TANZANIA MAINLAND POVERTY ASSESSMENT Tanzania’s Path to Poverty Reduction and Pro-Poor Growth PART 1 STANDARD DISCLAIMER This volume is a product of the staff of the International Bank for Reconstruction and Development/The World Bank. The findings, interpretations, and conclusions expressed in this paper do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. COPYRIGHT STATEMENT The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development/ The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA, ­ telephone 978-750-8400, fax 978-750-4470, http://www.copyright.com/. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA, fax 202-522-2422, e-mail pubrights@worldbank.org. TANZANIA MAINLAND POVERTY ASSESSMENT Tanzania’s Path to Poverty Reduction and Pro-Poor Growth PART 1 Table of Content Acknowledgement................................................................................................................................................................................................xiii Acronyms and Abbreviations.......................................................................................................................................................................... xv Overview ....................................................................................................................................................................................... 1 Executive Summary ................................................................................................................................................................... 3 Poverty has been falling for a decade, but recently the pace has slowed. .........................................................................4 Gradual improvement in living conditions and human capital helped reduce poverty. ....................................................6 Poverty reduction has not been responsive to Tanzania’s remarkable economic growth, and inequality has worsened..........................................................................................................................................................................11 Slow progress in poverty reduction has pushed up the number of poor people, leaving a noticeable share of the population still at risk of at least transitory poverty. ...........................................................................................................14 It is hard for the poor to achieve a better life because of a large number of dependents, low human capital, low-profile jobs, and limited access to basic services and assets. ....................................................................................16 The drivers of poverty are mutually reinforcing and carry across generations..................................................................19 Beyond the urban-rural divide, geographic disparities in poverty are substantial. .........................................................21 Structural transformation of the economy can help reduce poverty faster, but at the micro level the transformation is slow. ...........................................................................................................................................................24 Industry consists primarily of micro and informal enterprises, operating in precarious sectors and relying on unskilled labor...................................................................................................................................................................26 As with poverty, the distribution of firms and employment nationally is uneven..............................................................27 Although small firms dominate the economy and provide most of the jobs, medium and large firms could create jobs fast.............................................................................................................................................................29 Limited access to finance and the need to rely heavily on family networks and informal funding are barriers to emergence of a vibrant private sector..................................................................................................................................31 Implications for Policy............................................................................................................................................................33 Use a well-targeted life-cycle approach to enrich human capital......................................................................................34 Take advantage of the momentum created by investments in basic services to expand service delivery. ....................36 Accelerate creation of productive jobs by boosting the benefits of structural transformation and promoting opportunities for small firms to grow. ..................................................................................................................................37 Chapter 1: Poverty and Inequality Patterns........................................................................................................................ 39 I. Progress in Reducing Poverty................................................................................................................................................40 II. The Incidence of Growth and Patterns of Inequality...........................................................................................................51 III. The Structure of Inequality....................................................................................................................................................54 Chapter 2: Moving Up and Out of Poverty: From Vicious to Virtuous Cycles.............................................................61 I. Poverty Transitions from 2008 to 2015..................................................................................................................................62 II. Correlates and Determinants of Poverty Entry and Exit......................................................................................................67 III. Poverty Dynamics and Transitions Between Economic Sectors.........................................................................................71 IV. Summary and Policy Action...................................................................................................................................................79 Chapter 3: Profile of the Poor................................................................................................................................................. 81 I. Sociodemographic Characteristics of the Poor...................................................................................................................82 II. Community-Based Infrastructure and Services....................................................................................................................87 III. Perception of Poverty.............................................................................................................................................................89 Chapter 4: The Multiple Facets of Poverty......................................................................................................................... 95 I. Living Conditions and Assets Ownership.............................................................................................................................97 II. Human Capital......................................................................................................................................................................104 III. Multitude of Deprivations in Well-Being............................................................................................................................109 Chapter 5: Geographic Dimensions of Poverty................................................................................................................. 111 I. Geographic Disparities in Poverty......................................................................................................................................113 II. Drivers of Geographic Disparities in Poverty.....................................................................................................................120 Chapter 6: Agricultural Households and Nonfarm Enterprises.....................................................................................125 I. Overview...............................................................................................................................................................................126 II. Farm Households.................................................................................................................................................................129 III. Nonfarm Enterprise Households ........................................................................................................................................132 Appendixes...............................................................................................................................................................................137 Appendix A: Survey Description.................................................................................................................................................137 Appendix B: Drivers of Poverty Reduction.................................................................................................................................140 Appendix C: Distributional pattern of Growth in Urban and Rural Areas................................................................................143 Appendix D: Structure of Inequality...........................................................................................................................................145 Appendix E: Multivariate Regressions and Determinant of Consumption and Poverty.........................................................149 Appendix F: Multidimensional Deprivation and Multidimensional Poverty Index (MPI)........................................................154 Appendix G: Methodology for Generating Small-Area Poverty Estimates.............................................................................157 Appendix H: Market Accessibility...............................................................................................................................................162 Appendix I: Poverty Dynamics.....................................................................................................................................................164 References................................................................................................................................................................................203 List of Boxes Box 1.1: Poverty Measures ............................................................................................................................................................................................................................41 Box 1.2: Change in Poverty in Dar es Salaam...............................................................................................................................................................................................42 Box 1.3: Decomposition of Poverty ..............................................................................................................................................................................................................44 Box 2.1: Defining Chronic and Transient Poverty.........................................................................................................................................................................................63 Box 2.2: Poverty Decompositions and Sectoral Changes............................................................................................................................................................................74 Box 3.1: Tanzania Productive Social Safety Nets Program ..........................................................................................................................................................................91 Box I.1: Poverty Lines....................................................................................................................................................................................................................................173 Box I.2: Distribution Neutral and Growth Elasticity Approach..................................................................................................................................................................177 List of Figures Figure ES.1: Poverty Trends, National Poverty Line, 2007–18, Percent ........................................................................................................................................................4 Figure ES.2: Source of Lighting, Percent.........................................................................................................................................................................................................6 Figure ES.3: Access to Water, Percent.............................................................................................................................................................................................................6 Figure ES.4: Access to Sanitation, Percent.....................................................................................................................................................................................................6 Figure ES.5: Net Education enrollment rates 2012–18, Percent....................................................................................................................................................................7 Figure ES.6: Educational Achievements of Adults 15+, 2018, Percent.........................................................................................................................................................7 Figure ES.7: Human Capital Index...................................................................................................................................................................................................................8 Figure ES.8: Under-5 Children Nutrition Deficit, Percent..............................................................................................................................................................................8 Figure ES.9: Drivers of Poverty Reduction in Mainland Tanzania, 2012–18: Endowments and Returns Effects, Percent.........................................................................9 Figure ES.10: Drivers of Poverty Reduction in Rural Areas, 2012–18, Percent.............................................................................................................................................9 Figure ES.11: Drivers of Poverty Reduction in Urban Areas, 2012–18, Percent.........................................................................................................................................10 Figure ES.12: Growth Incidence Curves, 2012–18, Percent.........................................................................................................................................................................12 Figure ES.13: Growth and Redistribution Effects, 2012–18, Percentage points .......................................................................................................................................12 Figure ES.14: Number of Poor People, 2007–18, Million ............................................................................................................................................................................14 Figure ES.15: Consumption Density around the Poverty Line in 2018 (2011 PPP).....................................................................................................................................15 Figure ES.16: Transition In and Out of Poverty, 2010–15, Percent .............................................................................................................................................................15 Figure ES.17: Education Level of Household Head and Poverty, 2018, Percent........................................................................................................................................17 Figure ES.18: Sector of Employment of Household Head and Poverty, 2018, Percent.............................................................................................................................17 Figure ES.19: Intergenerational Mobility in Education, 2018, Percent.......................................................................................................................................................19 Figure ES.20: Contributions of Individual’s Circumstances to Inequality, 2018, Percent...........................................................................................................................19 Figure ES.21: Poverty Incidence by Region, 2018........................................................................................................................................................................................21 Figure ES.22: Poverty Incidence by District, 2018........................................................................................................................................................................................21 Figure ES.23: Market Access to Major Urban Centers.................................................................................................................................................................................22 Figure ES.24: Contribution of Economic Sectors to GDP, 1998–2017, Percent.........................................................................................................................................24 Figure ES.25: Working Hours, Primary and Second Jobs, 2010–15, Percent.............................................................................................................................................25 Figure ES.26: Productivity Differences Across Sectors, Per Worker, 2015..................................................................................................................................................25 Figure ES.27: Productivity Differences Across Sectors, Per Hour Worked, 2015.......................................................................................................................................25 Figure ES.28: Relationship Between Poverty and the Number of Firms....................................................................................................................................................27 Figure ES.29: Geographic Distribution of Firms...........................................................................................................................................................................................28 Figure ES.30: Employment by Gender and Firms Size ...............................................................................................................................................................................29 Figure ES.31: Value Added by Firm Size, Billion TZS and Percent..............................................................................................................................................................29 Figure 1.1: Poverty Trends, at the National Poverty Line, 2007–18, Percent..............................................................................................................................................40 Figure 1.1B: Extreme Poverty Headcount.....................................................................................................................................................................................................41 FIgure 1.1C: Poverty Depth and Severity......................................................................................................................................................................................................41 Figure 1.2: Poverty and Demographic Dynamics in Dar es Salaam, 2018..................................................................................................................................................42 Figure 1.3: Drivers of Poverty Reduction in Mainland Tanzania, 2007–18 .................................................................................................................................................45 Figure 1.4: Total Population and Number of Poor People, 2007, 2012 and 2018......................................................................................................................................48 Figure 1.5: Consumption Density around the Poverty Line in 2018 (2011 PPP).........................................................................................................................................49 Figure 1.6: International Poverty Headcount Ratio at $ 1.9 a day and GDP per capita (2011 PPP)..........................................................................................................50 Figure 1.7: Consumption Growth and Inequality 2007-2018.......................................................................................................................................................................52 Figure 1.8: Growth Incidence Curves, Percent.............................................................................................................................................................................................53 Figure 1.9: Determinants of Inequality Between Urban and Rural Areas...................................................................................................................................................56 Figure 1.10: Intergenerational Mobility among the Total Population and the Poor, 2018, Percent.........................................................................................................58 Figure 1.11: Overall Inequality and Inequality of Opportunity in Consumption, 2018, Percent...............................................................................................................59 Figure 1.12: Contributions of Individual’s Circumstances to Inequality, 2018, Percent.............................................................................................................................59 Figure 2.1: Poverty Transitions Across Three NPS Rounds by Location, 2008–2012, Percent...................................................................................................................64 Figure 2.2: Chronic and Transitory Poverty Over Three NPS Rounds.........................................................................................................................................................64 Figure 2.3: Poverty Status According to Main Source of Household Income, 2008-12, Percent..............................................................................................................66 Figure 2.4: Poverty Transitions According to Main Source of Household Income, 2008–2012, Percent..................................................................................................67 Figure 2.5: Marginal Effects Associated with Poverty Entry, 2008–2012.....................................................................................................................................................68 Figure 2.6: Marginal Effects Associated with Poverty Exit, 2008–2012.......................................................................................................................................................68 Figure 2.7: Poverty Exit and Unconditional Transition Rates by Individual and Household Characteristics, 2008–2012........................................................................69 Figure 2.8: Poverty Entry and Unconditional Transition Rates by Individual and Household Characteristics, 2008–2012.....................................................................69 Figure 2.9: Real Monthly Consumption Expenditure Per Adult Equivalent According to Poverty State.................................................................................................70 percent..............................................................72 Figure 2.10: Poverty Rates and Contribution of Households Occupation Sectors to Poverty: 2010, 2012 and 2015, ­ percent...................................................72 Figure 2.11: Population Shares and Poverty Contributions According to Main Source of Household Income, 2008–2015, ­ Figure 2.12: Share of Household Heads Engaged Primarily in Agriculture or Services by Age Cohorts, 2010......................................................................................73 Figure 2.13: Sectoral Composition by Consumption Deciles, 2010...........................................................................................................................................................73 percent........................................................................................................................................75 Figure 2.14: Sectoral Decomposition of Poverty Change, 2010–2015, ­ percent.....................................................................................................75 Figure 2.15: Decomposition According to Main Source of Household Income, 2008–2015, ­ percent......................................76 Figure 2.16: Share of Household Heads Changing Sectors Between 2010 And 2012 Across The 2010 Consumption Distribution, ­ Figure 3.1: Poverty Headcount According Number of Children, 2018 , Percent ......................................................................................................................................83 Figure 3.2: Poverty Headcount According to Gender of Household Head, 2018, Percent......................................................................................................................83 Figure 3.3: Poverty Headcount According to Household Head Education, 2018, Percent......................................................................................................................84 Figure 3.4: Educational Level of Household Members and Poverty Status, 2018, Percent......................................................................................................................84 Figure 3.5: Gross and Net Rates of Enrollment in School, 2018, Percent ..................................................................................................................................................85 Figure 3.6: Sector of Employment and Poverty, 2018, Percent...................................................................................................................................................................85 Figure 3.7: Status of Employment and Poverty, 2018, Percent....................................................................................................................................................................86 Figure 3.8: Access to Roads, 2018, Percent .................................................................................................................................................................................................87 Figure 3.9: Access to Health Facilities, 2018, Percent .................................................................................................................................................................................87 Figure 3.10: Access to Community Services and Infrastructure, 2018, Percent of Households................................................................................................................88 Figure 3.11: Subjective Poverty and Self-Assessment, 2018, Percent ........................................................................................................................................................89 Figure 3.12: Food Stress According to Poverty Status, 2018, Percent .......................................................................................................................................................90 Figure 3.13: Food Shortfalls According to Poverty Status, 2018, Percent..................................................................................................................................................90 Figure 3.14: PSSN Scheme, 2018, Percent ...................................................................................................................................................................................................93 Figure 4.1: Housing Conditions, 2012 and 2018, Percent............................................................................................................................................................................97 Figure 4.2: Access to Electricity, 2012 and 2018, Percent ...........................................................................................................................................................................99 Figure 4.3: Access to Water and Sanitation, 2012 and 2018, Percent ......................................................................................................................................................100 Figure 4.4: Ownership of Assets, 2012 and 2018, Percent ........................................................................................................................................................................101 Figure 4.5: Human Capital Index.................................................................................................................................................................................................................104 Figure 4.6: Gross and Net Enrollment Rates, 2012 and 2018, Percent ....................................................................................................................................................105 Figure 4.7: Repetition of Classes, 2018, Percent .......................................................................................................................................................................................106 Figure 4.8: Educational Achievements of Adults 15+, 2012 and 2018, Percent ......................................................................................................................................106 Figure 4.9: Health and Anthropometric Indicators, 2016...........................................................................................................................................................................108 Figure 4.10: Well-being Dimensions to Assess the Multitude of Deprivations.......................................................................................................................................109 Figure 4.11: Deprivation Levels According to Welfare Dimension, 2018, Percent .................................................................................................................................109 Figure 5.1: Poverty Headcount by Geographic Zone, 2018, Percent .......................................................................................................................................................113 Figure 5.2: Distribution of the Poor Population by Geographic Zone, 2018, Percent ............................................................................................................................113 Figure 5.3: Nighttime Lights, 2018 ..............................................................................................................................................................................................................114 Figure 5.4: Changes in Nighttime Lights, 2013–18 ....................................................................................................................................................................................114 Figure 5.5: Estimated Poverty Rate.............................................................................................................................................................................................................115 Figure 5.6: Estimated Share of Poor by District .........................................................................................................................................................................................117 Figure 5.7: Number of Poor and Poverty Rates in Districts.......................................................................................................................................................................117 Figure 5.8: Multidimensional Deprivation Rate by Region, 2018, Percent...............................................................................................................................................117 Figure 5.9: Number of Multidimensional Deprived by Region, 2018, Thousand....................................................................................................................................118 Figure 5.10: Access to Education and Public Services, 2018.....................................................................................................................................................................118 Figure 5.11: Relationship Between Poverty, Education and Public Services............................................................................................................................................119 Figure 5.12: Poverty According to Population Density .............................................................................................................................................................................120 