rates in Uganda and the initial period and GDP per capita growth rates in comparators, % Uganda and comparators, % Source: WDI, World Bank staff calculations. Note: Each country has its own initial period and the period of GDP per capita growth rate (Figure 103). 34 This is measured by the Standardised Precipitation-Evapotranspiration Index (SPEI) taking mean value below -0.5. 69 Figure 107. Value added in agriculture in per capita terms Figure 108. Fertility rate in the initial period and GDP in UGX constant prices and the share of dry months in per capita growth rates in Uganda and comparators, Uganda based on SPEI index in % births per woman and % Source: Global SPEI database accessed in October 2021 through Source: WDI, World Bank staff calculations. Goodman et al. 2019 and UBOS. Note: Each country has its own initial period and the period Note: Value added was taken for fiscal year to account for lagged of GDP per capita growth rate (Figure 103). impact of weather. For example, the observation for the year of 2006 was based on value added from 2006/2007, while SPEI index was based on 2006 year. II. The distributional aspects of structural change This section analyzes the key labor market indicators in Uganda, with a particular focus on structural change. The analysis is based on LFS datasets. Annual LFS datasets have the most comprehensive labor module based on the 7-day recall and are specifically designed to construct representative labor market indicators. Stand-alone labor force surveys were conducted in 2011/12, 2016/17, 2017/18, and 2018/19. Data was collected from all household members ages five and above to enable compilation of child labor statistics. The analysis is based on key indicators constructed using the same consistent methodology across rounds. Constructed numbers could differ from official Uganda Bureau of Statistics (UBOS) numbers,35 but trends remain qualitatively the same. From 2011/12 to 2018/19, the pace of growth in the number of people working in paid employment was much higher than those working in subsistence agriculture. Labor force statistics in Uganda separate two mutually exclusive types of working population: those engaged in paid employment and those engaged in subsistence agriculture. Subsistence agriculture is defined as work done on a household farm and the products produced are used mainly or only for own or family use. Any other employment is considered to be paid employment. Figure 109 shows the trend in the number of working age people (14- 64), those in paid employment, subsistence farming, and those not working (out of labor force or unemployed) during 2011/12 and 2018/19. The growth rates for the same indicators are shown in Figure 110. There are several important observations. The pace of job creation was identical to the pace of growth in the working age population, and more than 80 percent of the working age population was working. However, the pace of growth in paid employment and the pace of growth in subsistence 35 One potential reason for the divergence can be related to the definitions used to construct key labor market indicators. The report uses consistent definitions across years. 70 agriculture diverged substantially. The number of people working in subsistence agriculture grew by only nine percent, compared to 58 percent for paid employment. Figure 109. Working age population and working in Figure 110. Growth in working age population and paid employment and subsistence agriculture in working in paid employment and subsistence 2012/12 and 2018/19, millions of people agriculture in 2018/19 using 2012/12 as a base Source: LFS, World Bank staff calculations. An individual’s type of work is closely correlated with their socio-economic status. Figure 111 to Figure 113 show the structure of the working age population across areas, gender, education level, and age during 2011/12 and 2018/19. Overall, those more likely to have paid employment were urban residents, men, individuals above 24 years old, and those with higher levels of education. In contrast, rural residents, youth (ages 14 to 24 years), women, and those without formal education were more likely to be solely engaged in subsistence farming as their primary employment. The pace of structural change was faster among rural residents, men, elder individuals and those with at least some level of formal education. The structural change during the 2011/12 to 2018/19 period – measured by a falling share of subsistence agriculture and an increasing share of paid employment – was also pronounced among groups that already had higher levels of paid employment, with one notable exception – rural areas. Overall, the share of paid employment increased from 41 percent to 48 percent in rural areas and remained about 60 percent in urban areas. The share of paid employment among the youth (14 to 24 years of age) increased only by four percentage points from 31 to 35 percent, while among elder cohorts the growth was much higher. For example, the share of paid employment increased from 47 percent in 2011/12 to 60 percent in 2018/29 among those aged 44 to 64. The share of paid employment increased faster for men, widening even more the gender gap in favor of men. Working-age individuals with no education were the only group for whom the structure of work did not change in this period. Striking differences in the pace of structural change were observed across regions. Two of the poorest regions, the Eastern and the Northern ones, had the lowest share of paid employment in 2011/12: 32 percent and 40 percent, respectively. The share of subsistence farming fell in all regions, except for in the Eastern region, where it remained at about 50 percent. The fastest decline in subsistence farming was observed in the Northern region. It is not entirely clear what contributed to such a sharp reduction, but potential explanations include the high level of social assistance programs and donor support in this region in recent years. 71 Figure 111. Structure of the working age population by Figure 112. Structure of the working age population by type of working status across areas and time, % type of working status across education level and time, % Source: LFS, World Bank staff calculations. Note: The working age population is defined as individuals ages 14-64 years. Figure 113. Structure of the working age population by Figure 114. Structure of the working age population by type of working status across age groups and time, % type of working status across gender and time, % Source: LFS 2011/12 and 2018/19, World Bank staff calculations. Note: The working age population is defined as individuals ages 14-64 years. The role of agriculture declined as well following reduction in subsistence farming. For example, agriculture declined from 72 percent in 2011/12 to 61 percent in 2018/19, while the share of people working in industry and construction doubled from five percent to ten percent, and the share of those working in services increased from 21 to 29 percent. Interestingly, the share of people working in agriculture increased in urban areas at the expense of the services sector, while the share of people working in industry and construction remained the same over this period. Similar patterns were observed among those who engaged in paid employment only. Thus, the share of agriculture among paid employment declined from 48 percent in 2011/12 to 37 percent in 2018/19. Only for those living in urban areas did the share of agriculture in paid employment increase over time. Agriculture remained the sector with the lowest earnings among paid employees and the lowest growth rate in earnings during 2011/12 and 2018/19. Those working as paid employees were asked about their earnings (including commissions and earnings in kind). Median values in 2011/12 prices are reported in 72 Figure 115 for two rounds of data without any adjustment for hours worked. Paid employees who were female – working in agriculture, with lower than a primary complete education, belonging to the youth age group (14 to 24 years) – had the lowest earnings.36 The gap between the median earnings in agriculture and those in services and industry sectors during 2011/12 and 2019/20 increased because of stagnant earnings in agriculture and growing earnings in the services and industry sectors. The gap has also increased between those who have completed a primary education and those who have not. The gender gap in median earnings has narrowed. The largest growth in earnings was registered for employees with complete primary education and those ages 45 to 64 years. Earnings have also increased in all parts of the country, except in the Eastern region. Figure 115. Median total monthly earnings among employees in 2011/12 and 2019/20, UGX in 2011/12 prices Source: LFS 2011/12 and 2019/20, World Bank staff calculations. Note: Inflation during 2011/12 and 2018/19 was about 40 percent. 36 There was a large overlap between these categories. Thus, there were more younger individuals without education among those employees in agricultural sector. 73 Figure 116. Median total earnings per hour worked among employees in 2011/12 and 2019/20, UGX in 2011/12 prices Source: LFS 2011/12 and 2018/19, World Bank staff calculations. Note: Inflation during 2011/12 and 2018/19 was about 40 percent. Once controlled for hours worked, the sectoral gap between earnings among paid employees narrows substantially, but the trends during 2011/12 and 2018/19 do not change. It is important to check the earnings adjusted by actual hours worked because paid employees in agriculture and women may work significantly lower hours than those in other sectors and men. Indeed, paid employees in agriculture worked about 30 hours per week in 2018/19 compared to 53 hours worked in industry and 54 hours worked in services sectors. The gender gap between average weekly hours worked across all sections was much smaller (50 hours versus 47 hours in favor of men). Adjusting for hours worked significantly reduced the earnings gap between agriculture and industry and services sector (Figure 116). Once adjusted for hours, the gender gap in earnings increased slightly during 2011/12 and 2018/19. The trends in earnings per hour have not changed much. Similar to previous findings, the growth in earning per hour remained the lowest among paid employees working in agriculture, the youth, women and those living in the Eastern region. Despite falling share of employment in agriculture, income growth from non-agricultural activities was not universal and mostly observed for non-agricultural wages. Data come from Rural Livelihood Information System (RULIS) harmonized datasets constructed from the UNPS for 2013/14 and 2019/20 show evolution of income from different sources (Figure 117). The largest growth was observed for crop income, which was faster for the bottom 40 percent of the population than for the average. Non- agriculture wage income grew as well, but at a slower pace. The bottom 40 percent of the population demonstrated lower growth rates for this source of income. Income from livestock and agricultural wages demonstrated negative growth. Finally, the growth of non-farm self-employment income was negative for the total population and slightly positive for the bottom 40 percent. 74 Figure 117. Annualized real growth rates of labor income per capita by sources and across population groups between 2013/14 and 2019/20, % Source: RULIS database based on the UNPS 2013/14 and 2019/20, World Bank staff calculations. Note: Bottom 40 percent of population is based on consumption per adult equivalent spatially adjusted by poverty line. Income aggregates were spatially adjusted as well. Wealthier individuals experienced faster rates of structural change during the period from 2013/14 to 2019/20. One important question related to structural change is who benefits from it. As shown already, individuals without formal education were more likely to remain in subsistence agriculture. The UNPS data allows a look at the share of employment in agriculture across consumption per adult equivalent quintiles. As shown in Figure 118, the share of employment in agriculture among the working population declined across all quintiles from 2013/14 to 2019/20, but the rate of decline was higher among individuals from the wealthier quintiles. This may be related to constraints that the poorest face when trying to enter the non-agriculture sectors, such as low levels of human capital, limited access to infrastructure and markets, the up-front costs of transitioning from one sector to another, and so forth. Those households where the head switched from the agricultural sector to the non-agricultural sector benefited from substantial increases in consumption. This is evident from the panel component of the 2015/16 and 2019/20 rounds of the UNPS which allows comparing consumption per adult equivalent changes in households depending on the household head’s sector of employment. It turns out that households where the household head switched from the agricultural sector in 2015/16 to the non- agricultural sector in 2019/20 demonstrated the highest annualized consumption growth rate (Figure 119). 75 Figure 118. Share of agriculture in employment by Figure 119. Consumption per adult equivalent and its consumption per adult equivalent quintiles, % annual growth rate in 2015/15 and 2019/20 depending on head of household sector of employment, % Source: UNPS, World Bank staff calculations. Source: UNPS, World Bank staff calculations. Note: Cross-section weights are used. Consumption is Note: Panel population weights are used. spatially adjusted by using poverty lines. III. Impacts of COVID-19 on the labor market To date, Uganda has implemented two main lockdowns since the start of the COVID-19 pandemic. The timeline of the COVID-19 pandemic in Uganda includes the following key dates. During March 18–20 in 2020, all mass gatherings were suspended in Uganda and schools were closed for in-person learning. A National curfew was announced on March 30, 2020. Restrictions were eased on May 26th, 2020, allowing private cars with only three people and general merchandise shops to open with strict social distancing. Vaccination campaigns began in March 2021. Uganda re-entered a partial lockdown starting on June 7th, 2021. The second lockdown was slightly longer than the first one, but less strict. A presidential directive banned travel between districts, restricted gatherings, and closed selected classes which were opened in early 2021. Figure 120 shows the COVID-19 government response stringency index and the cumulative number of COVID-19 cases in the country since the beginning of 2020. In this section, the report explores the impacts of COVID-19 on the labor market using results from the high-frequency phone surveys (HFPS), supported by the World Bank. The first and seventh rounds of the HFPS were conducted after the first and second lockdowns, respectively, which allow us to compare their impacts on the labor market.37 37 Besides COVID-19 related restrictions, Uganda experienced droughts in most parts of the country in the first agricultural season of 2021 (March-May) negatively affecting germination, growth, and harvest of crops in particular cereals and legumes. Detailed analysis of the impact of COVID-19 on Ugandans can be found in a range of publications (World Bank 2020a; World Bank 2020b; World Bank 2020c; World Bank 2021a; World Bank 2021b; World Bank 2021c). Comparing the impact of COVID-19 on refugees and Ugandans is done by Atamanov et al. (2021). 76 Figure 120. Stringency index and cumulative number of COVID-19 cases in Uganda from January 2020 to December 2021 Source: WHO, Oxford COVID-19 Government Response Tracker, Blavatnik School of Government, University of Oxford. Note: Stringency index varies from zero to 100 with higher values meaning more stringent government policies. Significant work stoppages were observed after both lockdowns, but with different groups affected. Figure 121 shows the percentage of working respondents during the seven rounds of the HFPS (only those respondents participating in all rounds were kept for this analysis). The share of working respondents fell from 87 percent before March 2020 to 71 percent in June 2020. The most affected were those living in urban areas and working in the services sector before the lockdown, which is consistent with the more stringent restrictions in Kampala and urban areas and a lesser reliance on family farming in these areas (Figure 122). Thus, 28 percent of those working in urban areas before the first lockdown stopped working compared to 14 percent in rural areas. About 32 percent of those working in the services sector stopped working after the first lockdown compared to eight percent of those working in the agricultural sector. Employment recovered by July/August 2020 and remained high until the second lockdown introduced in June 2021. The seventh round of the HFPS was conducted after the second lockdown and is likely to capture its impact. It showed that the share of working respondents fell from 92 percent in March/April 2021 to 81 percent in October/November 2021, but work stoppages were more universally distributed across areas and sectors compared to the first lockdown. Female respondents were more likely than male respondents to stop working after the first lockdown. Based on the data from the first round conducted in June 2020, work stoppages after March 2020, when the first lockdown was introduced, were significantly higher among female respondents. The share of working male respondents declined by 19 percent in June 2020 compared to the pre-March 2020 level, while the share of working female respondents declined by 29 percent. The HFPS did not observe a gender difference in work stoppages after the second lockdown in June 2021. COVID-19 was not the only shock affecting Ugandans, and weather seemed to also contribute to work stoppages in the period after the second lockdown. Figure 122 shows the structure of work stoppages by the economic sector in which respondents were working. In June 2020, almost 80 percent of all work stoppages happened in the non-agricultural sector – particularly in commerce. In contrast, in October/November 2021 more than half of all work stoppages happened in agriculture. The disparity was also observed in different primary reasons for work stoppages (Figure 124). About 89 percent of work 77 stoppages reported in June 2020 were directly related to COVID-19 (such as businesses being closed due to COVID-19 restrictions, being ill/quarantined, movement restrictions, furlough and so forth), while in October/November 2021 about 62 percent were related to these reasons. This signals that work stoppages after the second lockdown were related to COVID-19 restrictions to a lesser extent than the first lockdown and were more prevalent in the agricultural sector. This may be related to drought in most parts of the country in the first agricultural season of 2021 (March to May), which negatively affected germination, growth, and harvest of crops, especially cereals and legumes. Figure 121. Working respondents across different Figure 122. Work stoppages among those who worked in rounds, % previous rounds across rural/urban areas and economic sectors, % Source: HFPS, World Bank staff calculations. Source: HFPS, World Bank staff calculations. Note: Only the same respondents across all seven rounds Note: Only the same respondents across the current and are kept. previous rounds are kept. Figure 123. Structure of work stoppages by economic Figure 124. Reasons for work stoppages across rounds 1 sector, % and 7, % of respondents who stopped working Source: HFPS, World Bank staff calculations. Source: HFPS, World Bank staff calculations. Note: Only the same respondents across the current and Note: All respondents are kept. previous rounds are kept. 78 COVID-19 might have slowed down the pace of structural change in Uganda, reversing momentum and pushing some individuals into the agricultural sector. Figure 125. Economic sector of current employment across There was a shift in the structure of work rounds, % among respondents to the HFPS. Before the COVID-19 pandemic, about half of all respondents were working in agriculture. After the first lockdown, this share increased to 62 percent (Figure 125). There were two main reasons behind this increase. First, work stoppages were more prevalent in the services sector after the first lockdown. Second, among those who worked before March 2020 and continued working in June 2020, more respondents shifted from services to agriculture than from agriculture to services. As a result, the share of agriculture Source: HFPS, World Bank staff calculations. increased, perhaps helping households cope Note: Only the same respondents across the current and previous with shocks to other forms of employment rounds are kept. during the COVID-19 economic crisis. Box 6. Economic impacts of COVID-19 on refugees Refugees lagged the native Ugandan population in host communities in terms of employment outcomes before COVID-19. According to the representative survey of refugees and host communities conducted in 2018 in Uganda, only 28 percent of refugees were employed compared to 64 percent of Ugandans. Unemployment was lower among refugees from early arriving cohorts, suggesting that refugees were able to adapt and join the labor market as time went on. However, the quality of jobs may still be dissimilar. For those in wage employment, refugees tended to have lower wages (40 percent lower) than Ugandans, even after controlling for underlying individual characteristics. Refugees were also less likely to own non-farm family enterprises than hosts (19 percent versus 27 percent). Interestingly, however, the profits of enterprises in Kampala were higher among refugees compared to Ugandans, perhaps due to the generally higher levels of education among refugees in Kampala. Figure 126. Share of working respondents among refugees Work stoppages after COVID-19 were more and Ugandans before COVID-19 and during the first year of pronounced among refugees than Ugandans pandemic, % and employment among refugees has not recovered to the pre-pandemic level. Figure 126 shows working status among refugees and Ugandans as reported in high-frequency phone surveys covering the pre-COVID-19 period (pre- March 2020) and two periods after it, which lasted approximately for one year. First, Ugandans had higher employment rates, with men more likely to be employed than women for both refugees and Ugandans. Second, work stoppages were more pronounced among refugees – the share of working respondents dropped from 56 percent before COVID-19 to 36 percent in October/November 2020 and Source: RHFPS and HFPS, WB staff calculations. further to 32 percent in February/March 79 Note: Numbers here are not comparable to results from the 2018 2021.38 Among Ugandans, though, work survey since phone surveys reported only the working status among stoppages were also pronounced, but respondents who were in many cases heads of households. employment has fully returned to the pre- pandemic level after June 2020. Income recovery was slower among refugees than Ugandans. Similar to the lagging recovery of employment, lower shares of refugee households reported income levels in February/March 2021 being the same or higher than the pre-COVID ones compared to numbers reported by Ugandans in February 2021. For example, 44 percent of refugee households reported farm income to be above or at the same level as the pre-March 2020 level compared to 61 percent of Ugandan households. About 32 percent of refugee households reported farm income to be above or the same as the pre-March 2020 level compared to 53 percent of Ugandan households. Finally, 43 percent of refugee households reported wage income to be above or the same as the pre-March 2020 level compared to 62 percent of Ugandan households (Atamanov et al. 2021). Figure 127. Share of households unable to cover basic needs Compared to Ugandans, the ability of refugees and access services among refugees and Ugandans, % to access sufficient drinking water, medicine and medical treatment was significantly limited. Atamanov et al. (2021) compared access to different basic goods and services among Ugandans and refugee households from high-frequency phone surveys, using similar time periods to the extent possible. Even though definitions were not fully comparable for all indicators, the difference between refugees and Ugandans was stark for selected basic goods and services (Figure 127). About 36 percent of refugee households did not have sufficient drinking water to meet household needs at least once in the last seven days. In Source: RHFPS and HFPS, WB staff calculations. contrast, only two percent of Ugandans faced Note: Definitions of some indicators differ slightly across the scarcity of drinking water. About 36 percent of refugee and national phone surveys. In addition, survey periods refugee households were not able to buy are also different for some indicators. medicine when needed, compared to 18 percent among Ugandans. Similar patterns were observed with regards to food security and the ability to buy food. IV. Internal migration In the previous section, the report focused on the changes in the structure of the labor market and the shift from agriculture to non-agricultural activities. In this section, the report continues examining structural changes in employment in Uganda by focusing on internal migration, which can help households diversify livelihoods and transition out of subsistence agriculture. 38 It is important to mention that not all work stoppages were directly related to COVID-19 and related restrictions. 