Figure 5.13: Poverty According to Nighttime Lights..................................................................................................................................................................................120 Figure 5.14: Market Access to Major Urban Centers in Tanzania..............................................................................................................................................................121 Figure 5.15: Poverty by Market Access and Distance to Dar es Salaam...................................................................................................................................................121 Figure 5.16: Poverty vs. Tropical Climate and Diseases.............................................................................................................................................................................123 Figure 5.17: Drought Hotspots....................................................................................................................................................................................................................124 percent..........................................................................................................................127 Figure 6.1: Socio-economic Status of Farm and Nonfarm Households, 2018, ­ Figure 6.2: Labor Intensity According to Crop, 2012.................................................................................................................................................................................130 Figure 6.3: Average Monthly Household Consumption According to Grown Crop Type, 2018, TZS....................................................................................................131 percent............................................................................................................................................................133 Figure 6.4: Top Five Nonfarm Enterprises Sectors, 2018, ­ percent............................................................................................................................................................134 Figure 6.5: Age of Nonfarm Enterprises, 2012 and 2018, ­ percent.......................................................................................................................135 Figure 6.6: Highest Level of Education Completed by Household Head, 2018, ­ Figure 6.7: Average Monthly Nonfarm Enterprise Revenue According to Owner Characteristics, 2018, TZS......................................................................................135 Figure B.1: Drivers of Poverty Reduction in Rural Areas From 2012 to 2018 ...........................................................................................................................................141 Figure B.2: Drivers of Poverty Reduction in Urban Areas From 2012 to 2018 .........................................................................................................................................141 Figure B.3: Drivers of Poverty Reduction in Dar es Salaam From 2012 to 2018 ......................................................................................................................................142 Figure B.4: Drivers of Poverty Reduction in Other Urban Areas From 2012 to 2018 ..............................................................................................................................142 Figure C.1: Growth Incidence Curves in Rural Areas, 2007–2012..............................................................................................................................................................144 Figure C.2: Growth Incidence Curves in Rural Areas, 2012–2018..............................................................................................................................................................144 Figure C.3: Growth Incidence Curves in Urban Areas, 2007–2012............................................................................................................................................................144 Figure C.4: Growth Incidence Curves in Urban Areas, 2012–2018............................................................................................................................................................144 Figure D.1: Intergenerational Mobility Poor Population, Father vs Son and Mother vs Daughter, Percent .........................................................................................147 Figure F.1: Welfare Dimensions and Deprivations Criteria........................................................................................................................................................................156 Figure G.1: CV of Direct vs. Small Area Estimates of Poverty Rates at the District level.........................................................................................................................161 Figure I.1: Levels and Trends of Poverty in Tanzania Mainland using HBS and NPS (Non-Harmonized), Percent................................................................................165 Figure I.2: Comparison of Basic Needs Poverty Rates Across Survey Rounds, Mainland, Percent........................................................................................................166 Figure I.3: Comparison of Basic Needs Poverty Rates Across Survey Rounds by Geographic Domain, Percent.................................................................................167 Figure I.4: Projected HBS Poverty Rates with NPS (Harmonized) and DHS (Imputed) Rates.................................................................................................................167 Figure I.5: Data and Assumptions Used for Analyzing NPS Poverty Trends.............................................................................................................................................168 Figure I.6: Comparison of Food Poverty Rates Across NPS Survey Rounds, Percent.............................................................................................................................172 Figure I.7: Comparison of Food Poverty Rates Across Survey Rounds by Geographic Domain, Percent.............................................................................................172 Figure I.8: Comparison of Basic Needs Poverty Rates Across Survey Rounds, Percent..........................................................................................................................172 Figure I.9: Growth Incidence Curves, 2008–2015.......................................................................................................................................................................................174 Figure I.10: Growth Incidence Curves by Geographic Domain, 2010–2015.............................................................................................................................................174 Figure I.11: Comparison of Growth Incidence Curves of Consumption Excluding and Including Clothing and Footwear, 2012–2015.............................................175 Figure I.12: Gini Index of Monthly Real Consumption Per Capita, Mainland..........................................................................................................................................175 Figure I.13: Lorenz Curve and Gini Index by Geographic Domain...........................................................................................................................................................176 Figure I.14: Actual and Projected Poverty Rates Using the Distribution Neutral and Growth-Elasticity Approaches..........................................................................178 Figure I.15: Kernel Density Function of Log Real Monthly Consumption Per Adult...............................................................................................................................178 Figure I.16: Data and Assumptions used for Analyzing Poverty Dynamics..............................................................................................................................................191 Figure I.17: Kernel Density Distributions of Real Consumption Expenditure..........................................................................................................................................191 Figure I.18: Share of Chronic and Transitory Poverty by Place of Residence...........................................................................................................................................192 List of Tables Table 1.1: Decomposition of Inequality by Household Attributes..............................................................................................................................................................54 Table 2.1: Transition Matrix, 2008–2012.........................................................................................................................................................................................................62 Table 2.2: Transition Matrix: 2012 Poverty Status Conditional on 2008 Poverty Status.............................................................................................................................63 Table 2.3: Poverty Dynamics based on Synthetic Panels, 2010–2015, Percent...........................................................................................................................................63 Table 2.4: Profiles According to Poverty Status............................................................................................................................................................................................65 Table 2.5: Inter-Temporal Sectoral Composition and Transitions, 2010–2012, Percent.............................................................................................................................77 Table 2.6: Sectoral Transition Matrices Conditional on Initial Sector of Employment, 2010–2012, Percent............................................................................................77 Table 2.7: Sectoral Transitions for Households Between 2010 and 2012, Percent.....................................................................................................................................77 Table 2.8: Transitions in Employment Type Between 2010 and 2012, Percent...........................................................................................................................................78 Table 2.9: Poverty Transition and Change in Annual Hours Worked by Sector.........................................................................................................................................78 Table 3.1: Sociodemographic Characteristics of Tanzanian Households, 2018.........................................................................................................................................82 Table 4.1: Change in Asset Ownership, 2012 and 2018, Percentage Points.............................................................................................................................................102 Table 4.2: Multidimensional Deprivations, 2018, Percent .........................................................................................................................................................................109 Table 5.1: Poverty District Estimates, 2012 and 2018.................................................................................................................................................................................116 percent.............................................................................................................................................127 Table 6.1: Participation in Farm and Nonfarm Activities, 2012–18, ­ percent.............................................................................................................................................127 Table 6.2: Participation in Farm and Nonfarm Activities, 2012–18, ­ percent.......................................................................................................................................................129 Table 6.3: Top Five Cash and Staple Crops, 2012 and 2018, ­ Table 6.4: Average Value Sold and Consumed of Top Five Cash Crops, 2012 and 2018, TZS................................................................................................................130 Table 6.5: Mean Days of Labor According to Crop, 2012 .........................................................................................................................................................................130 percent...................................................................................................131 Table 6.6: Prevalent Types of Farming Households in Areas with Markets and Roads, 2018, ­ percent.........................................................................................................................................132 Table 6.7: Top Five Sectors of Nonfarm Enterprises, 2012 and 2018, ­ Table 6.8: Most Predominant Sources of Start-Up Funds for Nonfarm Enterprises, 2012 and 2018, percentage of Total Funding ...................................................133 percent...........................................................................134 Table 6.9: Incidence of Nonfarm Enterprises in Communities with and without Markets and Roads, 2018, ­ Table E.1: Correlates of Consumption, 2018...............................................................................................................................................................................................150 Table E.2: Correlates of Poverty, 2018.........................................................................................................................................................................................................152 Table G.1: Beta Model Results (Variables Selected through Stepwise Process)......................................................................................................................................160 Table I.1: New Administrative Divisions.......................................................................................................................................................................................................169 Table I.2: Total Population Before and After Stratification.........................................................................................................................................................................169 Table I.3: Comparison of Consumption Per Adult Per Month in Tanzania...............................................................................................................................................170 Table I.4: Comparison of Total, Food and Non-Food Consumption Per Adult Per Month in Tanzania.................................................................................................171 Table I.5: Comparison of Food Lines Across NPS Survey Rounds in Tanzania.........................................................................................................................................171 Table I.6: Comparison of Basic Needs Poverty Lines Across NPS Survey Rounds in Tanzania...............................................................................................................172 Table I.7: Growth Elasticities and Pass-Through Rates...............................................................................................................................................................................178 Table I.8: New Regions created in March 2012 and Correspondence to Original Regions and Districts..............................................................................................180 Table I.9: Survey Sample Sizes......................................................................................................................................................................................................................180 Table I.10: Comparison of Means of Variables by Geographic Domain...................................................................................................................................................182 Table I.11: Share of Cellphone Ownership..................................................................................................................................................................................................184 Table I.12: Poverty Estimates and Imputations...........................................................................................................................................................................................184 Table I.13: Gini Coefficient...........................................................................................................................................................................................................................184 Table I.14: Poverty Rates by Age, 2010 and 2014.......................................................................................................................................................................................185 Table I.15: Poverty Dynamics based on Synthetic Panels, Tanzania 2010–2014.......................................................................................................................................186 Table I.16: Characteristics Associated with Poverty Mobility, Synthetic Panel Estimates, Tanzania 2010–2015: Joint Probabilities....................................................186 Table I.17: Characteristics Associated with Poverty Mobility, Synthetic Panel Estimates, Tanzania 2010–2015, Joint Probabilities....................................................187 Table I.18: Characteristics Associated with Poverty Mobility, BM* Synthetic Panel Estimates, Tanzania 2010–2015, Joint Probabilities...........................................187 Table I.19: Characteristics Associated with Poverty Mobility, Synthetic Panel Estimates, Tanzania 2010–2015, Conditional Probabilities........................................187 Table I.20: Characteristics Associated with Poverty Mobility, BM* Synthetic Panel Estimates, Tanzania 2010–2015. Conditional Probabilities................................188 Table I.21: Model Selection. DL, Weighted vs. Unweighted Regressions................................................................................................................................................189 Table i.22: Model Selection. Unweighted Regressions, DL vs. BM ..........................................................................................................................................................189 Table I.23: ρy for Different Age Ranges, Panel estimates...........................................................................................................................................................................189 Table I.24: ρy for Different Cohort Definitions. 20–75, Pseudo-panel estimates......................................................................................................................................189 Table I.25: ρy for Different Cohort Definitions. 25–60, Pseudo-panel estimates......................................................................................................................................190 Table I.26: ρy over 2 years in Different Periods, Panel estimates...............................................................................................................................................................190 Table i.27: Poverty Rates and Population Shares........................................................................................................................................................................................192 Table I.28: Proportion of Panel Members Poor and Non-Poor, 2008 to 2010..........................................................................................................................................192 Table I.29: Transition Matrix: 2010 Poverty Status Conditional On 2008 Poverty Status.........................................................................................................................192 Table I.30: Marginal Effects of Probit for Poverty Exit................................................................................................................................................................................192 Table I.31: Marginal Effects of Probit for Poverty Entry..............................................................................................................................................................................193 Table I.32: Inter-Temporal Household Income Source Composition and Transitions: 2008–2012..........................................................................................................194 Table I.33: Household Income Source Transition Matrices Conditional on Initial Poverty Status: 2008–2012.......................................................................................194 Table I.34: Marginal Effects of Probit Regressions on Poverty Status, 2010, 2012 and 2015...................................................................................................................195 Table I.35: Marginal Effects of A Probit Regression on Poverty Status In 2012 Conditional On 2010 Baseline Variables....................................................................195 Table I.36: Transitions in Employment Type Between W2 and W3 (Whole Sample)................................................................................................................................195 Table I.37: Transitions in Employment Type Between W2 and W3 (Escaped Poverty)............................................................................................................................196 Table I.38: Transitions in Employment Type Between W2 and W3 (Entered Poverty).............................................................................................................................196 Table I.39: Transitions in Employment Type Between W2 and W3 (Trapped in Poverty).........................................................................................................................196 Table I.40: Transitions in Employment Type Between W2 and W3 (Never Poor).....................................................................................................................................196 Table I.41: Transition Matrix for Sector of Employment Between W2 and W3 (Whole Sample).............................................................................................................197 Table I.42: Transition Matrix for Sector of Employment Between W2 and W3 (Escaped Poverty).........................................................................................................197 Table I.43: Transition Matrix for Sector of Employment Between W2 and W3 (Entered Poverty)..........................................................................................................197 Table I.44: Transition Matrix for Sector of Employment Between W2 and W3 (Trapped in Poverty)......................................................................................................198 Table I.45: Transition Matrix for Sector of Employment Between W2 and W3 (Never Poor)..................................................................................................................198 Table I.46: Poverty Transition and Change in Hours Worked by Sector...................................................................................................................................................198 Table I.47: Transitions Between Types of Agricultural Work (Whole Sample)..........................................................................................................................................199 Table I.48: Transitions Between Types of Agricultural Work (Escaped Poverty).......................................................................................................................................200 Table I.49: Transitions Between Types of Agricultural Work (Entered Poverty)........................................................................................................................................200 Table I.50: Transitions Between Types of Agricultural Work (Trapped in Poverty)...................................................................................................................................200 Table I.51: Transitions Between Types of Agricultural Work (Non-poor Non-Poor).................................................................................................................................200 Table I.52: Shocks Experienced in Wave 2..................................................................................................................................................................................................200 Table I.53: Shocks Experienced in Wave 3 .................................................................................................................................................................................................200 Table I.54: Poverty Transitions and Exposure to Shocks in Wave 2...........................................................................................................................................................201 Table I.55: Poverty Transitions and Exposure to Shocks in Wave 3...........................................................................................................................................................201 Table I.56: Transitions across consumption quintiles .................................................................................................................................................................................201 Table I.57: Change in Consumption between Wave 2 and Wave 3..........................................................................................................................................................202 Acknowledgement The members of the core team that prepared this report prepared by Ingela Alger (Professor, Toulouse School of are Nadia Belhaj Hassine Belghith (EA2PV, TTL), Wendy Economics). Karamba (EA1PV, co-TTL), Elizabeth Talbert (EA1PV), and Pierre de Boisseson (EA1PV). The report was prepared in close collaboration with the Ministry of Finance and Planning and the National Bureau of The analysis in the different volumes reflects the work of Statistics of the United Republic of Tanzania. We express our several researchers, including World Bank staff, academics, sincere thanks to Dr. Albina Chuwa, Sylvia Meku, and the and consultants. The core team worked closely with each Household Budget Survey team from the National Bureau of the teams to ensure the consistency of the analytical of Statistics for the important support, fruitful collaboration methods, findings, and main messages across the different and critical feedback provided throughout the preparation teams. of the report and for facilitating the exchanges. The data cleaning and construction of the consumption The report benefited from several consultations and insights aggregate using HBS 2017/18 was prepared by Sasun from groups of stakeholders, including senior officials from Tsirunyan (Consultant). The spatial analysis of poverty the Ministry of Finance and Planning, the Office of the was produced by David Newhouse (EA2PV) and Takaaki Prime Minister, Tanzania Social Action Fund, REPOA, and Masaki (EA1PV). Additional support for the analysis of development partners. climate effects and other hidden dimensions of poverty was provided by Jia Jun Lee (GENGE). The analysis The report also benefited from comments and advice of agricultural and nonfarm enterprises was produced from Yutaka Yoshino (AFCE1), Nobuo Yoshida (EA1PV), with the support of Foluyinka Fakoya (Master student, Tomomi Tanaka (EA1PV), and Mohamed Ihsan Ajwad Georgetown University) and Freeha Fatima (EA1PV). (HMNSP). The team gratefully acknowledges guidance The analysis of poverty dynamics was produced by from Bella Bird (Country Director AFCE1), Preeti Arora Wendy Karamba and Arden Finn (EA1PV). Additional (Country Program Coordinator, AFCTZ), and Pierella Paci analytical and research support was provided by David (Practice Manager EA1PV). Garcés Urzainqui (PhD, Vrije Universiteit Amsterdam and Tinbergen Institute), Kristen Himelein (EEAPV), The team offers its thanks to Loy Nabeta (AFREC), Elizabeth Jonathan Kastelic (DECPM) and Diana Winter (Master Howton (ECREF), Kafu Kofi Tsikata (AFREC), Anne Grant student, Georgetown University). The analysis of (Editor) and Jaime F. Alvarez (GCSSV) for their precious help structural transformation and productivity gaps was during the editing and communication process. The team produced by Marco Ranzani (EA1PV) with the support also extends its thanks to Martin Buchara (EA1PV), Diana of Christiane Wissa (Sr. Statistician, University of Cairo). Mpoki Mwaipopo (AFCE1), Anahit Poghosian (EA1PV), Firms performance was examined by Andre Marie Arlette Sourou (EA1PV), Anila Jane Mohan (AFCE1), Taptue (EA2PV) and the analysis of occupational choices Tsehaynesh Michael Seltan (EA1PV), and Santosh Kumar benefitted from the support of Diana Winter. The analysis Sahoo (EA2PV) for their invaluable assistance during the of economic choices by socially embedded individuals was preparation of the report. Acronyms and Abbreviations CPI Consumer Price Index DHS Demographic and Health Survey FDI Foreign Direct Investment GDP Gross Domestic Product HBS Household Budget Survey HE Household Enterprises HIV Human Immunodeficiency Virus MDI Multidimensional Deprivation Index NBS National Bureau of Statistics NFE Nonfarm Enterprise NGO Nongovernmental Organization NPS National Panel Survey pp Percentage points PSSN Productive Social Safety Net PPP Purchasing Power Parity ROSCA Rotating Savings and Credit Association SACCOS Saving and Credit Cooperative Society SSA Sub-Saharan Africa SDG Sustainable Development Goal TASAF Tanzania Social Action Fund TZS Tanzanian shilling UN United Nations WASH Water, sanitation, and hygiene WDI World Development Indicators WHO World Health Organization Vice President Hafez M. H. Ghanem Country Director Bella Bird Regional Director Asad Alam Practice Manager Pierella Paci Task Team Leader Nadia Belhaj Hassine Belghith Overview O ver the past decade Tanzania recorded the fastest-growing subsectors each employ on average no remarkable economic growth and a persistent more than 3 percent of the general population. They also tend decline in poverty. The country’s strategic to employ significantly more educated Tanzanians; their workers location, its rich and diverse resources, its socio- who have completed secondary and above exceed 60 percent political stability, and its economic reforms over the past four on average. Data from household surveys indicate a transition decades contributed to its economic success and serve as a of labor from low-productive agriculture to higher-productive foundation for further building up the economy. C ­ ontinued industry and services, but the transition is significantly slower government efforts to improve living conditions have than the transformation suggested by national accounts data. resulted in a sustained increase in access to basic services Lack of education and productive resources hold people back and improvement in human capital outcomes (though from a from improving their economic standing by moving to more pro- low base), which helped to reduce poverty. After plateauing ductive sectors. Only those equipped with more human capital between 2001 and 2007, in 2018 the poverty rate fell from and assets are able to benefit from the opportunities generated 34.4 to 26.4 percent. by economic growth; they increase their incomes and consump- tion much faster than the rest of the population. The result is However, Tanzania’s success is not unmitigated. Poverty was not more inequality. This pattern is partly driven by intergenerational reduced as much as the population grew, resulting in an increase transmission of poverty: Low parental education and economic in the absolute number of poor people. In 2018, about14 million status constrain the employment of their children, limit their people lived below the national poverty line of TZS 49,320 per upward mobility, and slow structural transformation. This contrib- adult equivalent per month and about 26 million (about 49 utes to perpetuate poverty and inequality across generations and percent of the population) lived below the $1.90 per person per deters efforts to eradicate poverty. day international poverty line. Vulnerability is also still high: For every four Tanzanians who moved out of poverty, three fell into it. This report provides a comprehensive analysis of poverty and A large number of nonpoor people living just above the poverty inequality in Tanzania and identifies some priority actions if pov- line are at risk of slipping below it. Beyond the persistent gaps erty is to be reduced. The analysis is contained in two parts. The between urban and rural areas, there are large disparities in the first part is based on the results of the Household Budget Surveys distribution of poverty across geographic regions. Poverty is (HBSs) for 2017/18, 2007, and 2011/12; several rounds of National highly concentrated in the western and lake zones, lowest in the Panel Surveys (NPSs); and Demographic Health Survey (DHS) eastern zones. data; it also combines spatial information from the population census and other sources with HBS data to (1) provide a rigorous The reduction in poverty is also low in relation to Tanzania’s analysis of the evolution, profile, and determinants of poverty remarkable economic growth. The growth elasticity of poverty and inequality; (2) explore movements in and out of poverty and indicates that a 10 percent increase in GDP growth per capita in their drivers; and (3) examine the distribution of poverty and Tanzania can be expected to reduce the proportion of the poor living conditions across the country at a detailed geographic by about 4.5 percent—low compared to estimates for other level. The second part examines the pattern of structural transfor- developing countries. This is due to both the concentration of mation, firm profiles, job creation, and financial inclusion using employment in slow-growing sectors and the dilatory trans- the rebased GDP figures released in February 2019 plus data formation of the economy. Based on national accounts data, from the Statistical Business Register (SBR), Census of Industrial industry and services are growing much faster than agriculture, Production (CIP), national accounts, NPS, Integrated Labor Force driving the growth and transformation of the economy. However, Surveys (ILFS), and other sources. 2 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Executive Summary Poverty has been falling for a decade, but recently the pace has slowed. Between 2007 and 2018 Tanzania’s national poverty rate went down from 13.3 to 9.7 percent in rural areas and from fell from 34.4 to 26.4 percent and extreme poverty fell 7.4 to 4.4 percent in urban areas. from 12 to 8 percent. The figures on poverty trends are drawn from the results of the Household Budget Surveys Poverty declined faster between 2007 and 2012 than (HBS) for 2007, 2011/12, and 2017/18. The poor are defined it has since. For 2007–12, poverty averaged a decline of as those whose consumption is below the national poverty percentage point (pp) a year, but the total reduction since 1 ­ line and who therefore were not able to meet their basic has been from 28.2 to 26.4 percent. In rural areas, between consumption needs; the extreme poor were not able to afford 2012 and 2018 poverty eased from 33.4 to 31.3 percent, while enough food to meet the minimum nutritional requirements urban poverty stagnated at around 16 percent. of 2,200 kilocalories (Kcal) per adult per day. The national basic needs poverty line for 2018 was TZS 49,320 per adult per Poverty in urban areas outside Dar es Salaam did not month and the food poverty line was TZS 33,748. Poverty fell really begin to fall until 2012. The reduction in urban pov- across the board but faster in rural areas, where poverty fell erty between 2007 and 2012 was driven entirely by a plunge from 39.1 to 33.1 percent, compared to a decline from 20.0 in the proportion of poor people in Dar es Salaam, from 14 to to 15.8 percent in urban areas (Figure ES.1). Extreme poverty 4 percent; in other urban areas the drop was marginal, from FIGURE ES.1: Poverty Trends, National Poverty Line, 2007–18, Percent 2007 34.4% 2012 28.2% Urban 20% Urban 2018 15.4% 26.4% Rural Rural 39.1% 33.4% Urban 15.8% Rural 31.3% Source: HBS 2007, 2011/12 and 2017/18. 4 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T 22.7 to 21.5 percent. In contrast, between 2012 and 2018, month for poor households in urban areas. The low poverty urban areas outside the metropolitan city saw poverty fall severity index, which was more than halved, from 4.5 in to 19.2 percent, while in Dar es Salaam it rose to 8 percent. 2007 to 2.1 in 2018, suggests that inequality between poor However, this increase is questionable: it may be driven by households is fairly low. changes in the sampling method as survey-to-survey imputa- tion results indicate a decline in poverty from 14 to 8 percent The robust record of poverty reduction is not affected by in 2007–12 and stagnation since then. changes in survey methodology. Assessing the changes in poverty over time is subject to issues of comparability The depth and severity of poverty also eased. For stemming from changes in survey design. The 2017/18 HBS 2007–18, the depth of poverty, which measures how far introduced computer-assisted personal interviews (CAPI), on average poor households are from the poverty line, lessened the length of the diary from 28 to 14 days, and decreased from 10 to 6 percent—in other words, a poor redesigned the sample for regional representation. Analysis household would on average need TZS 3,058 per adult and robustness checks, using survey-to-survey imputation equivalent per month to escape poverty. Since poverty methods, showed that the new survey design had little impact is deeper in rural areas, the amount needed there is TZS on changes in poverty levels and that Tanzania’s record of 3,650, far more than the TZS 1,726 per adult equivalent per poverty reduction is solid.  5 Gradual improvement in living conditions and human capital helped reduce poverty. Access to electricity has progressed somewhat but electrifi- with improved water sources has almost doubled. Piped cation of the whole country is still insufficient, particularly in water systems and household connections to them (mainly rural areas and for poor households. Although 29 percent of urban) contributed to the improvement (Figure ES.3). Water Tanzania’s households have access to electricity, access is avail- systems also helped to raise the percentage of households with able to just 10 percent of rural and 7 percent of poor households access to piped water, in or outside the dwelling, which has (Figure ES.2). The country’s strategy to diversify toward solar greatly reduced the time and distance needed to access water. energy has started to pay off, particularly in rural areas, where However, many households still lack access to a safe source of 33 percent of households use solar energy for lighting compared drinking water. In 2018, the drinking water of about 26 percent to 14 percent in urban areas. Despite some improvements, about of households was unimproved and unsafe—for urban house- 45 percent of households still rely on such inefficient lighting holds the rate was 12 percent and for rural 34 percent. sources as torches and kerosene. Use of efficient energy sources for cooking has also improved slightly, but over 80 percent of all Access to basic and limited sanitation improved con- households, and more than 90 percent of rural and poor house- siderably in urban areas but is still highly problematic holds, continue to rely on firewood and charcoal. in rural areas. Between 2012 and 2018 the percentage of urban households with improved sanitation rose from More Tanzanians now have safe drinking water, particu- 36 to 58 percent, but in rural areas from just 5 to a still low larly in urban areas, where the percentage of households 11 percent (Figure ES.4). In 2018, 65 percent of Tanzanian FIGURE ES.2: Source of Lighting, FIGURE ES.3: Access to Water, FIGURE ES.4: Access to Percent Percent Sanitation, Percent 100 100 100 7 3 1 8 12 12 13 10 10 19 26 25 18 18 24 27 34 31 31 10 80 40 44 80 40 80 41 48 46 14 38 61 43 14 60 60 65 60 40 51 72 79 81 61 16 15 78 25 79 1 40 45 60 40 26 20 59 40 70 69 49 47 18 19 64 11 33 20 7 35 20 33 20 48 28 8 2 16 17 4 8 17 3 17 2 29 8 23 25 34 6 7 2 3 8 18 0 0 2 2 7 42 10 3 1 7 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 0 2012 2018 2012 2018 2012 2018 2012 2018 Tanzania Rural Urban Poor Tanzania Rural Urban Poor Tanzania Rural Urban Poor Piped water inside dwelling Basic sanitation Electricity (grid) Piped water outside dwelling Limited sanitation Solar Improved drinking water Unimproved sanitation Kerosene/paraf n Unimproved drinking water Open defecation Other Source: HBS 2011/12 and 2017/18. Source: HBS 2011/12 and 2017/18. Note: Using SDG definitions, basic sanitation Note: Using WHO definitions, improved water Source: HBS 2011/12 and 2017/18. includes non-shared flush toilets or improved pit includes public tap, protected dug well. Unimproved latrines. Limited sanitation includes shared flush water includes unprotected sources and surface toilets or improved pit latrines. water. 6 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE ES.5: Net Education enrollment rates 2012–18, Percent 100 91 85 83 82 80 71 68 60 50 51 40 31 34 22 26 20 2 2 5 5 1 1 0 a l n a l n a l n ra ra ra i i i ba ba ba an an an Ru Ru Ru Ur Ur Ur nz nz nz Ta Ta Ta Primary Lower secondary Upper secondary 2012 2018 Source: HBS 2011/12 and 2017/18. households used unimproved sanitation systems; in rural areas FIGURE ES.6: Educational Achievements of Adults 15+, 10 percent of households still mainly use open defecation. 2018, Percent Better access to roads, markets, and public transportation 100 3 1 4 1 9 4 2 11 6 also helped account for the rise in household living stan- 17 19 16 19 80 29 dards. Access to roads, public transportation, and markets 47 48 improved throughout the country, but especially in rural areas 60 46 47 47 48 and for poorer households. Access to cell phone signals, 40 46 16 17 health centers, and to a lesser extent banks also expanded 13 11 20 12 15 among the poor, especially in urban areas other than Dar es 24 8 26 23 18 16 13 Salaam. However, improvements were from the low base and 8 0 Tanzania Rural Urban Nonpoor Poor Men Women are still far from sufficient. Area Poverty Gender No education Less than primary Completed primary As ownership of modern assets has risen, ownership of Lower secondary Upper secondary University traditional goods has declined. More Tanzanian households Source: HBS 2017/18. now have mobile phones, televisions, motorcycles, refrigera- tors, and to a lesser extent cars and laptops. Fewer now own more traditional items like radio sets or stoves. 17 percent; the education profile has improved mostly for the younger generation (Figure ES.6). Overall, primary and lower Human development outcomes, particularly in education, secondary education increased for the poor in both urban and are also better. Gross and net enrollment rates in primary rural areas, but enrollments in upper-secondary and university schools went up slightly between 2012 and 2018 in both rural education were significantly larger among the richest urban and urban areas. However, gross enrollment in both lower and households. Increased education for women seems to have upper secondary decreased, and net enrollment increased helped to slightly reduce the total fertility rate. Adult mortality only marginally. The combination of stagnant or decreas- and the probability of dying before age 50 have also fallen. ing gross enrollment with increasing net enrollment does, however, suggest improvements in enrollment within the qual- Despite these improvements, achievements in human ifying age categories and in fewer students repeating classes. capital are low. Tanzania was ranked 128th out of 157 coun- Yet in 2018 the lower secondary enrollment rate was only tries in the 2018 World Bank Human Capital Project with a low 34 percent, and upper secondary enrollment barely reached Human Capital Index (HCI) score of 0.4. Low expected years 2 percent (Figure ES.5). While more adults aged 15 and of schooling is among the main limiting factors to the HCI. older are now attaining lower secondary, the rate is still just The World Economic Forum Global Human Capital Report  7 of 2017 also ranked Tanzania 106th of 130 countries in HCI. Movement out of agriculture seems to have accelerated Tanzania trails countries with similar income levels, underper- since 2012, coupled with more jobs and higher earn- forming particularly in the know-how sub-index (109th) due to ings, especially in services and to a lesser extent industry the very low share of high-skilled employment, limited avail- Figure ES.9). Self-employment also increased. However, (­ ability of skilled employees, and low economic ­ complexity these positive changes occurred primarily among house- (Figure ES.7). holds in urban areas that were only moderately poor or already better-off. Poor rural households saw only a slight Over the last decade anthropometric indicators for chil- increase in engagement in self-employment and, while dren under 5 got better, but undernutrition is still chronic, remaining positive, their returns declined over recent years especially in rural areas. According to the 2015/16 DHS, (Figure ES.10). Poor urban households outside Dar es almost 35 percent of Tanzanian children under 5 are stunted Salaam saw a slight increase in wage employment and in and 12 percent are severely stunted, indicating a cumula- private businesses, coupled with a modest improvement in tive growth deficit. The problem is particularly acute in rural returns. In general, better-off households, which have more areas, where about 38 percent of under-5 children are stunted education and other assets, were better-positioned to take Figure ES.8). (­ advantage of the opportunities generated by economic growth and so were able to raise their consumption much There has been a progressive shift to more productive faster than poor households. work in services and industry, but the shift has mostly occurred in urban areas, and among better-off groups. These improvements helped to raise household consump- FIGURE ES.7: Human Capital Index tion, but the benefits were partly offset by the lack of opportunities for meaningfully higher economic returns. Human The reduction in poverty was driven by better access to basic Capital Index services, assets, and infrastructure; and more human capital, 100 which helped to raise both household endowments and living 80 Tanzania 60 Kenya standards. However, these endowments no longer generate 40 Uganda the same increase in consumption as previously, so that the Know-how Capacity 20 Rwanda reduction in poverty was slower than what would have been 0 South Africa achieved previously. The expansion of access to education and the increase in educational attainment in the general popula- tion has paralleled changes in labor market requirements, so Development Deployment that the rewards for years of schooling below a certain level have declined. In particular, the gains in income, and consump- Source: Human Capital Report, 2017. tion, associated with primary education have become minimal. Consequently, poorer households, whose heads are generally FIGURE ES.8: Under-5 Children Nutrition Deficit, Percent older and cannot access more education, have seen a marked decline in the returns to their years of schooling. As more 50 people have mobile phones and access to mobile signals and 38.1 the Internet, the value that such devices add to consumption 40 34.8 lessens. Popular for business purposes and fund transfers, 30 13 25.0 mobile phones continue to positively affect the livelihoods of 12 the poor, but since 2012 their marginal benefits have narrowed, 20 8 especially in urban areas and in moderately poor households, 23 25 for whom ownership of these assets has expanded rapidly but 10 17 opportunities for their productive use have not (Figure ES.11). 0 Similarly, returns from greater market access have been falling, Tanzania Rural Urban Stunting especially in rural areas. Possession of modern transportation Moderate Severe Total assets like motorcycles and cars has significantly reduced the returns to bicycles and other basic transport alternatives, which Source: DHS 2015/16. are still quite prevalent among the poor. 8 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE ES.9: Drivers of Poverty Reduction in Mainland Tanzania, 2012–18: Endowments and Returns Effects, Percent 100% 80% 74 67 60 60% 50 40% 29 30 30 17 19 20% 7 11 10 6 3 6 3 0% –1 –20% –9 –15 –40% –60% –80% –97 –100% Consumption Endowments Returns Basic Assets Education Nonfarm Assets Education Nonfarm services employment employment Total variation (%) Contribution to endowments (%) Contribution to returns (%) Poor Better-off Source: HBS 2011/12 and 2017/18. Notes:  - Results are based on the unconditional quantile regression, which decomposes changes in consumption over time into changes in people characteristics or endowments (i.e., increased education levels, ownership of assets, and access to productive employment) and the returns that they get for those endowments (i.e., returns to education, employment, and assets). The method applies the Oaxaca-Blinder decomposition to each unconditional decile of the consumption distribution to assess the amount of poverty reduction attributable to changes in the endowments of households and the amount due to changes in the returns to these endowments.  - The poor are those in the first three deciles and the better-off are those in the two highest deciles.  - Nonfarm employment groups self-employment as well as employment in services and industry. Improvements among the poor are due to higher engagement in self- employment alone, while improvements among the better-off are due to higher engagement in self-employment with others as well as better access (and higher returns) to wage employment and employment in industry and services. FIGURE ES.10: Drivers of Poverty Reduction in Rural Areas, 2012–18, Percent 100% 80 80% 60% 38 40% 29 27 17 20 20% 13 10 4 4 7 4 8 6 7 2 0% –5 –3 –9 –20% –40% –60% –80% –70 –100% Consumption Endowments Returns Basic Assets Education Nonfarm Assets Education Nonfarm services employment employment Total variation (%) Contribution to endowments (%) Contribution to returns (%) Poor Better-off Source: HBS 2011/12 and 2017/18.  9 FIGURE ES.11: Drivers of Poverty Reduction in Urban Areas, 2012–18, Percent 128% 140% 120% 100% 80% 55% 60% 42% 32% 32% 40% 25% 23% 17% 11% 13% 6% 10% 14% 20% 5% 4% 4% 0% –20% –12% –40% –31% –60% –80% –59% –77% –100% Consumption Endowments Returns Basic Assets Education Nonfarm Assets Education Nonfarm services employment employment Total variation (%) Contribution to Endowments (%) Contribution to Returns (%) Poor Better-off Source: HBS 2011/12 and 2017/18. 10 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Poverty reduction has not been responsive to Tanzania’s remarkable economic growth, and inequality has worsened. Considering Tanzania’s remarkable economic growth, poverty is due to the discrepancy between the price deflators reducing its poverty has been very slow. Growth in gross the GDP deflator and the consumer price index (CPI) used to domestic product (GDP) averaged 6.3 percent from 2007 convert nominal GDP and household consumption values to to 2017, dropping to 3.3 percent when adjusted by popula- real terms. The GDP deflator implies a much slower rate of tion size. These figures are based on the GDP series, base inflation than the CPI, which resulted in significantly higher year 2015, released in February 2019. The new series show a growth of real GDP per capita than of mean household con- slight increase in GDP and less volatility in economic growth sumption. In 2012–18, GDP and CPI deflators produced similar since 2012. The trend in the previous rebasing, released in inflation rates of about 38 percent, making the growth rates December 2014, base year 2007, was similar, with GDP growth more comparable, whether based on household consumption averaging 6.3 percent and GDP per capita growth 3.5 percent per capita (1.5 percent) or GDP per capita (3.3 percent). Thus, for 2008–13. However, persistent growth has had only a mod- no matter how growth is measured, the response of poverty est impact on poverty. to economic growth is very low (–0.73 using survey mean figures and –0.45 using GDP figures). The growth elasticity of poverty dropped from a low –1.02 in 2007–12 to –0.45 in 2012–18. Thus, a 10 percent increase in Poorer people therefore benefitted less from economic GDP growth per capita can be expected to produce a 4.5 percent growth. The pro-poor growth signs that emerged in 2012 decrease in the proportion of the poor. This is very low—for seem to have reversed in 2018—consumption growth in developing countries, on average poverty is expected to drop by 2012–18 was significantly lower for those at the bottom of over 20 percent when per capita GDP rises by 10 percent. the consumption distribution than among the better-off (Figure ES.12). The pattern replicated that of 2001–07 but not In 2007–12 the pattern of growth in household con- 2007–12, when growth mainly benefitted poorer groups. sumption diverged significantly from GDP growth due to a discrepancy between price deflators, but the differ- The beneficial effects of economic growth were partly ence narrowed in 2012–18. How much poverty reduction offset by worsening inequality. The poverty headcount fell responds to economic growth depends on whether economic faster in 2007–12 than in later years, despite a much lower growth is defined based on changes in GDP per capita in the increase in mean household consumption. In 2012–18, the national accounts or measured directly from the household positive impact of household consumption growth on poverty surveys on which poverty estimates are based. Economic reduction (the growth effect) was largely offset by the rise growth estimated using changes in mean household con- in inequality (the redistribution effect) (Figure ES.13). The sumption per capita calculated from HBS 2007 and 2012 was deterioration in the Gini coefficient supports this: having gone only 0.9 percent annually, significantly lower than growth down from 38.5 to 35.8 percent in 2007-12 it has risen to 39.5 in GDP per capita. Using survey-based mean consumption in 2018. Throughout the region, inequality was lower than it to measure growth shows an estimated growth elasticity of had been initially in rural areas (at 33.5 percent) but jumped poverty of –4, which implies that household consumption has in urban areas, essentially in Dar es Salaam where the Gini more impact on poverty reduction than GDP per capita. The coefficient reached 43 percent in 2018, up from 40 percent in difference between the estimates of the growth elasticity of 2007 and 36 percent in 2012.  11 FIGURE ES.12: Growth Incidence Curves, 2012–18, subsectors grew relatively fast at about 5 percent. However, Percent even there, few of the poor produce market-oriented crops and livestock; they mostly operate subsistence farms. 8 Tanzanians with more education and skills were thus 6 better positioned to benefit from fast- growing sectors. Better-off Tanzanians, who have more human capital and 4 productive assets, were better able to take advantage of the opportunities generated by the fast-growing sectors. Their 2 income and consumption rose significantly faster than for 0 those with less education and fewer endowments employed in sectors growing more slowly. As a result, inequalities –2 widened. Another result was the persistence of the urban-rural 0 20 40 60 80 100 welfare gap: urban dwellers with more education and assets Growth rate by percentile could better access productive jobs and maintain their higher Growth rate in mean economic status. Policies to empower the rural poor did partly reduce urban-rural gaps among poorer groups, but among Source: HBS 2011/12 and 2017/18. the better-off inequality worsened. FIGURE ES.13: Growth and Redistribution Effects, A greater proportion of households are operating their 2012–18, Percentage points own businesses, which could be a pathway out of poverty. In 2018, about 14 percent of households own nonfarm enter- prises (NFEs), up from 9 percent in 2012. Here, however, there 30 26.1 is a location and gender bias: the proportion of households 20 operating NFEs is about three times higher in urban areas and 10 fewer women own NFEs than men. Since 2012, the proportion of women engaged in NFEs has risen in rural areas but fallen 0 in urban areas. –1.8 –10 The profile of households operating NFEs closely –20 resembles that of the better-off households: (1) The –21.1 –30 household heads were more educated, with lower Change in poverty headcount secondary school completion rates of about 17 percent, Growth compared to 12 percent nationally and less than 6 percent Redistribution among farmers. (2) They had fewer children and thus fewer household members and lower dependency ratios. In 2018 Source: HBS 2011/12 and 2017/18. the average size of an NFE household was about 5, and of farming households 6. (3) Households that operated NFEs, Growth was driven by sectors where few in the general especially in urban areas, were at the top of the consumption population work, particularly the poor. The fastest-growing distribution, with average monthly consumption about 2.5 sectors are construction, information and communication times that of farming households. technology (ICT), real estate, nonmarket services (e.g., education, health, and public administration), and to a Despite being better-off, most NFE households continued lesser extent mining, transport, and trade. Each of these to operate in the informal sector, lacked access to sectors employs on average no more than 3 percent of the formal funding sources, and created few jobs. In 2018 population. However, their employees tend to be significantly the majority of NFEs were either mobile with no fixed more educated and better-off, who average 60 percent location (24 percent) or operated out of the owner’s house or more of their workers. These sectors employ over 20 (32 percent). Moreover, in 2012 only 11 percent had been percent of Tanzanian workers with lower secondary education registered with the business registration and licensing agency and above. Within agriculture, where most Tanzanians (BERLA), and in 2018 this had not changed. Moreover, in both work, particularly the poorer ones, the crops and livestock 2012 and 2018, most NFEs were no more than two years old. 12 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Despite initially relying on personal savings or agricultural from family or friends also grew. It will take a long time for proceeds to start their businesses, after 2012 more NFEs NFEs to create economies of scale and scope. Only a few began to use formal sources of credit to run their operations, employ workers who are not members of the household. In though the increase is from a very low basis. The share of 2018, 27 percent of NFEs were working proprietorships and NFEs funded by owner savings fell from 36 to 33 percent and 25 percent reported having unpaid household members the share funded by Credit Co-Operative Society (SACCOS) working in the business. Only 7 percent hired paid workers and bank loans rose from 2 to 3 percent. Reliance on loans who were not household members.  13 Slow progress in poverty reduction has pushed up the number of poor people, leaving a noticeable share of the population still at risk of at least transitory poverty. Because the population grew faster than poverty was number of poor by nearly 0.7 million. In rural areas, the total reduced, in 2018, about 14 million Tanzanians lived in population went up by only 26 percent, the number of poor poverty, up from 13 million in 2007 and 12 million in 2012. people was almost unchanged, and the number of extremely In 2018, 32 percent of the population lived in urban areas, poor declined. However, over 80 percent of the poor up from 29 percent in 2012. As urbanization accelerated, the (11.3 million) continue to live in rural areas, where 3.5 million increase of the poor was proportionately higher in urban of them suffer extreme poverty; in urban areas, 2.6 million live areas, but the majority still live in rural areas. In 2007–18, the in poverty and 745,000 in extreme poverty. urban population rose by about 7 million and the number of urban poor by 0.6 million—a result of both urbanization and A large proportion of the population is clustered around the dilatory pace of urban poverty reduction (Figure ES.14). the poverty line. Raising the consumption of poor house- Most of the changes occurred outside Dar es Salaam, where holds by just TZS 350 per adult equivalent per day would the population has gone up by nearly 5 million and the lift about half of the poor out of poverty. However, a quite significant proportion of Tanzanians who live just above the FIGURE ES.14: Number of Poor People, 2007–18, Million poverty line are at risk of being pushed into poverty by an economic shock (Figure ES.15). Mobility in and out of poverty is thus very high. 16 13.9 Between 2010 and 2015, the poverty status of nearly 13.2 12.3 30 percent of the population changed. Analysis of the 12 people (in million) Number of poor 11.2 11.3 dynamics of poverty based on National Panel Surveys (NPSs) 10.4 found that about 16 percent of Tanzanians escaped poverty 8 and about 12 percent fell into it. In both years, about 60 per- cent of the population were nonpoor and 12 percent stayed 4 poor (Figure ES.16). 2.6 2.0 1.9 Transition out of poverty has generally been coupled with 0 2004 2006 2008 2010 2012 2014 2016 2018 2020 shifts to more productive activities, within agriculture or outside to services. Over 70 percent of those who moved Tanzania Mainland Urban Rural out of poverty still considered agriculture their primary work Source: HBS 2007, 2011/12 and 2017/18. activity, even though many supplied more hours to services 14 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T and industry. Also, many moved from unpaid family helper They also tended to work fewer hours. Yet exposure to shocks status to self-employed or wage worker. The others moved does not seem to have significantly affected transitions into from agriculture to services or, to a lesser extent, industry. and out of poverty. The numerous initiatives to empower the On average, in all sectors they worked more hours, but even poor seem to have brought them closer to fulfilling basic more in productive ones. Meanwhile, those who fell into consumption needs, but the big jump out of poverty is yet to poverty moved to less productive work, such as unpaid family be achieved. helper, and continued to spend most of their time in farming. FIGURE ES.15: Consumption Density around the Poverty Line in 2018 (2011 PPP). 25 Int. poverty level 20 Nat. poverty level 15 10 5 0 0 10 20 30 40 50 60 70 80 90 100 Percentile Mean consumption/day/person US$ 1.9 (internat. poverty line) National poverty line US$ 3.2 Source: HBS 2017/18. FIGURE ES.16: Transition In and Out of Poverty, 2010–15, Percent NPS 2014–15 13% entered poverty (new poor) Nonpoor Poor Non-poor 60% 13% 73% NPS 2010–11 Poor 16% 11% 27% 76% 24% 100% 16% exited poverty (new non-poor) Source : NPS 2010/11 and 2014/15.  15 It is hard for the poor to achieve a better life because of a large number of dependents, low human capital, low-profile jobs, and limited access to basic services and assets. Poor households are burdened by large numbers of of women-headed households are particularly vulnerable to dependents and disadvantaged by too little education. poverty—single and divorced women are poorer than men by The number of children under 15 in poor households is almost about 11 pp. The gap is high in both rural and urban areas, double the number in nonpoor households. Poor households but particularly in the latter and among divorced households, also have significantly higher dependency ratios. About where it exceeds 22 pp. Urban widows are also poorer 44 percent of households with five or more children under 15 than urban widowers by about 14 pp. Ownership of assets, are poor, 18 pp higher than the national average and 28 pp especially mobility and communication equipment, is also more than the poverty rate for households with just one or significantly lower among women-headed households, which two children. About 29 percent of household heads have no indicates the limited access of women to productive assets. education and 19 percent did not complete primary school, with rates being highest among poor rural households. Only Less human capital and limited access to basic services 3.4 percent of the heads of poor households (and 1.3 percent limit opportunities for the poor to access productive of rural ones) went beyond primary education, compared to jobs. About 80 percent of the heads of poor households 20 percent for nonpoor households. Of households whose work on their own farms or to a lesser extent as unpaid family heads have no education or did not complete primary, farm helpers. In these households the poverty rate is more about 35 percent live in poverty. The poverty rate drops to than double that of those in other employment categories. 26 percent for those who completed primary schooling and Households whose head is an entrepreneur with others or to just 6 percent among households with lower secondary self-employed in nonfarm sectors are less likely to be poor. education and above (Figure ES.17). While education is still Similarly, poverty rates are lowest among households whose the best shield against poverty, primary education seems no head works in trade and services (Figure ES.18). Although longer sufficient to open up opportunities. so many poor households are engaged in agriculture, about 18 percent of them fear almost every month that they will Poverty seems to be more prevalent among women. It is run out of food, compared to 10 percent of nonpoor ones; difficult to understand poverty by gender because household another 37 percent of poor households estimate they might surveys assume equal distribution of consumption between run out in at least some months. In all, 50 percent of poor members of a household, and because of the status of households ran out of food if not every month in at least women who head households. However, there are indications some, compared to 32 percent of nonpoor households. that poverty is more prevalent among women. Particularly in urban areas, more women-headed households are poor Human capital and access to productive jobs are lower (20.3 percent) than men-headed ones (14 percent). Some types for women than men, particularly in poor households. 