80 Magnitude and direction of migration flows Uganda had the highest growth of urban Figure 128. Share of urban population in total population population among comparators during the last in Uganda and comparators, % decade, but the overall rate of urbanization is still much lower than the average in the region. Figure 128 shows the share of the urban population in Uganda and selected comparators in 2010 and 2020 years. Uganda had one of the fastest growth rates of urban population during the last decade. The urban population grew, on average, by six percent between 2010 and 2020.39 Source: LFS, World Bank staff calculations. Note: SSA stands for Sub-Saharan Africa. Nevertheless, only 25 percent of the total population in Uganda resided in urban areas in 2020 and Uganda lags behind the average observed among Sub-Saharan African countries (41 percent). According to projections based on Uganda’s high population growth rate and the estimated rural-urban migration trends, the urban population will surpass the rural population by 2060 (World Bank 2020d). Internal migration may play an important role in structural change and the shift out of subsistence agriculture into industry and services. It is widely accepted that the development process is closely linked to labor moving out of agriculture and into industry and services sectors, coupled with rural-to-urban migration. People may migrate for different reasons, such as pull factors like higher earnings in non- agriculture sectors or/and improved amenities and services. They can be also pushed to migrate by shocks and the lack of social protection systems. Households may also send one or several members to migrate to seek alternative employment elsewhere (and send remittances back to the rest of the household) to reduce the risk of depending solely on agriculture. While the decision to move or send migrants is affected by many reasons, it can also be constrained by a number of factors. For example, if internal migration is associated with high costs, the ability to finance the move can be a binding constraint. Distance is another obvious factor that may constrain migration due to costs or lack of information. Migration networks formed by a group of people from a given community can stimulate migration to a particular area by lowering the costs, such as sharing information, providing initial employment opportunities, informal credit, or housing. If there is not a critical mass of people with similar backgrounds and origins, the migration process is less likely to occur. Migration may also be hindered by opportunity costs if able bodied workers are scarce and the current contribution to household income from a potential migrant is high (Lucas 2015; Selod and Shilpi 2021). In order to explore the scope, direction and determinants of internal migration, data from the UNHS conducted in 2016/2017 and 2019/20 rounds are used.40 39 Some of the expansion can also be due to a redefinition of administrative boundaries for urban areas. 40 The question about internal migration was first introduced in the UNHS 2016/17 so we could not use UNHS 2012/13. 81 The UNHS 2016/2017 and UNHS 2019/20 capture the most recent migration experience of households by asking if any household member older than ten years has lived in another place, such as another village, town or country, for six or more months at least once during the last five years. A household is defined as an internal migrant if the head of a household lived in another place in Uganda, except the same district, for more than six months. This is an imperfect measure because it captures only those who had a migration experience during the last five years and focuses only on the household heads’ migration experience. Migration for all possible reasons – including economic, health, and family reasons – are covered by this definition. The share of households with a migration experience fell from nine percent in the 2016/2017 survey to three percent in the 2019/2020 survey. About nine percent of household heads in Uganda in 2016/17 reported migrating at some point during the last five years. These households accounted for about six percent of the population in the country. This share declined sharply in 2019/20 to just three percent. COVID-19 restrictions perhaps were partially responsible for this decline, but even if 2019/20 round is split into pre-COVID-19 and COVID-19, the reduction was observed in both periods. Due to the small sample size of migrant households in 2019/20 and the COVID-19 shock in that period, further analysis of internal migration is based on the 2016/17 year only. The majority of internal migrants migrated for economic and work reasons, but with some notable exceptions. About two-thirds of household heads who reported having migrated at some point in the last five years in the 2016/17 survey mentioned looking for work or other economic reasons as the main reasons (Figure 129). Looking for work and other economic reasons were more prevalent among those who moved from rural to urban areas or within urban areas. The poorest households in rural areas, those moving from urban to rural areas, within rural areas, and those from the Northern region were least likely to move for work or economic reasons. Many of them mentioned following/joining family members as the main reason to move. This may signal that migrating for work/economic reasons may be difficult from the point of view of costs and entry barriers. Figure 129. Main reasons for migrating across different groups as of 2016/17, % Source: UNHS 2016/17, World Bank staff calculations. 82 Rural to urban migration accounted for Figure 130. Structure of internal migrants by area of origin and the smallest share in overall internal destination as of 2016/17, % migration flows in 2016/17. Figure 130 shows the distribution of households and population among households with migrant household heads. The smallest group is formed by household heads migrating from rural to urban areas – only 16 percent of all migrant households moved from rural to urban areas. The differences between the remaining three groups of migrants were not large. The largest group of households was among those moving within urban areas (29 percent), followed by those moving within rural areas (28 percent), and those who Source: UNHS 2016/17, World Bank staff calculations. moved from urban to rural areas (27 percent). The Central region supplied the largest share of internal migrant household heads. Table 4 shows the distribution of migrant households by region of origin, with the Central region accounting for 46 percent of all population among internal migrants as of 2016/17. The Western region was in second place, followed by the Eastern and Northern regions. In fact, the majority of internal migration happened within the same regions. For example, 69 percent of all internal migrants in the Central region came from somewhere else within the Central region. If the Central region is split into rural, urban areas, and Kampala, it is seen that 46 percent of all migrant households in Kampala came from urban areas of the Central region, and 38 percent of all migrant households in the urban areas of the Central region came from Kampala. Interestingly, Kampala played an important role as a source of internal migrants to all regions. This could be migrants who returned to rural areas after being temporarily in Kampala. Table 4. Migration by region of origin among households whose household head was an internal migrant at some point in the five years prior up to 2016/17, % Destination Origin Central Eastern Northern Western All Central 69 28 24 29 46 Eastern 15 66 13 3 19 Northern 4 2 52 2 13 Western 13 3 11 66 22 Uganda 100 100 100 100 100 Source: UNHS 2016/17, authors calculations. Note: Household weights are used. The Central region was the top destination for internal migrant household heads in the five-year period prior to 2016/17. Table 5 shows the distribution of migrant households by destination. The Central region, which is the most urbanized and wealthiest region in Uganda, was the top destination, absorbing 46 percent of all migrant household heads. The Northern and Western regions were tied around second place, each absorbing 20 percent of all migrant households, followed by the Eastern region at 13 percent. Internal migrant household heads from rural areas within each region were more likely to stay in rural 83 areas of the same region compared to urban migrants, who were less likely to stay in urban areas. About 81 percent of all migrant household heads from the Northern region migrated within the Northern region. In contrast, less than half of migrant households from the Eastern region migrated within that region, while 36 percent moved to the Central region, probably due to its relative proximity and the generally better road connections to this region. Table 5. Migration by destination among households whose household head was an internal migrant at some point in the five years prior up to 2016/17, % Destination Origin Central Eastern Northern Western all Central 68 8 11 13 100 Eastern 36 47 14 3 100 Northern 14 2 81 4 100 Western 26 2 10 61 100 Uganda 46 13 20 20 100 Source: UNHS 2016/17, authors calculations. Note: Household weights are used. Characteristics of internal migration Migrant households and the characteristics of their household heads were different in many regards to non-migrant households. Characteristics of household heads and their respective households across migration status are shown in Table 6. As a robustness check, broader internal migration groups were added depending on whether any household member older than ten years lived in another place in Uganda, except the same district, for more than six months over the period from 2012 to 2016/17. Results remain qualitatively the same. Households with household heads who migrated internally had, on average, a lower household size and dependency ratio compared to households with heads that had not migrated. Internal migrant household heads were more likely to be younger and single. They had better education, were more likely to use the internet and own mobile phones. Migrant household heads were less likely to work in agriculture and were more likely to work as professionals or sales workers, which is perhaps a post-migration effect. Related to this, more migrant households ranked income from wage employment as their chief source of income while more non-migrant households ranked crop income as the main source of income. Only one-third of migrant households owned dwellings compared to more than 70 percent among non-migrant households. Table 6. Characteristics of households depending on internal migration status during the five-year period up to 2016/17 head is not head is internal No member is At least one migrant migrant migrant member migrated Demographic characteristics Household size 5 3 43 38 Demogr aphics Dependency ratio, % 46 30 46 34 Head age 43 33 43 38 Married monogamous, % 55 55 54 59 marital status Married polygamous, % 15 9 15 11 Divorced/separated, % 11 13 11 10 Widow/ Widower, % 13 3 13 8 Never married, % 7 20 7 13 Education of head, % Educatio Primary incomplete 17 9 17 10 Primary complete 14 5 15 6 n Secondary incomplete 39 27 39 30 84 Table 6. Characteristics of households depending on internal migration status during the five-year period up to 2016/17 head is not head is internal No member is At least one migrant migrant migrant member migrated Secondary complete 18 20 18 19 Post 19 26 19 24 University 2 6 2 5 Labor force status of head, % Professionals, technicians, managers 7 14 7 13 Service and sales workers 19 28 19 25 Occupation Skilled agricultural workers 46 25 46 32 Craft and related trade workers 6 10 6 9 Plant and machine operators 5 7 5 7 Elementary occupations 16 14 17 13 Other 1 1 1 2 Internet, % ern Int et Head is using internet 10 27 9 23 Housing conditions, rent Household owns a dwelling 74 34 74 52 Rent Household rent 20 51 20 37 Household rent for free 5 15 6 10 Income sources and savings, % Crop farming (small scale) 43 20 44 27 key income source Livestock farming (Small scale) 2 1 1 2 Commercial farming 2 2 2 2 (rank 1) Wage employment 24 42 23 35 Non-agricultural enterprises 20 24 20 23 Remittances 7 10 7 8 Other (specify) 3 2 3 2 Assets, % Motorcycle now 8 7 7 8 Motorcycle 12 months ago 7 6 7 7 Assets Mobile phone now 72 85 71 84 Mobile phone 12 months ago 68 80 66 80 Livestock now 44 30 44 37 Livestock 12 months ago 39 24 39 31 Source: UNHS 2016/17. Not: Household weighted estimates. Besides individual and household level characteristics, the decision to migrate can be affected by the characteristics of locations where migrants resided before the migration and the locations they migrated to. The UNHS 2016/17 administered a community questionnaire from which a range of variables could be constructed to capture potential factors affecting migration decisions. Since only the districts where migrants originated from and the districts they are currently living in are known, the analysis aggregates community variables up to a district level using the population size in each community. Multivariate regression models are run to check more formally and identify the key correlates of internal migration. Two models are run using district variables for districts of origins (push factors) in the first model and district variables for districts of destination in the second model (pull factors). Both models used the same household level variables.41 District variables included having a weather shock, new employment opportunities, new roads, and improved electricity during the last five years. These variables capture both “push” and “pull” factors 41 The multilevel logit was run to account for the hierarchical nature of the data (household and district levels). 85 behind internal migration. For example, having a weather shock in the district of origin can encourage migration, while a new road or improved electricity can discourage it by providing economic opportunities at home. At the same time, a new road or improved electricity in the district of origin can also facilitate migration by increasing connectivity and access to information. The same factors at the destination may affect migration decisions. Thus, improved electricity and new employment opportunities at the destination can pull households into migration. Additional variables include measures of the most common form of land tenure in the district and whether farming was the most important economic activity. Results from regression analysis confirm that internal migration may not be accessible for the poorest households living in lagging agricultural regions. Female, better educated, and younger heads with lower dependency ratios were more likely to migrate (Table 7). Urban areas were significantly more likely to be associated with a higher probability of in and out migration. Originating from agricultural districts with prevalent customary land tenure was negatively associated with internal migration, but households were more likely to migrate from the districts with improved electricity. Destination districts with a larger role of agriculture in the local economy were less likely to attract migrants. In contrast, the districts that experienced larger employment opportunities during the last five years were more likely to attract migrants. All these findings highlight how some households and individuals in Uganda may experience significant barriers to migrating, especially if they are not educated, have more dependents, and live in areas with predominant subsistence agriculture and inadequate infrastructure. Table 7. Multilevel mixed-effects logistic regression for a probability of having a head of household who migrated internally at some point in the five-year period up to 2016/17, marginal effects Community Community level level variables variables are for are for districts destination Variables of origin district Head of household is male -0.0291*** -0.0366*** Head of household age -0.00294*** -0.00287*** Head of household education, base = no education Primary incomplete 0.0127 0.0134* household level variables Primary complete 0.0239*** 0.0313*** Secondary incomplete 0.0174** 0.0322*** Secondary complete 0.0371*** 0.0711*** Post-secondary but not university 0.0470*** 0.0713*** University 0.0416*** 0.0766*** Dependency ratio -0.000664*** -0.000844*** Base=married monogamous Married polygamous -0.0119* -0.0195*** Divorced/ Separated -0.00816 -0.00873 Widow/ Widower -0.0318*** -0.0385*** Never married 0.00407 0.0114 Community experienced either flood or drought during last 5 years -9.39E-05 0.000840*** community level Community experienced new employment opportunities during last 5 years -5.93E-05 0.000374** variables Community experienced new road during last 5 years 0.000242 -9.49E-06 Community experienced Improved electricity during last 5 years 0.000425** 0.000247 Most common land tenure in the community is customary -0.000194* -4.31E-05 Main economic activity in the community is farming -0.000482*** -0.000487** Urban dummy 0.0169*** 0.0170*** Observations 15,553 15,553 Source: UNHS 2016/17. Note: Household weighted estimates. *** significant at one percent, **five percent, * ten percent. 86 Structural change, internal migration and monetary welfare Internal migration is correlated with structural change in employment for the household head who migrated. Figure 131 compares the sector of employment among heads of household who moved at least once during the five-year period up to 2016/17 and those who did not. In urban areas, household heads who moved from rural to urban areas or moved within urban areas were less likely to work in the agricultural sector than those who did not move. In a similar way in rural areas, household heads who moved from urban areas to rural or moved within rural areas were less likely to work in the agricultural sector compared to those who did not migrate. Given the nature of this data, it is not clear whether migration was undertaken by non-agricultural workers since they were more educated, or migrants were more likely to find non-agricultural jobs once they moved. Nonetheless, this analysis highlights the importance of internal migration for structural transformation in Uganda. Migration is often correlated with welfare gains, but identifying the causal impact is challenging. Migration can bring welfare gains if individuals or households can move to areas with higher returns to economic activities. Migrants can also help relieve credit constraints, diversify income sources and minimize risks for their households by sending remittances back home. Identifying the causal impact of migration on welfare is difficult because migrants may be systematically different from those who stayed in unobservable ways. If not controlled for, omission of these unobservable characteristics in the regression analysis can bias the impact of migration on welfare gains. Keeping this in mind, below are some descriptive statistics that compare consumption per adult equivalent among different migration groups using panel data and tracking the same households over time. Figure 131. Employment sector of household head using UNHS 2016/17, % a) rural areas b) urban areas Source: UNHS 2016/17. Rural to urban transitions generated welfare gains for migrants during 2015/16 and 2019/20, but it was comparable to the gains for those who remained in rural areas as well. 87 Figure 132. Monthly consumption per adult Figure 132 reports the levels and annualized growth equivalent and consumption annualized growth rates rates of consumption per adult equivalent adjusted across household heads’ migration status from the for price differences across rural and urban areas of UNPS 2015/16 and 2019/20, UGX and % different regions. The data is obtained from the panel component of the UNPS data from 2015/16 and 2019/20 and shows consumption for four different groups depending on internal migration status. Consumption grew in all groups, but the annual growth rate was the same among those households who remained in rural areas and those who moved from rural to urban areas (about four percent).42 This is also consistent with previous findings from Garlati et al. (2018) who also showed using panel data from UNPS 2009 and 2015/16 that consumption growth was not higher among those who moved from rural to urban areas compared to those who remained in Source: UNPS 2015/16 and 2019/20. rural areas. Note: N=2673 households IV. Agriculture Accelerating agricultural development remains an important pathway out of poverty in Uganda. As shown in previous sections, non-agricultural employment was correlated with higher monetary wellbeing, while the poor were more likely to engage in agriculture activities. Still, poverty reduction depends on agricultural performance simply because the absolute majority of poor households engage in subsistence agriculture and do not necessarily have access to non-farm opportunities. Figure 133 plots poverty versus several indicators by sub-regions in Uganda based on 2018 Agricultural Census and the UNHS 2016/17. Figure 133a shows that poverty rates were higher in subregions with a higher share of agricultural employment, which is consistent with findings in previous sections showing that the poorest households were more likely to be engaged in agriculture and to live in rural areas. Figure 133b shows that poverty rates were higher in sub-regions with a lower average size of arable land among agricultural households. Figure 133c and Figure 133d show that poverty rates were lower in subregions with higher maize and cassava yields. This provides some evidence for the correlation between agricultural development and poverty. 42 It is important to mention that definition of internal migrant household differs across the UNHS and UNPS data. In the UNPS a household is considered a migrant if it changed the residence from rural to urban areas and vice versa across two rounds . 88 Figure 133. Subregional poverty rates versus different agricultural indicators a) Poverty rate versus share of agricultural b) Poverty rate versus average crop area, % and ha employment, % c) Poverty rate versus maize yield, % and mt per ha c) Poverty rate versus cassava yield, % and mt per ha Source: 2018 Agricultural Survey from FAO and UBOS (2020) and the UNHS 2016/17, World Bank staff calculations. Agricultural growth in Uganda during the last decade was limited. Figure 134 shows the average growth rates in agricultural value added during the last decade versus the growth rates of rural population in Uganda and other low-income countries. Uganda is located on the 45 degrees line, with an average growth rate in agriculture of about three percent and a growth rate of the rural population of 2.7 percent. As a result, the average growth of agriculture per rural person was barely higher than zero during the last ten years. This performance was worse than the average performance among Sub-Saharan countries over the same period. 89 Figure 134. Growth rate in value added in agriculture versus growth rates in rural population during 2010 to 2020, % Source: World Development Indicators, World Bank staff calculations. Note: SSA stands for Sub-Saharan Africa. Favorable external conditions, such as weather, and little change in actual production practices were found to accompany the agricultural growth between 2005/05 and 2012/13. Previous studies showed that the sharp reduction of poverty before the 2016/17 spike in poverty was due to fast growing agricultural income (World Bank 2016; Hill and Mejia-Mantilla 2017; Hill, Mejia-Mantilla and Vasilaky 2018). The main factors behind increased agricultural income included stabilization of the conflict in the Northern region and infrastructure investments, but also good fortune in the form of favorable rainfall and positive trends in commodity prices. In contrast, only a small proportion of the growth in household agricultural income was explained by an increase in the adoption of better technology in agricultural production. In particular, it was shown that use rates of fertilizer, pesticides, and improved seeds in Uganda remained low, in part due to the low quality of some of these products as well as the cost. According to World Bank (2018), the free distribution of subsidized inputs has contributed to lower quality seed production by agribusinesses and a crowding out of the private sector. Even though extension services were correlated with higher usage of improved seeds and higher crop income (provided by National Agricultural Advisory Service), only a small share of households used them. Thus, the usage of extension services expanded from eight percent of households in 2006 to 12 percent of households in 2012. Moreover, there were institutional changes after 2014 and extensions services were performed by the military and were limited primarily to the distribution of subsidized inputs. Most recent data showed that value of production per hectare and agricultural income per capita increased during 2013/14 and 2019/20 years. RULIS data can be useful to understand the performance of the agricultural sector and the underlying factors behind agricultural production. The data show that median real value of crop production per hectare among households engaged in growing crops increased 90 by eleven percent annually with higher growth rate among the poorest households from the bottom consumption quintiles (Figure 135). The annualized real growth rate of agricultural income per capita was about eight percent. Growth in agricultural income transformed into consumption growth during this period. The main question of interest is whether the growth in agricultural income during the considered period was related to the adoption of better technologies or was mostly related to weather conditions. To answer this question, the report examines the changes in key factors of agricultural production such as fertilizers, pesticides, improved seeds and access to extension services. Fewer households were engaged in crop and livestock activities during 2013/14 and 2019/20, with a steeper reduction among wealthier households. Consistent with the declining share of agricultural employment discussed in the previous section, fewer households were engaged in crop and livestock activities (Figure 136). The reduction in participation in crop activities was observed across all consumption quintiles, but the magnitude of reduction was much larger among the top 60 percent of households. There was no reduction in the share of households engaged in livestock activities among the poorest households from the first quintile. This signals that the poorest households continued to rely heavily on agricultural activities. Figure 135. Real annualized growth rates of median crop production per hectare and median agricultural income per capita by consumption per adult equivalent quintiles during 2013/14 and 2019/20 among the population engaged in agricultural activities, % Source: RULIS database based on the UNPS 2013/14 and 2019/20, World Bank staff calculations. Note: Quintiles are based on spatially adjusted consumption using poverty lines. Agricultural income was spatially adjusted as well. Despite the reduction in the average share of households engaged in agricultural activities, the share of income from agriculture remained the same during 2013/14 and 2019/20 and even increased among the poorest households. Figure 137 demonstrates that the share of agricultural income in total income of households remained at about 49–51 percent.43 In fact, among the poorest households from the bottom quintile, the share of agricultural income increased from 60 percent in 2013/14 to 60 percent in 2019/20. The share of agricultural income declined only among households from the top two wealthiest consumption quintiles. 