16 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE ES.17: Education Level of Household Head and FIGURE ES.18: Sector of Employment of Household Head Poverty, 2018, Percent and Poverty, 2018, Percent 40 35.5 35.0 Agriculture 29.5 30 26.0 20 Manufacturing 19.3 10 6.3 7.9 1.1 0 Services 12.1 n y y ry ry y ar ar sit io da da im im at r ive on on uc pr pr Un c ec ed se Public administration 7.3 ed ed rs o et et r we pe N pl pl Up Lo m m co Co 0 10 20 30 40 an th ss Source: HBS 2017/18. Le Source: HBS 2017/18. Nationally, 23 percent of women have no education and rely on unimproved sanitation facilities or none at all. Only 19 percent more did not complete primary school, compared about 13 percent of poor households have access to tarmac to 13 percent of men with no education and 24 percent who roads; 44 percent lack any source of access. Among nonpoor completed less than primary. The gender gap is larger in poor households, the corresponding rates are 22 percent with households, where 32 percent of women have no education good access and 32 percent without any. Also, 41 percent compared to 19 percent of men. As a result, more women of households have no access to a health center, dispensary, than men are in unpaid household work and low-paying jobs. or hospital, whether public or private. Here, poverty status The gaps are particularly large among poor households, makes very little difference. where they exceed 7 pp. However, the gender gaps in edu- cation and employment are significantly lower in the younger For many, access to the Productive Social Safety Nets generation, suggesting that gender differentials are starting (PSSN) program is essential for meeting basic consumption to shrink and that policies to enhance girls’ education and needs, but its coverage is limited. The program is managed empower women are beginning to bear fruit. by Tanzania Social Action Fund (TASAF), which reports that the program covers 1.2 million households, of which 250,000 Poor households suffer from less access to infrastruc- benefit from public works program. HBS 2018 found that ture and community services, which minimizes their about 1 million households and 4.9 million people bene- opportunities. Social and community services, such as fit from the PSSN cash transfer program, of whom 291,000 electricity, water supply, health facilities, roads, markets, and households (1.4 million people) also benefit from the public communication networks, are the backbone of household work program. PSSN reached 15 percent of extremely poor development; they structure the household environment and households, 14 percent of poor households, and 8 percent promote emergence of opportunities. Moreover, the exis- of nonpoor ones. In line with the national distribution of the tence of essential services and infrastructure often exposes poor, the beneficiaries are mostly rural, constituting 78 per- the serious shortfalls in services for poor households, which cent of households benefiting; coverage reaches 11 percent are more likely to live in underserved communities, which of all rural and 5 percent of all urban households. The fact that perpetuates their dire lack of both cash and access to oppor- the PSSN benefit is mainly used to purchase food underscores tunities. Only 7 percent of poor households are connected to the high food stress in Tanzania, especially for the poorest the electrical grid and 28 percent use solar energy; the rest households. About 66 percent of beneficiaries reported using rely on inefficient energy sources for lighting. Over 90 percent PSSN income support to cover their food needs, 13 percent of poor households use firewood and charcoal for cooking. to cover education and health expenses, and 21 percent to About 30 percent of poor households still have access only invest in productive assets and improve housing (8 percent). to unsafe sources of drinking water and over 90 percent Of the poorest beneficiary households, only 4 percent invest  17 in productive assets. While most nonpoor beneficiaries also 5th invested the PSSN cash in productive assets compared to use the funds mainly for food, about 10 percent invest in only 4 percent of households in the poorest groups. Investing productive assets. in productive assets may have helped these households to improve their living standards faster and graduate to higher PSSN may have helped many beneficiaries to escape income groups. More than six years since the program was poverty. PSSN is intended to target the poorest Tanzanians. designed, its targeting needs to be reviewed, but the process According to HBS 2018, after the program had been in oper- needs to be managed very carefully because some nonpoor ation for three years, about 57 percent of PSSN beneficiaries beneficiaries may fall back into poverty if they are dropped were in the two poorest consumption quintiles. However, from the program. Even households in the highest quintiles about 69 percent of the households interviewed that reported are at risk as their productive investments could be affected, benefiting from PSSN were above the poverty line, though something that would be even more problematic if these still receiving PSSN support. For some this could be a tempo- investments are used to support other poor households. The rary change of status; that is, for households that were close review of program targeting and recertification of beneficiary enough to the poverty line, the income support from the households needs thorough analysis to identify appropriate PSSN cash transfer may allow them to afford consumption candidates, supported by measures to build the resilience of above the poverty line but they would likely fall back into pov- those who may no longer qualify so that they do not fall back erty if PSSN support is removed. For about 25 percent of them into poverty. It needs also to be guided by processes that are the average consumption was only 20 percent higher than the objective and standardized. poverty line; these households are at high risk of f­ alling back into poverty if income support is taken away. In other cases, Without PSSN, basic needs and extreme poverty the PSSN may have allowed some households to move sus- would have been higher. Without PSNN income sup- tainably above the poverty line. About 23 percent of current port, poverty would have been about 2 pp higher, which beneficiary households are in the two upper consumption translates to an additional 1 million poor people, and quintiles (15 percent in the 4th quintile and 8 percent in the extreme poverty would also rise from 8 to 9.2 percent, 5th), which makes them 7 percent of all households in the 4th equivalent to 700,000 more people. Expanding coverage quintile and 3 percent of all in the 5th. Of PSSN beneficiaries, of the program and its targeting would help to accelerate 10 percent of those in the 4th quintile and 18 percent in the poverty reduction. 18 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T The drivers of poverty are mutually reinforcing and carry across generations. The poor start life at a disadvantage, and many pass vulnerability and gender inequality. Intergenerational mobility poverty on to their offspring. They are hobbled by, among across economic sectors seems very limited; Tanzanians often other deficits, limited resources, malnutrition and health have the same employment status and jobs in similar sectors problems, poor access to social services and health care, and as their parents. low education and skills. They lack income, can save little for the future, are vulnerable to shocks, and have limited cop- Estimates of inequality of opportunity have found that ing strategies. Lacking the skills to take advantage of most about 20 percent of total inequality in consumption is work opportunities, they are generally limited to vulnerable due to circumstances outside the individual’s control and and low-productivity jobs. These deficits limit the upward about 16 percent is explained by family background. mobility of their children, perpetuating poverty for follow- Parents’ education and father’s employment have the most ing generations. Tanzanians of less-educated parents are influence on their children’s outcomes and opportunities for more likely to be less-educated themselves, and those with economic mobility (Figure ES.20). This is a much larger share better-educated parents are more likely to achieve higher than in other Sub-Saharan African countries, where inequality education (­Figure ES.19). Educational mobility is lower among of opportunity is significantly lower. Without additional policy the poor and women, so that low human capital perpetuates actions, future generations of the poorest Tanzanians will likely be trapped in persistent poverty. FIGURE ES.19: Intergenerational Mobility in Education, FIGURE ES.20: Contributions of Individual’s 2018, Percent Circumstances to Inequality, 2018, Percent 1 2 4 100 15 3 2 12 16 16 8 6 8 14 80 10 Education level of 35 26 26 children (%) 49 23 60 44 22 21 18 20 40 16 15 5 19 18 11 8 10 20 41 20 21 29 27 24 0 6 n y y ry ry y 2 ar ar ar tio 1 a a 0.3 im im nd nd nd a uc pr pr co co co ed se se se e d m te o e ed e So e Tanzania Urban Rural N m ov pl et So Ab m pl Mainland Co m Co Education level of the father (%) Gender Birth place No education Some primary Family background Completed primary Some secondary Completed secondary Above secondary Source : HBS 2017/18. Source : HBS 2017/18.  19 Inequality of opportunity is two times higher in urban areas extent that unobserved circumstances and institutional measures than in rural ones. The share of inequality of opportunity in total (e.g., family composition, parental financial situation, supply inequality is 25 percent in urban areas compared to 11 percent and quality of schooling, and labor and land market institutions) in rural zones. This reflects two facts: (1) Family background shape opportunities for rural Tanzanians, estimates of inequality variables have more influence on households and individuals of opportunity that do not take these circumstances into account who have more education and are engaged in more diversified are significantly biased downward. occupations and jobs as it is the case in urban sectors. (2) To the 20 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Beyond the urban-rural divide, geographic disparities in poverty are substantial. There are major disparities in the incidence of poverty people in extreme poverty. Estimates for districts reveal broad and the distribution of poor people across the country. By pockets of poverty in the north, west, and south. Poverty was region, poverty ranges from as high as 45 percent in Rukwa to most pervasive in Longido in the Arusha Region to the north, as low as 8 percent in Dar es Salaam (Figure ES.21). Overall, followed by Kibondo in Kigoma Region and Sumbawanga about 33 percent of the poor are concentrated in the highly Rural in Rukwa Region to the west, where poverty exceeds rural lake zone, where less-productive and subsistence activ- 50 percent (Figure ES.22). Generally, regions with the highest ities are common. In the lake zone, 4.6 million live in poverty poverty rates also host the largest numbers of poor, particu- and 1.3 million in extreme poverty; in the northern and larly in the area around Lake Victoria and Simiyu region in the eastern zones, less than 1.4 million live in poverty and 420,000 north; Kigosi, located between Shinyanga and Tabora regions; FIGURE ES.21: Poverty Incidence by Region, 2018 FIGURE ES.22: Poverty Incidence by District, 2018 Source: HBS 2017/18 and auxiliary variables. Source: HBS 2017/18 and auxiliary variables.  21 and Moyowosi in Kigoma region. However, there are also sig- to jobs and trade opportunities are hard to overstate. While nificant numbers of poor people in Lushoto district in Tanga the Dar es Salaam administrative region accounts for only a and Ilala district in Dar es Salaam because their populations small fraction of mainland Tanzania’s land area (0.16 percent) are larger. Conversely, Longido district in Arusha region and and only about 8 percent of its population, the region Nanyumbu district in Mtwara region, where poverty is high, accounts for about 40 percent of its manufacturing jobs and are sparsely populated and therefore have fewer poor people. 53 percent of manufacturing value. Dar es Salaam also han- dles 95 percent of its port traffic, dwarfing the importance of Urbanization and its associated structural transformation other cities as sources of imported goods. led poverty to drop faster in leading areas and gaps to widen in lagging ones. For the past decade, urban growth Across the country, climate is also a driver of geographic has averaged 5.5 percent—higher than the national average inequalities in welfare and poverty. In Tanzania, poverty of 3 percent population growth. In 2018, 32 percent of all Tan- is more pronounced in tropical savannah zones, which are zanians lived in urban areas and by 2045 this share is expected dominant in the northwest and southeast. Agriculture in the to reach 50 percent. Poverty is lower in larger and more tropical zones tends to be less productive due to poor soil densely populated districts and in areas with a greater con- quality because of high temperatures and heavy rain. These centration of lights at night, which demonstrate both greater zones are not suitable for producing wheat, which grows only urbanization and more economic activity. In general, success- in temperate climates, or maize and rice, which prefer tem- ful urbanization translates into poverty reduction through perate and subtropical climates. Thus, production of maize, the structural transformation in which workers shift from the most productive crop and a critical driver of poverty low-productivity agricultural activities to higher-productivity urban jobs. In large cities, more productive jobs are available, and the productivity gains come partly from the benefits of FIGURE ES.23: Market Access to Major Urban Centers agglomeration economies, such as resource-sharing, quicker and more accurate job matching, and greater knowledge spillovers. Tanzania is no exception. Access to markets is limited, particularly in the northwest and southeast, areas typically characterized by severe pov- erty. Farmers in rural areas with limited market access suffer from relatively higher prices for fertilizers because of higher transportation costs and fewer product options (Figure ES.23). These farmers also have little access to output markets and must take less competitive prices. Lack of market access thus traps rural farmers in poverty and exacerbates inequalities between rural and urban areas. Road infrastructure and transport services also affect access to markets. The poor quality of rural roads deters con- nectivity between rural areas and urban markets. Rural roads tend to be rudimentary, and transport services and facilities are unreliable and inadequate; in many remote parts of the country high post-harvest losses are an estimated 35 percent of total production. Thus, few and poorly maintained rural roads are a serious deterrent to the development of commer- cial agriculture. Areas surrounding Dar es Salaam, the commercial and eco- nomic center of the country, tend to have better access to markets. The region’s road network is wide and dense, which Source: World Bank’s estimates based on OpenStreetMap. allows people living near the city to benefit from its agglomer- Notes: The Open Source Routing Machine (OSRM) algorithm is used to compute ation economies. The positive economic dividends of access travel time between each village and major cities. 22 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T reduction in Tanzania, is concentrated in the southern high- to such risks. Drought is the most frequent natural disaster lands (e.g., Iringa and Mbeya); the southwest (e.g., Shinyanga in poverty-stricken areas. Parts of Tanzania that were previ- and Rukwa); and Arusha, where the climate is nontropical and ously highly productive, such as the southern and northern the land is most fertile. The fact that tropical zones also host highlands, will increasingly become tropical areas because of a variety of agricultural diseases, particularly the vector-borne declining rainfall, frequent droughts, and significantly more diseases prevalent in tropical countries, has serious implica- spatial and temporal rainfall variability. In Tanzania drought tions for health and labor productivity. has significantly worsened food insecurity, livestock and crop losses, and transmission of infectious diseases. Drought is The climate also has a major influence on the vulnerability more common now in parts of Arusha, Manyara, Shinyanga, of the poor to natural disasters. Although there is no direct Simiyu, and Dodoma regions. Among the most drought- correlation between poverty and natural disaster risks, many stricken districts is Longido in Arusha region, where poverty is of the pockets of poverty in Tanzania are repeatedly subject also estimated to be high.  23 Structural transformation of the economy can help reduce poverty faster, but at the micro level the transformation is slow. Structural transformation, an integral part of the develop- labor in agriculture remains very high, people tend to supply ment process, will determine whether Tanzania can pull more hours to services, followed by industry (Figure ES.25). itself out of poverty. A central aspect of any policy for curbing poverty and reducing the absolute number of poor would be Because of the slow transition of labor, the enormous creation of enough productive jobs to absorb the emerging potential of structural change for productivity growth young generation of workers. The need for more—and more has yet to be realized. There are sizable productivity gaps productive—jobs is even more pressing in Tanzania given ram- between economic sectors. Compared to agriculture, in pant demographic expansion and the fast-growing workforce. services productivity measured by value of output per worker Structural transformation, which is at the core of this process, is is an estimated 9.5 times higher and in industry it is 5.6 times a dynamic process that refers to both reallocation of labor from higher (Figure ES.26). When productivity is measured by value less to more productive sectors, and the productivity and job of output per hour worked, compared to an hour worked in gains associated with this move. The larger the productivity gap agriculture, one hour worked in services is estimated to be between sectors—especially between agriculture and manu- on average 3.2 times as productive, and one hour worked in facturing and services—the greater the opportunity to achieve large gains as labor shifts across sectors. FIGURE ES.24: Contribution of Economic Sectors to GDP, 1998–2017, Percent National accounts show that agriculture is contributing 60 far less to value-added than services and industry. In 1998, agriculture accounted for 40 percent of GDP; in 2017, it accounted for 28 percent. Meanwhile, industry accounted for 29 percent and services for 43 percent. Although ser- 40 vices account for the largest share of GDP, in the past two decades its contribution to the economy went up by just 3 pp. Since 2016 industry has grown much faster; since 2000 its contribution to GDP has gone up by more than 50 percent 20 (Figure ES.24). While micro data show similar trends, people are shifting more slowly from agriculture to services and indus- try. Integrated Labor Force Surveys (ILFS) show that between 2006 and 2014, employment in agriculture fell from 76 to 0 67 percent. Labor appears to be slowly shifting to services, 1998 2001 2004 2007 2010 2013 2016 where employment went up by 9 pp, and to a lesser extent industry, up by 1.2 pp. HBS 2012 and 2018 indicate similar Agriculture trends: a decline of employment in agriculture from 75 to 58 Industry Services percent and increase of employment in services and industry by, respectively, 12 pp and 4 pp. But while the participation of Source: National Accounts, 2019. 24 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE ES.25: Working Hours, Primary and Second Jobs, FIGURE ES.27: Productivity Differences Across Sectors, 2010–15, Percent Per Hour Worked, 2015 100 Sector output ratio (output per hour worked) 4 34 34 28 80 45 44 42 42 9 64 66 3.2 60 8 8 11 9 11 9 40 3 58 58 63 12 9 20 44 46 47 49 2.4 24 24 0 2011 2013 2015 2011 2013 2015 2011 2013 2015 2 Working hours Working hours Working hours all jobs primary jobs secondary jobs Agriculture Industry Services 1.0 1 Source: NPS 2010/11, 2012/13, and 2014/15. 0 Agriculture Industry Services FIGURE ES.26: Productivity Differences Across Sectors, Source: NPS 2014/15. Per Worker, 2015 industry about 2.4 times (Figure ES.27). Clearly, if there was a 10 9.5 faster transition of labor away from agriculture, the productiv- Sector output ratio (output per worker) ity gains would be enormous. However, the transition of labor 8 is low because of both worker capacity and skills limitations and the limited capacity of service and industrial firms to 5.6 absorb the large and fast-growing workforce. 6 In a more granular disaggregation, mining, transport, 4 and trade are the most productive sectors. On average, a mining worker is 8 times more productive than an agricul- 2 tural worker, a manufacturing worker 4.8 times, and a worker 1.0 in construction or utilities 5.4 times. When productivity is measured in terms of output per hour worked, the gaps are 0 Agriculture Industry Services smaller and less varied: for example, one hour of mining is 3.1 times more productive than one hour of farming, and Source: NPS 2014/15. manufacturing is just 1.7 times more productive.  25 Industry consists primarily of micro and informal enterprises, operating in precarious sectors and relying on unskilled labor. About 96 percent of Tanzanian firms have fewer than wearing apparel, and leather (30 percent); and furniture 10 workers and only 1 percent have more than 50. Among (14 percent). Only 1 percent are in high-value-added and the smallest firms, 60 percent have only one or two workers. knowledge-intensive industries. The rest operate essentially Because nearly half of the firms are not registered anywhere, in services, especially nonmarket services. Less than 1 percent they are considered informal. Tanzanian firms tend also to are in agriculture; most people working in that sector run their be fairly young, with a median age of four years. In general, own farms without creating a business. particularly in wholesale and retail trade and manufacturing, micro and small firms are likely to be informal and younger. Less than 30 percent of the workers in industry are skilled. Their small startup capital is financed essentially from per- About 13 percent of industrial firms cite shortage of qualified sonal income. Meanwhile, large businesses tend to be older labor as their main problem; for medium and large firms, the and mainly engaged in formal nonmarket services such as proportion rises to 33 percent. This suggests that as firms grow, education or health (many are State owned); it may be that they engage in more sophisticated production that requires formal firms operating in public services have higher chances higher-skilled workers, who are rare in the local market. of surviving and growing. This shortage of qualified labor severely undermines the performance of large industries, particularly mining, utilities, Two-thirds of Tanzanian firms are in manufacturing and more technically advanced machinery, electric equipment, and trade. About 35 percent are in manufacturing and 34 and medical and pharmaceutical industries. However, nearly percent in wholesale and retail trade. Manufacturing firms 40 percent of large manufacturers of furniture also consider the primarily produce food and beverages (39 percent); textiles, shortage of qualified labor to be a major concern. 26 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T As with poverty, the distribution of firms and employment nationally is uneven. There are pronounced disparities in where firms are The more businesses in a district or region, of any type, the located. The highest number are in the eastern zone, less poverty there is. Although micro and small firms are 27 ­percent, and the lake zone, 17 percent. In contrast, known to not pay well, their presence nevertheless contrib- the southern and western zones each have no more than utes considerably to improving living standards and reducing 5 ­ percent of Tanzanian businesses. Regional disparities are poverty. However, poverty declines more when there are even more pronounced; nearly 20 percent of firms are based more medium and large businesses (Figures ES.28 and ES.29). in Dar es Salaam. The results are supported by the 2015–16 DHS findings, which show that in the eastern zone, which has most of the busi- The geographic distribution of firms resonates with the nesses, 7 in 10 people are in the two highest wealth quintiles. distribution of poverty. There seems to be a negative and Conversely, in the southern and western zones, which have statistically robust correlation by region and district between the fewest businesses, around half of the population is the the incidence of poverty and the number of businesses: two lowest wealth quintiles. FIGURE ES.28: Relationship Between Poverty and the Number of Firms 100 Poverty headcount (%) 80 60 40 20 0 0 0.1 0.2 0.3 0.4 0.5 Number of medium and large businesses (thousand) Source: Statistical Business Register 2014/15, HBS 2017/18, and Tanzania Jobs Diagnostic (2017).  27 FIGURE ES.29: Geographic Distribution of Firms Source: Statistical Business Register 2014/15, HBS 2017/18, and Tanzania Jobs Diagnostic (2017). 28 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Although small firms dominate the economy and provide most of the jobs, medium and large firms could create jobs fast. Despite the predominance of micro firms in the economy, In industry large firms contribute the highest propor- they account for just 24 percent of employment; large tion of value-added. In 2013, total industrial value-added firms account for 28 percent. About half of Tanzanian jobs was TZS 8,220,560 million, 84 percent of it from large firms (49 percent) are in medium and large firms, which account for (­Figure ES.31). Value-added per worker in large firms is about 4 percent of all businesses (Figure ES.30). Thus, even ­ double that of medium firms and about 20 times higher than though fewer than 0.5 percent of all firms employ more than in micro-firms. Apparently, over time firms that manage to 100 workers, they account for about 28 percent of all employ- grow become markedly more productive. Yet they account ment—their average number of workers is over 370, compared for only 0.5 percent of industrial firms; that may explain why to no more than 2 in micro firms and 9 in small ones. New firms industry in general is persistently unproductive. in business for less than 5 years account for less than 25 percent of jobs, compared to 36 percent in firms that have lasted more Net job creation is mostly attributable to larger, older than 20 years. Though only 50 percent of firms are formal, they firms. In 2010–13, small firms created 34 percent of new jobs account for over 80 percent of jobs. In general, firms tend to and medium and large firms about 67 percent. However, employ twice as many men as women; the gender discrepancy micro firms lost jobs; they were down by about 1 percent. is slightly higher in small and young firms. The average number Firms in business less than 5 years created 8 percent of new of men is two times higher than that of women in these firms, jobs. Firms more than 15 years old accounted for more than while it is 1.6 times higher in larger and older firms. 50 percent of the jobs created. FIGURE ES.30: Employment by Gender and Firms Size FIGURE ES.31: Value Added by Firm Size, Billion TZS and Percent 40 400 Average number of workers 223 Contribution to total 3% 320 30 300 employment (%) 4% 20 200 731 9% 10 100 0 0 Micro Small Medium Large Percent of employment Micro Average number of workers 6,948 Small 84% Medium Average number of women Large Average number of men Source: Census of Industrial Production 2013. Source: Census of Industrial Production 2013.  29 Small firms are hemmed in by the economic environment support to the private sector, and insufficient demand and and the lack of support for entrepreneurs, large firms by production capacity as major concerns. However, more than factors related to competitiveness. Both small and large 30 percent of medium and large enterprises report inade- firms perceive the high cost of production as the leading quate physical infrastructure, currency fluctuations, unfair barrier to better performance. However, they differ in their competition, and lack of raw materials and qualified labor as perception of the importance of other problems. More their main problems. Inadequate financial services also rank than 30 percent of micro- and small firms cite the uncertain high as a challenge to industry, but the problem affects small economic environment, inadequate technology, lack of firms more than large ones. 30 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Limited access to finance and the need to rely heavily on family networks and informal funding are barriers to emergence of a vibrant private sector. Tanzania’s formal firms perceive access to finance as a household enterprises and household services businesses. major and increasingly important obstacle. In the 2013 The credit comes mostly from micro-credit entities or infor- Enterprise Survey, about 70 percent of respondents identified mal credit channels like SACCOS, private money lenders, access to finance as an important problem for their current or informal microcredit institutions. About 40 percent of the operations, far higher than in 2006, when about 54 percent household enterprises that borrowed used the micro- and of respondents identified access to finance as a moderate or informal credit channel; just 18 percent secured loans from major obstacle. a traditional bank. Most household enterprises are financed from personal savings or income, or from other households as The larger the firm, the less is access to finance a con- gifts or loans. straint. In 2013, only 28 percent of Tanzanian companies with fewer than 10 employees saw access to finance as either not Financial inclusion is only moderate; many Tanzanians an obstacle or only a minor one, but 69 percent of companies lack formal financial relationships. Owning a bank account, with more than 200 employees were not at all concerned having formal savings, and being able to access credit from about it. Similarly, only 23 percent of large companies saw a financial institution are the three main indicators for mea- access to finance as a major or severe obstacle, compared to suring an individual’s access to basic and critical financial 39 percent of micro and 52 percent of small firms. tools. According to the 2017 Findex survey, 47 percent of the population had a formal bank account, just 6 percent had Financial constraints do not seem to be suppressing formal savings, and 5 percent had access to formal credit. the growth of Tanzanian firms. Even though a majority of HBS 2018 found that in the previous 12 months, 12 percent of Tanzania’s firms are perceived as financially constrained, from household members had savings or current accounts and only the results of the 2006 and 2013 Enterprise Surveys, lack of 2 percent took out loans. access to finance does not seem to be a major determinant of growth: as firms grow larger, access to finance tends to be Financial inclusion is heavily determined by socioeconomic less of an obstacle—perhaps because larger firms tend to characteristics: Older, better-off, and educated men are have more access to evolved financial tools, such as overdraft more likely to be financially included than young, uned- facilities and credit lines in formal financial institutions. ucated, and poor women. Education and living standards are thus very significant in explaining financial inclusion, Informal household enterprises have very little access to which suggests that financial literacy matters to the choices loans. Only 10 percent secured a loan to develop or support individuals make with regard to saving income. The better-off business operations. The rate rises to 12 percent for urban individuals are, the more likely they will have a formal account,  31 formal savings, and access to formal credit. According to 2017 a tertiary education had a formal bank account, compared to Findex, only 32 percent of individuals in the lowest wealth only 41 percent of those who had completed no more than quintile had a formal bank account, compared to 63 percent primary education. HBS 2018 found significantly lower rates: of those in the richest quintile. The differences are similar for only 1.4 percent of individuals in the poorest quintile had a formal savings and formal credit. Similarly, the more educa- savings or current account compared to 35 percent of those tion Tanzanians have, the more likely they are to have access in the richest. to major financial tools; for instance, all respondents who had 32 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Implications for Policy. This report provides a comprehensive analysis of poverty creation and reduce poverty and inequality. Besides strategies in Tanzania. The analysis compares data from the 2017/18 to accelerate growth to transform Tanzania into a middle-in- HBS with those of the 2007 and the 2011/12 HBSs. It also come, semi-industrialized economy and to strengthen human incorporates information from such other sources as waves capital, the government has initiated reforms to build up 1 to 4 of the NPS, the 2006 and 2013 Enterprise Surveys, infrastructure, strengthen fiscal management, and improve the 2013 CIP, the 2014/15 SBR, the rebased GDP statistics, the business environment. In recent years the government the 2006 and 2014 ILFSs, the 2015/16 DHS, the 2017 Findex, has also done a great deal to support the poorest Tanzanians: the 2012 Population and Housing Census and satellite data. it has opened access to free primary education; increased Drawing on all these sources gives a truly comprehensive pic- incomes and reduced vulnerability through productive social ture of poverty in Tanzania, addressing issues that could not safety nets; and expanded delivery and coverage of basic be captured from a single source. The rich analysis illustrates social and community services. Although these efforts have the diverse, multisectoral nature of poverty and its dynam- indeed helped to reduce poverty, they need to be intensified ics—useful information for prioritizing poverty reduction if they are to bring about sustained improvements in the lives strategies. The report identifies areas where concerted efforts of Tanzanians. by the government and other stakeholders would yield the highest payoffs for poverty reduction and more sustainable Policy now needs to be directed to ending the vicious and inclusive development. The policy pointers that follow are cycles of unequal opportunity and vulnerability and focused on making growth more inclusive through continued putting in place mutually reinforcing interventions to development of human capital, furthering structural transfor- build capacity and foster better livelihoods. The basic mation, and promoting growth of labor-intensive firms and commitments should be to improved service delivery and creation of more productive jobs. infrastructure for all; expanded employment opportunities and higher productivity; investments in human capital to Tanzania has solid fundamentals for combatting poverty. help people develop the skills they need; and protection With its rich and diverse resources, strategic location, effec- for the most vulnerable. The design of priority interven- tive planning, and political will, Tanzania is well-positioned tions should take into account the specifics of Tanzania’s to use a variety of policy tools to promote productive job poverty.  