43 We have kept only households with zero or positive agricultural income and total income. 91 Figure 136. Share of households engaged in crop and Figure 137. Share of agricultural income in total livestock production activities by consumption per adult income by consumption per adult equivalent equivalent quintiles, % quintiles, % Source: RULIS database based on the UNPS 2013/14 and 2019/20, World Bank staff calculations. Note: Share of agricultural income is calculated only for those with zero or positive agricultural and total income. Income values were spatially adjusted using poverty lines. Even though there was some increase in the proportion of households using organic fertilizers and pesticides during 2013/14 and 2019/20, it was relatively marginal and only observed among wealthier households. Figure 138 shows the percentage of households engaged in planting activities who used fertilizers at least once in any of the agricultural seasons in 2013/14 and 2019/20. The share of households who used organic fertilizers increased from 11 percent to 14 percent, pesticides from 14 percent to 17 percent, while usage of inorganic fertilizers remained at about seven percent. On top of only marginal or zero change, the usage of fertilizers and pesticides remained at very low levels among households from the poorest bottom quintile and without changes over time. For example, the share of households using organic fertilizers increased among the richest households from 20 percent in 2013/14 to 27 percent in 2019/20, while among the poorest households the share only increased from five to six percent during the same period. This means that low quality of inputs is not the only barrier preventing uptake of fertilizers, but costs may also play a limiting role. Figure 138. Share of households engaged in planting activities who used fertilizers at least once on any land plot during any of the agricultural seasons in 2013/14 and 2019/20, % Source: UNPS 2013/14 and 2019/20, World Bank staff calculations. 92 Note: Cross-sectional household weights are used. There was a significant decline in the usage of improved seeds and getting advice from extension services. Figure 139 shows information about the usage of improved seeds. In contrast to the usage of fertilizers, a lower share of households used improved seeds in 2019/20 compared to 2013/14. The largest drop in the usage of improved seeds was observed among the poorest households from the first quintile. This decline may be related to the substandard quality of seeds, which reduced the yield gain by more than 75 percent as shown in Hill, Mejia-Mantilla and Vasilaky (2018). Cost considerations may be another reason for the lower usage of improved seeds, particularly among the poorest households. In addition to a lower usage of improved seeds, there was a drastic decline in the share of households who received advice about agricultural/livestock activities from organizations offering extension services (Figure 139). Thus, the average share of households who received advice fell from 20 percent in 2013/14 to seven percent in 2019/20 and the decline was the most pronounced among households from the poorest quintiles. Given that extension trainings and knowledge were correlated with the usage of improved inputs such a sharp decline in these services is concerning and requires further investigation. Figure 139. Share of households engaged in planting activities who used improved seeds on any land plot during either of the two agricultural seasons and who received any advice from extension services during the last 12 months in 2013/14 and 2019/20, % Source: UNPS 2013/14 and 2019/20, World Bank staff calculations. The adoption rate of new technologies measured by the level of mechanization remains very low, in particular among the poorest households. Previous studies indicated low levels of mechanized traction in Uganda and associated with this, a subsistence nature of agricultural production (World Bank 2018). The report checked the usage of mechanized equipment44 in 2013/14 and 2019/20 (Figure 140). Overall, about 17 percent of households engaged in agricultural activities used mechanized equipment in 2013/14 and this share increased slightly to 22 percent. Increased usage of mechanized equipment was not distributed equally across the consumption quintiles. Thus, among the poorest households from the bottom consumption quintile the usage of mechanized equipment has actually declined from 11 percent in 2013/14 to eight percent in 2019/20. 44 Mechanized equipment is measured as usage of ploughs, tractors, chain/band saws, shellers, harrows/cultivators, weeders, planters and sprayers during the last 12 months. 93 Figure 140. Share of households engaged in planting activities who used mechanized equipment during the last 12 months in 2013/14 and 2019/20, % Source: RULIS database based on the UNPS, World Bank staff calculations. Overall, it seems that the growth in agricultural income and crop production was to a lesser extent related to modernization of production practices during 2013/14–2019/20, but more to favorable external environment – in particular, favorable weather after the drought in 2016. This analysis did not reveal significant positive changes in modernization of production by using improved seeds, more fertilizers and mechanization, in particular among the poorest households. Getting advice from extension services demonstrated a declining trend as well. Given these findings and the robust growth of agricultural income observed during this period, favorable external conditions were likely to play a positive role. Thus, except for a drought in 2016, weather conditions were favorable, in particular in the years of 2018 and 2019. Prices of agricultural products, including maize and beans were volatile, but did not decline. These factors could at least partially explain robust growth of agricultural income and crop production during the considered period. V. Policy implications Economic transformation in Uganda will require a fundamental shift in the nature of production – from low-investment, informal activities – to higher-capital, more productive employment. In Uganda, the agricultural sector – where the majority of the population currently works – has the lowest productivity level compared to services and industry. However, it was the only sector where productivity increased after 2016/17. In contrast, productivity in services and industry sectors, which absorbed excessive labor from agriculture, has been in decline since 2011/12 – 2018/19. In agriculture, the majority of employment is in subsistence production, which is characterized by much lower productivity relative to wage employment. Similarly, a high degree of informality in the services sector is also associated with lower productivity. Therefore, accelerating wage job creation is key for faster economic transformation. Opportunities to reallocate from the less productive agricultural sector to more productive sectors are not widely accessible and are especially limited for those with low human capital and those who live in lagging regions. Households that transitioned out of agriculture into other sectors and those who migrated from rural to urban areas tended to benefit from increases in consumption, demonstrating the poverty reducing effects of structural change. However, the pace of structural change was faster for 94 wealthier individuals and internal migration was not widely accessible for the poorest households living in lagging agricultural regions, constraining their abilities to access better opportunities. Results from regression analysis revealed that, after controlling for other factors, households with less educated heads, households with more dependents, and households living in areas with predominantly subsistence agriculture and inadequate infrastructure were less likely to migrate. Therefore, reducing the barriers and costs to migration could enable more vulnerable households and individuals in Uganda to take advantage of opportunities in other sectors and in more economically vibrant places. COVID-19 may have slowed down the pace of structural change in Uganda, reversing the momentum of economic transformation. The lockdowns and mobility restrictions imposed to limit the spread of COVID- 19 had a disproportionally higher impact on urban areas, women and those working in the services sector. This prompted a shift of labor into the agricultural sector, which continues to be the lowest productivity sector and is highly vulnerable to weather shocks. Therefore, revitalizing the industry and service sectors will be an important part of the COVID-19 recovery strategy. In addition, providing targeted support to refugee and host community businesses and self-employed can contribute to the mitigation of shocks while fostering job creation in refugee hosting districts. For example, providing reskilling technical and vocational training can help refugees get better jobs and wages right from the start. Accelerating agricultural development remains an important pathway out of poverty in Uganda, but growth in agricultural income observed between 2013/14 and 2019/20 was to a large extent associated with favorable weather conditions rather than fundamental changes in the nature of production. Thus, there was just a marginal increase in the usage of fertilizers and mechanized traction during the last several years and was mostly among wealthier households. In addition, there was a significant decline in the overall usage of improved seeds and getting advice from extension services. Without improving access and usage of these important inputs, it will be hard to modernize production practices, improve productivity and make agriculture less dependent on such exogenous factors as weather. 95 References Atamanov, A.; Beltramo, T.; Reese, B. C.; Rios Rivera, L. A.; Waita, P. (2021). One Year in the Pandemic: Results from the High-Frequency Phone Surveys for Refugees in Uganda. World Bank, Washington, DC. FAO (Food and Agriculture Organization) and UBOS (Uganda Bureau of Statistics). (2020). Annual Agricultural Survey 2018. FAO (Food and Agriculture Organization). Rural Livelihoods Information System. Garlati, P., Mejia-Mantilla, C., Merotto, D., and Webber, M. (2018). “Towards a Better Labor Market i n Uganda”, mimeo, The World bank Group. Goodman, S.; BenYishay, A.; Lv, Z.; & Runfola, D. (2019). GeoQuery: Integrating HPC systems and public web-based geospatial data tools. Computers & Geosciences, 122, 103-112. Hill, R.; Mejia-Mantilla, C. (2017). With a Little Help: Shocks, Agricultural Income, and Welfare in Uganda. Policy Research Working Paper; No. 7935. World Bank, Washington, DC Hill, R.; Mejia-Mantilla, C.; Vasilaky, K. (2018). Vanishing Returns? Potential Returns and Constraints to Input Adoption among Smallholder Farmers in Uganda. Policy Research Working Paper; No. 8627. World Bank, Washington, DC. Lucas, R. E. B. (2015). “Internal migration in developing economies: An overview.” KNOMAD Working Paper 6, Global Knowledge Partnership on Migration and Development, World Bank, Washington, DC. May. Selod, H. & Shilpi, F. (2021). Rural-Urban Migration in Developing Countries: Lessons from the Literature. Policy Research Working Paper; No. 9662. World Bank, Washington, DC. Uganda Bureau of Statistics (UBOS) and ICF. (2018). Uganda Demographic and Health Survey 2016. Kampala, Uganda and Rockville, Maryland, USA: UBOS and ICF World Bank. (2016). The Uganda Poverty Assessment Report 2016: Farms, Cities and Good Fortune – Assessing Poverty Reduction in Uganda from 2006 to 2013. World Bank, Washington, DC World Bank. (2020a). COVID-19 Impact Monitoring: Uganda, Round 1. World Bank, Washington, DC. World Bank. (2020b). COVID-19 Impact Monitoring: Uganda, Round 2. World Bank, Washington, DC. World Bank. (2020c). COVID-19 Impact Monitoring: Uganda, Round 3. World Bank, Washington, DC. World Bank. (2020d). Tackling the Demographic Challenge in Uganda. World Bank, Washington, DC. World Bank. (2021a). COVID-19 Impact Monitoring: Uganda, Round 4–5. World Bank, Washington, DC World Bank. (2021b). COVID-19 Impact Monitoring: Uganda, Round 6. World Bank, Washington, DC. World Bank. (2021c). COVID-19 Impact Monitoring: Uganda, Round 7. World Bank, Washington, DC. 96 Chapter 3. Vulnerability in Uganda I. Shocks and coping strategies Exposure to shocks before COVID-19 Communities and people in Uganda were exposed to many different shocks related to climate, health, and forced displacement during the last two decades. Many of them were severe. According to the EM- DAT – a global database on natural and technological disasters – there have been 141 disasters in Uganda between 1999-2020.45 Of these 141 disasters, 77 were natural and 64 technological.46 The main natural disasters recorded in Uganda were flooding and epidemics, but droughts affected the greatest number of people (Figure 141). Indeed, more than 3.5 million people were affected by droughts in Uganda between 1999 and 2020. Figure 141. Total number of natural disasters and Figure 142. Experience with climate conditions for population affected* in Uganda during 1999–2020 agricultural production in Uganda and selected comparators compared to 10 years ago in circa 2016– 2018, % Source: Centre for Research on the Epidemiology of Disasters Source: AfroBarometer, round 7. World Bank staff (CRED), accessed on 1/18/2022. calculations. *Note: The EM-DAT database defines “affected” as “people requiring immediate assistance during a period of emergency, i.e., requiring basic survival needs such as food, water, shelter, sanitation and immediate medical assistance.” Uganda is one of the most vulnerable and least adapted countries to climate change. According to World Bank (2020), average temperatures in Uganda have increased by 1.3°C since the 1960s, while Uganda has experienced a statistically significant reduction in annual as well as seasonal rainfall.47 While trends in 45 A shock should meet at least one of these criteria to be included in the database: 10 or more people dead, 100 or more people affected, the declaration of a state of emergency, and/or an official government request for international assistance. 46 Technological disasters include industrial and transport accidents also including fires, collapse of buildings and explosions. 47 Uganda’s climate is largely tropical with two rainy seasons per year – March to May and September to December. The Northern region, which forms one quarter of the country, lies outside the tropical belt, and hence experiences only one rainy season – March to October. 97 extreme rainfall conditions are more difficult to define due to the lack of data and seasonal variability, droughts have increased in Uganda over the past 60 years. Extreme events leading to disasters such as floods, droughts, and landslides have also increased over the last 30 years. According to the Notre Dame Global Adaptation Initiative (ND-GAIN), Uganda is one of the most vulnerable and least adapted countries to climate change. As of 2019, Uganda ranked the 10th most vulnerable country to climate change and the 35th least prepared country to deal with the impacts of climate change. Respondents to the seventh round of the Afro Barometer Survey, conducted in 2016, were asked about their experience with climate conditions for agricultural production in Uganda compared to ten years ago.48 About 59 percent of respondents indicated that climatic conditions were much worse, and 26 percent indicated that climatic conditions were worse than ten years ago (Figure 142).49 In other comparator countries, except Malawi, perceptions were much more positive. This is consistent with data above that shows that climate shocks and weather changes have indeed been worsening in recent decades in Uganda. Climate change is expected to increase the risk and intensity of natural disasters . Increased temperatures are expected for Uganda in future as well. Under a high-emission scenario, monthly temperature change is expected to increase by 1.8°C for the 2050s and this will increase aridity and the length and severity of the dry season. Under the same scenario, monthly annual precipitation is expected to increase in some areas of the country, with decreases in others, notably the northern and north-eastern areas. Climate change is expected to increase the risk and intensity of flooding as well as increase the likelihood of water scarcity for certain areas of the country. Given the high dependence on rain-fed agriculture and overall large role agriculture for food security and economic prosperity, climate change may have a significant negative impact on the development of this sector. Exposure to shocks increases the volatility of economic growth and limits the progress in poverty reduction. One way in which experiencing shocks manifests itself is through a high volatility of economic growth. This is observed in Uganda, where GDP per capita has dropped to less than one percent multiple times during the last 20 years – which is typically associated with a major shock. Most recently, the significant increase in poverty in 2016/17 was to the largest extent due to severe drought and pests, with large negative impacts on agricultural households’ production (World Bank 2018). The inflow of refugees and the COVID-19 pandemic also stand out as prominent covariant shocks to have recently hit Uganda. Indeed, poverty reduction stagnated in 2019/20 largely due to the economic consequences of the COVID- 19 pandemic and related restrictions. This chapter explores the types of shocks experienced by Ugandan households, and the strategies they used to cope with them. The report distinguishes between pre-COVID- 19 and COVID-19 periods and uses two different data sources. Analysis of shocks before COVID-19 is based on the UNPS data, while High-Frequency Phone Surveys are used to explore shocks after March 2020 during the COVID-19 pandemic. According to the UNPS data, the share of households that experienced shocks during the last decade ranged from 30 to 40 percent in Uganda, with rural and the poorest households affected the most. Figure 143 shows the share of households that experienced shocks during 2013–2020 (before COVID-19) using UNPS data. About 41 percent of households experienced at least one shock in 2013/14, with the share dropping to about 30 percent in 2015/16 and 2019/20. In all years, rural inhabitants were 48 Representativeat the national level. 49These results may have been particularly influenced by the recent experience of drought, which occurred in Uganda in 2016/17. 98 significantly more likely to face a shock. Residents of the Northern and Eastern regions were more likely to experience shocks in 2015/16 and 2019/20 (Figure 144). In all years, the poorest households, based on consumption per adult equivalent, were more likely to be exposed to shocks regardless of the area considered (Figure 145). Figure 143. Share of households with at least one shock during last the 12 months across rounds and by rural and urban areas, % Source: UNPS, WB staff calculations. Figure 144. Share of households that experienced at Figure 145. Share of households that experienced at least one shock during the last 12 months across years least one shock during the last 12 months across years and regions, % and rural and urban consumption per adult equivalent quintiles, % Source: UNPS, WB staff calculations. Note: Quintiles are based on spatially adjusted by poverty lines consumption per adult equivalent. Regional poverty lines were used for spatial adjustment. Drought was listed as the most frequent shock experienced by households in 2019/20. Before COVID- 19, drought was the most common shock – experienced by 10 percent of households – in Uganda in 2019/20.50 The exposure to drought was very different across regions during this period (Figure 146). About 20 percent and 16 percent of households experienced drought in the poorest Northern and Eastern regions, respectively, compared to only six percent and three percent in the Western region and Kampala. All other shocks were less frequent. About four percent of households experienced irregular rains, about 50 Share of households experienced droughts was 26 and 15 percent in 2013/14 and 2015/16 accordingly, so the overall decline in exposure to shocks over time was driven to a large extent by weather shocks – in particular droughts. 99 three percent experienced floods, and three percent experienced an illness of an income earner or household member. Similar to droughts, irregular rains and floods were more likely to affect households living in the Northern and Eastern regions. All other shocks were experienced by fewer than three percent of households. Figure 146. Incidence of different shocks in 2019/20 across regions, % of households Source: UNPS, WB staff calculations. Shocks in Uganda frequently result in a decline in income. Those households that experienced shocks during the last 12 months were asked if they experienced a decline in any of the following dimensions: income, assets, food production, and food purchases. Figure 147. Shocks and decline in different dimensions of wellbeing in Results from different rounds of the 2019/20, % UNPS data showed that most reported shocks resulted in a decline in at least one dimension. Results from the UNPS 2019/20 are shown in Figure 147. A decline in income was observed after most of the shocks. In particular, the loss of employment, low prices of agricultural outputs, livestock diseases, and landslides led to a decline in income in 100 percent of cases. A decline in food production was often observed as a result of irregular rains and drought. A decline in assets was mostly associated with fires. A decline in food purchases was associated with reduction in earnings and the death of income earners. Source: UNPS, WB staff calculations. 100 Box 7. Impact of weather on the economic wellbeing of households in Uganda According to Hill and Mejia-Mantilla (2017), per capita crop income was significantly higher among those who farm more land and apply more labor, fertilizer and pesticides. However, these were not the key factors explaining substantial growth in agricultural incomes in Uganda during 2006 –2013. During this period, much of the gains in agricultural income growth were associated with good weather, peace, and favorable prices, and not technological change or profound changes in agricultural production. The positive impact of rainfall was observed in the Eastern and Northern regions and was also stronger among households from the bottom 40 percent of population. Besides crop income, rainfall shocks were also found to be correlated with other variables. Thus, wage employment and self-employment out of agriculture were inversely and significantly correlated with rainfall. The results suggest that diversification of productive activities can be an important risk hedging strategy for households in Uganda, particularly the poorest. Households could at least partially offset the negative impact of climate shocks by engaging in non-farm activities and increasing non-farm income. Finally, the authors showed that higher levels of education of the household head were associated with lower negative effect of rainfall shocks on both crop income and per capita consumption. In addition to a change in incidence, there was a change in duration and severity of shocks across years. Figure 148 plots three dimensions using data from UNPS 2013/14 and UNPS 2019/20. It shows the duration of key shocks, the share of households that experienced these shocks during the last 12 months, and the share of those who had a decline in all four dimensions due to the shock: income, assets, food production, and food purchases. The incidence of all shocks was lower in 2019/20 compared to 2013/14, but the duration and severity of some shocks increased. For example, droughts and floods lasted longer and were slightly more likely to affect multiple dimensions in 2019/20 compared to 2013/14. It important to note, however, that the period considered is too short to make any conclusions about climate change in Uganda. Figure 148. Characteristics of shocks between 2013/14 and 2019/20 Source: UNPS 2013/14 and 2019/20, WB staff calculations. 101 Displaced refugees and internal climate migrants pose additional challenges to building resilience among Ugandan households and communities. While Uganda has often been lauded for its progressive refugee policy, the decline in humanitarian finance has caused reductions in basic services, including a cutback in food rations to below minimum requirements. This illustrates the requirements for further planning to continue providing the necessary services for both refugee and host communities, including through more shock-responsive national development systems. Climate change is also expected to accelerate out-migration from certain regions, such as the rural northeast, while other regions may become the epicenters of climate in-migration, creating new pressures and demands for public services and income-generating opportunities. The impact of climate change in neighboring countries may also further exacerbate regional conflicts, creating new inflows of refugees into Uganda over the next decade. Coping strategies before COVID-19 Savings, unconditional help from relatives, and changing dietary patterns were the three main coping strategies in 2019/20. Households were asked which coping strategy they followed for each reported shock (Figure 149). Respondents were allowed to report up to three coping strategies and rank them by importance. Among all coping strategies practiced in 2019/20, the most frequent was using savings (23 percent). Unconditional help provided by relatives was in second place, accounting for 20 percent of all strategies, followed by involuntary changes in dietary patterns, accounting for 19 percent of all coping strategies. Doing nothing was also mentioned by many respondents and this strategy was observed in 14 percent of all cases. In about six percent of cases, household members were doing more non-farm work and obtained credit. Box 8 discusses social assistance programs in Uganda before and during COVID-19. The poor in Uganda were more likely to change dietary patterns, while more affluent households were more likely to use savings to cope with shocks. Figure 150 looks at all the coping strategies followed by Ugandan households in rural/urban areas and across gender of a head of household and rural and urban consumption per adult equivalent quintiles. Significant differences in coping strategies were observed depending on the gender of the household head. Female headed households were more likely to get unconditional help provided by relatives, while male headed households were more likely to rely on savings. Regardless of the residence, the poorest households from the bottom first consumption quintile were more likely to follow inferior coping strategies, such as an involuntary change in dietary patterns compared to households from the top richest quintile. The largest gap between the poorest and the richest was observed in urban areas, with only seven percent of households from the bottom urban quintile using savings compared to 38 percent among the richest top urban quintile. 