33 Use a well-targeted life-cycle approach to enrich human capital. Invest in human capital and increase skills to heighten Preventing stunting for new generations is possible productivity and incomes and sustainably reduce pov- with a sound combination of targeted social services erty. Improvement in human capital is among the factors (nutrition, income support), community monitoring, that helped ease poverty. However, the reduction in poverty and parental education. An additional priority for has been quite slow compared to Tanzania’s remarkable early childhood development is advancing govern- economic growth—which was driven by sectors that are not ment efforts related to maternal and infant health. always open to the poor and mainly benefited those with Essential interventions are expanding access to health higher education and more endowments. Building human care; universal provision of safe water and adequate capital is critical to ensure more inclusive growth and faster sanitation; and mainstreaming health and nutrition poverty reduction. Because such investments are structural, interventions. Investments in the supply and quality of they are likely to be long-term interventions. Meeting the both pre-school and basic education will also help to goals will require a four-pronged reform strategy: (1) expand build cognitive skills in early childhood, enhance abil- provision of early childhood development services to build ities and motivation for learning, and sustain learning the foundational capabilities, cognitive and noncognitive, of throughout schooling and beyond. tomorrow’s workers; (2) empower vulnerable families; (3) invest more in education and education quality; and (4) improve • Draft and apply a policy to empower families the capability of those new to the labor market and upgrade that have a large number of dependents and the skills of current workers. The first prong has a long-term strengthen the resilience of vulnerable pop- agenda, but the second through fourth can be accomplished ulations. Poverty is significantly higher among in the short to medium term. The reforms should be part of a households with a large number of children and cohesive and sequenced policy agenda, guided by economic dependents. Improving human capital in these development needs, that addresses current and coming households could eventually lower the number needs for skills. They should be complemented by interven- of children and dependents. The policy should tions to improve the quality and accessibility of primary health incorporate factors that promote the education of care and social services so as to raise labor productivity and girls and the participation of women in the labor empower poor people. Evidence from the report points to the force, both of which have implications for fertility following priorities throughout the life-cycle: and poverty. The PSSN program has had impact in strengthening resiliency and reducing poverty; • By investing in their early years, move children without the program both the proportion and num- to high-development trajectories. Poor children ber of poor people would have been much higher. under 5 suffer from acute malnutrition, manifested However, only 14 percent of poor households by high stunting, particularly in rural areas. Depri- and 9 percent of total households benefit from vations not only in nutrition but also in such basic PSSN. Today, several years since the program was amenities as safe water and sanitation impair their designed, targeting of the cash transfer program current and future learning and development and needs to be revisited, but a mechanism should also will have long-lasting effects on their socioeconomic be put in place to strengthen the resilience of those achievements. While some losses are irreversible, who graduated from the program to ensure that others can be partly mitigated by early stimulation. they do not fall back into poverty. 34 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T • Provide universal access to education beyond the to train mid-level skilled workers for the immediate primary level, particularly for girls. Currently, 18 per- needs of the labor market but might not equip gradu- cent of Tanzanians have no education and of those with ates with a solid foundation of general skills that allow some schooling, 60 percent did not go beyond primary them to adapt as labor market requirements change. school; and the rates for women are 23 and 58 percent. It is also probable that young graduates from the Enrollment in lower secondary and beyond is very low general track who do not enter tertiary education lack generally and strikingly so among poor families. This many job-relevant skills. It is important that all tracks suggests that deficiencies in education will perpetuate provide the right skill-mix and that tracks are perme- over time. Because those with secondary or higher able enough to ensure that graduates have a range education were better able to benefit from the returns of paths for continuing to acquire skills. Improving the of economic growth and escape poverty, a policy that access and relevance of technical and pre-employ- expands access to secondary and higher education ment training is the most direct way to build the skills could ultimately open up job opportunities in more of the current workforce. Better coordination with productive sectors. Policies need to give special atten- private employers is necessary to design market-rel- tion to expanding access to higher education for girls, evant curricula and course offerings and to provide particularly girls in rural areas and in poor families. financial and technical support that better responds to the needs of growing sectors of the economy. It • Build the capacity of current workers and bridge is also important to incorporate gender equity into skills gaps through technical training. In expanding development programs and projects to ensure that access to higher education, it is essential that techni- all citizens have equitable access to human develop- cal and vocational tracks provide graduates with the ment efforts. This could open up job opportunities in solid general skills the labor market demands. Techni- sectors where women do not traditionally work and cal and vocational schooling could also be a fast way help to empower them.  35 Take advantage of the momentum created by investments in basic services to expand service delivery. Investment in social and community services could make more rural households; promote solar energy as these services more accessible in rural areas, which would a source of lighting; expand the water system maximize their impact on poverty reduction. Access to to improve access to safe drinking water and electricity, safe water, sanitation, and roads has helped reduce bring water closer to users to minimize the time poverty, but services in rural areas and for the poor are still and ­distance required to access it; and expand inadequate. Access to electricity is still low, but in rural areas ­sanitation systems. use of solar energy as a source of lighting is expanding. Access to safe water has doubled in urban areas but there • Expand the road network and improve the quality has been limited changes in rural areas, which also suffer from of roads and their links to better link rural farmers minimal change in access to improved sanitation. Most rural to urban markets. Access to roads has been iden- roads are still rudimentary. It is therefore necessary to: tified as crucial if rural incomes are to grow. Better rural roads will reduce the costs of moving agricultural • Increase coverage and access to basic services. products and promote agricultural marketing and For instance, expand the electricity grid to reach commercialization. 36 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Accelerate creation of productive jobs by boosting the benefits of structural transformation and promoting opportunities for small firms to grow. Tanzania’s economy is transforming, but too slowly. • Build the capacity of micro- and small enterprise The decrease in agriculture’s contribution to GDP and to survive and grow in order to support income the shift of labor from agriculture to services and industry growth and creation of better jobs. For some poor suggest that the structure of the economy is being trans- workers, self-employment is the most viable way out formed, but the large productivity gaps between sectors of unemployment and poverty. Targeted interven- suggest that faster transformation would provide signifi- tions to improve returns to self-employment by, e.g., cant productivity gains. Skills shortages are a formidable facilitating access to productive assets, frequent and barrier to transformation, but so are other factors, such sustained coaching, pre- and post-business creation as the small size of Tanzanian firms and the difficulty they advisory services for entrepreneurs, and specific social have in surviving and growing; the prevalence of informal- support measures could help push up their profitabil- ity; the lack of productive capacity; their limited access to ity and economic returns. financial resources and technology; and the general low value-added of small firms. Micro- and small enterprises • Remove barriers to accessing finance. If private need to be e ­ mpowered to further structural transformation businesses are to grow and poor households to and increase creation of productive jobs. Because agricul- generate income to break the cycle of poverty, both ­ ture remains the mainstay of the vast majority of the poor, need external financing. it needs to be made more productive. The following could help address these issues: • Promote financial literacy. Besides expanding access of the poor and microenterprises to credit, inclusion • Boost the productivity of agriculture. Despite may depend on empowering them by educating the continuing transition of agricultural workers to them on the best available financial options, how to services and industry, the livelihoods of a significant save for life events, and how to use insurance or simi- number of Tanzanians still depend on agriculture. lar products to prepare for the unexpected. Among efforts to make agriculture more productive that should continue are supplying farmers with the • Improve the general business environment, which inputs necessary to increase fertilization and improve is crucial to private sector growth and job creation. irrigation and soil management; and providing credit, Since job creation is the most immediate avenue to pro- transport, and marketing facilities. Efforts should also moting pro-poor growth, and the private sector employs be directed to supporting a transition to cash crops most of Tanzania’s workforce, accelerating growth and and commercial agriculture, where productivity and promoting shared prosperity will depend on significant economic returns are higher. improvements in the national business environment.  37 CHAPTER 1 Poverty and Inequality Patterns I.  Progress in Reducing Poverty Tanzania has seen a progressive decline in poverty over 2007–2018. The proportion of Tanzanians living below the national Not only was the proportion of the population living in basic needs poverty line, set at Tanzanian shilling (TZS) poverty reduced, but so was the depth and severity of 49,320 per adult per month based on the 2018 Household poverty. From 2007 to 2018, the depth of poverty (or poverty Budget Survey (HBS), declined from 34.4 percent in 2007 gap) decreased from 10 to 6 percent and the severity of pov- to 26.4 percent in 2018. The reduction was faster between erty was more than halved, from 5 to 2 percent. This implies 2007 and 2012 than thereafter (Figure 1.1). The basic needs that in 2018 poor households would require an average of poverty headcount fell nationwide but most dramatically in TZS 3,058 per adult equivalent per month to escape poverty. rural areas. In the past decade the proportion of Tanzani- The amount averages TZS 3,650 in rural areas and TZS 1,726 ans who are extremely poor and cannot afford to buy basic in urban areas. Furthermore, the severity of poverty index foodstuffs to meet their minimum nutritional requirements of indicates limited inequality in consumption between poor 2,200 kilocalories (Kcal) per adult per day also declined from households. approximately 12 percent to 8 percent (Box 1.1). FIGURE 1.1: Poverty Trends, at the National Poverty Line, 2007–18, Percent A. Basic Needs Poverty Headcount 2007 34.4% 2012 28.2% Urban 20% Urban 2018 15.4% 26.4% Rural Rural 39.1% 33.4% Urban 15.8% Rural 31.3% 40 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 1.1B. Extreme Poverty Headcount FIGURE 1.1C. Poverty Depth and Severity 14 13.3 14 11.7 11.3 11.7 12 12 10.3 9.7 9.7 10 10 8.0 7.9 8 7.4 8 7.4 6.2 6.7 6.2 6.1 6 4.4 6 4.5 5.0 3.9 3.5 4 4 2.8 2.6 2.3 2.7 2.1 2 1.5 1.2 2 0 0 2007 2012 2018 2007 2012 2018 2007 2012 2018 Depth of Poverty Severity of Poverty Tanzania Mainland Rural Urban Tanzania Mainland Rural Urban Sources: HBS 2007, 2011/12 and 2017/18. BOX 1.1 Poverty Measures In Tanzania, poverty is measured by comparing a house- The following four measures are commonly used to assess hold’s consumption per adult equivalent with the national poverty. The basic needs headcount poverty rate (“poverty poverty line using Household Budget Survey (HBS) data. rate” in the text) measures the proportion of the population The consumption aggregate comprises food, including whose monthly spatially- price-adjusted total household food that households produce themselves, and expendi- consumption per adult equivalent is below the basic needs tures on a range of other goods and services (e.g., clothing, poverty line. The extreme headcount poverty rate (“extreme utilities, transportation, communication, health, education). poverty rate”) measures the proportion of the population It excludes rent and other housing-related expenditures and living below the food poverty line. The depth of poverty (or spending on exceptional events (e.g., marriages, funerals) poverty gap) indicates how far, on average, poor households and larger consumer durable items (e.g., cars, TVs). Price are from the poverty line. Capturing the mean consump- deflators are used to adjust consumption per adult equiva- tion shortfall relative to the poverty line across the whole lent for price differences in different locations and over the population, it is measured as the sum of the consumption course of the HBS fieldwork. Poverty lines are based on the deficit from the poverty line for the poor (the nonpoor cost of basic needs: The food poverty line (TZS 33,748 per have a shortfall of zero) divided by the total population. adult per month in the 2018 HBS) is based on the cost of The depth of poverty shows the total resources needed a food basket containing 2,200 calories per adult per day per adult equivalent to eliminate poverty, assuming that all given consumption patterns in a reference population. The poor individuals have exactly the same shortfall between basic needs poverty line (TZS 49,320 per adult per month) their consumption and the poverty line. The severity of adds an allowance for basic nonfood necessities to the food poverty (or squared poverty gap) captures both how far the poverty line. poor are from the poverty line and consumption inequality among the poor. Rural poverty declined steadily over 2007-18, but urban in other urban areas the drop was marginal, from 22.7 to poverty outside Dar es Salaam did not really begin to fall 21.5 ­percent. In contrast, between 2012 and 2018, poverty fell until 2012. The reduction of urban poverty between 2007 to 19.2 ­percent in urban areas outside the metropolitan city, and 2012 was driven entirely by a plunge in the proportion whereas it increased in Dar es Salaam, to 8 percent, although of poor people in Dar es Salaam, from 14 to 4 percent; this increase is questioned (Box 1.2). C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 41 BOX 1.2 Change in Poverty in Dar es Salaam Is the recent increase in poverty real? Poverty in Dar es Migration patterns do not indicate substantial changes Salaam declined from 14 percent to 4 percent from 2007 that could explain a substantial variation in poverty. to 2012 but seems to have increased to 8 percent in 2018. Nearly 80 percent of heads of households in Dar es Salaam Changes in sampling methods or overestimation of the migrated from other regions—a slight increase from reduction in poverty in 2012 may be behind this increase. 74 percent in 2012. According to the data, the poverty rate The survey-to-survey imputation results do not support increased slightly among migrants from 2012 to 2018 but to these results and indicate a decline in poverty from 14 to a lesser extent than the increase in poverty among nonmi- 8 percent from 2007 to 2012 and stagnation since then. grants. Overall, the poverty rate was lowest in individuals who migrated over the past five years, followed by those Changes in living conditions do not point to an increase in who migrated during the past 5-15 years, and highest in poverty from 2012 to 2018. Access to safe drinking water, those who migrated over 15 years ago or those who did not improved sanitation, efficient cooking fuels, and electricity migrate. increased during this period, although to a lesser extent than in other parts of the country. Human capital outcomes Changes in economic activity and disparity in poverty improved, as did employment in more productive sectors. across the city do not support a potential doubling of pov- All these changes support the fact that poverty may have erty between 2012 and 2018. There are large variations in not increased during this period. poverty across the districts and wards in Dar es Salaam. The least impoverished wards include Kivukoni, Upanga Mashariki, Deterioration of returns to education in poor households Kurasini, Kawe, and Upanga Magharibi, which all have a may have aggravated poverty. With the expansion of edu- poverty rate of less than 1 percent (Figure 1.2).2 These wards cation, opportunities for well paid jobs for individuals with are clustered around the central business district of Dar es primary education and lower decreased, resulting in deterio- Salaam and the vicinity. Among the most impoverished areas ration of returns for poorer households. Although this offset are Chanika, Msongola, Makurumla, Vingunguti, and Kivule, the improvements in living conditions, it cannot explain the with poverty rates greater than 20 percent. These areas observed doubling of poverty from 2012 to 2018. are clustered around the southwestern part of Ilala District. FIGURE 1.2: Poverty and Demographic Dynamics in Dar es Salaam, 2018 A. Poverty Incidence at Ward Level     B. Nighttime Lights in 2018    C. Changes in Nighttime Lights, 2013–18 Sources: World Bank estimates based on 2018 HBS and auxiliary variables. Nighttime lights are based on Visible Infrared Imaging Radiometer Suite (VIIRS) Version 1. Notes: The nighttime light data are all downloaded from: ngdc.noaa.gov. Fig. 1.2 B. shows the sum of monthly nighttime light luminosity for 2018. The nighttime light data for June 2018 are not available and were excluded from the analysis. Fig. 1.2C. shows the differences in the sum of nighttime light luminosity between 2013 and 2018. continued 2 See Appendix G for details on the estimation method of ward-level poverty rates in Dar es Salaam. Numbers should be treated with caution, as they are less precise than district-level poverty estimates (coefficient of variation exceeds 20). 42 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T BOX 1.2 (Continued) Urbanization and growth patterns of Dar es Salaam, as areas, including Gongolamboto, Kinyerezi, Chamazi, Ukonga, measured according to intensity of nighttime light and its and Vijibweni. There are some signs that growth is spreading changes over time, suggest that the poorest areas in Dar outward beyond the city core, although it remains to be seen es Salaam (mainly in the southeastern part of Ilala district) to what extent the agglomeration effects of urbanization have seen only marginal growth in economic activity; growth benefit the poorest within Dar es Salaam, who seem still to be appears to be most pronounced in peri-urban and suburban disconnected from such benefits. Tanzania’s record of poverty reduction is robust and is not Christiaensen et al. (2012). The results support the finding affected by changes in survey methodology. Assessments of declines in poverty and extreme poverty over the past of changes in poverty over time are subject to problems decade. Although the imputed results support the observed with comparability stemming from changes in design and reduction in both rural and urban poverty, they suggest that, methodological improvements in the 2007, 2012, and the in Dar es Salaam, poverty declined at a slower pace (than that 2018 HBS (see Appendix A for details). These challenges observed from original data) between 2007 and 2012, whereas were addressed using survey-to-survey imputation pro- the poverty rate stagnated at approximately 8 percent cedures based on the small-area estimation approach of between 2012 and 2018. Greater access to basic services and to productive assets is the main cause of the reduction in poverty. Tanzanian households have seen a marked increase in households in poorer income groups in both 2007–12 and access to basic services and infrastructure, which has 2012–18 (Figure 1.3). The endowments rose much higher accelerated since 2012. The analysis of the drivers of in the second period, primarily due to improved access to poverty reduction is based on the decomposition described basic services and infrastructure, which accounted for about in Box 1.3. The results indicate that poverty reduction was 60 ­percent of consumption growth. entirely explained by improvements in the endowments of C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 43 BOX 1.3 Decomposition of Poverty ˆi − Q Qθ ˆi' = Q θ { ˆi − Q θ ˆ* + Q θ } { ˆ* − Q θ ˆi' θ } To explore the basic factors behind the decline in poverty, changes in household consumption have been decomposed ( = X −Xi i' ) ˆi + X i ' β βθ ( ˆi − β θ ˆi ' θ ) into (1) improvements in household characteristics or endow- where Q ˆ is the θth unconditional quantile of log real per θ ments, such as more education of the head of the household, adult monthly household consumption, X the vector of ownership of assets, and access to employment opportunities characteristics averages, and β ˆ the estimate of the uncondi- θ and basic services; and (2) changes in the rewards or returns tional quantile partial effect. Superscripts i, i’ and * designate that they get for those characteristics like returns to education, respectively the final year (2018 or 2012), initial year (2012 or assets productivity, and return or profit to business. The two 2007), and counterfactual values. components have themselves been decomposed to identify specific attributes that contribute to changes in consumption, Q ˆi is the counterfactual quantile of the uncon- ˆ* = X i 'β θ and the decomposition has been applied to each decile of ditional counterfactual distribution; it represents the the consumption distribution to understand differences in the distribution of welfare that would have prevailed if the patterns of change for different income groups. relationship between endowments and consumption had remained constant over time. It is used to determine which The approach is based on the Recentered Influence changes in endowments could have helped to reduce pov- Function (RIF) and unconditional quantile regression erty, and how poverty reduction could have changed as a method proposed by Firpo, Fortin, and Lemieux (2009), result of a changing relationship between consumption and in which traditional Oaxaca-Blinder decompositions are endowments. Changes in return to endowments represents applied to the consumption distribution by percentile. This the variation of the conditional correlation between a given makes it possible to assess the amount of poverty reduction endowment and consumption over time. The decomposition attributable to changes in the endowments of households works as follows: and the amount due to changes in the Tanzanian economy and economic returns to people’s endowments: Counterfactual: Change in Poverty if only Change in the Poverty in 2012 relationship btw poverty Poverty in 2018 endowments endowments change and endowments Expansion of access to improved lighting sources fol- improved cooking fuels, safe drinking water, and basic san- lowed by greater access to roads, markets, and public itation went up for both rural and urban poor households, transportation accounted for the increase in household but the increase was faster in urban zones other than Dar es endowments and consumption. The most significant con- Salaam. Access to roads, public transportation, and markets tributor to the increase in access to basic services is the use expanded throughout the country, but especially in rural areas of solar energy, which since 2012 has expanded particularly in and for poorer households. Access to cell phone signals and, rural areas and in poorer households. For instance, between to a lesser extent, health centers and banks also expanded 2012 and 2018, use of solar energy in rural areas soared from among the poor, especially in urban areas other than Dar es less than 2 to over 30 percent. Use of electricity for lighting, Salaam. 44 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 1.3: Drivers of Poverty Reduction in Mainland Tanzania, 2007–18      A. Endowment and Return Effects, 2007–12        B. Endowment and Return Effects, 2012–18 0.4 0.4 Difference in log real per adult consumption Difference in log real per adult consumption 0.2 0.2 0 0 –0.2 –0.2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.5 0.4 0.6 0.7 0.8 0.9 Quantiles Quantiles Con dence interval/endowment effect Con dence interval/endowment effect Con dence interval/returns effect Con dence interval/returns effect Endowment effect Endowment effect Returns effect Returns effect C. Estimates of Endowment and Return Effects, 2007–18 EXTREME POOR POOR MIDDLE CLASS RICHEST 2007–12 2012–18 2007–12 2012–18 2007–12 2012–18 2007–12 2012–18 Total 0.167*** 0.038*** 0.078*** 0.059*** 0.040*** 0.097*** -0.077*** 0.253*** Endowments 0.062** 0.145*** 0.120*** 0.155*** 0.116*** 0.171*** 0.067* 0.263*** Access to basic services 0.040*** 0.108*** 0.080*** 0.115*** 0.067*** 0.095*** 0.055*** 0.02 Education of head 0.00 0.003* 0.004** 0.005*** 0.004** 0.010*** 0.010*** 0.029*** Assets 0.114*** 0.037*** 0.093*** 0.029*** 0.089*** 0.047*** 0.056*** 0.177*** Head nonfarm employment -0.023** 0.005** -0.02* 0.005* -0.01 0.009*** -0.03 0.017*** Demographic structure -0.016*** 0.001* -0.024*** 0.002* -0.025*** 0.003*** -0.016*** 0.004*** Returns 0.105*** -0.108*** -0.041* -0.096*** -0.075*** -0.073*** -0.144*** 0.01 Access to basic services 0.055* -0.141 0.05*** 0.025* -0.177*** 0.018* -0.140*** 0.09 Education head -0.086*** -0.08*** 0.014 -0.093*** 0.012 -0.024 0.006* 0.014*** Assets -0.166*** 0.023 -0.069*** 0.022* -0.156*** 0.055** -0.201*** 0.049** Head nonfarm employment 0.123** 0.005* 0.062** -0.013** -0.011 0.004** 0.012 0.014** Demographic Structure 0.257*** 0.031** -0.014 0.038** 0.115* -0.003 0.292*** -0.02 Sources: HBS 2007, 2011/12, and 2017/18. Note: Extreme poor are population groups in the bottom 10 percent of the distribution; the poor, in the third decile; middle class, in the fifth decile, and the richest, in the top decile. C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 45 Ownership of communication and transportation assets in educational attainment, particularly among the younger also helped reduce poverty. In general, rural and poor generation, but the improvements benefited primarily households added fewer assets than urban and better-off better-off households whose heads are younger. Although ones (Appendix B, figures B.1-B.4). The increase was also primary and lower secondary education increased noticeably slower in 2012–18 than in 2007–12. However, in the second for the poor in urban and rural areas, improvements in upper period ownership of assets like mobile phones, motorcycles secondary and university education were significantly greater and motorbikes rose significantly faster for rural households in for the richest urban households. The size and composition lower-income groups than for the rest of the population; and of households has not changed in recent years, but although these increases were higher compared to the previous period. having many children seems to be a continuing constraint Meanwhile, ownership of more sophisticated assets (e.g., cars, on household well-being, the negative effect seems to have computers, satellite dishes, TV, or fridges) rose faster in urban diminished. These changes are particularly important among and better-off households. poorer households, as is apparent from the positive variations in the returns to demographic structure. This may be the Greater access to basic services and infrastructure has result of the decisions to grant free access to primary educa- helped to heighten the productivity of poorer households. tion and improved health conditions. Greater access to roads and public transportation significantly increased the productivity and economic returns of poorer A shift to more productive employment encouraged households, especially those in secondary urban centers, consumption, but mainly in urban areas and among the followed by those in rural areas. The latter also benefitted richest groups. The shift out of agriculture seems to have from more productive livestock. Investments in road mainte- accelerated since 2012 and was coupled with more jobs and nance, access to local and regional markets, and expansion higher returns to employment, especially in services and to of agricultural production through, for example, irrigation a lesser extent industry. However, these positive changes schemes and credit for small producers and businesses seem were more apparent among moderately poor households to have helped expand household endowments and their and those already better-off. Poor rural households saw only economic returns in some regions, especially Dodoma, Kili- a slight increase in engagement in self-employment and, manjaro, Tanga, Iringa Mbeya, and Kigoma. Although access while remaining positive, their returns declined over recent to microfinance such as Savings and Credit Co-Operative years. Poor urban households outside Dar es Salaam saw a Society (SACCOS) and other informal financial institutions has slight increase in wage employment and in private businesses, not changed much since 2012, these institutions have had an essentially self-employment, coupled with a modest improve- increasingly positive effect on the consumption of urban poor ment in returns. Better-off households in Dar es Salaam saw a households other than in Dar es Salaam and to a lesser extent marked rise in the returns to their education and employment, of rural households. Better hospital services may also have especially among public employees and the self-employed. enhanced the productivity of the poor. These households, which have higher education and endow- ments in productive assets, may have been better positioned Better education and demographic changes have helped to take advantage of the opportunities generated by eco- increase consumption, but only slightly. The expansion of nomic growth. access to education in Tanzania brought about an increase Lack of opportunities for meaningful increases in productivity and economic returns have partly offset efforts to curb poverty. The changing relationship between endowments and con- While the reduction in returns was slower than the increase in sumption was the main drag on consumption growth and endowments—which explains higher consumption and less on the speed of poverty reduction. Except for the highest poverty—it seems to have interfered with efforts to diminish decile, changes in that relationship, holding endowments con- poverty. On average, for 2012–18 the reduction of poverty stant, was negative across the consumption distribution. This would have been about 6 percentage points (pp) higher if suggests a general deterioration in economic returns over the the benefits from greater household endowments had not past decade, which seems to have accelerated since 2012. been partly offset by the deterioration of economic returns. 46 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Only the returns for urban households at the top of the con- As more people have become educated, the correlation sumption distribution seem to have marginally gone up. between educational attainment and poverty has fallen. Expansion of access to education and increases in educa- In 2018, means of basic transportation, ownership of a tional attainment in the general population have occurred mobile phone, and access to markets no longer generated alongside changes in labor market requirements so that the same increase in consumption as before. As more peo- the rewards for fewer years of schooling than a certain level ple have mobile phones and access to mobile signals and the have declined. In particular, the gains in income and con- Internet, the value they add to consumption lessens. Increas- sumption associated with primary education are no longer ingly used for business purposes and fund transfers, mobile as large as they had been. Consequently, households in the phones continue to positively affect the livelihoods of the poorest groups, whose heads are generally older and cannot poor, but their marginal benefits have narrowed since 2012, increase their education, experienced a marked decline in especially in urban areas and in moderately poor households, returns to their years of schooling. This decline, although for whom ownership of these assets has expanded rapidly but apparent in rural and urban areas, was significantly larger opportunities to foster their productive use have not. Similarly, in the latter, especially Dar es Salaam, where educational returns from greater access to markets have been declining, levels in general have increased faster. However, the correla- especially in rural areas. The possession of modern transpor- tion between consumption and postsecondary education tation assets like motorcycles and cars has brought about a increased significantly, again benefitting the richest urban significant decline in returns to bicycles and other basic alter- households. natives, which are still quite prevalent among the poor. C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 47 The slow reduction of poverty resulted in an increase in the number of poor people. Poverty declined more slowly than the population grew, increased during the first period, leading to a slight reduc- so the absolute number of poor Tanzanians went up. tion in the number of poor people from 13.2 to 12.3 million, Poverty declined annually by about 3.6 percent in 2007–12, in 2018 the situation was reversed and the number of poor when population growth averaged 2.8 percent, and by reached about 14 ­ million (Figure 1.4). However, from 2007 to 1.2 percent in 2012–18, when it averaged 3.5 percent.2 2018 the number of Tanzanians who were extremely poor did Although poverty declined slightly more than the population decline, though very slowly, from 4.5 to 4.2 million. FIGURE 1.4: Total Population and Number of Poor People, 2007, 2012 and 2018         A. Population Size (million)          B. Poor Population (million) 70 18 Number of poor people (in miilion) 60 Total population (in million ) 52.7 16 50 13.2 13.9 43.6 12.3 40 38.3 12 35.9 11.2 11.3 10.4 30 31.0 28.4 8 20 16.8 12.6 10 9.9 4 2.6 0 2.0 1.9 2004 2006 2008 2010 2012 2014 2016 2018 2020 0 Tanzania Mainland Urban Rural 2004 2006 2008 2010 2012 2014 2016 2018 2020 Tanzania Mainland Urban Rural C. Distribution of the Total Population by Geographic Area, Percent  D. Distribution of the Poor Population by Geographic Area, Percent 19.0 31.8 Rural Rural Urban Urban 68.2 81.0 Sources: HBS 2007, 2012 and 2018. 2 These figures are based on estimates from HBS 2007 to 2018 and projections from the 2012 census. 48 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Urbanization has been accompanied by a faster by nearly 5 million (72 percent) and the number of poor increase in the urban population, and consequently the by nearly 0.7 million (45 percent). In rural areas, the total number of urban poor, but most of the poor are still population went up by only 26 percent, the number of rural. During 2007-18, the urban population increased poor people was almost unchanged, and the number of by about 7 million (70 percent) and the number of urban extremely poor declined. However, over 80 percent of the poor grew by 0.6 million (34 percent). The increases are poor (11.3 million) continue to live in rural areas, where the result of both urbanization and the dilatory pace of 3.5 million of them suffer from extreme poverty; in urban urban poverty reduction.3 Most of the changes occurred areas, 2.6 million live in poverty and 745,000 in extreme outside Dar es Salaam, where the population has gone up poverty (Figure 1.4). Many Tanzanians are clustered around the poverty line. Given the large proportion of the population at or near the national poverty line. For instance, increasing the daily the poverty line, economic shocks or policy interven- consumption of poor households by just TZS 350 per adult tions could quickly push the poverty rate either up or equivalent would lift about half of the poor out of poverty. down. The curve of average consumption by percentile in Meanwhile, a quite important proportion of Tanzanians figure 1.5 appears relatively flat at the bottom end of the live just above the poverty line, which underscores their distribution, which implies that many people are very near vulnerability. Around 20 percent of the nonpoor have con- sumption levels that are no more than TZS 550 per adult equivalent per day above the poverty line and are at risk of falling back into poverty if they experience unexpected FIGURE 1.5: Consumption Density around the Poverty economic shocks. The pattern in 2012 was similar, but in Line in 2018 (2011 PPP) 2018, concentration around the poverty line seems to have intensified. Numerous initiatives to empower the poor seem to have brought them closer to fulfilling their basic Mean consumption per capita per 25 consumption needs, but as yet there has been no huge Int. poverty level leap out of poverty. day – US$ 2011 PPP 20 Nat. poverty level 15 The concentration of the population around the pov- erty line translates into a significantly higher level of 10 poverty according to international standards. Tanza- nia’s national poverty line is lower than the international 5 line of US$1.90 per person per day in 2011 Purchasing Power Parity (PPP) exchange rate. Using the international 0 line, about 49.1 percent of the population were living 0 10 20 30 40 50 60 70 80 90 100 in poverty in 2018—25.9 million Tanzanians, among the Percentile largest numbers of poor people in Africa (Figure 1.6). The Mean consumption/day/person share of the poor in Tanzania’s population is also signifi- US$ 1.9 (Internat. poverty line) cantly higher than would be expected given the country’s National poverty line per capita Gross Domestic Product (GDP). There is a US$ 3.2 22.7 pp difference between the international and national Source: HBS 2017/18. poverty rates, which represents around 12 million people Notes: Fig. 1.5 plots the average consumption (per capita per day at US$ 2011 PPP) because of the large share of the population clustered for each percentile of the consumption distribution. The national poverty line is converted in per capita per day terms (at US$ 2011 PPP). The national poverty line is around the national poverty line. equivalent to $1.35 PPP per day per capita. 3 The urban population increased by about 70 percent from 2007 to 2018 and the urbanization rate rose from 26 to 32 percent. C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 49 FIGURE 1.6: International Poverty Headcount Ratio at Households close to the poverty line are highly likely to $ 1.9 a day and GDP per capita (2011 PPP) transition in and out of poverty. Based on data from the National Panel Survey (NPS), between 2010 and 2015 only 80 about 16 percent of Tanzanians significantly improved their economic status and moved out of poverty; and 13 percent Poverty headcount ratio at $ 1.9 60 of middle-class households fell back in. Nearly 12 percent a day in 2011 PPP TZA of those at the bottom of the consumption distribution LIC are trapped in chronic poverty. Movements into and out 40 SSA of poverty appear to be higher in rural than in urban areas, suggesting that rural residents are more vulnerable 20 to transitory (as well as chronic) poverty. These dynam- LMIC WLD ics are explored more in depth in the second chapter of UMIC this report. 0 0 20000 40000 60000 GDP per capita (constant 2010 US$) Sources: HBS 2017/18 and WDI 2019. Notes: Fig. 1.6: TZA, LIC, LMIC, UMIC, SSA, WLD stand respectively for Tanzania, Low Income Countries, Lower Middle-Income Countries, Upper Middle-Income Countries, Sub-Saharan Africa and World. 50 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T II.  The Incidence of Growth and Patterns of Inequality The patterns of consumption growth and distribution have changed considerably since 2012. Considering Tanzania’s remarkable economic growth, its than growth in GDP per capita. Using survey-based mean reduction in poverty is very slow. In February 2019, Tanzania consumption to measure growth shows an estimated growth released the revised GDP figures with a base year of 2015. elasticity of poverty of –4.0, which indicates that household From 2007 to 2017 GDP growth averaged 6.3 percent per consumption affected poverty reduction more than GDP year, dropping to 3.3 percent when adjusted by population per capita did. The difference between the estimates of the size. The new series showed that after 2012, GDP growth growth elasticity of poverty found using the different mea- accelerated slightly and was less volatile.4 The previous sures of economic growth is due to the discrepancy between rebasing, released in December 2014 with 2007 as a base the price deflators used to convert nominal GDP and house- year, showed quite similar trends, with GDP growth averaging hold consumption values into real terms. The first measure 6.3 percent per year and GDP per capita growth averaging uses the GDP deflator, which indicates a much slower rate of 3.5 percent for 2008–13. However, the persistence of that inflation than the consumer price index (CPI) or price indices growth rate had only a modest impact on poverty reduction. based on survey unit values.5 This resulted in a significantly The growth elasticity of poverty was estimated at –1.02 for higher growth rate in real GDP per capita than in survey real 2007–12 and dropped (in absolute value) to –0.45 in 2012–17. mean household consumption and thus a lower response of Thus, a 10 percent increase in GDP growth per capita can be poverty to per capita GDP growth. expected to produce a 4.5 percent decrease in the proportion of the poor. This is very low compared to estimates for other This difference resolved during 2012 to 2018, indicat- developing countries, which show poverty can be expected to ing that poverty responded slowly to growth no matter decline by over 20 percent when per capita GDP growth goes how economic growth is defined. During this period, GDP up by 10 percent. and CPI deflators showed relatively similar inflation rates of approximately 38 percent, resulting in more comparable For 2007 to 2012, the pattern of growth in household con- growth rates, whether based on household consumption sumption diverged significantly from GDP growth because per capita (1.5 percent) or GDP per capita (3.3 percent), than of the discrepancy between price deflators. How much during the previous period.6 Consequently, no matter how poverty reduction responds to economic growth depends on growth is measured, the response of poverty to economic whether economic growth is defined based on changes in growth remains very low (–0.73 using survey mean figures and GDP per capita in the national accounts or measured directly –0.45 using GDP figures). from the household surveys on which poverty estimates are based. Economic growth estimated using changes in mean Worsening inequality partly offset the beneficial effects of household consumption per capita calculated from HBS 2007 economic growth on poverty. The slow response of poverty and 2012 was only 0.9 percent annually, significantly lower to growth will be discussed in more detail later in the report, 4 During 2007 to 2012 GDP growth averaged 6 percent with a standard deviation of 1.1, compared with 6.3 percent growth and a standard deviation of 0.9 during 2012 to 2017. 5 From 2007 to 2017, inflation was approximately 70 percent based on the GPD deflator but more than 90 percent based on the CPI or survey unit values. 6 This is only slightly higher than the inflation rate from the survey price deflator, estimated at 35 percent. C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 51 but one important cause may have been deterioration in the Signs of pro-poor growth observed in 2012 seem to have distribution of household consumption from 2012 to 2018. As reversed thereafter. The rate of consumption growth from is apparent from figure 1.7, the poverty headcount decreased 2012 to 2018 was significantly lower for Tanzanians at the more from 2007 to 2012 than in subsequent years, despite a bottom of the consumption distribution than for those who much smaller increase in mean household consumption. From were better off, indicating that poorer people benefitted 2012 to 2018, the increase in inequality (redistribution effect) less from economic growth (Figure 1.8). This pattern is simi- seems to have largely offset the positive effect of growth in lar to that of 2001 to 2007 but contrasts with that of 2007 to household consumption on poverty reduction (growth effect), 2012, when the country’s poorer groups benefited most. The as the deterioration in the Gini coefficient, which after having Appendix C, patterns for rural and urban areas were similar (­ declined in 2012, rose to levels slightly higher than in 2007, figures C.1 to C.4), with on average a larger increase in con- indicates. Throughout the region, inequality remained lower sumption for the urban than the rural poor from 2012 to 2018 than its initial levels in rural areas but increased significantly but also more volatility in consumption changes among the in urban areas, essentially in Dar es Salaam, where the Gini urban poor. coefficient fell from 40 percent in 2007 to 36 percent in 2012 and then grew to 43 percent in 2018. FIGURE 1.7: Consumption Growth and Inequality 2007-2018 A. Growth and Redistribution Effects on Poverty Reduction in B. Redistribution Effects on Poverty Reduction 2012–2018 2007-2012 (in percentage points) (in percentage points) 0 30 26.1 –1 20 –2 10 –3 –2.5 0 –4 3.7 –1.8 –10 –5 –6 –20 –21.1 –6.2 –7 –30 Change in poverty headcount Growth Redistribution Change in poverty headcount Growth Redistribution C. Gini Coefficients 2007–2018 45 40.3 42.2 38.5 39.1 39.5 40 35.5 35.8 35 33.5 29.9 30 25 20 15 10 5 0 2007 2012 2018 Tanzania Mainland Rural Urban Sources: HBS 2007, 2011/12 and 2017/18. Note: The contribution of growth and redistribution to poverty reduction is based on the decomposition method of Datt and Ravallion (1992). 52 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 1.8: Growth Incidence Curves, Percent A. 2007– 2012 B. 2012–2018 8 8 6 6 4 4 2 2 0 0 –2 –2 0 20 40 60 80 100 0 20 40 60 80 100 Growth rate by percentile Growth rate in mean Growth rate by percentile Growth rate in mean Sources: HBS 2007, 2011/12 and 2017/18. C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 53 III.  The Structure of Inequality Widening gaps between groups in education and employment were the primary cause of the increase in inequality. Over time inequality between households sorted accord- TABLE 1.1: Decomposition of Inequality by Household ing to the educational attainment of their head rose to Attributes more than one-fifth of total inequality.7 The shares of SHARE OF INEQUALITY EXPLAINED BY (%) inequality explained by the differences between population 2007 2012 2018 subgroups according to individual household attributes are THEIL-L THEIL-T THEIL-L THEIL-T THEIL-L THEIL-T summarized in Table 1.1.8 Households whose head had com- Education of head 14.70*** 15.40*** 20.80*** 21.10*** 22.60*** 21.60*** pleted post-secondary education were best able to benefit (0.010) (0.012) (0.015) (0.014) (0.016) (0.016) from opportunities generated by economic growth. Their Gender of head 0.000 0.000 0.001 0.001 0.04 0.03 (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) mean real consumption increased significantly faster (about Age of head 1.19*** 1.04*** 1.32*** 1.08*** 0.50 0.39 27 percent) during 2012–18 than that of households whose (0.003) (0.003) (0.003) (0.003) (0.003) (0.002) head completed only lower secondary school (16 percent). Empl. sector of head 12.60*** 12.10*** 13.70*** 12.60*** 20.90*** 17.70*** Households whose head had primary education or less (0.010) (0.010) (0.010) (0.009) (0.017) (0.01) experienced only a minimal increase in their consumption, Family type 10.50*** 11.20*** 10.60*** 10.30*** 11.70*** 11.40*** supporting the argument that labor market requirements have (0.008) (0.009) (0.008) (0.009) (0.014) (0.014) been changing. More productive jobs and more sophisticated Urban/rural status 8.69*** 8.27*** 19.10*** 17.40*** 16.70*** 14.60*** occupations (e.g., in public administration and manufacturing, (0.007) (0.007) (0.012) (0.012) (0.015) (0.014) and senior and professional occupations) are becoming open Regional location 11.50*** 10.50*** 18.40*** 16.60*** 16.20*** 14.0*** (0.010) (0.009) (0.011) (0.011) (0.015) (0.014) only to households with higher education attainment, while Sources: HBS 2007, 2011/12 and 2017/18. those with education levels lower than secondary school are Note: Significance: * At the 10 percent level; ** at the 5 percent level; *** at the 1 percent level. Numbers in trapped in low-productivity jobs. parentheses are bootstrap standard deviations based on 100 replications. Because rewards to wage employment and self-­ employment activities increased much faster in industry, jobs experienced only marginal changes in average con- followed by services, than for other work, inequalities sumption. This widened the gaps in consumption between between employment sectors widened. Households run- employment groups and accounted for nearly 20 percent of ning their own economic activity are gradually operating more inequality in 2018, compared to 13 percent and less, earlier. in services and industry and less in agriculture, and the tran- sition seems to have helped push up their economic returns.9 Inequality between geographic regions persisted in In 2012-18 their average consumption went up by over 2018. Consumption gaps between urban and rural areas and 60 ­percent, especially for those in industry and, to a lesser between geographic regions widened substantially in 2012, extent, services. Average consumption of wage employees despite the general decline in inequality up to then, and by in public administration and other services also went up by 2018 it had declined marginally. Differences between urban about 15 percent. Households employed in other sectors and and rural areas and between geographic locations accounted 7 The analysis of the contribution of households’ characteristics to overall inequality is based on the method described in Box D.1 in appendix D. 8 Seven household attributes are considered: the gender, age, educational level, and sector of employment of the household head, and regional location, urban/rural status, and household demographic composition. 9 These changes occurred mainly in households whose head was an entrepreneur working with others. 54 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T for about 16 percent of inequality in the most recent survey, numbers of dependents. The explanatory powers of the gen- compared to around 18 percent in 2012 and about 10 percent der and age of the household head barely exceed 1 percent. in 2007. Total consumption inequality is overwhelmingly a matter of inequality within household groups. The low share of gender Differences in household demographic composition in these decompositions can be explained by the low propor- inequality. account for a relatively important share of total ­ tion of woman-headed households in the sample, less than The share has held steady over the past decade at about 20 percent, and the particular status of women who head their 11 percent, due to the persistent gaps between households own households, who benefit from wide family support. whose members are all over 14 years old, and those with large Differences in endowments are the primary cause of inequality between urban and rural households. Urban households are better off and consume more than Differences in education between the poorest groups their rural counterparts because they have more human according to location also declined, although the differences and physical assets. As shown in figure 1.9, the urban-rural increased for those in lower-middle- and upper-income gap in consumption is lower among households in lower-­ groups (third decile and above). Nevertheless, more-­ income groups than among those in richer groups, but the productive jobs and sectors (e.g., wage jobs, employment in inequality in endowments is larger among poorer households. industry and services) are significantly more available to urban These patterns have persisted for the past 20 years (World than to rural poor households. Access to markets and informal Bank, 2015). financial services increased faster for the rural than the urban poor, reducing the differences between the two areas, but Inequality between urban and rural poor households differences in access to roads, health centers, and secondary declined slightly in 2018, but gaps in access to productive schools persist. Rural-urban inequality in ownership of mod- jobs and basic infrastructure remain large. Efforts to expand ern assets (e.g., mobile phones, cars, refrigerators) declined electricity and improved lighting sources, safe drinking water, significantly for poor households but increased markedly for and improved sanitation to rural poor households have begun households in median and richer groups. to pay off, helping the poorest people access basic services. C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 55 FIGURE 1.9: Determinants of Inequality Between Urban and Rural Areas A. 2012 B. 2018 1.0 1.0 Difference in log real per Difference in log real per adult consumption adult consumption 0.5 0.5 0 0 –0.5 –0.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Quantiles Quantiles CI / endowment CI /returns CI / endowment CI /returns Endowment effect Returns effect Endowment effect Returns effect 2012 2018 POOREST QUINTILE MEDIAN HOUSEHOLD RICHEST QUINTILE POOREST QUINTILE MEDIAN HOUSEHOLD RICHEST QUINTILE Total Consumption Gap 0.289*** 0.470*** 0.560*** 0.238*** 0.443*** 0.603*** Endowments Gap 0.505*** 0.470*** 0.478*** 0.497*** 0.461*** 0.303*** Returns Gap -0.216*** -0.072*** 0.082** -0.259*** -0.018* 0.300*** Sources: HBS 2011/12 and 2017/18. Note: * Significant at the 10 percent level; ** significant at the 5 percent level; *** significant at the 1 percent level. Numbers in parentheses are bootstrap standard deviations based on 100 replications. Inequality in economic returns matters mostly for bet- Some investment programs to enhance the capacity ter-off households. The difference between urban and rural of the poor have more impact in rural areas, others in areas in economic returns to household characteristics does urban areas. Investments in maintenance and construction not seem to be important for poorer households engaged in of schools and roads, irrigation and rainwater harvesting activities that pay only slightly above subsistence level. How- schemes, credit and savings schemes for small producers and ever, moderately poor urban households employed in wage businesses such as SACCOS, and cash transfer and public jobs or working in industry and trade have significantly higher work schemes all improve both the capacity and productivity economic returns to their activities than their rural counter- of the poor, whether rural or urban. But investment in local parts who work in the same sectors. These differences make agricultural production, services to improve livestock pro- it difficult for rural households to catch up with their urban duction and food storage, and contract farming schemes all counterparts and to overcome spatial inequalities. Urban-rural contribute to higher economic returns and profits for rural inequality in economic returns is significantly wider among than urban poor. Yet their positive impact on the endowment households in upper quantiles than poorer ones, indicating or capacity of the rural poor was limited. Surprisingly, while a that even though all urban households continue to have supe- larger proportion of households in rural areas benefit from the rior endowments compared to their rural counterparts, the Productive Social Safety Nets (PSSN) program managed by contribution of differences in returns to households’ attributes Tanzania Social Action Fund (TASAF), more urban poor house- to inequality is gaining importance for most well-off house- holds than rural ones seem to benefit from the program.10 Yet holds. The gap among urban and rural poor households is in general the program has a larger positive impact on the narrowing over time, while it is widening among the better-off, returns and productivity of the rural poor. Other services, such driven mainly by larger differentials in economic returns to as community heath schemes, have more impact on the pro- households attributes for the wealthiest. ductivity of the urban poor, and likewise, investment programs 10 Overall, 11 percent of rural households (13 percent of rural poor households) and 6 percent of urban households (17 percent of urban poor households) benefit from the PSSN program. 56 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T to facilitate access to local and regional markets tend to be employment and profit opportunities are still low for the more prevalent in areas where the urban poor are located and urban, they are expanding faster than in rural areas. Urban contribute more to raising their economic returns. households that were initially better- educated and had more assets than rural households were better positioned to take Efforts to empower the poor have begun to bear fruit and advantage of opportunities to expand their endowments and help narrow the gaps between urban and rural vulnera- leverage their returns. Investment programs and local initia- ble households. However, poor households in rural areas tives to empower the poor may help to close the gaps, but find it difficult to access better job opportunities or obtain more efforts will be necessary to enhance real diversification higher returns for their work and assets. Although productive and the economic transformation of local communities. Characteristics of the poor affect economic mobility across generations. Intergenerational transmission of parental educational Parental economic status seems to severely constrain attainment limits the upward mobility of their children. the employment of their children – intergenerational Tanzanians of less-educated parents are more likely to be less mobility across economic sectors seems very limited. The educated and those of better-educated parents are likely to vast majority of individuals whose father is a farmer are also have more education, suggesting relatively low intergener- farmers – 65 percent of the general population and 82 percent ational education mobility (Figure 1.10A and Appendix D). of the poor. When the father is self-employed, children tend Education mobility is lower among the poor and among also to work on their own farm but around 25 percent (and women; it appears that low human capital perpetuates 15 percent among the poor) are also self-employed in the vulnerability and gender inequality in future generations. nonfarm sector (Figure 1.10B). Employment of fathers in, for Only 7 percent of Tanzanian adults, and less than 3 percent example, trade, transport, and accommodation, generally of the poor, achieve education beyond primary when the increases the chances that their children will be employed in father has no education. This rate drops to less than 2 percent more productive sectors, although more than 20 percent are for daughters of poor mothers who had no education but is engaged in agriculture even though their father works in the nearly 4 percent for boys. Individuals whose father is educated nonfarm sector (Figure 1.10C). beyond primary school have more education; 62 percent in the general population, and 21percent in poor households, have some secondary schooling or more. C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 57 FIGURE 1.10: Intergenerational Mobility among the Total Population and the Poor, 2018, Percent A. Educational level of individuals vs father Total population Poor 100 0.4 0.4 1.0 100 2.7 2.3 0.4 4.8 0.7 8.4 2.7 2.1 5.9 1.5 4.0 12.1 0.3 5.3 1.3 6.3 0.2 11.3 9.5 0.3 90 14.4 7.7 90 14.6 Education level of individuals (%) Education level of the poor (%) 10.5 29.9 80 26.4 80 35.0 25.6 37.7 14.8 24.0 70 70 44.7 49.1 23.0 49.8 60 43.5 60 21.6 20.8 21.1 33.9 22.0 50 18.5 50 19.6 25.4 40 40 21.1 15.6 14.8 13.0 30 19.2 18.2 10.9 30 40.6 46.3 42.3 20 20 34.3 28.9 26.9 27.7 30.3 27.4 10 20.2 20.7 24.0 10 0 0 No Some Completed Some Completed Above No Some Completed Some Completed Above education Primary Primary Secondary Secondary Secondary education Primary Primary Secondary Secondary Secondary Education level of the father (%) Education level of the father (%) No education Some Primary Completed Primary Some Secondary Completed Secondary Above Secondary B. Employment status of individuals vs father Total population Poor 100 100 Employment status individual (%) 90 90 31.0 Employment status poor (%) 80 39.9 80 70 66.0 70 64.7 60.9 60 78.3 76.7 76.3 60 82.5 87.9 86.3 90.4 31.6 50 50 29.0 40 40 8.6 30 6.5 30 18.4 24.9 20 20 36.7 11.7 11.2 13.2 24.5 4.0 28.9 11.6 1.7 10 3.0 4.2 3.0 10 3.6 5.3 4.0 9.8 2.2 0.0 1.5 11.7 7.9 7.5 5.6 7.1 6.8 4.2 2.8 1.8 2.4 2.5 0 0 Employee Self Unpaid Unemployed Retired Never Employee Self Unpaid Unemployed Retired Never Employed Household Worked Employed Household Worked worker worker Employment status of the father (%) Employment status of the father (%) Employee Self employed with others Self employed alone Unpaid family helper/own farm C. Employment status of individuals vs father Total population Poor 100 3.0 6.0 100 1.2 0.0 0.0 0.6 4.5 7.3 9.5 13.0 6.7 90 12.6 90 4.4 4.9 15.8 8.7 5.6 4.3 24.7 18.6 24.0 80 4.0 80 31.7 Industry individual (%) Industry poor (%) 13.6 22.6 12.7 70 50.5 19.7 70 5.7 7.5 60 21.0 60 10.6 23.8 19.1 50 41.3 50 10.6 40 82.5 40 15.2 12.0 9.7 30 65.1 30 63.5 62.8 9.8 53.4 20 38.5 20 34.9 29.1 33.3 10 20.8 10 0 0 Agriculture Industry Trade & Other Public Agriculture Industry Trade & Other Public Transport Services Administration Transport Services Administration Industry father (%) Industry father (%) Agriculture Industry Trade & Transport Other Services Public Administration Source: HBS 2017/18. 58 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T The family background of poor households contributes to Unless action is taken, it is likely that future generations of the the intergenerational persistence of poverty and inequality. poorest Tanzanians will be trapped in persistent poverty. Inequality of consumption or income reflects differences in effort and in circumstances that may be beyond an individual’s Inequality of opportunity is two times higher in urban control. The circumstances may include both family back- areas than in rural ones. This reflects two facts: (1) Family ground, such as parental education and economic status, and background variables have more influence on households such other factors as gender and region of birth. Strategies for and individuals who have more education and are engaged directly equalizing outcomes may come at the cost of weaken- in more diversified occupations and jobs as it is the case ing incentives for individual effort, investment, and innovation. in urban sectors. (2) To the extent that unobserved circum- However, inequality in opportunities due to circumstances stances and institutional measures (e.g., family composition, beyond an individual’s control perpetuates the lack of capa- parental financial situation, supply and quality of schooling, bilities and opportunities for large parts of society, wastes and labor and land market institutions) shape opportunities productive potential, and contributes to institutional frailty. for rural Tanzanians, estimates of inequality of opportunity In Tanzania, individual circumstances explain approximately that do not take these circumstances into account are sig- 20 percent of total inequality in consumption, and family back- nificantly biased downward. This is supported by how little ground explains approximately 16 percent (see Appendix D parental employment and education affect rural consump- for the methodology). Parental education and father’s employ- tion. Although these factors are significant determinants of ment have the most influence on their children’s prospects and inequality compared with the observed circumstances, their opportunities for economic mobility (figures 1.11 and 1.12). role is very weak in rural areas, where more than 83 percent of This is a more sizable share than in other countries in sub-­ the population have their fathers employed in agriculture and Saharan Africa (SSA), where inequality of opportunity is lower.11 nearly 50 percent have parents with no education. FIGURE 1.11: Overall Inequality and Inequality of FIGURE 1.12: Contributions of Individual’s Circumstances Opportunity in Consumption, 2018, Percent to Inequality, 2018, Percent 35 30 30 25 25 20 20 15 15 10 10 5 5 0 0 Tanzania Mainland Urban Rural Tanzania Mainland Urban Rural Overall inequality Opportunity inequality Opportunity share Gender Birth place Family background TANZANIA MAINLAND URBAN RURAL Overall inequality 25.7*** 29.7*** 18.4*** Opportunity inequality 5.5*** 7.3*** 2.1*** Opportunity share 21.4*** 24.6*** 11.3*** Gender 0.7*** 2.2*** 0.3*** Mother education 6.7*** 7.7*** 0.7 Father education 6.0*** 7.6*** 1.9** Mother employment status 1.0 2.2 0.13 Father employment status 1.7 0.4 0.1 Mother industry 1.5 2.7 0.01 Father industry 2.0 0.8 0.01 Mother sector of employment 3.1 4.5 0.02 Father sector of employment 2.6* 3.3 0.4 Birth place 7.9*** 9.6*** 6.0*** Family Background 16.1*** 16.3*** 5.2*** Source: HBS 2017/18. Note: * Significant at the 10 percent level; ** significant at the 5 percent level; *** significant at the 1 percent level. Numbers in parentheses are bootstrap standard deviations based on 100 replications. 