102 Figure 149. Types of coping strategies among households experienced shocks during the last 12 months in 2019/20, % Source: UNPS 2019/20, World Bank staff calculations. Note: All coping strategies are taken into account regardless of their rank. Figure 150. Distribution of coping strategies during the last 12 months in 2019/20 across residence, head of household gender and rural/urban consumption quintiles. % Source: UNPS 2019/20, World Bank staff calculations. Note: All coping strategies are taken into account regardless of their rank. Quintiles are based on spatially adjusted by poverty lines consumption per adult equivalent. Involuntary changes in dietary patterns were closely related to shocks affecting agricultural activities. Figure 155 shows the main strategies households used to cope with the key shocks in 2019/20. Households experienced weather shocks such as drought, irregular rains and those who experienced a high level of crop pests were more likely to involuntarily change their dietary patterns. This can happen because of the high level of subsistence agriculture, with households consuming most of the agricultural produce from their farms. In this case, weather shocks disrupt agricultural production and affect food consumption patterns. In addition, those rural households who are engaged solely in agricultural activities and more prone to weather shocks, are typically poorer, which limits their resilience to shocks. Finally, 103 weather shocks are more likely to be covariate, affecting all households in the area. This can explain the low share of unconditional help among those who experienced weather shocks. In contrast, unconditional aid played a very important role for coping with idiosyncratic or household specific shocks such as illness of income earner or other household members. Box 8. Social assistance in Uganda The importance of social protection is articulated in the main development programs of Uganda. Uganda’s Vision 2040 mentions the limited existing social protection and support programs as important challenges for Uganda’s development. Indeed, the provision of social protection for citizens is indicated as one of the factors that will drive social transformation of Uganda’s society. The national deve lopment plans (NDP) also prioritize social protection as one of the key strategies for transforming Uganda from an agrarian society to a modern and prosperous country. In particular, the most recent NDPIII (2020–2025) prioritizes merging, modifying and/or expanding existing social protection programs to cover more beneficiaries as a key component for achieving inclusive growth. Direct income support in Uganda is composed of two major and several minor programs . The main direct income support programs currently in Uganda are the Senior Citizens Grant (SCG) and The Northern Uganda Social Action Fund (NUSAF 3). The SCG is the main program to mitigate poverty among the elderly – it provides a bimonthly cash grant to those above the age of 80 (60 and 65 in some areas). The government has announced a national roll-out of this program, a vision that effectively transforms the SCG into a social pension for everyone above the age of 80. The NUSAF3 aims to provide effective income support and build the resilience of poor and vulnerable households in Northern Uganda. It also helps beneficiary households and communities to build assets and improve their capacity to adapt to shocks. The project has three primary components, namely: (i) Labor- Intensive Public Works (LIPW) combined with a disaster-risk financing element, (ii) a sustainable livelihoods pilot program, and (iii) a component focusing on strengthening transparency, accountability, and anticorruption systems (World Bank 2020). Figure 151. Incidence of beneficiaries of SCG during Figure 152. Incidence of beneficiaries of NUSAF during the last 12 months in 2019/20 among individuals 60 the last 12 months in 2019/20 among individuals 15 years+ by consumption quintiles, % years+ by consumption quintiles, % Source: UNHS, WB staff calculations. Despite reasonable targeting accuracy and effectiveness for the program beneficiaries, the current levels of expenditure on social protection are rather low. According to the World Bank (2020), the allocation to social development, which includes social protection expenditures, was only 0.7 percent of the overall government budget in FY19/20. Expenditure on the two largest direct income support programs, the SCG and the NUSAF3, 104 was just 0.14 percent of GDP in FY17/18, which is lower than in neighboring countries like Kenya and Rwanda who spend 0.4 percent and 0.3 percent of GDP, respectively, on direct income support. Figure 153. Getting any type of social assistance based Figure 154. Incidence of social assistance by types of on HFPS, % of respondents aid based on HFPS, % of respondents Source: HFPS, WB staff calculations. Note: Cash for work was added to the list of social assistance in the 5th round only. The most recent data confirms very low coverage of SCG and NUSAF programs, but also demonstrates their pro-poor incidence. The UNHS 2019/20 asked if individuals 60 years old and above were beneficiaries of SCF and whether individuals 15 years old and above were beneficiaries of the NUSAF programs during the last 12 months. As shown in Figure 151, about nine percent of all individuals aged 60 years and above benefited from SCG, and the coverage rate was twice as high among the poorest individuals from the bottom consumption quintiles. About 0.6 percent of individuals aged 15 years and above benefited from NUSAF and, similar to SCG, the coverage rate was higher among the poorest individuals (Figure 152). There were significant changes in the magnitude and structure of selected types of social assistance during the pandemic with in-kind non-food transfers having the largest coverage. According to seven rounds of the HFPS, the share of respondents who got any type of social assistance (cash transfers, food aid and other in-kind transfers) ranged from four percent in March/April 2021 to 25 percent in September/October 2020 (Figure 153) The types of social assistance varied a lot as well – with the largest incidence of in-kind non-food transfers, such as mosquito nets, masks and soap – with two exceptions in June 2020 and October/November 2021. After the first lockdown introduced in March 2020, the Government of Uganda distributed food aid, targeting urban areas near Kampala – some of the most affected by the pandemic-related mobility restrictions. This led to a higher incidence of food aid (nine percent of respondents), as reported in June 2020. After the second lockdown introduced in June 2021, the incidence of cash transfers and food aid increased again, as reported in October/November 2021 (Figure 154). Those households who experienced shocks were more likely to be present among chronic poor and those who moved from non-poor to poor status. Using the same households from the UNPS 2015/16 and 2019/20 it is possible to check the relationship between poverty transitions and incidence of shocks (Figure 156). Thus, about 36 percent of the population experienced at least one shock during the last 12 months among those who remained poor across two rounds and moved from being non-poor to poor. This was higher compared to 27 percent among those who remained non-poor and 30 percent among those who moved from being non-poor to poor. 105 Figure 155. Main coping strategy used across key shocks Figure 156. Share of population with at least one shock experienced by households during the last 12 months in during last 12 month across poverty transition status 2019/20, % during 2015/16-2019/20, % Source: UNPS 2019/20, WB staff calculations. Source: UNPS 2019/20, WB staff calculations. Note: Only the first key coping strategies are considered in this Note: The same households were considered during two figure. rounds. Results are population weighted using panel weight for 2015/16 year. The poverty line for the UNPS data was calibrated to give roughly 30 percent poverty in 2019/20. Shocks and coping strategies during the COVID-19 pandemic In order to better understand shocks and coping strategies during the COVID-19 pandemic, data from high-frequency phone surveys was used. Given that the latest round of data collection for the UNPS 2019/20 survey was completed before the pandemic began (in March 2020), the report used results from the first two rounds of the Uganda High-Frequency Phone Survey (HFPS) to explore the shocks and coping strategies households used after Mach 2020 when the first lockdown was introduced. The majority of households in the HFPS reported experiencing at least one shock during March-June 2020, with the poorest and those living in the Northern region having the highest chances of facing shocks. Figure 157 shows that almost 60 percent of the households that participated in the first round of the HFPS in June 2020 experienced at least one shock after March 2020.51 Every fifth household had two or more shocks. The incidence of shocks was higher among the poorest households from the first pre- COVID-19 consumption per adult equivalent quintile (65 percent) and those living in the Northern region (69 percent). According to Figure 158, the most frequent four shocks during the considered period included increase in prices of major food items households consumed (30 percent), followed by non-farm business failure (15 percent), falling output prices (11 percent), and illness of income earning members (7 percent). 51 The incidence of shocks is not comparable across pre-COVID-19 UNPS data and HFPS. Both surveys used a different list of shocks. 106 Figure 157. Incidence of shocks among Ugandans Figure 158. Types of shocks faced by Ugandans during during March–June 2020, % March–June 2020, % Source: HFPS round 1, World Bank staff calculations. Note: Quintiles are based on spatially adjusted by poverty lines consumption per adult equivalent. Similar to the pre-COVID-19 period, the poorest households were less likely to rely on savings as a coping strategy and more likely to reduce food consumption when they experienced a shock.52 Figure 159 shows the structure of all coping strategies used by households to cope with the shocks experienced during March–June 2020. Relying on savings was the most popular strategy, followed in 26 percent of all cases. The second popular strategy was reducing food consumption in 17 percent of all cases. Households from the richest pre-COVID-19 consumption quintile were more likely to use savings compared to households from the poorest quintile (35 percent versus 20 percent, respectively). Every fifth household reported a need to borrow to cope with the COVID-19 emergency. Respondents of the HFPS were asked if they had to borrow to cope with the COVID-19 emergency during March- July/August 2020. About 23 percent of households reported that they had to borrow, with the highest share among households from the Eastern region (36 percent) and the lowest share among households from the Western region (8 percent). 52One important caveat here is that recall period in the HFPS is much shorter than in the UNPS and this may affect coping strategies as well. 107 Figure 159. Type of coping strategies used by households in Uganda during March–June 2020 across different dimensions, % of all strategies Source: HFPS round 1, WB staff calculations. Note: All coping strategies used are included. Quintiles are based on spatially adjusted by poverty lines consumption per adult equivalent. The main reasons for borrowing were drastically different across different socio-economic groups. About 27 percent of those who had to borrow did this because they did not get assistance from family or neighbors, 25 percent had a decline in sales, 24 percent were not able to sell their produce, 18 percent had their business closed, and 6 percent had to borrow because they had lost their job (Figure 160). The stark difference in reasons was observed across different groups of households. Rural households, those in the Northern region, and from the poorest pre-COVID-19 quintile were more likely to report not being able to sell the produce. Female headed households were more likely to borrow because they did not get any assistance from family and neighbors. A decline in sales was mentioned more frequently among households living in the Central region, while a closed business was more frequent among households living in the Western region. Figure 160. Reasons for borrowing to face COVID-19 Figure 161. Level of worry that the household will not emergency during March–July/August 2020, % be able to repay all the money borrowed within the repayment period, % of those who borrowed Source: HFPS round 2, World Bank staff calculations. Note: Quintiles are based on spatially adjusted by poverty lines consumption per adult equivalent. 108 The majority of those who borrowed to face the COVID-19 emergency worried that they would not be able to repay all the money within the repayment period. As shown in Figure 161, about 44 percent of those who had to borrow to face COVID-19 emergency were very worried and 24 were just worried about being unable to repay the money in time. A higher level of worry was observed among households residing in the Western region and among those from the poorest pre-COVID consumption quintile compared to other groups of households. Box 9. Shocks among refugees in Uganda There was no large difference in the incidence of shocks between refugees and Ugandans before COVID-19 in 2018. According to the Uganda refugee survey conducted in 2018, about 63 percent of Ugandan households experienced at least one shock during the last 12 months compared to 59 percent among refugee households (Figure 162). In both groups, shocks affecting agricultural activities prevailed, affecting 52 percent of Ugandans and 42 percent of refugees. Health shocks were in second place (17 percent among hosts and 10 percent among refugees).53 Figure 162. Incidence of shocks among Ugandans during Exposure to shocks among refugees during March-June 2020 and refugees during March-Oct/Nov 2020, the COVID-19 pandemic became almost % of households universal. According to the Uganda Refugee High-Frequency Phone Survey (RHFPS), almost 90 percent of refugee households experienced at least one shock between March and October/November 2020. The most frequent shock was related to an increase in food prices. Even though this is not strictly comparable to the national HFPS as it covered a shorter time period, the incidence of shocks among Ugandans was much smaller, with about 58 percent of households experiencing any shock during March–June 2020. Source: HFPS and RHFPS rounds 1, World Bank staff calculations. The main coping strategies for refugees included reducing food consumption, engaging in additional income generating activities, receiving assistance from NGOs, and doing nothing. Compared to Ugandans, the distribution of coping strategies followed by refugees was very different. Thus, among all coping strategies followed by refugees during March–October/November 2020, using savings was reported by only 3 percent of respondents. For comparison, among all coping strategies followed by Ugandans during March-June 2020, using savings accounted for 26 percent. 53 Comparing impact of COVID-19 on refugees and Ugandans can be found in Atamanov et al. (2021). 109 II. Quantifying vulnerability to poverty54 The concept of vulnerability to poverty As discussed in the previous section, the impact of shocks and the availability of various coping mechanisms has important implications for poverty reduction in Uganda. This is particularly relevant given the scope and nature of vulnerability to poverty among Ugandan households, which will be explored further in this section. The main difference between the concept of poverty and vulnerability to poverty is related to risk and the reference period. Poverty is an ex-post measure of household wellbeing, reflected in the inability to satisfy some basic material needs. In contrast, vulnerability to poverty measures the probability of a household’s future wellbeing, which depends on the exposure to shocks and the ability to cope with them. Without the consideration of future risks, the vulnerability and ex-post poverty measures would be identical. Therefore, vulnerability is always a function of the expected mean and variance of households’ consumption. The mean of consumption depends on household and community characteristics, while variance depends on the shocks that a household faces and the coping mechanisms it can use. Figure 163. Vulnerability to poverty is characterized by the mean and variance of household consumption Figure 163 illustrates the concepts of vulnerability relative to poverty by plotting the expected consumption and variance of five households (adopted from Gao, Vinha & Skoufias 2020). The average consumption associated with many shocks or different periods of time is depicted by red circles, while the variance of consumption is shown by the width of the horizontal black lines around the mean value. Therefore, all households differ from each other based on their mean consumption and its variance. The figure also shows actual consumption at a particular point in time (blue squares) and the poverty line, which allows us to see that households A and B are poor. As such, the poverty rate is 40 percent since two out of five households have actual consumption below the poverty line. Vulnerability is defined as the probability of a household falling below the poverty line in the near future, which is a function of both average consumption and the variance. Let us assume that a household is 54 This section draws heavily from Atamanov, Mukiza and Ssennono (2022). 110 vulnerable if the probability of falling below the poverty line at some point in the next two years is higher than 50 percent (in Figure 163 this would imply a household with more than 50 percent of its variance line being below the poverty line or/and average consumption below the poverty line). Using this definition, households A, D and E are vulnerable, which equates to a vulnerability rate of 60 percent. Households A and D are vulnerable because their average consumption is below the poverty line (poverty induced vulnerability driven by low endowments). Household E is vulnerable as well even though its average consumption is above the poverty line because it has a high estimated variance in consumption with about 70 percent of its variance line below the poverty line, which is higher than the vulnerability threshold of 50 percent. Therefore, household D’s vulnerability is risk induced. Two remaining households, B and C, are not vulnerable. Household C is not vulnerable because neither average consumption nor variance fall below the poverty line. Household B is not vulnerable either, even though its actual consumption was below the poverty line because in the future it is very unlikely to have consumption falling below the poverty line again. Being a graduating student can be viewed as an example for household B. A student may be poor now, but once graduating he or she is very likely to have high and stable income in the near future once he or she gets a job. Distinguishing between risk and poverty induced vulnerability has important policy implications. Addressing poverty induced vulnerability caused by low human capital endowments may require direct transfer programs or programs improving access to basic services. Addressing risk induced vulnerability driven by income fluctuations may require a very different approach based on an insurance type of social assistance. Knowing the relative role of household specific (idiosyncratic) or common (covariate) shocks is also important. This might help policy makers to set up insurance priorities, as insurance mechanisms to address household specific and common shocks can differ significantly. Estimating vulnerability to poverty in Uganda in 2019/20 Poverty and vulnerability rates Half of the population in Uganda was vulnerable to poverty in 2019/20. Households are considered vulnerable if they have a 50 percent or higher probability of falling below the poverty line in the next two years,55 which is equivalent to a 29 percent probability in any given year. Utilizing this definition, the vulnerability to poverty rate in Uganda was estimated to be 50 percent as of 2019/20 (Figure 164). This is much higher than the observed poverty rate of 30 percent in 2019/20. Rural areas are, as expected, characterized by both higher poverty and vulnerability rates (34 percent and 59 percent, respectively) compared to urban areas (20 percent and 26 percent, respectively). Among regions, the highest vulnerability rates were found in the poorest Northern and Eastern regions (66 percent and 64 percent, respectively). In relative terms though, the largest difference between poverty and vulnerability rates was in the Western region, where the vulnerability rate was twice as high as the poverty rate. The lowest difference was observed in the Central region. Vulnerability to poverty was very different across subregions. Figure 165 shows poverty and vulnerability at the subregional level. The highest vulnerability rates were observed in one of the poorest subregions, Karamoja, followed by Acholi and Teso subregions. Several subregions, such as West Nile, Ankole, Bunyoro, Tooro, Elgon and Lango, stand out as having very large differences between poverty and 55 Please note that we used a revised poverty line based on updated consumption basket from UNHS 2016/17. Updated poverty rates were reported first using UNHS 2019/20 in 2021. 111 vulnerability rates, reaching more than 100 percent in all these areas, and reaching even 200 percent in the case of West Nile. Thus, only every fifth person was poor in the West Nile, but every second person in this subregion was vulnerable, or in other words had non-negligible chances to become poor any given year. Figure 164. Poverty and vulnerability rates in Uganda in Figure 165. Poverty and vulnerability rates in Uganda in 2019/20 across rural and urban areas and regions, % 2019/20 across subregions, % Source: UNHS 2019/20, authors’ calculations. Poverty versus risk induced vulnerability Vulnerability rates can be decomposed into the sources of vulnerability. Vulnerability can be either driven by permanent low consumption (poverty-induced) or high consumption volatility (risk-induced). Poverty- induced vulnerability happens when the expected mean of consumption already lies below the poverty line. Households face risk-induced vulnerability when their expected consumption is higher than the poverty line, but high estimated variance leads to a probability higher than the established threshold of 29 percent for a given year. Figure 166. Vulnerability decomposition into poverty- and Figure 167. Vulnerability decomposition into risk-induced components across rural and urban areas in poverty- and risk-induced components across 2019/20, and regions, % subregions in 2019/20, % Source: UNHS 2019/20, authors’ calculations. 