11 Inequality of opportunity is 18 percent in Comoros, 12 percent in Ghana, 15 percent in Ivory Coast, and 21 percent in Madagascar. C h a p t e r 1 P O V ERT Y A N D I N E Q U A L I T Y PATTER N S 59 CHAPTER 2 Moving Up and Out of Poverty: From Vicious to Virtuous Cycles I.  Poverty Transitions from 2008 to 2015 The analysis of poverty transitions is based on panel data poverty rate between 2008 and 2012, which is presumably to identify households that remained poor or nonpoor and linked to similar shares of Tanzanians escaping (13 percent) households that moved into or out of poverty. In addition and entering poverty (14 percent) during this period.3 to HBS, Tanzania has a second survey series that collects con- sumption data and is hence suitable for poverty analysis, NPS, The increase that resulted from those who entered poverty with four rounds so far (2008/09, 2010/11, 2012/13, 2014/15). between 2008 and 2012 offset the reduction in poverty NPS is a longitudinal survey (tracking individuals) conducted that resulted from those who exited poverty. Transitions every two years. Unlike the HBS, which only covers mainland conditional on initial poverty status show that more than half Tanzania, NPS is representative of the whole United Republic of the poor in 2008 escaped poverty by 2012 and that one-fifth of Tanzania (including Zanzibar), although it has a smaller sam- of the nonpoor became poor (Table 2.2). Even though the con­ ple size than HBS. Although HBS is used to produce official ditional exit rate from poverty was much higher than the con­ poverty estimates, the panel nature of NPS data makes it a ditional entry rate into poverty, the smaller population size of particularly attractive survey for studying poverty dynamics and the poor than of the nonpoor in 2008 moderated the effect transitions. Differences in the methodology and estimation on overall poverty: 24 percent of the sample was poor in 2008, procedures between HBS and NPS resulted in differences in and 52 percent of this group exited poverty by 2012.4 In isola­ poverty levels and trends. Harmonization of estimation meth- tion, this decreased the poverty rate by approximately 13 per­ ods helped partially address the mismatches, but some differ- centage points (pp). But 76 percent of the sample was nonpoor ences in poverty estimates remain because of differences in in 2008, and 19 percent of them became poor in 2012. survey instruments and other idiosyncrasies that are difficult to adjust.1 Moreover, although the first three rounds of NPS are TABLE 2.1: Transition Matrix, 2008–2012 real panel data, the last round (2014/15) was implemented as ROUND 3 a cross-sectional survey based on a new redrawn sample and NON-POOR POOR was therefore converted into a synthetic panel to estimate Non-poor 61.9 14.1 75.9 poverty dynamics. More technical details about surveys meth- Round 1 Poor 12.5 11.6 24.1 ods and estimation approaches are in Appendix I. 74.4 25.6 100 Source: NPS 2008/09 and 2012/13. High mobility between poverty states was observed Note: The transition matrix compares the poverty status of individuals in the base during 2008 to 2012. The first three rounds of NPS reveal period and the final period. The cells of the matrix indicate the proportion of the population in each poverty state at the base and final period. The diagonal shows that approximately 27 percent of Tanzanians transitioned into the proportion whose poverty status did not change between the two periods, and out of poverty between 2008 and 2012 and that 12 per- and the off-diagonals show the proportion that transitioned into and out of cent remained poor (Table 2.1),2 indicating that poverty in poverty. The bold numbers at the end of each row and column are the poverty and non-poverty rates in each round. Tanzania is largely transient, as opposed to chronic (Box 2.1). In addition to providing a dynamic picture of poverty, these estimates help make sense of the small change in the overall 1 The harmonized methodology shows slightly lower poverty estimates based on NPS than on HBS. When using cross-sectional weights, NPS data show a consistent declin- ing trend in poverty since 2010, whereas when using panel weights, the data show a marginal increase in poverty in 2012/13 of 1 pp, followed by a decline of approximately 3 pp in 2014/15. 2 This underestimates the true amount of movement into and out of poverty over five years, with households (or individuals) who were not poor in NPS1 (2008/09) and NPS3 (2012/13) possibly experiencing multiple episodes of poverty between the two rounds. This is investigated in the following sections. 3 This results in a marginal increase of 1 pp in the overall poverty rate from 2008 to 2012, although this increase is observed only when panel weights are used. When using cross-sectional weights, poverty appears to have marginally declined. 62 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T BOX 2.1 Defining Chronic and Transient Poverty Several approaches to distinguishing chronic from transient One application of the components approach is to think poverty have been proposed. Two of the most commonly about classification of poverty into chronic and transient found are the spells approach (McKay and Lawson 2003) over three rounds of data by using four characterizations and the components approach (Jalan and Ravallion 1998). (adapted from Hulme and Shepherd 2003): The spells approach defines chronic poverty according • Always poor: Consumption expenditures below the pov- to how many times (or how long) a household or individ- erty line in all three NPS rounds. ual has been below the poverty line over a given period of • Usually poor: Average of consumption expenditures over time. For instance, a household could be defined as being the three NPS rounds below poverty line, but household in chronic poverty if it was poor in at least two of the three or individual is not poor in at least one round. rounds of the National Panel Survey (NPS). • Occasionally poor: Average of consumption expendi- The components approach involves estimating the tures over the three NPS rounds of above poverty line, chronic and transient components of some measure of but household or individual is poor in at least one round. permanent welfare. The fluctuating nature of household poverty • Never poor: Consumption expenditures above ­ welfare over time is thought of as containing transient line in all three NPS rounds. and permanent components. Variability in household consumption levels generates the transient component, In this chapter, households and individuals who are always or and the permanent component gives the long-run aver- usually poor are considered to be chronically poor, and those age of consumption. who are occasionally poor are considered to be transiently poor. TABLE 2.2: Transition Matrix: 2012 Poverty Status TABLE 2.3: Poverty Dynamics based on Synthetic Panels, Conditional on 2008 Poverty Status 2010–2015, Percent ROUND 3 JOINT PROBABILITIES CONDITIONAL PROBABILITIES NON-POOR POOR POVERTY STATUS ESTIMATE POVERTY STATUS ESTIMATE Non-poor 81.5 18.5 100 NP / NP 59.6 NP / NP 81.5 Round 1 Poor 51.9 48.1 100 P / NP 16.4 P / NP 60.8 Source: NPS 2008/09 and 2012/13. NP / P 13.5 NP / P 18.5 Note: The transition matrix shows end-year poverty status conditional on initial- P/P 10.5 P/P 39.2 year poverty status. For example, the Figure in the right upper corner indicates that 18.5 percent of those who were non-poor in 2008/09 became poor in 2012/13. Source: Synthetic panel of NPS 2010/11 and 2014/15 using Dang and Lanjouw (2013) methodology. Predictions are obtained using population weights. N=3,529 for NPS 2010/11; The synthetic panel analysis supports the high mobility N=3,066 for NPS 2010/11. The sample includes household heads aged 20 to 75 by the time of NPS 2010/11. Estimations used Model 1, BM calibration (adapted for into and out of poverty during 2010 to 2015 but suggests residual weights) and simulation procedure to address residuals (with 1,000 more poverty exits than entries. Table 2.3 presents the joint simulations in the simulation step), and ρy from the NPS 2008/09 and 2012/13 panels. P, poor; N, nonpoor. and conditional probabilities of poverty transitions between 2010 and 2015 using the synthetic panel data. Overall pov- erty mobility was slightly higher during 2010 to 2015 (30 per- 2010 to 2015, irrespective of whether an actual or synthetic cent) than during 2008 to 2012 (27 percent). The poverty exit panel is used. The higher rate of exit from than entry into pov- rate was also higher during 2010 to 2015 than in the previous erty during 2010 to 2015 led to a decline in poverty of approx- period. Conditional on being poor in 2010, the probability of imately 3 pp. Overall, 16.4 percent of the population escaped transitioning out of poverty in 2015 was 61 percent. This is sig- poverty during 2010 to 2015, and 13.5 percent fell into it. nificantly higher than the transition rate obtained using the actual panel of 2008 to 2012 (52 percent). At the same time, Rural areas account for the majority of the poor and the transitions into poverty conditional on being nonpoor in the large movements into and out of poverty. Movements initial period are similar (~19 percent) for 2008 to 2012 and into and out of poverty are significantly higher in rural C h a p t e r 2 M o v i n g U p a n d Ou t o f P o v e r t y : F r o m V i c i o u s t o V i r t u o u s C yc l e s 63 FIGURE 2.1: Poverty Transitions Across Three NPS FIGURE 2.2: Chronic and Transitory Poverty Over Three Rounds by Location, 2008–2012, Percent NPS Rounds 100 100 90 80 80 44 Percent of population Percent of population 44 53 53 59 70 59 60 78 60 78 95 95 50 10 8 40 11 4 40 32 9 4 28 30 7 9 22 7 7 4 20 6 5 5 20 4 4 7 6 17 24 10 5 6 6 1 19 7 9 7 4 5 0 0 Overall Dar es Other Rural Zanzibar Overall Dar es Other Rural Zanzibar Salaam Urban Salaam Urban PPP PPN PNP NPP PNN NPN NNP NNN Chronic poor Transient poor Never poor Sources: NPS 2008/09, 2010/11 and 2012/13. Source: NPS 2008/09, 2010/12 and 2012/13. Notes: Geographic locations refer to household locations in NPS 2008/09. P, poor; N, nonpoor. For example, NPN means nonpoor in 2008/09, poor in 2010/11, and nonpoor in 2012/13. ­ opulation, areas (47 percent), which host a majority of the p Zanzibar and fewer than 2 percent in Dar es Salaam. Because followed by Zanzibar (34 percent), and lowest in Dar es ­ of the large share of the rural population, 92 percent of those Salaam (5 ­percent) (Figure 2.1).4 Approximately 30 per- who were always poor lived in rural areas. That said, more cent of rural dwellers were poor in at least two of the three than half of those who were never poor also resided in rural rounds between 2008 and 2012, compared with 25 percent in areas. Who are the chronically poor, transiently poor, and never poor? Almost half of the Tanzanian population was transiently Zanzibar, 5 percent in other urban areas, in Dar es Salaam). or chronically poor between 2008 and 2012. Based on the Given that three-quarters of Tanzanians live in rural areas on definitions of poverty in box 2.1, the share in transient pov- the mainland, these areas also host the largest number of erty was higher (28 percent) than the share in chronic poverty chronically and transiently poor individuals. (20 percent), which means that 48 percent of the popula- tion was classified as being in chronic or transient poverty Differences in asset possession contributed the most during 2008 to 2012 (Figure 2.2). This is much higher than the in distinguishing chronically poor from transiently poor cross-sectional snapshot poverty estimates and highlights the households. The asset index is lowest for households in value of having longitudinal data for investigating poverty. chronic poverty, followed by those in transient poverty There are important differences in poverty dynamics accord- (Table 2.4).5 This index is three times as high, on average, for ing to geographic area (Appendix I, Figure I.18). Transient the never poor than for the chronically poor. That said, chron- poverty was highest in mainland rural areas (32 ­ percent vs ically poor households tend to own land for cultivation, and 22 percent in Zanzibar, 17 percent in other urban areas, many own their dwellings, but their inability to escape poverty 4.4 percent in Dar es Salaam). Chronic poverty was also high- may be because of the lower quality or marketability of their est in rural mainland areas (24 percent, vs 19 percent in assets. Chronically poor households are generally engaged in less-productive agricultural activities with limited returns 4 Eight possible combinations over three rounds of data are used (PPP, PPN, PNP, NPP, PNN, NPN, NNP, NNN) (P=poor; N=nonpoor). 5 The asset index is estimated by multiplying an indicator variable of asset ownership (e.g., household owns a refrigerator) by the proportion of households that own the asset and then summing these products for all assets at the household level to generate the share index. The index includes assets such as refrigerator, sewing machine, computer, radio, bicycle, car, cellphone, television, and stove. 64 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Although the average number of income earners is highest TABLE 2.4: Profiles According to Poverty Status in chronically poor households (2.6), the ratio of earners to (1) (2) (3) household members is lowest in these households. On aver- CHRONICALLY TRANSIENTLY NEVER POOR (1) VS (2) (1) VS (3) POOR POOR age, there are 0.41 income earners per household mem- Household ber in chronically poor households, 0.44 in transiently poor Household size, n 6.4 5.4 4.7 *** *** households, and 0.42 in never poor households. The close- Children, n 3.4 2.6 2.0 *** *** ness of the ratios suggests that one critical factor distin- Adults, n 2.8 2.7 2.6 * ** guishing households according to poverty state is job quality Older adults, n 0.3 0.3 0.2 and returns to employment rather than overall employability Income earners, n 2.6 2.4 2.0 ** *** and number of dependents. This is further explored under Rural, % 94.4 88.7 63.5 *** *** the discussion on the main economic activity of the house- Own dwelling hold head. unit, %, % 94.4 89.4 74.3 *** *** Own land for cultivation 91.6 88.4 73.2 *** Chronically poor households have the poorest access Asset index 0.4 0.6 1.2 *** *** to water, sanitation, and hygiene (WASH) infrastructure Piped water, % 15.1 16.3 30.1 *** and other basic services than those who are transiently Electricity, % 0.3 2.8 20.8 *** *** poor or never poor. Access to better WASH facilities and Flush toilet, % 0.5 2.3 8.7 ** *** electricity is limited in Tanzania, especially for the poor; Household head for the chronically poor, access to basic services is virtually Age 47.6 47.2 45.1 *** zero. In the case of WASH services, for instance, 0.5 percent Female, % 26.4 27.2 23.2 of chronically poor households report having a flush No education, % 34.9 32.4 17.6 *** toilet, compared with 2 ­ percent of transiently poor and Primary, % 62.0 62.1 62.9 Secondary, % 0.9 2.5 10.0 ** *** 9 percent of never poor households. Similarly, 15 percent of University, % 0.0 0.1 0.8 *** chronically poor and transiently poor households reported Other, % 2.2 2.9 8.7 *** having access to piped water, which is half the access rate Main household income source, % for the never poor. None of the chronically poor households Food crop sales 68.6 63.7 37.4 *** reported having access to electricity. Transiently poor Cash crop sales 8.9 9.5 7.7 households did not fare much ­ better, with 3 percent having Livestock sales 2.2 3.4 2.9 access to electricity, compared with 20 percent for the Business income 6.2 7.9 22.6 *** never poor. Wages 7.1 8.7 20.9 *** Remittances 2.9 4.5 5.5 *** Educational attainment and source of household income Other 4.1 2.3 3.1 play a crucial role in determining poverty status over Source: NPS 2008/09, 2010/12 and 2012/13. Notes: *** p<0.01, ** p<0.05, * p<0.1. The last two columns provide a test of the time. The chronically poor have lower educational endow- statistical significance of the difference between chronically and transiently poor ments than the never poor. One percent of chronically poor people, and chronically and never poor people, respectively. households are headed by someone with secondary educa- tion, compared with 10 percent of never poor households. and may not have the ability to move to areas with more-­ Similarly, 35 percent of chronically poor household heads productive employment opportunities. In contrast, an import- have no education, compared with 18 percent of never poor ant proportion of transiently poor people live in urban areas household heads. There are no differences in the share of outside Dar es Salaam, where they can work in the nonfarm household heads with primary education between these sector, exploiting more-productive employment opportunities groups, which suggests that access to primary education because they are not limited by being attached to the land. may not be enough to lift people out of poverty. The chron- ically poor also rely more on agricultural sources of income Chronically poor households are less likely to earn an than the never poor, who rely more on business and wage income than those who are transiently poor or have income. Approximately 80 percent of chronically poor house- never been poor. Chronically poor households have on holds report sales of crops or livestock as their main source of average 6.4 members, compared with 5.4 for transiently poor income, compared with 48 percent of never poor households; and 4.7 for never poor households. The average number of 13 percent of chronically poor households report business children, which is 1 more in chronically poor households than income or wages as their main source of income, compared in other households, is the source of most of this difference. with 44 percent of nonpoor households. C h a p t e r 2 M o v i n g U p a n d Ou t o f P o v e r t y : F r o m V i c i o u s t o V i r t u o u s C yc l e s 65 Households for which the sale of agricultural produce was FIGURE 2.3: Poverty Status According to Main Source of the primary income source were more likely to be tran- Household Income, 2008-12, Percent siently or chronically poor. In 2008, more than 54 percent of individuals lived in households in which sale of food crops was 100 the main source of income. Approximately 35 percent of these 90 Percent of population 80 people were transiently poor, and an additional 26 percent 70 were chronically poor (Figure 2.3). Similarly, for the 10 percent 60 of individuals living in households that depended on cash 50 crop income in 2008, there was a high incidence of transient 40 30 and chronic poverty, although fewer were chronically poor. For 20 the 30 percent of people living in households that depended 10 on business or wage income, there was a low incidence of 0 transient or chronic poverty; more than 75 percent were never s s s e es es er le le le om th ag nc sa sa sa O c poor. Only 3.7 percent of individuals lived in households a W op op ck in itt o m ss cr cr st Re in which remittances were the main source of income: Of e od sh ve sin Ca Li Fo Bu these, 22 percent were transiently poor, and 14 percent were Chronic poverty Transitory poverty Never poor ­chronically poor. Source: NPS 2008/09, 2010/12 and 2012/13. 66 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T II.  Correlates and Determinants of Poverty Entry and Exit To examine the correlates and determinants of poverty classified as “remained poor” if they were poor in 2008 and entry and exit between 2008 and 2012, this section 2012 and remained nonpoor if they were not poor in both defines groups according to their poverty transition sta- periods. Those who were poor in 2008 and became not poor tus between two rounds. Instead of classifying households in 2012 “escaped poverty,” and those who were not poor in and individuals according to their poverty status over three 2008 and became poor in 2012 “entered poverty.” rounds—chronically, transiently, and never poor—they are Poverty transitions are highly correlated with the main source of household income and the level of education of household heads. Households that depend on agricultural or livestock income move frequently into and out of poverty. Poverty rates for households relying on agricultural or livestock and FIGURE 2.4: Poverty Transitions According to Main cash crop sales as their main source of income increased more Source of Household Income, 2008–2012, Percent than 2.9 pp between 2008 and 2012. Rates of entry into pov- erty were significantly higher than rates of exit from poverty 100 Percentage of the population 90 for these households than for those relying on other sources 80 of income (Figure 2.4). The difference between entry and exit 70 rates was highest for households that depend on agricultural 60 or livestock sales. Households relying on food crop sales wit- 50 nessed high entry and exit rates, although they accounted 40 for the largest share of those who remained poor through- 30 out 2008 to 2012. In contrast, households that depend on 20 business or wage income were less likely to move into and 10 out of poverty, approximately 1 and 3 pp, respectively. Eighty 0 percent of individuals in households whose main source of –10 income was business or wages remained nonpoor throughout s s s e es es er le le le om th ag nc sa sa sa O 2008 to 2012. nc a W op op k itt oc si m cr cr es st Re d sh ve sin o Ca Li Fo Households that depend on agriculture are least likely to Bu exit poverty and most likely to enter poverty over time. Remained poor Escaped poverty Probit estimates show that households whose main income Entered poverty Remained non-poor % point change in poverty was derived from the sale of cash crops were less likely to enter poverty than households relying on food crop sales Source: NPS 2008/09 and 2012/13. (Figure 2.5). The largest effect can be seen in households Note: The composition of the poverty groups in this Figure focuses on transitions from NPS1 to NPS3 and is different from the composition in Figure 2.3, which with business income as the main source of income; their focuses on poverty status over all three rounds. C h a p t e r 2 M o v i n g U p a n d Ou t o f P o v e r t y : F r o m V i c i o u s t o V i r t u o u s C yc l e s 67 FIGURE 2.5: Marginal Effects Associated with Poverty FIGURE 2.6: Marginal Effects Associated with Poverty Entry, 2008–2012 Exit, 2008–2012 Male Male Married Married Individual Individual Primary edu. Primary edu. Secondary edu. Secondary edu. Other edu. Other edu. Moved distr. Moved distr. Household Household Male Male Primary edu. head head Primary edu. Secondary edu. Secondary edu. Other edu. Other edu. No. of children No. of children Household Household No. of adults No. of adults No. of elders No. of elders Other urban Other urban Rural Rural Zanzibar Zanzibar Cash crops Cash crops Main income Main income Livestock Livestock source source Business Business Wages Wages Remittances Remittances Other Other –20 –10 0 10 20 –40 –30 –20 –10 0 10 20 30 40 Marginal effect (percentage points) Marginal effect (percentage points) Base categories: No education; Did not move in last 5 years; Base categories: No education; Did not move in last 5 years; Household head has no education; Dar es Salaam; Household head has no education; Dar es Salaam; Main source of income is food crop sales. Main source of income is food crop sales. Source: NPS 2008/09 and 2012/13. Source: NPS 2008/09 and 2012/13. Note: Marginal effects (dots) are presented along with their 95 percent confidence Note: Marginal effects (dots) are presented along with their 95 percent confidence intervals (lines). Marginal effects to the left of the dashed vertical line are associated intervals (lines). Marginal effects to the left of the dashed vertical line are associated with lower probabilities of poverty exit or entry, and those to the right are with lower probabilities of poverty exit or entry, and those to the right are associated with higher rates of exit or entry. associated with higher rates of exit or entry. household members were approximately 10 pp less likely to than those with no schooling and those who had not moved. enter poverty than those in food crop or agricultural house- Overall, living in urban settings, including Dar es Salaam, and holds, although households that depend on business income moving districts within the five years before being surveyed may also get stuck in low productivity enterprises. Members were associated with higher probability of transitioning out of households for which the main source of income was from of poverty. Probit estimates show that the probability of exit- businesses were also, on average, less likely to exit poverty ing poverty for those living in Dar es Salaam was significantly than those in households with the sale of food crops as the higher than for those in rural areas and in Zanzibar. Given that main source of income (Figure 2.6). In contrast, households most of the economic growth in Tanzania has been concen- that reported wages as the main source of income were no trated in urban centers, these seem to have provided more more likely to exit poverty than agricultural households, hold- economic opportunities than rural areas. ing all else constant. Lack of education and location in certain areas are Higher education and migrating for better economic strong correlates of poverty entry. Overall, individu- opportunities are strong correlates of poverty exit.6 This als with secondary education entered poverty at a much positive correlation holds at the individual level and for lower rate (4 percent) than those without any schooling the household head (Figure 2.7). Individuals whose house- (25 percent) (Figure 2.8). Probit estimates show that indi- hold heads attained secondary education and individuals viduals with secondary education are 11 pp less likely to who moved districts were 25 pp more likely to exit poverty enter poverty than those with no schooling, while those 6 Useful insights can be drawn from plotting different correlates together and can help uncover their ordering of importance. The major challenge is the potential overlap between groups; those with more education, for instance are more likely to live in Dar es Salaam. See Dang et al. 2017 for details. 68 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 2.7: Poverty Exit and Unconditional Transition FIGURE 2.8: Poverty Entry and Unconditional Transition Rates by Individual and Household Characteristics, Rates by Individual and Household Characteristics, 2008–2012 2008–2012 80 30 70 Percentage 20 Percentage 60 10 50 40 0 H ead ead fem e. d pri no le. Fe ale nd y e . O ry e u. H er u. ea m . D ast r ed . es ye . er am n Se rim o e le H on ry u. ov o ry . ar O Sa ars H ea ea fem ale Ur m n d pr no ale H on ry u. ov o ry u. D last r ed . e y . th la s Fe ale Se rim o e le O ry e u. . ar nd y e . H er u. nz l co ar du H ad edu l e u ar 5 u M ead da edu nz al e u ar 5 u H ead edu Za ra co ar du ba O s Sa ear ba H H d al a d th d ed th ed P N a se m ed ib P N a c a d M ead da ed ed th ed er a a d a th d ib Za ur M Ru m th la H H ad m M m se im e Ur R a e ea d d e c ea Individual Household Household Individual Household Household head head Above average Below average Above average Below average Source: NPS 2008/09, 2012/13. Source: NPS 2008/09, 2012/13. Note: Dashed line represents average poverty exit and poverty entry rate of Note: Dashed line represents average poverty exit and poverty entry rate of 51.9 percent, corresponding to transition matrices presented earlier. Dark green 18.5 percent, corresponding to transition matrices presented earlier. Dark green triangles above the dashed line indicate above-average probability of transition, and triangles above the dashed line indicate above-average probability of transition, and Light green dots below the dashed line indicate below-average probability of transition. Light green dots below the dashed line indicate below-average probability of transition. with primary education were 4 pp less likely to enter individuals living in Dar es Salaam entered poverty, com- poverty on average.7 The ­ protective effect of living in pared with 7 percent in other urban areas and 12 percent a household whose head has ­ secondary schooling is in Zanzibar. In contrast, rural households had a poverty even larger (~15 pp). The correlation between poverty entry rate of 23.5 ­percent—5 pp higher than the overall entry and location is also high; fewer than 1 percent of average poverty entry rate of 18.5 percent. Transiently poor people remain very close to the poverty line. Poor people who remained poor and those who escaped overlap in consumption distributions, and the clustering just poverty had comparable levels of average consumption under the poverty line suggest that those trapped in long- in the initial year, highlighting their closeness to the pov- term poverty remain close to the poverty line and can be erty line and suggesting that a strong downward push can moved out of poverty with the right support. be exerted on poverty with the right policies. Figure 2.9A compares the real monthly consumption expenditure per The prevalence of vulnerability, as the clustering of a large adult equivalent for those who remained poor and those who share of nonpoor population right above the poverty line escaped poverty. The average level of consumption for those demonstrates, indicates the need for safety nets to build who escaped poverty (TZS 22,941) was only slightly higher resilience. This can be seen in the differences in average con- than the average for those who remained poor (TZS 22,270). sumption of various groups between the two periods. Overall, The similar levels in mean consumption, the considerable the difference between the average consumption of those who remained nonpoor and those who entered poverty was 7 Full results can be found in the Appendix I along with corresponding estimates for transitions between NPS1 (2008/09) and NPS2 (2010/11). C h a p t e r 2 M o v i n g U p a n d Ou t o f P o v e r t y : F r o m V i c i o u s t o V i r t u o u s C yc l e s 69 much larger (Figure 2.9B) than the difference between the only a small increase in their average consumption level and mean consumption of those who remained poor and those remained right above the poverty line, remaining at risk of fall- who exited poverty (Figure 2.9A), suggesting the prevalence ing back into poverty, which suggests that measures to pre- of vulnerability among the nonpoor. Those who entered pov- vent people falling into poverty combined with measures to erty in 2012 had an average consumption level just above the improve welfare could help keep millions of Tanzanians out of poverty line in 2008. Those who escaped poverty in 2012 saw poverty. FIGURE 2.9: Real Monthly Consumption Expenditure Per Adult Equivalent According to Poverty State A. Those Who Remained Poor Versus Those Who Exited Poverty   B. Those Who Remained Nonpoor Versus Those Who Entered Poverty Remained in poverty Exited in poverty Remained non-poor Entered poverty Wave 1 to wave 3 Wave 1 to wave 3 Wave 1 to wave 3 Wave 1 to wave 3 Poverty line Poverty line 75,000 75,000 70,000 70,000 65,000 65,000 25,000 25,000 60,000 60,000 Real expenditure in wave 1 Real expenditure in wave 1 Real expenditure in wave 1 Real expenditure in wave 1 55,000 55,000 20,000 20,000 50,000 50,000 45,000 45,000 40,000 40,000 15,000 15,000 35,000 35,000 Poverty Poverty line line 10,000 10,000 25,000 25,000 20,000 20,000 15,000 15,000 5,000 5,000 10,000 10,000 5,000 5,000 0 0 0 0 Average expenditure Average expenditure Source: NPS 2008/09 and 2012/13. Note: The top of each panel indicates the poverty line, and each dot represents the adult equivalent consumption of a panel member who was (A) poor and (B) nonpoor in 2008. 70 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T III.  