112 Rick-induced vulnerability prevailed in urban areas, while poverty-induced vulnerability prevailed in the poorest Eastern and Northern regions. Figure 166 shows that, at the national level, risk-induced vulnerability slightly prevails over poverty-induced vulnerability. The difference widens in urban areas, where high consumption volatility accounts for almost 70 percent of all vulnerability. In contrast, in rural areas vulnerability is equally driven by low consumption and its high volatility. At the regional level, vulnerability in the poorest Eastern and Northern regions is poverty-induced to a larger extent, while in the Central and Western regions vulnerability was more associated with high consumption fluctuations. At the subregional level, risk-induced vulnerability prevails over poverty-induced vulnerability in most places with some notable exceptions (Figure 167). Thus, in subregions with high vulnerability rates such as Karamoja, Teso, and Acholi, vulnerability is mostly associated with permanent low consumption, which can be related to few household assets and low human capital endowments. Idiosyncratic and covariate shocks Further analysis focuses on the impact of idiosyncratic and covariate shocks on vulnerability to poverty by decomposing consumption variance. Idiosyncratic shocks include household specific shocks such as health issues, job losses or business closures and so forth, which are weakly correlated across households living in the same community. Covariate shocks, in contrast, are correlated across households within communities, or in other words, households living in a community were exposed to similar shocks. These may include price and weather shocks, political crises and so forth. Figure 168. Ratio of idiosyncratic to covariate shocks Figure 169. Ratio of idiosyncratic to covariate shocks across rural and urban areas and regions in 2019/20, % across subregions in 2019/20 Source: UNHS 2019/20, authors’ calculations. The impact of idiosyncratic shocks was consistently higher than the impact of covariant shocks, but in relative terms the role of idiosyncratic shocks is less prevalent in rural areas and in the poorest Northern and Eastern regions. Figure 168 and Figure 169 present the ratios between the percentage of households that would fall below the poverty line from an idiosyncratic shock and the percentage of households that would fall below the poverty line from a covariate shock. The estimates reveal that both at the national and regional levels, the impact of idiosyncratic shocks is consistently higher than the impact of covariate shocks. However, in relative terms, the role of idiosyncratic shocks is much more prevalent in urban areas and the Central and Western regions, compared to rural areas and the Northern and Eastern regions. This implies that covariate shocks such as weather, locust, or price shocks play a more important role for rural residents. High heterogeneity was observed in the role of covariate and idiosyncratic shocks across 113 subregions as well. For example, in Buganda South, the impact of idiosyncratic shocks was dominant, while in Karamoja, vulnerability was affected almost equally by both idiosyncratic and covariate shocks.56 Profile and correlates of vulnerability Households can be categorized into four mutually exclusive groups based on estimated vulnerability and actual poverty status: poor-vulnerable, not poor and not vulnerable, poor and not vulnerable, and the last group of non-poor, but vulnerable. Several figures below show the distribution of four groups depending on the household characteristics. Thus, only 34 percent of the population in households where the household head is employed in agriculture are not vulnerable and non-poor compared to 94 percent among those with household heads employed in financial and business services (Figure 170). Having the head of a household employed in manufacturing or construction sectors is also associated with a high level of poverty and vulnerability. Vulnerability and poverty depend as well on the head of household education. If a head of household has higher than a secondary education, this household is almost fully out of risk of being poor or vulnerable, while among households with uneducated heads this risk is higher than 80 percent (Figure 171). Figure 170. Poverty and vulnerability nexus by head of Figure 171. Poverty and vulnerability nexus by head of household’s sector of employment in 2019/20, % household’s education level in 2019/20, % Source: UNHS 2019/20, authors’ calculations. Education, assets and working status were found to be the main determinants of vulnerability. In order to check the correlates of vulnerability to poverty, the probit regression is estimated to see the marginal effects for each variable, keeping the impact of other variables constant. Results are shown in Figure 172. Education of the head of household remains the key determinant of vulnerability. Thus, households where the household head has higher than a secondary education (the smallest group accounting for about seven percent of all households) have lower probability of being vulnerable by 57 percent compared to households where heads do not have any education. For households where the household head has a primary incomplete education (the largest group accounting for more than 35 percent of all households) the probability of being vulnerable drops by 15 percent, relative to households where the head has no education. The findings also include a much higher probability of being vulnerable among households 56 Robustness of the obtained results was checked by changing the assumption about the share of measurement error in the estimated individual variance in consumption. Even in the presence of very high measurement error (25 and 50 percent), the results do not change qualitatively. There is still a relatively higher impact of covariate shocks on rural households’ consu mption relative to urban households. Results are available upon request. 114 living in the Eastern, Northern, and Western regions compared to those living in the Central region. Ownership of selected assets such as TVs and motorcycles, used as a proxy of monetary wellbeing, were also associated with significantly lower chances of being vulnerable. Having the head of household out of the labor force or in subsistence farming increases the probability of being vulnerable compared to having the head of household in paid employment, but the relationships are not as strong as for head of household education level or asset ownership. Figure 172. Probability of being vulnerable to poverty with regards to non-vulnerable in 2019/20, marginal effects Source: UNHS 2019/20, authors’ calculations. Note: Base category is in parentheses. 115 III. Policy implications Using only the poverty rate may not be enough to understand the dynamics of poverty in Uganda, and shocks – if not properly addressed – will undermine sustainable poverty reduction in the country. Consistent with the high incidence of shocks, half of the population in Uganda was estimated to be vulnerable to poverty in 2019/20, which was much higher than the poverty rate of about 30 percent recorded in the same period. The types of vulnerability differ substantially across rural and urban areas. Vulnerability in urban areas and the Central and Western regions is predominantly risk-induced or due to high volatility of consumption. Vulnerability in rural areas can be equally attributed to high volatility (risk- induced) and poverty-induced (due to low human/physical capital and assets), while in the Northern and Eastern regions, vulnerability to a larger extent is associated with low consumption or poverty-induced. Higher risk-induced vulnerability among urban households in relative terms calls for social security type policy measures. The almost equal importance of risk- and poverty-induced vulnerability in rural areas calls for policy measures that can tackle both structural poverty and high consumption volatility. Expanding safety nets is an important strategy to help the vulnerable face shocks without falling (deeper) into poverty. With the data showing that the poor and vulnerable experience the greatest number of shocks and have limited coping mechanisms to deal with them, targeted social safety net schemes could mitigate some of the worst impacts of temporary shocks. In particular, efforts should be made to ensure that households do not have to resort to coping mechanisms that jeopardize their long- term wellbeing and future prospects, such as reducing food consumption and forgoing investments in the health and education of children. Rural areas are characterized by both higher poverty and vulnerability rates compared to urban areas, highlighting the need to increase the resilience of the agricultural sector. Almost 60 percent of rural Ugandans are vulnerable to poverty compared to just 26 percent of urban Ugandans. Furthermore, the Northern and Eastern regions, which depend mainly on agriculture, have particularly high vulnerability rates at almost two-thirds of the population. Importantly while idiosyncratic shocks are more prevalent on the national level, covariate shocks, such as weather or/and price shocks, are more important for rural residents. Therefore, increasing the resiliency of the agricultural sector could go a long way to reducing the overall levels of vulnerability for rural Ugandans. More broadly speaking, diversifying household incomes and increasing access to non-farm jobs is fundamental to improving resilience and reducing vulnerability to shocks. In order to achieve this, investing more in human capital accumulation, particularly education, is essential. A recurrent finding in the vulnerability analysis is that households with higher levels of education were less likely to work in agriculture, were less vulnerable to shocks, and have better coping mechanisms for dealing with shocks. Investments in education take time to pay off, and thus more immediate strategies should also be employed to help households better cope with shocks in the near term. Nevertheless, for Uganda’s long- term future, especially in light of climate change, structural change (reducing the share of labor in agriculture) and increasing education levels are vital for the country to successfully reduce poverty and vulnerability. Refugees experienced significantly more shocks than Ugandans after COVID-19. The most common shock among refugees was an increase in the price of key food items, followed by illness, injury or death of income earning household members. Reducing food consumption remained the most frequently used 116 coping strategy among refugees, while using savings as a coping strategy was almost non-existent among refugees, suggesting a lack of savings to draw from. Continuing the roll out of shock-responsive cash- based assistance to help refugees increase their consumption options remains a crucial policy measure. 117 References AfroBarometer data. Uganda, round 7. Atamanov, A.; Mukiza, C.N.; Ssennono, V.F. (2022). Quantifying Vulnerability to Poverty in Uganda. Washington, DC: World Bank. © World Bank. Atamanov, A.; Beltramo, T.; Reese, B. C.; Rios Rivera, L. A.; Waita, P. (2021). One Year in the Pandemic: Results from the High-Frequency Phone Surveys for Refugees in Uganda. World Bank, Washington, DC. Centre for Research on the Epidemiology of Disasters. The EM-DAT. The international disaster database. Chaudhuri, S. (2003). Assessing vulnerability to poverty: Concepts, empirical methods and illustrative examples. New York: Mimeo, Department of Economics, Columbia University. Gao, J.; Vinha, K.; Skoufias, E. (2020). World Bank Equity Policy Lab Vulnerability Tool to Measure Poverty Risk. Poverty and Equity Notes; No. 36. World Bank, Washington, DC. Goldstein, H. (1997). Multilevel models in educational and social research. London: Griffin. Gunther I. & Harttgen, K. (2009) Estimating Household Vulnerability to Idiosyncratic and Covariate Shocks: A Novel Methods Applied in Madagascar, World Development, Vol. 37, No. 7, pp.1222-1234. Hill, Ruth; Mejia-Mantilla, Carolina. (2017). With a Little Help: Shocks, Agricultural Income, and Welfare in Uganda. Policy Research Working Paper; No. 7935. World Bank, Washington, DC Naudé, W.; Santos-Paulino, A.; & McGillivray, M. (2009) Measuring Vulnerability: An Overview and Introduction, Oxford Development Studies, 37:3, 183-191, DOI: 10.1080/13600810903085792 Notre Dame Global Adaptation Initiative (ND-GAIN). ND-GAIN country index. Skoufias, E.; Vinha, K.; Beyene, B.M. (2021). Quantifying Vulnerability to Poverty in the Drought-Prone Lowlands of Ethiopia. Policy Research Working Paper; No. 9534. World Bank, Washington, DC. World Bank. (2019). Detour from the Poverty Reduction Path: Uganda Poverty Update Note - Uganda National Household Survey 2016-17. World Bank, Washington, DC. World Bank. (2020). Uganda Economic Update, 14th Edition, February 2020: Strengthening Social Protection to Reduce Vulnerability and Promote Inclusive Growth. World Bank, Washington, DC World Bank (2021). Climate Risk Profile: Uganda. World Bank, Washington, DC 118 Annex 2 In most household level studies that measure the vulnerability to poverty, vulnerability is related to the propensity or probability of a household experiencing an episode of poverty or a shock that brings its welfare below a socially accepted level (often established as a poverty line). As a result, assessing vulnerability requires information on the mean and variance of the household welfare collected ideally from many repeated observations in different periods of time. In reality, there are very few countries, particularly in the developing world, where panel data tracks the same households for more than 15–20 years, and this prevents the construction of a reliable estimate of the mean and variance. Therefore, the most feasible approach is to estimate vulnerability using cross-sectional data by making some strong assumptions (Naudé, Santos-Paulino & McGillivray 2009; Skoufias, Vinha and Beyene 2021). In this study, the method used is based on Gunther and Hartgen (2009) who integrated a two-level hierarchical model (Goldstein, 1999) into Chaudhuri’s (2002) method. This allows one to differentiate unexplained variance at the household level (related to idiosyncratic household specific shocks) from unexplained variance at the community level (covariate community level shocks). This method also corrects for inefficient estimators that might occur if Chaudhuri’s (2002) method is applied to hierarchical data with household and community level variables used simultaneously in consumption modelling (Gunther and Hartgen, 2009). The main hypothesis which the method makes is that unexplained variance in the consumption of otherwise equal households captures the impact of both household-specific and community-specific shocks on households’ consumption. This variance is assumed to be correlated and can be explained by observable household and community characteristics. Following Gunther and Hartgen (2009) and Skoufias, Vinha and Beyene (2021), let =(1,… , n) denote households at level one and =(1,… , j) denote communities at level two, with households being nested within communities. Consumption, c, of household i in community j is specified as = 0j + 1j + (1) The coefficients (both constant terms and slopes) of each community are assumed to be affected by community observed characteristics (Z) and community unobserved factors denoted by 0 and 1 0j = 00 + 01 + 0 (2) 1j = 10 + 11 + 1 (3) Substituting (2) and (3) into (1) gives a regression equation = 00 + 01 + (10 + 11 ) + 0 + 1 + (4a) Thus, there are three error terms to be estimated 0 , 1 and where captures the idiosyncratic shocks and 0 and 1 capture the community covariate shocks. In particular, 0 captures the direct effect of covariate shocks affecting the intercept of each community and all households in the same community in the same manner, while 1 captures the indirect effect of covariate shocks. Equation (4a) may be rewritten as equation (4b) = 00 + 01 + (10 + 1 ) + 0 + 11 + (4b) 119 During the first stage, the equation (4b) is estimated using Stata’s command for mixed-effects maximum likelihood regression. In the second stage, the squared residuals and their squares sum from equation (4a) may be regressed on and : 2 = 0 + 1 + 3 (5a) 2 = 0 + 1 (5b) 2 2 (0 + ) = 0 + 1 + 3 (5c) The estimated coefficients of equations (4a), (5a), (5b) and (5c) may then be used to estimate the expected 2 2 2 mean and the expected idiosyncratic , covariate , and the total + of a household’s consumption based on the observed characteristics of both the household and the community. Assuming that consumption is log-normally distributed, one can estimate the probability of consumption falling below the poverty line z and using a threshold for this probability (for example 50 percent) to define who is vulnerable to poverty. The probability of consumption falling below the poverty line z is estimated by assuming that consumption is log-normally distributed, where X and Z are household and community level characteristics i.e. ⌃ −⌃ = ( < |, ) = ∅( 1/2 ) (6) ̂ (2 0+ ) Where ∅ denotes the cumulative density of the standard normal distribution function; lnz, the log of ⌃ poverty line; the expected mean of per capita (log) consumption. Expression (6) can be used to derive an estimate of the vulnerability to poverty from covariate or ̂ ) by ̂ community level shocks by replacing ( 2 2 , while an estimate of the vulnerability to poverty 0+ 0 ̂ from idiosyncratic shocks by using 2 ̂ 2 in place of (0+ ). The last step to identify a vulnerable household involves a selection of a threshold for the probability of being poor. Following previous empirical literature (Gunther and Hartgen, 2009; Skoufias, Vinha & Beyene, 2021) households are considered vulnerable if they have a probability of 50 percent or higher to fall below the poverty line at least once in the next two years. This is equivalent to a 29 percent or higher probability in any given year. This threshold is chosen in the current study. 120 Table A.1 Descriptive statistics for the variables selected for the consumption model by areas and across consumption per adult equivalent quintiles Source: UNHS 2019/20, authors’ calculations. Note: Initial list of variables was larger, but this table contains only highly significant variables used in the final model. Table A.2 Regression results: estimates of consumption equation (log consumption per capita) Variable Log Per Capita Consumption agriculture is the main activity in the community -0.190*** (0.031) the community is connected to the national grid 0.051*** (0.017) community with migration (people were coming or leaving community) -0.297*** (0.045) community benefited from new road during last 3 years 0.059*** (0.016) household owns fridge 0.388*** (0.029) household owns television 0.297*** (0.016) household owns radio 0.049** (0.025) household owns motorcycle 0.136*** (0.041) household owns bicycle 0.067*** (0.013) household size -0.102*** (0.005) age of household head 0.004*** (0.000) share of children 0-13 and elderly 60 plus in total household size -0.059*** (0.019) each member has a pair of shoes 0.202*** 121 Table A.2 Regression results: estimates of consumption equation (log consumption per capita) Variable Log Per Capita Consumption (0.011) each member had two pairs of clothes 0.120*** (0.018) at least one household enterprise 0.049*** (0.011) head of household is not working -0.097*** (0.013) head of household is in subsistence farming -0.048*** (0.013) household head has incomplete primary education 0.104*** (0.014) household head has complete primary education 0.158*** (0.017) household head has some secondary education 0.207*** (0.017) household head has post-secondary education 0.417*** (0.025) energy used for cooking is charcoal -0.280*** (0.039) energy used for cooking is firewood -0.412*** (0.038) community with migration*charcoal 0.197*** (0.046) community with migration*firewood 0.224*** (0.048) agricultural community*household size 0.027*** (0.005) agricultural community*motorcycle 0.196*** (0.046) agricultural community*radio 0.074*** (0.027) Constant 11.309*** (0.050) Observations 11,895 Number of groups 1,424 Source: UNHS 2019/20, authors’ calculations. Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses. 122 Figure A10. Predicted consumption distribution among Ugandans Source: UNHS 2019/20, authors’ calculations. 123 Chapter 4. Inequality of opportunities Building human capital is crucial for economic development. Human capital broadly consists of the knowledge, skills, and health that people accumulate over their lives, enabling them to realize their potential as productive members of society. This section focuses on the most recent performance of Uganda in achieving equality of opportunities in basic services for children. I. Background Equality of opportunities is related to the idea that some critical human capital inputs should be available to all people regardless of their background. Opportunities are described as access to basic goods and services such as primary or/and secondary education, health services, clean water, adequate sanitation and so forth (OECD and World Bank 2006). In most societies, these opportunities play an essential role in shaping a person’s life. According to Roemer (1998), equality of opportunities requires that an individual’s opportunities do not depend on his/her circumstances such as gender, place of birth, wealth of parents, religion and so forth. In other words, the outcomes and achievements in an individual’s life should only depend on his/her effort and abilities. This concept is less debatable than the concept of equality of outcomes, which is often measured by the degree of inequality in incomes. Equality of opportunities within a country forms the basis for growth and shared prosperity. Several studies showed that improving inequality of opportunities was not only about fairness and building a just society, but also about enhancing society’s aspirations of economic prosperity. Thus, Molina, Narayan, and Saavedra (2013) showed that inequality of opportunities explains a lot of differences in income levels across countries. Marrero and Rodríguez (2013) found a negative relationship between inequality of opportunities and economic growth in the United States, while Grimm (2011) found a strong negative relationship between health inequality and growth in low- and middle-income countries. The Human Opportunity Index (HOI) is often used to measure inequality of opportunities among children. The HOI was developed as a measure of a society’s progress toward equitable provision of opportunities for all children (Barros et al. 2009). The HOI explores how the personal “circumstances” for which a child cannot be held accountable – like location, gender, household composition, or parental wealth – can affect the child’s probability of accessing basic services that are necessary to succeed in life, like timely education, clean water, electricity, decent housing and so forth. In other worlds, the HOI is an inequality-sensitive coverage rate, which “penalizes” the extent to which different circumstance groups have different coverage rates for basic or critical services – the penalty is zero if coverage rates across different circumstance groups are equal and positive and increasing as differences in coverage among circumstance groups increase. More details about the method are provided in Box 10. The HOI is a useful tool for policy makers. The HOI indicates which socio-demographic characteristics influence a child’s likelihood of access to a particular opportunity and identifies the ones with the largest contributions to inequality. Changes in the HOI over time can also be decomposed into two components: change in coverage and equalization effect. Therefore, the HOI can be used to monitor progress towards two important goals: (1) ensuring that as many people as possible access opportunities, and (2) allocating new opportunities first to those who are at a disadvantaged position because of their circumstances. 124 Calculating the HOI is of particular interest for Uganda, where the future path of prosperity depends heavily on investment in the youth. According to the World Bank (2020), Uganda is entering the early stage of its demographic transition, with declining child mortality and decreasing (but still relatively high) fertility rates. As a result, the population – currently estimated at about 46 million – is projected to rise to around 106 million by 2060. With the right set of policies and conditions in place, the country has the opportunity to capitalize on the demographic dividend, as around 70 percent of that population will be of working age. This will, however, require that the young population is able to accumulate the human capital necessary to reach its full productive potential, regardless of the conditions into which one is born (such as gender, location, etc.). It makes sense, therefore, to focus on inequality of opportunities among children in Uganda using the HOI. Analysis of inequalities in opportunities in this section focuses on the period from 2012/13 to 2019/20. It utilizes three traditional household surveys to measure inequalities before the COVID-19 shock by constructing the HOI and uses results from the phone survey to measure the impact of COVID-19. Where possible, additional information is used to better understand low access/usage of selected opportunities. II. Human Opportunity Index for Ugandan children Uganda’s most recent performance in human capital Before COVID-19, Uganda demonstrated average performance on the Human Capital Index (HCI). First launched in 2018, the HCI measures the amount of human capital that a child born today can expect to attain by age 18. It is designed to highlight how improvements in current health and education outcomes shape the productivity of the next generation of workers, assuming that children born today experience the same educational opportunities and health risks over the next 18 years as children currently in this age range. According to the HCI-2020 (Figure 173), a child born in Uganda today will be only 38 percent as productive when she grows up as she could be if she enjoyed complete education and full health.57 This performance is slightly better than the average in low-income countries, but lower than the average observed in Sub-Saharan African countries. Compared to the HCI-2010, which was equal to 34 percent, the progress Uganda has made in the various components of the HCI include an increase in the share of children who are not stunted, increases in harmonized test scores, and higher survival rates for children and adults. 57 Full health implies no stunting and 100 percent adult survival, while compete education implies 14 years of high-quality school by age 18. 125 Figure 173. Human capital index in Uganda and selected comparators in 2020 Source: World Bank. Note: SSA stands for Sub-Saharan Africa, LIC for low-income countries and ECA for Europe and Central Asia. Uganda performed better in 2020 than countries at comparable levels of economic development in terms of harmonized test scores, children’s probability of survival, and stunting rates,58 but lagged behind comparator countries in indicators related to years of schooling and adult survival rates. Performance of Uganda in different HCI components is shown in Figure 174. Uganda had better than average performance in the probability of children to survive – 95 out of 100 children born in Uganda survived to age 5 compared to 93 out of 100 children in Sub-Saharan Africa countries on average. About 71 out of 100 children were not stunted nor at risk of cognitive and physical limitations that can last a lifetime – performance was slightly better than average for African countries (69 children). Students in Uganda scored 397 on a scale where 625 represented advanced educational attainment and 300 represented minimum educational attainment. That performance was better than the average score of 374 among SSA countries. Two dimensions of HCI in Uganda were significantly behind comparators: expected years of schooling and learning adjusted years of schooling. Specifically, as of 2020, the expected years of schooling in Uganda was equal to 6.8 years, compared to 8.3 years on average among African countries. The learning-adjusted years of schooling was even lower, at just 4.3 years in Uganda, compared to the African average of 5 years. Human capital achievements in Uganda were not distributed equally and masked a lot of variation by socio-economic status and location. National averages of HCI components reported in Figure 174 do not provide information on inequalities in human capital within each country and may mask persistent and even rising inequality among groups. For example, the percentage of children who were not stunted ranged from 67 percent to 83 percent between the poorest and the richest quintile of households. The subregional disparities were even wider, ranging from 59 percent to 86 percent. These examples highlight the importance of looking beyond national averages and exploring inequalities across different groups of the Ugandan population. This is done in the next section by constructing the HOI across several indicators for children in Uganda. 58It is important to remember that many subcomponents used in the HCI are quite dated. For example, information on expected years of schooling go back to 2013 year, stunting from 2016 year. The Human Opportunity Index constructed in this section uses the most recent data from the latest household survey 2019/20, and changes/trends focus on the period from 2012/13 to 2019/20 including the middle point of 2016/17. 126 Figure 174. Selected components of HCI in Uganda and other countries in 2020 versus logarithm of GDP per capita 2017 PPPs Source: World Development indicators, accessed on 11/23/2021. Constructing Human Opportunity Index for Uganda Defining the opportunities Opportunities are defined as access to basic goods and services for children ages 0–16 years. In this analysis, access means utilization. For example, if there is a school nearby but a child does not attend it because parents do not value education, for our purpose this will imply that a child does not have access to schooling. The assumption is that the society and the family are jointly responsible for establishing all the necessary conditions for the child to actually utilize the goods or service. Another important thing is the quality of services. For some opportunities, it is possible to take this into account, for example, for improved water or sanitation, but for some opportunities related to education the analysis will have to rely on imperfect proxy information. For example, completion of primary school on time will serve as a proxy for quality of school but will also capture student achievements. 127 Table 8 provides a list of the indicators selected as opportunities for the HOI analysis. It is not exhaustive and does not cover a much wider list of opportunities that should be available to any child to achieve his/her potential in life. The list in Table 8 was formed by selecting key opportunities and following the data availability. In selecting and defining the opportunities, the analysis was guided by the Sustainable Development Goals as based on the global consensus on priorities for developing countries and previous studies (Barros et al. 2009; Dabalen et al. 2015; Krishnan et al. 2016). The Uganda National Household Survey (UNHS) was selected as a main data source for constructing the HOI. The UNHS is a large survey, representative at the national, regional, and subregional levels. It is also the only survey used to monitor official poverty. Five indicators were created to measure opportunities in education, following Dabalen et al. (2015) in selecting them. Education opportunities measure attendance and achievements. School attendance is measured for children ages 6–12 years (primary school age in Uganda starts at age 6, lasts seven years, and is free and compulsory) and for children ages 13–16 (lower secondary school age in Uganda starts at 13, lasts four years, not compulsory, but free for those passing minimum threshold in examination). Two other indicators measure achievement or school quality and child’s ability by using proxy indicators, as UNHS does not collect information on test scores. In particular, the analysis uses indicators for starting primary school on time (for children ages 6–7 years) and finishing primary school (for children ages 13–16 years). Children starting primary school on time are more likely to get necessary educational inputs at an early age. Similarly, children who complete primary school on time are more likely to have achieved minimum learning to progress through grades without repetition. The fifth education opportunity considered here assesses the home environment for conducive learning by checking availability of desk and chair for doing homework in households with at least one child between the ages of 6–16 years. This is not the most important input for a child to complete homework, but the only objectively measured indicator available in 2016/17 and 2019/20 surveys. Health opportunities are limited to only one indicator in this analysis. Typically, opportunities in health services measure the absence of stunting, malnutrition and wasting. They may also include whether mothers received any pre-natal care and medical assistance during the birth. Unfortunately, this information is not collected in the UNHS. The only imperfect measure of access to health services is based on the question whether the household with a child between the ages of 0–16 years was able to visit a health facility when ill and got all medication prescribed to treat the illness. This indicator can only be constructed for 2016/17 and 2019/20 surveys. Table 8. List of opportunities education school attendance (children ages 6–12 years) school attendance (children ages 13–16 years) started primary school on time (children ages 6–7 years) finished primary school on time (children ages 13–16 years) availability of desk and chair at home for homework (children ages 0–16 years) health visit to a health facility when ill (children ages 0–16 years) basic services improved drinking water (children ages 0–16 years) improved and unshared sanitation (children ages 0–16 years) dwelling has a hand washing facility next to the toilet (children ages 0–16 years) electricity from and off-grid (children ages 0–16 years) electricity from grid (children ages 0–16 years) 128 Opportunities in basic services include access to improved drinking water,59 improved sanitation, electricity from the national grid, electricity from and off-grid, and availability of hand washing facility with water and soap or just with water next to toilet.60 Water and sanitation are key drivers of public health, as they are correlated with the incidence of diarrhea and serious health consequences. Electricity is an important determinant of the quality of life and facilitates access to other opportunities, including access to digital technologies, information and studying. Box 10. Explaining the intuition of HOI using a simple example The HOI is defined as the difference between two components: (i) the overall coverage rate of the opportunity (C); and (ii) a “penalty” for the share of access to opportunities that are distributed in violation of the equality of opportunity principle (P). Computing P requires identifying all circumstance groups with coverage rates below the average rate. If all groups have the same coverage rate, the penalty will equal to zero. Using a penalty allows one to calculate the inequality-adjusted coverage rate, or HOI. Thus, the HOI=C-P. The HOI can also be calculated using the coverage rate and dissimilarity index (D). HOI=C(1-D). D is a measure of relative inequality in the allocation of the opportunities and can be obtained by dividing the penalty, P, by the overall coverage rate, C. The example below shows how the HOI, P and D are calculated using a simple example. Number of children ages 6-10 years enrolled in primary school Groups by circumstance Country A Country B Group I, top 50 percent richest households 40 35 Group II, bottom 50 percent poorest households 20 25 Total 60 60 Imagine two countries each having a population of 100 primary-school aged children. In each country, children are grouped into the 50 poorest and 50 richest groups based on per capita household income. The average primary enrollment rate in both countries is 60 percent, however, educational opportunity is not distributed equally across groups. The principle of equality will hold only if each group has 30 kids enrolled in primary school and the same coverage. In reality, children from the bottom 50 percent of income in country A only have 20 kids enrolled in primary school, and the children from the bottom 50 percent of income in country B only have 25 kids enrolled in primary school. This suggests inequality in opportunity in both countries and that country A is more unequal than country B. More formally, the D-index for country A is 10/60 and 5/60 for country B. The HOI index for country A is equal to 0.5 and for country B 0.55. In sum, despite equal coverage rates, the HOI is lower and the D-index is higher for country A, signaling that inequality in access to education opportunities is higher than in country B. HOIa = Ca (1-Da) = 0.6 * (1-10/60) = 0.50 and Pa = Ca*Da = 0.6 * (10/60) = 0.10; HOIb = C b (1-D b) = 0.6 * (1-5/60) = 0.55 and P b = C b * D b = 0.6 * (5/60) = 0.05 Calculating the D-index when there are multiple circumstances is done econometrically. The HOI has a number of appealing features. For example, the HOI will increase by a factor k if coverage for all groups increases by a factor 59 We do not take into account the time needed to get the drinking water. Official indicator does not view the source as improved if collection time exceeds 30 minutes for a roundtrip, including queuing, therefore our definition of access presents an upper bound and the coverage rate is higher than the number reported in the WDI. 60 Access to basic services is available only at the household level, so we assume that all children living in the household enjoy identical access to these services. 129 k. If coverage for one group increases without decreasing coverage for other groups, the HOI increases. If inequality declines and overall coverage remains constant, or overall coverage increases while inequality remains constant, the HOI will always improve. Defining the circumstances In order to achieve equality of opportunities, the circumstances – defined as exogenous characteristics of the child – should not be correlated with having access to a basic good or service. Many of them do affect access to basic services, and the analysis will identify how much they matter. For this analysis, we have selected the following set of circumstances: household demographic composition, child’s gender, head of household’s education level, monetary wellbeing, and location (Table 9). Household composition includes the number of children (ages 0–16 years), presence of at least one elderly person (ages 60 years and above) and the presence of both household head and spouse in the household. Child characteristics are represented by one variable – gender. Head of household characteristics include gender and education level. Monetary wellbeing is measured by average household consumption per adult equivalent.61 Finally, location variables include rural/urban and regional dimensions. It is important to remember that the HOI changes if circumstances change, so the results in this section hold for a selected set of circumstances and are subject to change if circumstances are changed. If new circumstances are added, the HOI will be always lower, meaning that provided the HOI/inequality always serve as an upper/lower bound of the “true” HOI/inequality based on all possible circumstances. Table 9. List of circumstances household members’ characteristics number of children in the household (ages 0–16 years) presence of elderly (ages 60 years and above) household head and spouse both present child characteristics child’s gender head of household characteristics head’s gender head of household’s education level monetary wellbeing average household consumption per adult equivalent location urban areas regions Progress towards opportunities for all children in 2019/20 Opportunities in access to education and health There is a substantial gap between opportunities related to school attendance and those related to completion and start of school. Figure 175 shows the coverage rates (dots) and HOI (bars) with the gap between the two reflecting the penalty due to inequality of opportunity among children of different circumstances. Opportunities in education are split into two types related to attendance and those related imperfectly to the quality of education (starting school and finishing primary school on time). The first observation is that none of the opportunities have universal coverage, which can be viewed as an aspirational goal for a society. The average HOI for primary school attendance was the highest, at close to 88 percent. School attendance for the elder cohort of children (ages 13–16 years) was slightly lower, but 61 Consumption is spatially deflated by using poverty lines. 130 still quite high, at close to 84 percent. The picture changes dramatically for indicators related to the quality of education. Specifically, the HOI for starting school on time for children ages 6–7 years was only about 51 percent and the HOI for finishing primary school on time was much lower, close to 18 percent. The inequality, measured by the D-index, was also much higher for opportunities related to the quality of education. Finally, the HOI for having an environment conducive to study by having a desk and chair was the lowest, reaching only 11 percent among children ages 6–16 years. The HOI for this opportunity was also much lower than the coverage (11 percent versus 17 percent), which, along with the highest D-index, signals substantial inequality of access across selected circumstances. Figure 175. The coverage, HOI and D-index for access to education and health among children in 2019/20, % Source: UNHS 2019/20, World Bank staff calculations. Note: Age of children is shown in brackets. Inequality in education opportunities was explained by location and monetary wellbeing to a large extent, but with some important differences across types of opportunities. Figure 176 shows relative contributions of circumstances for each education opportunity in 2019/20. Location explains more than 40 percent of all inequality in school enrollment for primary age children, with monetary wellbeing being the second largest contributor and accounting for 29 percent of all inequality. As shown in Table 3, the Northern region stood out as having much lower coverage in school enrollment among primary age children. The role of circumstances in school participation changed dramatically for children of lower secondary school age of 13–16 years. Household composition accounted for the largest contribution to inequality. Among household composition, the most important circumstances were the presence of both parents in a household and the gender of the household head. This may be related to the demand for child labor in households with only one parent present or with a female head of household. Inequality in starting school on time was driven by monetary wellbeing and head of household education levels. Thus, only 45 percent of children ages 6–7 years had started schooling among households with an uneducated head compared to 75 percent of children ages 6–7 years in households where the household head had completed secondary education. Inequality in finishing school on time was related the most to location differences, with the Northern region’s coverage rate four times lower than that of the Central region (10 versus 43 percent, respectively). Location and monetary wellbeing were the key contributors to inequality of having a desk and chair to study among children ages 6–16 years. Indeed, variation in access to this opportunity across consumption quintiles was striking. It varied from five percent among the poorest bottom quintile to 38 percent among the richest fifth quintile (Table A.5). 131 Figure 176. Contribution of each circumstance to inequality of opportunities in education and health in 2019/20, % Source: UNHS 2019/20, World Bank staff calculations. Note: age of children is shown in brackets. The child’s young age and the school being too expensive were the most frequently mentioned reasons for not attending school. Besides constructing the HOI, additional data from the UNHS is used to explore the reasons why children did not attend school. This information will help to better understand the reasons behind low education opportunities in Uganda. Figure 177 reports the main reasons for not attending school across regions, area, gender of a child and consumption quintiles. The most common reason for children not attending school at the national level was being too young (33 percent), even though only school age children were considered. Schools being too expensive were the second most frequently cited reason (22 percent). Other reasons roughly accounted for 7–13 percent each. Some interesting patterns in answers emerged across groups. For example, being expensive played a more important role for the richest children (45 percent) than the poorest (22 percent). Being too young was mentioned in 50 percent of cases in the Eastern region compared to only 19 percent in the Central region. Need to help at home, on the farm or with a business was important for children in the Northern region (16 percent) and among female children (12 percent) but was ineligible in the Western region (1 percent) and less important for male children (5 percent). Disability played an important role for no attendance in the Central region (24 percent) but only accounted for 4 percent of cases in the Northern region. Visits to health facilities and access to medication when needed were accessible for only 67 percent of children, with significant inequality across groups with monetary wellbeing playing the key role. As shown in Figure 175, the HOI in access to health services was close to 61 percent which was 6 percentage points lower than the average coverage rate of 67 percent. The main contributing factor for inequality in access to health services was monetary wellbeing measured by consumption per capita quintiles. It accounted for more than 55 percent of inequality compared to only 17 percent related to location. Only 48 percent of the poorest children from the bottom quintile were able to visit health facilities when ill and access medication, compared to very high access rate of 91 percent among children from the richest quintile (Table A.5). 132 Figure 177. The reasons for not attending the school among children 6–16 years old by different circumstances in 2019/20, % Source: UNHS 2019/20, World Bank staff calculations. Opportunities in access to basic services Access to all infrastructure services in Uganda except drinking water was very limited and very unequal. Figure 178 shows the coverage rates and the HOI for a selected set of key services, including improved drinking water, improved sanitation, hand washing facility, and electricity. For all services, except drinking water, the coverage in 2019/20 was very low. Access to electricity improves substantially if electricity off grid (mostly solar) is counted in, but still the coverage was below 60 percent. This is also an upper bound given that the quality of the electricity supply off-grid is not known. Access to improved unshared sanitation and hand washing facilities was the lowest. Only 17 percent of children ages 0–16 years lived in households with a hand washing facility next to the toilet. The HOI for each service was also significantly lower than the overall coverage rate, signaling substantial inequality in access among selected groups as also shown by the D-index. The most unequal access was observed in for opportunity to get electricity from the national grid. Getting access to electricity off-grid reduces inequality substantially. Figure 178. Coverage, HOI and D-index for access to basic services among children in 2019/20, % Source: UNHS 2019/20, World Bank staff calculations. Note: Age of children is shown in brackets. 133 Location was the key contributor to inequality in access to basic services in Uganda, with one exception for hand washing facility. Figure 179 shows the relative contribution of each circumstance to inequality of opportunities. Location, especially regional disparities, was responsible for the largest chunk of inequality in access to drinking water (78 percent). The role of location prevailed in inequality in access to sanitation as well (41 percent), although monetary wellbeing also played a significant role (27 percent). Location played a more important role in explaining the inequality in access to electricity from the national grid (55 percent) compared to access to electricity on and off-grid (44 percent). Indeed, the gap between rural and urban areas, and across regions narrows for opportunities to get electricity on and off-grid. Opportunities in sanitation and electricity were consistently the lowest in the Northern region (Table A.5). Inequality in access to a hand washing facility was an exception, with monetary wellbeing being the most important contributor to inequality (45 percent) and head of household education and location accounting for equal, but much smaller shares in inequality (25 percent each). Figure 179. Contribution of each circumstance to inequality of opportunities in basic services in 2019/20, % Source: UNHS 2019/20, World Bank staff calculations. Note: Age of children is shown in brackets. Households did not connect to the national grid mainly due to lack of the grid nearby, while ignorance and negative attitude were mentioned as important reasons for not using sanitation. Households were asked why they were not connected to the national grid. For 63 percent of households the grid was too far away to connect, while 23 percent of households cited that the initial costs of connection were too high. About eight percent of households in the Central region mentioned that the service was too unreliable to connect. Interviews with community leaders were used to better understand why some groups of the population do not use any type of sanitation in the community. Low income was mentioned in about 25 percent of communities, while negative attitude and ignorance was mentioned in 44 percent of communities. Poor soil type and terrain were mentioned as a main reason for people not using sanitation in 22 percent of communities. Explaining changes in HOI between 2012/13 and 2019/20 Human opportunity indexes vary over time due to changes in coverage and inequality. As the HOI takes into consideration both coverage and inequality of access to a given basic opportunity, one can explore 134 the role of these two factors behind the changes in the HOI over time. For example, if the HOI has improved, it can be related to better coverage (scale effect) or a decline in inequality (distribution effect) or a combination of both factors. Figure 180. Trends in the HOI between 2012/13, 2016/17 and 2019/20, % Source: UNHS, World Bank staff calculations. Note: visit to health facility and access to desk and chair were not possible to construct in 2012/13. The largest positive changes in children’s access to basic opportunities between 2012/13 and 2019/20 occurred with electricity off grid, improved and unshared sanitation and finishing primary school on time. Figure 180 shows changes in the HOI for all eleven opportunities. No positive changes were observed in the opportunity for school attendance among children of secondary school age and the opportunity to start school on time. Actually, both indexes declined over time. Modest positive changes in the HOI were observed with regards to school attendance among children of primary school age (an increase of three percentage points), access to electricity from grid (an increase of three percentage points), improved drinking water, and hand washing facility (an increase of four percentage points each). Human opportunity indexes for access to a health facility when ill and the learning environment at home were only available for the period from 2016/17 to 2019/20 and both showed improvement by four and five percentage points respectively. Substantial increase in the HOI were observed for finishing primary school on time (increase by 10 percentage points) and access to improved sanitation (increase by 11 percentage points). The largest and the most impressive positive change occurred with the HOI in access to electricity on and off-grid, which increased by 44 percentage points between 2012–2020. It is important to mention that for all three opportunities with the largest progress achieved, an increase occurred from a very low base and HOI remained below 50 percent. All improvements in the HOI were associated to a large extent with better coverage, with one noticeable exception when equitable distribution played a more important role for expanded access to off-grid electricity. Figure 181 shows contributions of scale and distribution effects for changes in 11 opportunities between 2012/13 and 2019/20. In all cases, except access to on and off-grid electricity, the scale effect dominated. Thus, the opportunity to access electricity (on and off-grid) increased by 44 percentage points, with 21 percentage points coming from the scale effect and 23 percentage points coming from more equal distribution. For other opportunities, redistribution played a marginal role and 135 any improvement observed was mostly due to a scale effect. For example, more children were able to visit a health facility when ill, but this change was fully due to a scale effect. Much higher coverage for children from the wealthier quintiles shown in Table A.5 contributed to this result. Similarly, improved school attendance for children ages 13–16 years, access to improved drinking water and hand washing facilities were fully driven by the scale effect, but not by any improvements in reducing inequality. Figure 181. Decomposing trends in the HOI between 2012/13 and 2019/20, by scale and distribution effects, change in percentage points Source: UNHS, World Bank staff calculations. Note: Changes are calculated for 2016/17 and 2018/19. Age of children is shown in brackets. III. Impact of COVID-19 on human capital Analysis in previous sections provided information on access to opportunities before COVID-19, but the ongoing pandemic has stalled the progress Uganda had been making in improving human capital accumulation. This subsection looks at the results from the National High-Frequency Phone Survey (HFPS), which aimed to measure the socio-economic impacts of the COVID-19 pandemic on Ugandans. School closures, due to the COVID-19 pandemic, have widened pre-existing inequalities in access to and quality of schooling within Uganda. During the period from March 2020 to February 2021, schools in Uganda were fully or partially closed for 83 weeks which was longer than in any other county in the world (UNESCO). This resulted in substantial lost learning for students. Participation in schooling declined dramatically (Figure 182): about 90 percent of children ages 6–18 years participated in schooling before the lockdown was introduced in March 2020, but by March/April 2021, this had dropped to only 46 percent of children participating in any learning activities, with substantial gaps across different groups, exacerbating previously existing inequalities in education opportunities. Overall, participation in learning activities was much higher in urban areas compared to rural areas (55 percent versus 43 percent, respectively). Regional disparities also widened substantially during the pandemic, with children in the Northern region having a much lower likelihood of participating in learning activities compared to other regions. Substantial and more pronounced differences in participation in learning activities were also observed across the wealth distribution based on the pre-COVID-19 consumption quintiles. Thus, only 33 136 percent of children from the poorest pre-COVID-19 quintile were participating in any learning activities in March/April 2021, compared to 64 percent among children from the richest quintile. Children who did not engage in distance learning, most frequently listed reasons such as lack of learning materials from the school or government, no interest, and lack of access to radio and TV. The inability to get learning materials from the government and schools was particularly important in the Northern region and in rural areas. Children from the poorest households, measured by pre-COVID-19 consumption quintiles, were more likely not to engage in learning from home because of increased household chores. Lack of access to radio and TV played a more prominent role in constraining children from learning in the Eastern and Northers regions. Among those who engaged in distance learning, about 51 percent revised textbooks from past classes, 24 percent studied materials provided by parents, 17 percent used private tutors, 14 percent had a lesson with a teacher, 12 percent completed assignments provided by a teacher, and eight percent used to read materials provided by the government. Lessons from private tutors and learning materials provided by parents were used more among children from the top 40 percent of the pre-COVID-19 consumption distribution. It is not very clear which types of home learning activities have the highest education potential, but higher usage of private tutors and better access to information technologies among children from wealthier households most probably amplified pre-existing inequalities in learning outcomes and likely affected performance when students resumed in-person classes. Review of health facility data points to disruptions in health services after the lockdown, while results from the phone survey indicates that access to medicine and medical treatment was lower in rural areas, with low income and a lack of transportation being key limiting factors. A detailed review of health facility data from Uganda’s health management information system shows that, compared to pre- pandemic trends and seasonality, Uganda has experienced significant disruptions in service volumes since the outbreak of COVID-19. As compared to March–December 2019, the average monthly drop in service utilization across all the key health indicators ranged from one to 42 percent over the period from March– December 2020 (World Bank, International Finance Corporation, Multilateral Investment Guarantee Agency 2021). Respondents to the HFPS were asked about access of household members to medicine and medical treatment when needed. In June 2020, two months after the first lockdown, about 33 percent of households in Uganda did not have access to medicine (Figure 183) and 18 percent did not have access to medical treatment when needed (Figure 184).62 Access to medicine and medical treatment was higher in urban areas. Access to medicine was the lowest in the poorest Eastern and Northern regions, while access to medical treatment was the lowest in the Central region. According to respondents to the HFPS, the main reasons responsible for the lack of access to medical treatment included lack of money (63 percent) and lack of transportation (27 percent). Lack of transportation was more important in rural areas, whereas almost 20 percent of households in the Western region were not able to access medical treatment because authorities prohibited traveling (Figure 185). 62 Unfortunately, we do not know the baseline numbers before the lockdown. 137 Figure 182. School attendance and participation in any type of schooling among children ages 3–18 years in March 2020 (before lockdown) and in March/April 2021 by area, region and pre-COVID-19 consumption quintiles, % Source: HFPS, World Bank staff calculations. Note: Age range in the HFPS data was from 3–18 years and differs from the age range used for the HOI. Quintiles are based on spatially adjusted consumption by poverty lines. Figure 183. Access to medicine when needed in the Figure 184. Access to medical treatment when needed HFPS, % of households in the HFPS, % of households Source: HFPS, World Bank staff calculations. 138 Figure 185. Main reasons for inability to access medical treatment in June 2020 according to the HFPS, % of households without access Source: HFPS, World Bank staff calculations. IV. Policy implications The HOI is a useful tool for policy makers, measuring inequality in access to opportunities and indicating which socio-demographic characteristics influence a child’s likelihood of access to a particular opportunity and identifying the ones with the largest contributions to inequality. Decomposing changes in the HOI over time can also indicate how improvements in access to opportunities have come about, namely through changes in coverage or through equalization effects. This is important for Uganda as the quality of opportunities within a country forms the basis for growth and shared prosperity. For education, inequality was found to be higher in opportunities that captured the quality of services, calling for greater attention to this component. Whereas opportunities for more simple indicators – such as school attendance – were more equal, others that matter for learning and education outcomes – such as completion and at what age the child started school – had bigger gaps. As such, policies and investments focusing on expanding access to education should also focus on factors that improve the quality of those services to deliver more significant gains in education outcomes. Location and monetary wellbeing were the most significant factors explaining the inequality in most indicators, highlighting the importance of targeted approaches to reduce regional gaps and improve opportunities for the poorest. For example, location explains more than 40 percent of all inequality in school attendance for primary age children, with monetary wellbeing being the second largest contributor and accounting for 29 percent of all inequality. Given the strong correlations between education level, poverty status and economic mobility, increasing opportunities for children in poorer and more vulnerable regions could be an important pathway for breaking the cycle of poverty. In Uganda, improvements in the HOI indicators between 2012/13 and 2019/20 were due primarily to better coverage rates, suggesting that further progress could be made by also factoring in distribution effects for expanded service delivery. The largest positive changes in the HOI indicators occurred with electricity on- and off-grid, improved and unshared sanitation, and finishing primary school on time. The increase in the opportunity to access electricity (on- and off-grid) was the only one coming from both scale 139 and distribution effects. For other opportunities, redistribution played a marginal role and any improvement observed was mostly due to a scale effect. While improvements in the HOI indicators so far have been starting from a very low base in terms of coverage, as coverage expands, more attention will need to be placed on distribution to ensure that expansion is equitable and continues to increase opportunities for those who need it the most. 140 References Barros, R.; Ferreira, F.; Molinas Vega, J.; & Saavedra, J. (2009). Measuring Inequality of Opportunities in Latin American and the Caribbean. Washington, DC: World Bank. Dabalen, A.; Narayan, A.; Saavedra-Chanduvi, J.; & Hoyos Suarez, A. (2015). Do African Children Have an Equal Chance? A Human Opportunity Report for Sub-Saharan Africa. Washington, DC: World Bank. Grimm, Michael. (2011). “Does Inequality in Health Impede Economic Growth?” Oxford Economic Papers 63 (3): 448–74. Nandini, K.; Lara Ibarra, G.; Narayan, A.; Sailesh, T.; & Vishwanath, T.. (2016). Uneven Odds, Unequal Outcomes: Inequality of Opportunity in the Middle East and North Africa. Directions in Development-- Poverty. Washington, DC: World Bank. Marrero, G. A.; & Rodriguez, J. G. (2013). “Inequality of Opportunity and Growth.” Journal of Development Economics 104: 107–22. Molina, E.; Narayan, A. & Saavedra-Chanduvi, J. (2013). “Outcomes, Opportunity and Development: Why Unequal Opportunities and Not Outcomes Hinder Economic Development.” Policy Research Working Paper 6735, World Bank, Washington, DC. (OECD) Organization for Economic Co-operation and Development; World Bank. (2006). Liberalization and Universal Access to Basic Services: Telecommunications, Water and Sanitation, Financial Services, and Electricity. OECD trade policy studies. OECD and the World Bank, Paris. Roemer, J. (1998). Equality of Opportunity. Cambridge, MA: Harvard University Press. World Bank. 2020. Tackling the Demographic Challenge in Uganda. World Bank, Washington, DC. World Bank; International Finance Corporation; Multilateral Investment Guarantee Agency. (2021). Uganda Systematic Country Diagnostic Update. World Bank, Washington, DC. 141 Annex 3 Table A.3 Coverage of human opportunities across areas, regions, and consumption quintiles in 2012/13 Areas Regions Consumption quintiles rural urban Central Eastern Northern Western first 2 2 4 fifth Total school attendance (children ages 6–12 years) 86 91 89 89 81 87 80 87 88 91 91 87 school attendance (children ages 13–16 years) 88 89 87 94 84 86 85 90 91 87 89 88 started primary school on time (children ages 6–7 years) 58 66 62 65 57 53 54 58 60 64 65 60 finished primary on time (children ages 13–16 years) 9 26 23 11 4 13 6 10 13 16 24 13 households have a desk and chair for homework (children ages 0–16 years) n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a visit to a health facility when ill (children ages 0–16 years) n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a improved drinking water (children ages 0–16 years) 69 87 61 86 77 62 73 75 71 70 75 73 improved and unshared sanitation (children ages 0–16 years) 13 23 26 17 4 11 9 12 15 15 26 15 hand washing facility (children ages 0–16 years) 11 18 18 15 4 12 6 9 12 14 23 12 electricity from and off grid (children ages 0–16 years) 4 31 23 5 3 6 1 3 4 11 32 9 electricity from grid (children ages 0–16 years) 1 28 19 4 2 3 1 2 3 8 26 7 Source: UNHS, World Bank staff calculations. Table A.4 Coverage of human opportunities across areas, regions, and consumption quintiles in 2016/17 Areas Regions Consumption quintiles rural urban Central Eastern Northern Western first 2 2 4 fifth Total school attendance (children ages 6–12 years) 88 94 96 91 77 92 81 87 92 94 97 89 school attendance (children ages 13–16 years) 83 86 83 91 78 81 80 84 85 87 84 84 started primary school on time (children ages 6–7 years) 52 66 60 61 47 50 48 53 54 55 69 55 finished primary on time (children ages 13–16 years) 15 38 40 15 5 19 9 15 20 27 38 20 households have a desk and chair for homework (children ages 0–16 years) 8 15 14 5 5 16 2 5 7 14 27 10 visit to a health facility when ill (children ages 0–16 years) 60 71 71 59 48 70 49 57 62 68 82 62 improved drinking water (children ages 0–16 years) 74 90 74 90 82 63 79 77 77 74 82 77 improved and unshared sanitation (children ages 0–16 years) 19 32 43 14 10 20 11 17 21 25 43 22 hand washing facility (children ages 0–16 years) 11 23 23 9 11 12 8 9 12 16 30 14 electricity from and off grid (children ages 0–16 years) 29 64 59 25 18 42 13 27 34 49 72 37 electricity from grid (children ages 0–16 years) 6 50 41 8 3 10 2 6 11 22 48 16 Source: UNHS, World Bank staff calculations. 142 Table A.5 Coverage of human opportunities across areas, regions, and consumption quintiles in 2019/20 Areas Regions Consumption quintiles rural urban Central Eastern Northern Western first 2 2 4 fifth Total school attendance (children ages 6–12 years) 90 95 96 94 79 93 82 90 93 96 97 91 school attendance (children ages 13–16 years) 86 88 88 92 82 83 79 87 88 91 91 87 started primary school on time (children ages 6–7 years) 53 63 59 58 50 54 44 50 55 65 68 56 finished primary on time (children ages 13–16 years) 20 44 43 21 10 28 11 19 24 34 50 26 households have a desk and chair for homework (children ages 0–16 years) 14 24 23 10 5 26 5 10 15 21 38 17 visit to a health facility when ill (children ages 0–16 years) 64 76 73 63 62 70 48 60 68 77 91 67 improved drinking water (children ages 0–16 years) 74 86 74 94 84 57 77 75 76 75 85 77 improved and unshared sanitation (children ages 0–16 years) 25 39 41 30 10 28 18 23 26 33 48 28 hand washing facility (children ages 0–16 years) 15 24 19 11 22 18 9 10 15 20 34 17 electricity from and off grid (children ages 0–16 years) 53 73 78 50 34 66 36 49 59 69 85 58 electricity from grid (children ages 0–16 years) 5 43 38 7 2 7 1 4 9 20 44 14 Source: UNHS, World Bank staff calculations. 143 Chapter 5. The role of the telecommunications sector for poverty reduction I. Development of the telecommunications sector Trends in the telecommunications sector during the last decade Uganda has identified ‘digital transformation’ as one of the key drivers that will enable the growth and transition of its economy to the modern state under the latest National Development Plan (NDPIII). The digital sector represents one of the fastest growing sectors in Uganda, with positive spillover effects on other sectors of the economy, the combination of which can play a key role in post-COVID-19 recovery (World Bank 2020). Recent analysis by the World Bank (WB) Africa Region Chief Economist’s Office found that closing the digital infrastructure gap in the eastern and southern Africa regions could result in a 1.5 percentage point increase in economic growth per capita (Calderon et al. 2019). If complemented by expansion in human capital development, the growth effect could increase to 3.9 percentage points. Providing affordable and ubiquitous telecommunication services is particularly important to promote technology-based empowerment, which is especially relevant for Uganda’s fast growing, young population. Population coverage by basic mobile infrastructure is extensive in Uganda but shows regional disparities. According to International Telecommunication Unit (ITU), almost 100 percent of the population in Uganda was covered by a mobile-cellular network. About 74 percent was covered by a 3G mobile network and 40 percent was covered by a 4G network. Coverage by a 4G network in Uganda was comparable to the African average, but lower than in Kenya (77 percent) and Rwanda (98 percent). The geographic coverage is different from the population coverage, however. Rural areas, particularly the Northern region of the country, have more limited coverage, as sparsely populated communities and lower income levels make it more challenging for telecommunications companies to deploy conventional network infrastructure (World Bank 2020). Figure 186. Mobile-cellular subscriptions per 100 Figure 187. Active mobile-broadband subscriptions per inhabitants based on ITU 100 inhabitants based on ITU Source: International Telecommunication Union. 144 There was a sharp increase in access to mobile phones and mobile services in Uganda during the last decade based on the estimates from the ITU. As shown in Figure 186 and Figure 187, mobile-cellular subscriptions and mobile-broadband subscriptions per 100 inhabitants in Uganda increased significantly during the last decade. As a result, the penetration rate in 2020 was 61 percent for mobile subscriptions and 44 percent for mobile broadband. The GSMA Mobile Connectivity Index, which is based on four pillars such as infrastructure, affordability, consumer readiness, and content and services, was equal to 49 percent in Uganda in 2019 compared to 36 percent in Ethiopia, 43 percent in Rwanda and 50 percent in Kenya. These scores placed all mentioned countries in the ‘emerging cluster,’ which indicates that a county performs well on one or two pillars but shows room for improvement on others (GSMA). Uganda currently performs relatively better in terms of consumer readiness and infrastructure and performs worse in terms of affordability and content and services. Despite notable progress in ICT penetration rates provided by the ITU, affordability remained one of the main constraints for Uganda to realize the full potential of digital transformation. The ITU monitors affordability of selected standardized baskets of ICT measured in prices expressed as a percentage of Gross National Income (GNI) per capita. Figure 188 shows average prices on five baskets of different services as a percentage of GNI for Uganda and compared them with the median prices for the African countries and Least Developed countries (LDCs). Figure 188. Price of ICT baskets in percent of GNI per capita in Data and telecom services are considered 2020 relatively unaffordable if they cost above 10 percent of GNI per capita (the two percent threshold is also frequently used as a more conservative measure to check affordability of the basic services). For all ICT baskets, the price as a share of GNI per capita in Uganda was higher than two percent. For four out of five ICT baskets, the price as a share of GNI per capita in Uganda was higher than the median for African and least developed countries. Moreover, for two ICT baskets that included high and low usage for data and voice,63 the price as a share of GNI per capita was more than twice as high as the 10 Source: International Telecommunication Union. percent threshold. All this evidence Note: ITU ICT Price Baskets, historical data series, March 2021 release. highlights the importance of affordability as a barrier towards access to ICT. The ICT sector is highly concentrated, which affects the level of competition of the telecom operators and prices. Despite many licensed operators, two main players – MTN Uganda and Airtel Uganda – had a combined market share of about 80 percent, while the other three players – Africell, Uganda Telecom and Smile Communications – accounted for the remaining share of the mobile telephony market segment as 63 Mobile data and voice low-usage basket includes 70 minutes of voice, 20 SMS and 500 megabytes of data. High usage basket includes 140 minutes of voice, 70 SMS and 1.5 gigabytes of data. 145 of September 2019 (World Bank 2020). In the third quarter of 2020, MTN Uganda and Airtel also accounted for more than 85 percent of wireless subscribers (Statista 2022). Improving competition in the market may have a positive impact on prices and increase the number of users of telecommunication services (World Bank 2020). Distribution of telecommunication services In this subsection, the report focuses on the distributional angle in access and usage of telecommunication services in Uganda during the last five years. Descriptive analysis is based on two rounds of the UNHS conducted in 2016/17 and 2019/20. The numbers on usage of internet obtained from the surveys can differ from the numbers reported by the ITU, which often rely on supply-side ICT data from national regulatory authorities and imputations. According to Frankfurter et al. (2020), national representative surveys in Africa provide considerably lower estimates for home internet access than prevailing ITU estimates. One important advantage of using the national representative surveys is that they can show access to ICT across the consumption distribution, education level and other socio-economic groups. There was no increase in the usage of computers, internet, and ownership of mobile phones in Uganda between 2016/2017 and 2019/2020, with significant rural/urban gaps remaining. Several comparable indicators using UNHS 2016/2017 and 2019/2020 are constructed to check the most recent trends in individual usage of computers during the last three months, individual usage of the internet, and ownership of at least one mobile phone in a household.64 None of the indicators increased significantly over this time period, and the usage of computers actually declined (Figure 189). Thus, ownership of at least one mobile phone among households remained at 73 to 74 percent, individual usage of internet was about 8 to 9 percent, and individual usage of computers ranged from 3 to 5 percent during 2016/17 and 2019/20. In both periods, there was a notable rural/urban gap, with only one and three percent of rural individuals using a computer or the internet in 2019/20 compared to seven and 19 percent among urban individuals, respectively. Little progress in the usage of internet from the household survey differs from mobile-broadband subscriptions data from the ITU and indicates that using supply-side data may overestimate the penetration rates and should thus be treated with caution. Figure 189. Usage of computers, internet, and ownership of mobile phones in 2016/2017 and 2019/2020, % Source: UNHS 2016/17 and 2019/20, World Bank staff calculations. 64 We had to use age threshold 16+ for the descriptive analysis since questions about access to ICT in the UNHS 2016/17 was asked for individuals 16+, but for simulation analysis we relied on data from the UNHS 2019/20 and used age threshold 15+. 146 The poorest individuals, those who are less educated, those living outside of the Central region, and women were less likely to use and have access to ICT in 2019/2020. Figure 190 compares individual usage of computers, internet, and ownership of mobile phones at the household level in 2019/2020 across different groups of individuals and households. Striking gaps exist across all groups. For example, literally nobody from the poorest consumption per adult equivalent quintile and from individuals with incomplete primary education, used the internet. This was in sharp contrast with 19 percent of individuals in the Central region who used the internet, 26 percent of individuals from the top consumption quintile, and 48 percent of individuals with complete secondary education and above. The same gaps were observed in other indicators, including access to mobile phones in the household. Striking regional gaps were observed as well, with the highest usage and access to ICT in the Central region and the lowest access in the Northern region. Female individuals and female headed households had consistently lower usage and access to ICT in 2019/2020. The largest improvements in access and usage of ICT were observed for those two groups of individuals who completed primary education and those who completed secondary education and above, while the difference between those with primary incomplete and primary complete education was minimal. Figure 190. Usage of computers, internet, and ownership of mobile phones in 2019/2020, % Source: UNHS 2019/20, World Bank staff calculations. Note: For household ownership of mobile phones, the head of household’s education level was used. Lack of confidence, knowledge or skills was the most frequently mentioned reason among individuals not using the internet. When individuals (16 years+) were asked about reasons why they did not use the internet, 63 percent of individuals mentioned lack of confidence, knowledge, or skills, about 51 percent mentioned high equipment costs, and 47 percent did not have a need to use the internet (Figure 191). Demand for internet and lack of knowledge, confidence and skills were closely correlated with individual education level. Thus, among those without complete primary education and not using internet, more than half did not need internet and almost 70 percent did not have skills and knowledge. This was in sharp contrast to those individuals who had complete secondary education and above: 35 and 38 percent, respectively. 147 Figure 191. Reasons for not using internet among individuals 16 years+ in 2019/2020, % Source: UNHS 2019/20, World Bank staff calculations. Note: Multiple answers were allowed. The internet was most frequently used for social networking and telephoning, with a non-negligible share using the internet for business and academic work and increasing rapidly with education level. Among those individuals (16 years+) who used the internet, about 85 percent used it for social networking and 40 percent used it for telephoning. About 20 percent used it for academic reasons and 17 percent used it for business. These shares varied dramatically depending on the education level of individuals. Thus, only two percent of individuals with incomplete primary education used the internet for business compared to 15 percent among those who completed primary education and 23 percent among those who completed secondary education and above. Access to mobile phones in 2019/2020 varied across groups in Uganda, with lower penetration rates among the poorest individuals, females, and those with lower levels of education. There was a detailed module in the UNHS 2019/2020 about use and ownership of mobile phones at the individual level. Figure 192 shows access to mobile phones among individuals (16 years+), distinguishing between those who owned mobile phones and those who used someone else’s phone. Overall, 59 percent of individuals ages 16 and above had access to mobile phones in 2019/2020. Among these 59 percent, 52 percent used their own mobile phone and seven percent used someone else’s mobile phones. Access to mobile phones among the poorest consumption quintile was half the level of the top quintile: 36 percent versus 78 percent, accordingly. Access to mobile phones was much lower in rural areas compared to urban ones (54 percent versus 72 percent, respectively). Residents of the Northern region were least likely to access mobile phones (40 percent) compared to other regions, and the average for the country (59 percent). Males were more likely to access and own mobile phones than women. As of 2019/2020, the majority of Ugandans owned basic phones that allowed calls and texting only. Individuals were asked about the type of mobile phones owned: basic, feature, and smartphones. About 70 percent of all owned mobile phones were basic ones with text and calling functions only (Figure 193). About 16 percent were feature phones, that had a camera and a radio. Only 14 percent of phones were smartphones. On top of lower usage and ownership of mobile phones, the poorest individuals from the bottom consumption quintile were more likely to own basic phones compared to the richest individuals 148 from the top quintile: 83 percent versus 58 percent, accordingly. No significant difference in terms of types of phones was found across gender. Figure 192. Access to mobile phones among individuals Figure 193. Distribution of phones by types among 16 years+: owned or shared in 2019/2020, % individuals 16 years+ in 2019/2020, % Source: UNHS 2019/20, World Bank staff calculations. The majority of individuals who used airtime or data on mobile phones during the last 30 days in 2019/2020 mentioned expensiveness as the main challenge in using these services followed by unreliability and the distance to the services. Actual usage of mobile phones during the last 30 days was lower than penetration rates reported in the previous paragraph. About 48 percent of individuals (16 years+) used airtime and six percent used data on mobile phones. When asked about the main challenge they faced using these services, 78 percent and 63 percent mentioned expensiveness of data and airtime, respectively. About 17 percent and 15 percent mentioned unreliability of data and airtime, accordingly. The remaining main challenges included services being located too far from the households. Mobile money played an important and increasing role for financial inclusion in Uganda. Financial inclusion is an important prerequisite for transforming Uganda from a peasant to a prosperous modern society. According to the Uganda National Financial Strategy 2017–2022 (Bank of Uganda 2017), the increase in financial inclusion in Uganda over this period was driven by the rapid expansion of mobile money users. This strategy aims to increase mobile money users in Uganda to 60 percent in 2022. According to the World Bank (2020), the number of mobile money transactions increased from 694 million in 2015 to 2841 million in 2019, while the number of registered customers increased from 21 million to 27 million during the same period. The UNHS 2019/2020 data can help identify the gaps in access and usage of mobile money among different population groups in Uganda. Almost half of the population in Uganda ages 16 years and above had at least one registered mobile money account in 2019/2020, with substantial gaps among the poor and the rich. Individuals with mobile phones were asked in the UNHS 2019/2020 if they had registered mobile money accounts. About 49 percent of all individuals ages 16 years and above had at least one registered account, which was equivalent to 80 percent of individuals with access to mobile phones. Expectedly, the gaps in access to mobile accounts closely resembled those gaps in access to mobile phones. Rural residents, the poor, women, and those who were less educated had lower rates of owning mobile money accounts. The gaps narrowed significantly once focused only on individuals with mobile phones. For example, the gap in 149 access to mobile money between the poorest and the richest consumption quintile was about 200 percent, but among those with a mobile phone, the gap shrank to less than 40 percent. This suggests that improving access to mobile phones could help narrow the gap in access to mobile money services. Still, the gap in access to mobile services existed among individuals with phones as well. For example, access to mobile accounts among individuals with phones was 72 percent among those with primary incomplete education and 92 percent among those with secondary complete education and above. These gaps may be related to the lack of skills and knowledge among poor individuals with low levels of education. Figure 194. Access to mobile money registered accounts among individuals 16 years+ in 2019/20, % Source: UNHS 2019/20, World Bank staff calculations. Most individuals with registered mobile money accounts used them at least once in the past 90 days prior to the survey, with the poorest having higher chances of owning inactive accounts. Among those individuals with registered mobile money accounts, 85 percent used them during the last 90 days, which can serve as an indicator of active account. However, among the poorest individuals from the bottom consumption quintile this share was much lower – about 70 percent compared to 92 percent among individuals from the top quintile. The poorest individuals were least likely to use registered mobile money accounts to borrow, save, get wages, receive money from the government, and pay fees and taxes. About half of individuals with registered accounts used them for borrowing or savings, but the share was much lower among the poorest individuals from the bottom consumption quintile: only 33 percent. Even fewer people used mobile money accounts to get wages and government payments or pay taxes and fees – 13 percent. The poorest individuals were least likely to use accounts for this purpose: 5 percent among the poorest quintile compared to 20 percent among the top richest quintile. II. Welfare impacts of competition in the mobile telecommunications market in Uganda Description of data and methodology According to Foster, Cull and Jolliffe (2021), the gap between availability of mobile telecommunications services (coverage) and its usage in Africa were related to lack of literacy and affordability. Affordability of services played a particularly important role in Uganda. As discussed above, for four out of five ICT baskets assessed, the price as a share of GNI per capita in Uganda was higher than the median for African and least developed countries. Moreover, two ICT baskets that included high and low usage for data and 150 voice were relatively unaffordable for the average consumer (above 10 percent of GNI per capita). At the same time, recent studies suggest that promoting competition in mobile telecommunication markets can help reduce prices of overall mobile services, contributing to reduce this usage gap (Rodriguez Castelan et al. 2021). The approach proposed by Rodriguez Castelan et al. (2019) is followed to assess the distributional impacts of competition policies in mobile telecommunication markets. Assumptions are made about the relationship between the structure of the market and competition intensity in a partial equilibrium model, to produce simulations of the impact of a reduction in market concentration (a proxy of an increase in competition in the market) on prices and consumers’ welfare under mild data requirements (see annex for details). To conduct the simulation, the report uses the World Bank–developed tool on welfare and competition, WELCOM. The tool allows the simulation of the distributional effects of changes in market competition through its impact on prices. In simple terms the tool (i) estimates the expected change in prices resulting from the increased competition, by assuming that after new entrants come into the sector, competition pushes prices toward the marginal cost (the result observed in perfectly competitive markets); (ii) identifies current users of telecom services and the share of their total expenditure allocated in these services; and (iii) applies the estimated price decrease to households that are currently users of telecom services to estimate their gain in welfare. In addition, the tool can model consumer behavior by estimating the probability of adoption based on sociodemographic characteristics, and subsequently, identifies potential consumers and their monetary gains (relying on the welfare gains calculated in the first part and the change in price of ICT services). The tool requires selecting a market structure for the ICT market and an appropriate price elasticity for telecom services. In the case of Uganda, the model also uses a target number of new users of mobile services. The study assumes Partial Collusive Oligopoly Structure (PCO) of the telecommunication market in Uganda. PCO involves few firms with a significant share of the market and multiple smaller firms with no market power and a small market share. The analysis estimated relative welfare impacts of reducing the market share of the two largest mobile network operators from 83 percent to 55 percent, using individual level expenditure data, district-level coverage information, and a demand price elasticity of -1.4. In addition to the direct impact of lower prices for telecommunication services, the analysis assumes that besides inducing a price reduction, competition will increase the numbers of users who can access the services now that the prices are lower. More precisely, it is assumed that coverage of mobile services will expand to 42 percent of the population (a middle point between coverages in 54.1 percent and 28.6 percent observed in South Africa and Kenya, respectively). New users are distributed following the geographical distribution of current users, but within each subregion, new users are determined based on their estimated probability of using ICT services conditional on individual characteristics and network coverage data at the village level. Then, a different model is used to impute ICT expenditure for these households and estimate the impact of reduced prices on their wellbeing. The simulation exercise used a detailed module on individual expenditure on telecommunications in the most recent Uganda National Household Survey (UNHS) from 2019/20, as well as detailed data on service coverage from Collins Bartholomew (see annex for details). Specifically, the report uses information on how much individuals spent over the last 30 days on data and airtime for mobile phones. Approximately 46 percent of the population 15 years and older (or 24.4 percent of the entire population) were consumers 151 of the reported ICT services, based on non-zero expenditure, however there was a significant digital divide across consumption quintiles, especially with regards to data usage. For example, about zero percent of individuals 15 years of age reported expenditure on data and 11 percent on airtime in the poorest quintile compared to about 12 percent and 42 percent among those from the top quintile, accordingly. Figure 195. Expenditure share on ICT services by consumption The shares of expenditure on ICT services quintiles in 2019/20 across quintiles are shown in Figure 195. Notably, the richest population from the top quintile spent 13 times more on ICT services than those from the poorest quintile. In relative terms, the gap is less pronounced, but still visible. The poorest population from the bottom quintile spent about 1.8 percent of their budget on ICT services, predominantly on airtime, while the richest from the top quintile spent about three percent of their budget on ICT services. Source: UNHS 2019/20, World Bank staff calculations. Note: Quintiles are constructed using official welfare aggregate spatially adjusted by regional poverty lines. Results of the simulation Improving competition in the telecommunications market would have positive equalizing welfare effects for new users, but positive dis-equalizing effects among current users. Figure 196. Relative welfare gains after improved A reduction in market concentration from 85 competition, percent percent to 44 percent is associated with an average relative welfare gain (increase in consumption per capita) of the Ugandan population by 1.7 percent, with 0.9 percent coming from the current users and 0.8 percent coming from new users. Among current users, the positive impact increases by consumption quintile, given their larger expenditures and coverage. In contrast, among new users the positive impact of improving competition is pro-poor, with larger consumption growth observed among the population from the bottom wealth quintiles. Source: UNHS 2019/20, World Bank staff calculations. Note: Quintiles are constructed using official welfare aggregate spatially adjusted by regional poverty lines. As a direct result of increased competition in the telecommunications sector, the reduction in average ICT prices, and the entrance of new users of telecommunication services, poverty is expected to decline 152 by about 0.8 percentage points, without significant changes in inequality. Consistent with increased consumption, poverty is expected to fall by about 0.8 percentage points, with almost equal shares coming from the current and new users. The change in inequality, as measured by the Gini index, was close to zero. The inequality increasing effect on current users was offset by the inequality decreasing effect on the new users. Figure 197. Effect of improved competition on national Figure 198. Effect of improved competition on poverty (change in percentage points) inequality as measured by the Gini index (change in Gini coefficient points) Source: UNHS 2019/20, World Bank staff calculations. Source: UNHS 2019/20, World Bank staff calculations. Note: Poverty is measured using the updated poverty line with the poverty rate at about 30 percent in 2019/20. This positive impact found from the simulated increase in competition is most likely a lower bound to the actual welfare gains that would result in practice, as our simulation is focused only on the direct impact of improved competition coming from lower prices and the entrance of new users. Stronger competition in mobile telecommunication services could also bring a larger variety, better quality of services, increased access to information, new markets (of services and labor) and new technologies (e.g., mobile money). In addition, notice that the literature finds that access to mobile telephony can improve individual welfare through better labor market outcomes, heightened financial inclusion, improved learning outcomes, and increased access to information and new markets through mobile broadband internet (Aker & Mbiti 2010; Aker, Ksoll, & Lybbert 2012; Maudi & Dubus 2020; Granguillhome Ochoa et al. 2022). While household wellbeing could be improved through these channels as well, the analysis of these alternative mechanisms goes beyond the scope of this study, but strongly suggests that our findings correspond to lower bound estimates. III. Policy implications Accelerating digital transformation can enable faster economic growth for Uganda. The digital sector represents one of the fastest growing sectors in Uganda, with positive spillover effects on other sectors of the economy, the combination of which can play a key role in post-COVID-19 recovery. However, despite an impressive increase in penetration rates, the poor, women, rural residents and individuals from selected regions continued to have significantly lower access to information communications technologies with affordability being one of the main constraints for greater utilization. As such, providing affordable 153 and ubiquitous telecommunication services is particularly important to promote technology-based empowerment, which is especially relevant for Uganda’s fast growing, young population. In addition, expanding digital access can also facilitate the delivery of more efficient social protection programs for Ugandans and refugees alike. Improving competition in the market can have a positive impact increasing access to ICT services by reducing prices and thereby increasing the number of users. The ICT sector in Uganda is highly concentrated, which affects the level of competition of the telecom operators and prices. Therefore, as demonstrated by the simulation, increased competition in the telecommunications sector could directly increase the number of users and reduce poverty. Furthermore, indirect benefits, such as a greater diversity of services, could also prompt further gains in economic growth and household wellbeing. Actions to promote this greater competition may therefore be reasonable policy approaches. 154 References Aker, J. C.; & Mbiti, I. M. (2010). "Mobile Phones and Economic Development in Africa." Journal of Economic Perspectives 24 (3): 207-232. doi:10.1257/jep.24.3.207. Aker, J. C.; Ksoll, C. & Lybbert, T. J. (2012). "Can Mobile Phones Improve Learning? Evidence from a Field Experiment in Niger." American Economic Journal: Applied Economics 4 (4): 94-120. doi:10.1257/app.4.4.94. Bank of Uganda (2017). Republic of Uganda: National Financial Inclusion Strategy 2017-2020. Calderon, C.; Kambou, G.; Korman, V.; Kubota, M.; & Cantu Canales, C. (2019). Africa's Pulse, No. 19, April 2019: An Analysis of Issues Shaping Africa’s Economic Future. Washington, DC: World Bank. Foster, V.; Cull, R. J. & Jolliffe, D. M. (2021). World Development Report 2021: Data for Better Lives. Washington, D.C.: World Bank Group. Frankfurter, Z.; Kokoszka, K.; Newhouse, D.; Silwal, A. R.; & Tian, S. (2020). Measuring Internet Access in Sub-Saharan Africa. Poverty and Equity Notes; No. 31. World Bank, Washington, DC. Fudenberg, D. and J. Tirole, (1991). Game Theory, M.I.T. Press. Granguillhome Ochoa, R.; Lach, S.; Masaki, T.; & Rodríguez-Castelán, C. (2022). "Mobile internet adoption in West Africa." Technology in Society 68: 101845. doi:doi.org/10.1016/j.techsoc.2021.101845. GSMA. GSMA Mobile Connectivity Index. Hasbi, M. & Dubus, A. (2020). "Determinants of mobile broadband use in developing economies: Evidence from Sub-Saharan Africa." Telecommunications Policy 44 (5). doi:doi.org/10.1016/j.telpol.2020.101944. Jackson, L. F. (1984). Hierarchic Demand and the Engel Curve for Variety. The Review of Economics and Statistics, 66(1), 8–15. https://doi.org/10.2307/1924690. International Telecommunication Unit (2022). https://www.itu.int/itu-d/sites/statistics/ Koutsoyiannis, A. (1975). Collusive Oligopoly. In: Modern Microeconomics. Palgrave Macmillan, London. https://doi.org/10.1007/978-1-349-15603-0_10. Rodríguez-Castelán, C.; Araar, A.; Malásquez, E. A.; & Granguillhome Ochoa, R. (2021). "Competition reform and household welfare: A microsimulation analysis of the telecommunication sector in Ethiopia." Edited by Elsevier. Telecommunications Policy. https://doi.org/10.1016/j.telpol.2021.102243. Rodríguez-Castelán, C.; Araar, A.; Malásquez, E. A.; Olivieri, S.; & Vishwanath, T. (2019). "Distributional Effects of Competition: A Simulation Approach." Policy Research Working Paper (World Bank Group). https://imagebank2.worldbank.org/search/31025713. Statista (2022). Share of mobile subscriptions Uganda 2015-2020, by operator. World Bank (2020). Digital Economy for Uganda Diagnostic Report. Washington, DC: World Bank. 155 Annex 4 Network coverage data The analysis uses data on spatial coverage of mobile networks from Collins Bartholomew to assess the role of network coverage in consumption of ICT services. The network coverage data from Collins Bartholomew relies on self-reported information by mobile providers worldwide from 2020 in roster format. Although the self-reported data might be incomplete, the information from Collins Bartholomew depicts updated and accurate data on network coverage. Collins Bartholomew data show coverage of GSM, 3G, and 4G networks. The network coverage used in this study comprises only 3G technologies. The information retrieved from the Collins Bartholomew coverage maps was matched with household survey data using the coordinates of households surveyed collected by the Ugandan Bureau of Statistics (UBOS). Thus, if a household’s point coordinates lie in an area known to have 3G coverage (based on the raster polygon), the households identified to be part of that village would thus have 3G coverage. Results from the merge show that in 2019/20, approximately 63.2 percent of individuals surveyed lived in areas covered by 3G.65 Theoretical framework This section discusses the theoretical background for estimating the welfare effects of a price reduction driven by an increase in market competition and the impact on users who were previously priced out of the market. The analysis carries out the simulations in two steps. First, the analysis focuses on measuring the welfare effects of increasing competition. For this, the analysis simulates the impact of a change in market structure on the market price, next the tool simulates the expected impact on welfare (relying on the Laspeyres variation method). For the second step, the marginal increase of new consumers and their corresponding monetary gains are predicted using a probit regression. Finally, the analysis estimates the change in welfare as the product of the change in the probability of consumption and the expected consumption.66 Next, the Annex discusses alternative market structures used for simulations; the proportion of entrants relying on household survey data; and how to estimate the expenditures of entrants after a change in price due to competition and the change in welfare through the Taylor approximation approach. Market structures and prices This section shows the equilibrium conditions of the two market structures modeled: Cournot competition and partial collusive oligopoly (PCO). The equilibrium outcomes of each market structure are used to estimate the change in the price given a change in the market structure. For instance, to calculate the price change of going from a monopolistic competition to a Cournot competition, the tool measures the difference between the equilibrium price in the monopolistic market and the equilibrium price in the Cournot competition. Cournot competition 65 This estimate likely serves as a lower-bound estimate of 3G coverage as Collins Bartholomew data do not cover all mobile operators. As such, estimates differ in magnitude from what GSMA Intelligence reports, where 3G coverage as a share of the population stood at 75.0 percent in 2019. 66 We use the WELCOM Stata tool to perform all the steps of our simulations. 156 In an oligopolistic market with Cournot competition, each firm chooses simultaneously the level of quantity to produce taking as given the opponents’ decisions (Fudenberg and Tirole 1991) and the demand function. For instance, a firm n maximizes = () ⋅ − . (1) When the marginal cost is constant ( = ) and firms are homogeneous, the market price is − = ⋅ 1 + Next, it is possible to write the price change of going from a concentrated market to a competitive one as −1 (2) − = 1+, where represents the total number of firms. Partial collusive oligopoly (PCO) Koutsoyiannis (1975) identifies two types of collusions in oligopolistic markets, cartels and price leadership. This analysis models the latter. The economic problem of the collusive group is to maximize the function that depends on the group production ( ), the production of the rest of firms in the market ( ), the market prices (p), and the cost function ((. )): = ( = + ) ⋅ − ( ). (3) Solving the group’s problem yields 1 (4) = ( ⁄ ) ⋅ , 1+ where is the market share of the oligopolistic group of firms that is colluding. If → 0, then = , the classical result in competitive markets. Then, going from the competitive market outcome to the partial collusive oligopoly implies a proportional price increase of67 − (5) = =− . + Probability model and proportion of entrants The analysis uses the Welfare and Competition (WELCOM)68 tool developed by the Poverty and Equity Global Practice of the World Bank to simulate the change in the probability of consumption associated to a price change, and the expected new consumers. The analysis uses the variability on prices (and/or on 67 Note that this model does not require an elasticity higher than one, as the monopoly model. 68 https://worldbankgroup.sharepoint.com/sites/Poverty/Pages/ToolsWELCOM-07092019-165156.aspx 157 welfare) to estimate the probability of consumption of the goods of interest through a probit regression.69 The sample is split into urban and rural areas to consider spatial differences in adoption and expenditure patterns. Specifically, the probability of consumption is: ( = 1) = Φ(0 + , ln() ⋅ + ⋯ + ), (6) where is an indicator variable that takes the value of one if the household belongs to the group g and, otherwise, zero and Φ(. ) denotes the normal cumulative distribution function. In general, the income elasticity is: ( = 1) ̅̅̅̅̅̅̅̅̅̅ = ⋅ , (7) ̅̅̅̅̅̅̅ where ̅̅̅̅̅̅̅̅̅̅ ̅̅̅̅̅̅̅ refer to the average income and average probability of having positive and consumption, respectively. Let the absolute change in probability, given a change in income, is: = (8) () () = . . . () where is the vector of independent variables. We assume that the change in probability represents the proportion of new entrants. As shown above, the change in the probability is defined as70: = () ⋅ ⋅ . (9) ̅̅̅̅̅̅̅̅̅̅ Expenditure estimates In addition to estimating the proportion of entrants, the analysis estimates the new consumers’ expenditure on the relevant service. For this, a first order Taylor approximation71 on demand (consumption expenditure or welfare) given a price change is used, assuming continuity of the demand function. The impact of a price change on welfare can be approximated by 69 In Stata, the stepwise prefix can be used to automatically perform the selection of explanatory variables according to their significance levels. The module also enables the user to implement five other variations to the probit model, please see the user manual for more information. 70 Another method to estimate the proportion of entrants is to calculate the expected change in probabilities. Because on the nonlinearity, the formula above may be less precise. An alternative approach is to compute the difference between averages of the predicted probabilities with initial and final welfare (which could lead to more accurate estimates), resulting in the estimated proportion of entrants. Formally, the change in probability of use of the household ℎ is: ,ℎ = ℎ ( = 1|′ℎ ) - ℎ ( = 1|ℎ ) The expected change in probability is equal to the expected predicted probability after the increase in welfare ( ′), minus the expected probability under the initial values of welfare. The advantage of this second method is that we do not need to evaluate the density for the reference individual, which will give more accurate results, as the two measurements will capture the main part of the change in probability. 71 Jackson (1984) introduces the foundations of the hierarchic demand system where the change in a bundle of goods can be modelled. In what follows, a Selective Demand Model is introduced to link the selection of goods with real income. 158 = − ⋅ , (10) where denotes the change in welfare, represents the expenditure on good , and the observed proportional change in the price of good i. However, besides measuring the impact of price changes on the welfare of current consumers, the analysis estimates the change on the welfare of new consumers. Denoting the average expected expenditure on good of entrants on group (e.g., decile 2) as , , and that of old users in the 72 same group as , . Then − (11) , = , , where , represents the proportion of new entrants on group for item . The total change on welfare on group is (12) = −, − , . Considering that the household survey databases tend to be large (approximately 20,000 observations), the entrant’s expected expenditures are calculated by running a quantile regression model.73,74 72 Current users are users with positive expenditure under a concentrated market (status quo). Non-users would be the remainder of users who don’t have positive expenditure under concentration (status quo). 73 Please refer to MCEMA User Manual for a detailed explanation of the different estimation techniques available in WELCOM (six total). 74 The quantile regression is carried out at the median. 159