Poverty Dynamics and Transitions Between Economic Sectors Economic growth over the last decade combined with of household members to household welfare in the latter, rural-to-urban migration has led to changes in the sec- which therefore may also reflect some of the underlying sec- toral composition of Tanzania’s labor force. The three eco- toral shifts and attachment to the labor market within the nomic sectors (industry, services, agriculture) experienced household. positive real GDP growth between 2008 and 2013. Industry grew the fastest (7.8 ­percent per year), followed by services The relationship between employment sector and house- (6.6 ­ percent per year) and agriculture (4.2 ­ percent per year) hold poverty status is explored from various vantage (World Bank 2015). The most recent projections show that points in each subsection below: descriptive statistics industry will keep growing at 11 ­ percent, services at 6 ­ percent, regarding the cross-sectional composition of the household and agriculture at 3.5 ­percent per year over the medium term heads’ occupational status and households’ main source (World Bank 2018). This section examines sectoral mobility of income during 2010 to 2015, analysis of the relationship over time and the effect on poverty of these changes between between the sectoral composition of households and their 2008 and 2015. The analysis focuses on the sector of employ- welfare, decompositions of the overall change in the poverty ment of the household head and changes in the household’s rate according to changes between sectors versus changes primary income source.8 Because of data limitations, only sec- within sectors, and analysis of sectoral transitions and their toral changes from 2010 onward can be presented, whereas implications for the longitudinal welfare of the household. changes in the main source of income can be presented Additional information on the data used and assumptions for all NPS rounds.9 It is easier to include the contributions made are shown in Appendix I. Poverty status and main occupation of household heads. With the decline in the national poverty rate from 2010 to poverty rate declined from 35 ­ percent for agri- percent to 33 ­ 2015, poverty became more concentrated in agricultural cultural households and from 10 ­ percent for percent to 5 ­ households and less concentrated in households in the households engaged in services and stayed at approximately services sector. Overall, approximately two-thirds of house- percent for households working in industry. Because of the 9 ­ holds depended on agriculture in 2015, and one-­ quarter small share of households engaged in industry (7 ­ percent), were engaged in services. In 2015, those living in agricul- 10 those living in these households accounted for only approx- tural households accounted for 88 ­percent of the poor, up imately 3 ­percent of the poor in 2015, and those living in percent in 2010 (Figure 2.10). From 2010 to 2015, the from 82 ­ households working in services accounted for 5 ­ percent. 8 Viewing dynamics in this way may underestimate the true amount of movement into and out of sectors over a period of several years. For example, a household head who was recorded as being occupied in the agricultural sector in consecutive rounds of data could have experienced episodes of formal or informal employment in industry or services over the period. 9 The scale of missing data on the main occupation of the household head in NPS1 is so large as to prevent meaningful analysis. 10 Where possible, each household is defined as being primarily in agriculture, industry, or services in NPS2, 3, and 4. This variable reflects the sector of employment of the household head in his or her primary occupation. Mining, manufacturing, utilities, and construction are grouped into the industry category. The services category comprises retail and trade; transport, storage, and communications; finance; and general services. C h a p t e r 2 M o v i n g U p a n d Ou t o f P o v e r t y : F r o m V i c i o u s t o V i r t u o u s C yc l e s 71 FIGURE 2.10: Poverty Rates and Contribution of FIGURE 2.11: Population Shares and Poverty Households Occupation Sectors to Poverty: 2010, 2012 Contributions According to Main Source of Household and 2015, ­percent Income, 2008–2015, ­percent 100 80 88 68 90 82 84 70 80 60 51 49 70 50 Percent 38 60 40 Percent 50 30 23 24 22 40 35 33 20 1514 10 10 12 33 9 10 4 6 7 8 3 30 22 4 2 3 3 3 3 21 20 0 10 10 15 Population Population Poverty 10 9 97 78 77 Poverty 2 2 5 35 4 share share rate contribution 0 2008/2009 2014/15 te n te te n n io io io ra ra ra ut ut ut Food crop sales Cash crop sales Livestock sales rty rty rty rib rib rib ve ve ve nt nt nt Business income Wages Remittances Po Po Po co co co rty rty rty ve ve ve Source: NPS 2008/09 and 2014/15. Po Po Po 2010/11 2012/13 2013/15 Agriculture Industry Services NEA/Unknown main occupational sector of the household head. Younger household heads are far less likely to be engaged in agricul- Source: NPS 2010/11, 2012/13 and 2014/15. ture than older household heads (Figure 2.12). Of the oldest percent reported agriculture as their main occu- cohort, 69 ­ pational sector, compared with 43 ­ percent of the young- There was a sharp decline in the share of households est cohort. Younger household heads were more likely to be whose primary income was food crop sales, combined engaged in the services sector than older household heads, with an increase in the share of households that primar- although their overall share of employment in services was ily depended on business income or wages.11 In 2008, lower than their share in agriculture. For the youngest cohort, 51 ­percent of the population was in households that earned aged 16 to 24, the share in agriculture was 5 pp higher than their main income from the sale of food crops, and 29 ­ percent percent vs 38 ­ the share in services (43 ­ percent). This contrasts were in households that earned their main income from busi- with a gap of approximately 40 pp or more for the oldest ness income or wages (Figure 2.11). By 2015, the share with three cohorts. income from the sale of food crops declined to 38 ­ percent, and the share with business or wages income increased to Sectoral composition of household heads ­ varied 47 ­percent. These significant compositional shifts induced ­ significantly according to consumption decile. changes in the contributions of these groups to the propor- Approximately 80 ­ percent of households in the b­ ottom tion of poor people. In 2008, more than two-thirds of the percent of the consumption distribution were engaged 40 ­ poor were in households that earned income from food crop primarily in agriculture in 2010, whereas fewer than ­ sales. By 2015, this share had declined to half. In contrast, the percent were primarily engaged in services (Figure 2.13). 10 ­ share among the poor of people living in households that Moving up the consumption distribution, the share of depended on wages rose from 7 ­ percent to 22 ­ percent. The households engaged in agriculture decreases steadily contribution to overall poverty of business income households while the share in the services sector increases. Of the increased but not as sharply as in wage income households. percent of households, fewer than 20 ­ richest 10 ­ percent These households include small businesses that are possibly were primarily engaged in agriculture, and 60 ­ percent were informal and operating in retail and services. primarily working in services. There were also some relatively large changes in the sectoral composition of households The difference in the share of households that depend between 2010 and 2015. on income from food crop sales is related to the age and 11 In the module on finance in the household questionnaire, the responding sample member was asked what the main source of cash income for the household was. Options included food crops, cash crops, livestock and livestock products, business income, wages and casual cash earnings, remittances, and fishing. 72 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 2.12: Share of Household Heads Engaged Primarily in Agriculture or Services by Age Cohorts, 2010 80 69 70 65 66 63 62 60 55 54 Percent of households 50 43 40 38 30 28 30 25 25 26 24 21 20 10 0 16–24 25–29 30–34 35–39 40–44 45–49 50–54 55–60 Age cohorts of household heads in NPS2 2010/11 Share in agriculture Share in services Source: NPS 2010/11. FIGURE 2.13: Sectoral Composition by Consumption Deciles, 2010 100 8 8 6 9 8 6 9 12 12 13 90 5 8 12 8 1 20 1 19 80 2 4 14 27 4 Percent of household heads 70 4 7 40 60 8 60 50 3 85 83 40 79 79 69 71 68 30 53 47 9 20 10 19 0 Poorest 2 3 4 5 6 7 8 9 Richest 2010/11 consumption deciles Agriculture Industry Services Other/Unknown Source: NPS 2010/11. C h a p t e r 2 M o v i n g U p a n d Ou t o f P o v e r t y : F r o m V i c i o u s t o V i r t u o u s C yc l e s 73 Sectoral structure and poverty decompositions. This section investigates the role that shifts within and generate the decomposition (Box 2.2). The aim is to decom- between sectors play in determining changes in the over- pose changes in poverty into an intra-sectoral effect (changes all poverty rate. As shown earlier, the national poverty rate in poverty within each sector, holding the size of the sec- fell by about 3 pp between 2010 and 2015. This time interval is tor unchanged from baseline), and a population shift effect relevant for poverty decompositions as it is the longest period (changes in the distribution of the population across sectors for which we have data for both consumption and occupa- between the baseline and end period). Decomposing poverty tional sector. In addition, the change in the poverty is large changes in this way allows us to determine which of the two enough that it can be meaningfully decomposed. factors was more responsible for driving poverty changes over a given period. The decomposition method most suited to attributing changes in poverty to intra-sectoral versus population Decompositions based on primary occupation sector of shift effects is based on Ravallion and Huppi (1991). This the household head shows that intrasectoral effects rather approach exploits the additive decomposability of the stan- than movements between sectors caused most of the pov- dard Foster-Greer-Thorbecke (FGT) measures of poverty to erty decline between 2010 and 2015. The intrasectoral BOX 2.2 Poverty Decompositions and Sectoral Changes The poverty decompositions used here are based on decomposition is purely a statistical exercise that should be Ravallion and Huppi (1991). The decompositions will reflect used to understand past changes, rather than a tool that can changes in the poverty headcount rate but extend natu- be used to estimate future trends in poverty. rally to the poverty gap and poverty gap squared mea- sures. Poverty at time t is given as Pt . The change in poverty A generalized example adapted from Valderrama and between t and t + 1 is composed of the following effects: Viveros (2014) explains the decomposition as relevant to sectoral shifts in Tanzania. Consider that, in period 1, n the household head is employed in one of the three main Pt + 1 − Pt = ∑α (P s=1 st s,t + 1 ) − Ps,t Intra-sectoral effect sectors—primary, secondary, and tertiary. The poverty rate ­ in each of these sectors is different, with the highest poverty n + ∑P (α i =1 st s,t + 1 ) − α s,t Population shift effect rate in the primary sector and the lowest poverty rate in the tertiary sector. Suppose a group of households shifts from n the primary sector to the tertiary sector. If, after the shift, the + ∑ (P i =1 s,t + 1 )( ) − Ps,t α s,t + 1 − α s,t Interaction effect within-sector poverty rates remain the same, it must be that the national poverty rate went down. This is a pure popula- tion shift effect, with the population shift from the primary where s represents the specific sector, and n is the number to the tertiary sector causing the entire decline in national of sectors. In this case, there are three main sectors: agricul- poverty. Consider now a situation in which the poverty rate ture; industry; and services. A fourth sector and unknown, in the primary sector falls, but no households change sec- may include not economically active, or unemployed. Ps,t is tors.1 Again, the national poverty rate would decrease, but the poverty rate of sector s in period t. as,t is the population in this situation, the intrasectoral effect would have caused share of sector s in period t. the entire decrease. In practice, as shown later, the overall poverty change will be a combination of the population shift ­ Poverty is defined at the household level, so we use the effect and the intrasectoral effect, along with an interaction occupational characteristics of the household head as effect to balance the accounting exercise. the characteristics of the entire household. Finally, the 1 This could also be the case if the poverty rate decreased in one sector, but there was no net mobility between sectors. 74 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 2.14: Sectoral Decomposition of Poverty Change, FIGURE 2.15: Decomposition According to Main Source 2010–2015, ­percent percent of Household Income, 2008–2015, ­ 3 4 2 3 1 2 Change in poverty rate Change in poverty rate 0 1 –1 0 –2 –3 –1 –4 –2 –5 –3 –6 Tanzania Dar es Other Rural Zanzibar –4 Salaam urban Tanzania Dar es Other Rural Salaam urban Overall Intra-sectoral effect Overall Intra-sectoral effect Population-shift effect Interaction effect Population-shift effect Interaction effect Source: NPS 2010/11 and 2014/15. Source: Authors’ calculations based on NPS 2008/09 and 2014/15. effect is far higher in absolute terms than the population-shift Decompositions based on main source of household effect nationally and in each geographic area (Figure 2.14). income show that intrasectoral and population-shift If there had been no mobility between sectors, the national effects accounted for the poverty change between 2008 poverty rate would have fallen approximately 3 pp anyway. and 2015. In contrast to the sectoral decomposition pre- Zanzibar had the greatest reduction in poverty (almost 5 pp). sented earlier, the main driver in the income source decom- The reduction in poverty in mainland rural areas was also sig- position is the population-shift effect (Figure 2.15).12 Shifts nificant. Even where a very small increase in the poverty rate between sectors were significantly poverty reducing in urban was observed, as in Dar es Salaam and other urban areas, the areas other than Dar es Salaam and in rural areas. As shown overall effect of population shifts between sectors was always above, a shift from food crops to wages and business income smaller than the intrasectoral effect. was the primary cause of this population-shift effect in rural areas, although poverty-increasing intrasectoral effects offset the population-shift effect, particularly in rural areas. Sectoral mobility and poverty between 2010 and 2012. Sectoral transitions according to consumption decile between 2010 and 2012. A larger share of rich households show that richer household heads were more likely to switched occupational sectors than of poor households, par- transition between sectors than poorer household heads. ticularly those that started in agriculture. Approximately Exploiting the longitudinal nature of the NPS, the relation- one-fifth of the richest household heads initially employed ship between sectoral mobility and poverty between 2010 in agriculture transitioned into services or industry, although and 2012 is explored. Occupational sector of the house- their initial share of employment in agriculture was minimal hold head could only be tracked in two consecutive rounds (figures 2.13 and 2.16). In the bottom 40 ­ percent, a very small 12 Decompositions based on primary household income allow for a slightly longer period of analysis—starting from 2008 instead of 2010. One drawback is that the national poverty rate was stagnant during 2008 to 2015, with an increase in rural areas offsetting the decrease in Dar es Salaam. Thus, intersectoral and intrasectoral effects generally cancelled each other out at the national level. C h a p t e r 2 M o v i n g U p a n d Ou t o f P o v e r t y : F r o m V i c i o u s t o V i r t u o u s C yc l e s 75 FIGURE 2.16: Share of Household Heads Changing Sectors Between 2010 And 2012 Across The 2010 Consumption Distribution, ­percent 60 50 Percent of household heads 40 30 20 Any change of sector (n = 15,440) 10 Shift out of agriculture (n = 9,906) 0 Poorest 2 3 4 5 6 7 8 9 Richest 2010/11 consumption deciles Source: NPS 2010/11 and 2012/13. Note: The upper line shows the share of the 15,440 household heads that switched sectors regardless of initial sector of employment. The bottom dotted line shows the share of household heads that shifted out of the agricultural sector. proportion of agricultural households transitioned out of Sectoral transitions conditional on the initial sector of agriculture between 2010 and 2012. Fewer than 10 ­ percent employment show that transitions out of agriculture are of household heads in the bottom 40 ­percent transitioned relatively uncommon (Table 2.6). This complements what was ­ between sectors—which is half of the share of the top presented in table 2.5, in which each cell represents the total 20 ­percent. share of households in each category; 90 ­ percent of initially poor and 86 ­ percent of initially nonpoor households stayed All of the household heads that were poor in 2010 started in agriculture over time. The crucial difference between in agriculture and remained there. Three-quarters of initially the groups was the higher proportion of poor than of non- poor households remained engaged in agriculture in 2010 poor households engaged in services that transitioned into and 2012, whereas 8.5 ­ percent moved out of agriculture while agriculture. Approximately 41 ­ percent of poor households 6.5 ­percent moved in (Table 2.5A). Among the initially poor engaged primarily in services in 2010 transitioned into agri- who transitioned out of poverty, about 7.3 ­ percent moved culture by 2012, compared with 15 ­ percent of initially non- in agriculture and 9.4 ­percent moved out by 2012. Almost as poor households. many initially nonpoor households transitioned into agricul- ture as transitioned out (Table 2.5B). However, the propor- Transitions based on household head’s primary income tion of households that were initially nonpoor and remained source also show higher dependence on crop income of in agriculture throughout 2010 to 2012 was significantly lower the poor than of the nonpoor but large transitions into (46 ­percent) than the proportion of households that were ini- food crop, business, or wage income for both groups. tially poor and remained in agriculture during the same period Approximately 41 ­ percent of the initially poor relied mainly on (74 ­percent), whereas the proportion of the initially nonpoor income from food crop sales, compared with 27 ­ percent of the who remained engaged in services in both periods was sig- initially nonpoor (Appendix I, table I.32). Almost 60 ­ percent nificantly larger (20 ­ percent) than that of the initially poor of food crop households continued to earn income from the (4 ­percent). sale of food crops, and 25 ­ percent transitioned to wage or business income, with no significant difference according Agricultural households tend to remain locked in agricul- to initial poverty status (Appendix I, table I.33). Most house- ture over time, regardless of their initial poverty status. holds e ­ arning income from other sources transitioned into 76 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE 2.5: Inter-Temporal Sectoral Composition and Transitions, 2010–2012, Percent A. INITIALLY POOR IN 2010 B. INITIALLY NON-POOR IN 2010 2012/2013 2012/2013 AGRIC. IND. SERV. N/A AGRIC. IND. SERV. N/A Agric. 74.4 2.0 2.3 4.2 Agric. 46.4 1.6 2.9 3.4 Ind. 0.6 0.7 0.0 0.6 Ind. 1.1 3.7 1.8 0.5 2010/11 2010/11 Serv. 3.5 0.5 4.1 0.6 Serv. 4.3 2.2 19.7 2.7 N/A 2.4 0.3 1.0 3.0 N/A 3.8 0.9 2.0 3.1 Source: NPS 2010/11 and 2012/13. TABLE 2.6: Sectoral Transition Matrices Conditional on Initial Sector of Employment, 2010–2012, Percent A. INITIALLY POOR B. INITIALLY NON-POOR 2012/2013 2012/2013 AGRIC. IND. SERV. N/A AGRIC. IND. SERV. N/A Agric. 89.8 2.4 2.8 5.0 100 Agric. 85.6 2.9 5.4 6.2 100 Ind. 31.3 36.1 0.0 32.7 100 Ind. 15.5 52.2 25.5 6.8 100 2010/11 2010/11 Serv. 40.6 6.2 46.7 6.4 100 Serv. 15.0 7.7 68.0 9.4 100 N/A 35.7 3.9 14.7 45.8 100 N/A 39.3 9.2 20.1 31.4 100 food crop sales regardless of initial poverty status, although TABLE 2.7: Sectoral Transitions for Households Between initially nonpoor households reliant on wages or business 2010 and 2012, Percent income were far less likely to transition to other income cate- 2012/13 gories than initially poor households, indicating that the lat- AGRICULTURE INDUSTRY SERVICES NEA/UNKNOWN TOTAL ter may have experienced employment and income volatility. A. Escaped poverty The importance of having a connection to the labor market is Agriculture 70.6 2.5 2.8 4.1 79.9 clear; 18 ­percent of the initially nonpoor had wages or busi- Industry 1.0 1.2 0.0 0.4 2.7 ness income as their main income source in both periods, Services 4.1 0.4 5.0 0.9 10.4 compared with 5 ­ percent of the initially poor. 2010/11 NEA/Unknown 2.2 0.1 1.2 3.5 7.0 B. Entered poverty Most households that escaped poverty remained in the Agriculture 68.1 0.8 3.0 5.3 77.2 Industry 2.3 1.1 1.2 0.0 4.6 same sector but increased their participation in paid Services 3.8 0.0 5.1 1.3 10.2 employment and worked more hours. As stated earlier, 2010/11 NEA/Unknown 4.3 0.2 1.3 2.2 8.0 approximately 90 ­ percent of poor households in 2010 were Source: NPS 2010/11 and 2012/13. engaged primarily in agriculture. The majority of house- holds that escaped poverty by 2012 remained in agriculture. who escaped poverty worked on average longer hours than At the same time, there was a positive net movement out those who entered poverty (Table 2.9), and those who fell of agriculture (Table 2.7). In comparison, the net movement into poverty worked fewer hours than those who remained out of agriculture was negative for households that entered poor between periods. ­ poverty. As for employment status, those who escaped pov- erty increased their participation in paid employment and The locking-in effect of the agricultural sector is corrob- reduced their participation in unpaid work within nonagri- orated in the cross-sectional probit regressions, which cultural employment (Table 2.8). Those who entered poverty indicate that agricultural households were more likely to appear to have significantly increased their participation in be poor in all NPS rounds. Controlling for household char- unpaid agricultural work while reducing their participation acteristics, probit regressions show that agricultural house- in self-employed agricultural work. There was also mobil- holds were between 12 and 17 pp more likely to be in poverty ity between paid and unpaid agricultural work for those who than households engaged in industry or services (Appendix I, escaped and those who entered poverty. In addition, those table I.34). C h a p t e r 2 M o v i n g U p a n d Ou t o f P o v e r t y : F r o m V i c i o u s t o V i r t u o u s C yc l e s 77 TABLE 2.8: Transitions in Employment Type Between 2010 and 2012, Percent 2012/13 2010/11  PAID EMPLOYEE NON-AG SELF EMPLOYED NON-AG UNPAID WORKER AG SELF-EMPLOYED AG UNPAID WORKER UNPAID APPRENTICE TOTAL A. Escaped poverty Paid Employee 1.7 0.6 0.2 1.9 0.9 0.0 5.3 Non-Ag Self Employed 0.7 2.8 0.1 1.3 2.3 0.0 7.0 Non-Ag Unpaid Worker 0.5 0.3 0.0 1.5 5.2 0.2 7.8 Ag Self-Employed 4.1 2.4 0.2 19.6 10.9 0.0 37.2 Ag Unpaid Worker 5.6 3.5 0.8 10.7 21.7 0.3 42.7 Total 12.7 9.6 1.3 35.1 40.9 0.5 100.0 B. Entered poverty Paid Employee 2.5 0.8 0.0 2.2 2.1 0.0 7.6 Non-Ag Self Employed 0.6 2.7 0.6 1.7 2.8 0.0 8.3 Non-Ag Unpaid Worker 0.8 0.9 0.6 0.9 4.0 0.1 7.3 Ag Self-Employed 4.9 2.5 0.3 24.2 11.4 0.0 43.3 Ag Unpaid Worker 3.7 0.8 1.3 7.4 20.3 0.0 33.5 Total 12.4 7.7 2.8 36.4 40.6 0.1 100.0 Sources: NPS 2010/11 and 2012/13. Note: NPS 2010/11 did not solicit information on unpaid apprenticeship, thus the missing category for 2010/11. TABLE 2.9: Poverty Transition and Change in Annual Hours Worked by Sector A. ESCAPED POVERTY B. ENTERED POVERTY SECTOR 2010/11 MEAN HOURS 2012/13 MEAN HOURS CHANGE SECTOR 2010/11 MEAN HOURS 2012/13 MEAN HOURS CHANGE Agriculture and Fishing 533 669 137 Agriculture and Fishing 545 657 111 Mining 530 1449 919 Mining 96 1689 1593 Manufacturing 1123 1684 561 Manufacturing 997 577 -420 Electricity & utilities 1540 1920 380 Electricity & utilities 1103 1506 403 Construction 2086 664 -1422 Construction 1474 2326 853 Commerce 726 1442 716 Commerce 1288 1440 152 Transport, storage 2191 2519 328 Transport, storage 224 224 Finance, insurance 0 2806 2806 Finance, insurance 1305 2037 732 Other Services 1026 1827 801 Other Services Total 575 801 226 Total 608 747 139 Source: NPS 2008/09, 2010/11, 2012/13. Initially poor household heads who changed their physical mobility between districts between NPS rounds. ­ sectoral occupation were 18 pp less likely to be living Although the overall effect is statistically insignificant, it is in poverty in 2012 than those who did not. Exploiting large in absolute terms for initially poor households. On the panel dimension of the data, the marginal effects average, households that relocated, in many cases prob- of a household changing sectors on the probability ably in response to better perceived or actual economic of the household being poor in 2012 were also strong opportunities, were 21 pp less likely to remain in poverty (Appendix I, table I.35). Changing sectors significantly than households that did not relocate between 2010 and increased the probability that a household would exit 2012. Finally, initially nonpoor household heads with some ­ poverty between 2010 and 2012. For the ­ initially nonpoor, education were less likely than those with no education to there was no statistically significant poverty effect asso- be poor. The fact that very few initially poor households ciated with the household head changing his or her main were headed by someone with a higher than primary edu- sector of occupation. Another interesting effect is that of cation was the main reason for this effect. 78 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T IV.  Summary and Policy Action Understanding poverty dynamics and their drivers is cru- percent of chronically poor house- low across the country; 0.5 ­ cial in designing and implementing effective, targeted, percent of transiently poor households, and 9 ­ holds, 2 ­ percent sustainable antipoverty programs and policies. Such pol- of never poor households reported access. icies and programs can reduce the risk of people falling into poverty, end the cycle of long-term poverty, and increase Regression analyses show that poverty exit is largely opportunities for upward mobility. For those living in chronic associated with higher education of the household poverty, a comprehensive set of interventions may be needed head and migration to a new district. The probability of to address systemic challenges preventing them from upward exiting poverty for those living in other urban areas was not mobility. For those who recently fell into poverty, temporary significantly different from the base case of Dar es Salaam, but assistance may be needed. An appropriate policy response those living in rural areas and in Zanzibar were significantly to chronic poverty could focus on increasing the attainment less likely to exit poverty over the period. Households in which of and returns to physical and human capital of the poor, the main source of income was from businesses were 17 pp whereas transient poverty would be better addressed through less likely to exit poverty than households with the sale of initiatives that focus on insurance and income stabilization food crops as their main income source. In contrast, holding (Lipton and Ravallion 1995). all else constant, households that reported wages as their main source of income were no more likely to exit poverty Panel data analysis reveals important mobility between than agricultural households. poverty states in Tanzania, largely driven by transitions in rural areas. Almost 88 ­percent of the transiently poor and Poverty changed primarily because of within-sector mobil- 93 ­percent of the chronically poor resided in rural areas in ity, as opposed to between-sector mobility. The Ravallion- 2012. One-fifth of panel members were estimated to be in Huppi sectoral decomposition of poverty suggests that, chronic poverty, 29 ­ percent were in transient poverty, and even if there had there been no mobility between eco- 53 ­percent were never poor over three consecutive survey nomic sectors, the national poverty rate would have fallen rounds between 2008 and 2012. Households are defined as approximately 3 pp between 2010 and 2015. Changes in the being in chronic poverty when their average level of con- composition of employment within the agricultural sector sumption expenditure was below the national poverty line in explain almost all of the poverty dynamics. Those who exited all three rounds. poverty participated more in paid employment and worked more hours, whereas those who entered into poverty shifted The chronically poor resemble the transiently poor except toward unpaid agricultural work rather than agricultural for having slightly larger families, a lower ratio of income self-employment. earners to family size, fewer assets, and less access to basic services. Furthermore, the average per-adult-equivalent The key to sustainable poverty and vulnerability reduction consumption of those who remained poor between 2008 and in Tanzania is addressing rural poverty and focusing on the 2012 is similar to that of those who exited poverty, suggest- welfare of the population clustered around the poverty ing opportunities for upward mobility, although chronically, line. Large sections of the population in Tanzania are clus- transiently, and never poor households had very large differ- tered around the poverty line, offering hope that many can be ences in access to WASH and basic services. Approximately lifted out of poverty with the right policy prescriptions. With 15 ­percent of the chronically and transiently poor had access most transiently poor living in rural areas and household wel- to piped water, which is half of the never poor. Slightly more fare primarily based on agricultural output, household wel- than one-fifth of never poor households had access to elec- fare can be prone to volatility stemming from weather-related tricity in the main dwelling, compared with virtually none of shocks and shocks to agricultural input and output markets. the chronically poor. Access to a flush toilet was generally very This would suggest two policy prescriptions for those clus- tered around the poverty line. C h a p t e r 2 M o v i n g U p a n d Ou t o f P o v e r t y : F r o m V i c i o u s t o V i r t u o u s C yc l e s 79 First, safety nets and agricultural insurance could prevent There is also hope for lifting Tanzanians out of a vicious millions of Tanzanians from falling into poverty. Because cycle of poverty. There were few differences in the charac- agriculture is the largest source of employment in Tanzania, teristics of the chronically poor and the transiently poor. The providing agricultural insurance could help Tanzania by pro- critical differences were low levels of educational attainment, tecting farmers against loss of or damage to crops and live- physical endowments, and access to services. Improving the stock. Social safety nets targeted to vulnerable people and physical and human capital endowments of the poor can linked to employment could help them avoid unpaid work push them out of the poverty trap and help break the vicious and offer them more hours of paid employment. cycle of poverty. This requires not only helping the poor attain more education, but also facilitating their school-to-work tran- Second, more economic opportunities and greater mobil- sitions such that they can access more-productive employ- ity could help many poor Tanzania move out of poverty. ment opportunities. In addition, increasing access to basic Increasing opportunities for more stable employment oppor- infrastructure and universal services, especially in rural areas, tunities in agriculture and increasing mobility to urban areas can improve the welfare of poor people who do not have the through improvements in infrastructure could help increase resources to pay for these services. opportunities for employment outside agriculture. CHAPTER 3 Profile of the Poor I. Sociodemographic Characteristics of the Poor Heads of poor households tend to be independent farmers who are older and less educated. Poverty in Tanzania is overwhelmingly rural, and location The more children and other dependents a household has, has significant effects on consumption. About 33 percent of the poorer it is. Poverty is significantly correlated with more the poor are concentrated in the lake zone, which is rural and children younger than 15 in the household.1 Poor ­households where less-productive and subsistence activities are common. tend to have nearly twice as many children as nonpoor In the lake zone, 4.6 million live in poverty and 1.3 million in households and thus have higher dependency rates than the extreme poverty; in the northern and eastern zones, fewer national average (Table 3.1). Approximately 44 percent of than 1.4 million live in poverty and 420,000 in extreme poverty. households with five or more children younger than 15 are TABLE 3.1: Sociodemographic Characteristics of Tanzanian Households, 2018 AREA POVERTY TANZANIA RURAL URBAN NONPOOR POOR RURAL POOR URBAN POOR EXTREME POOR Household size, % 4.6 4.9 4.2 4.3 6.1 6.2 5.6 6.6   <15 years 2.0 2.3 1.5 1.7 3.0 3.2 2.4 3.4   15–64 years 2.4 2.4 2.5 2.3 2.9 2.8 3.0 3.0   >64 years 0.2 0.2 0.1 0.2 0.2 0.3 0.2 0.2 Dependency ratio 1.03 1.17 0.77 0.94 1.36 1.45 1.04 1.44 Age of household head 46.5 47.6 44.5 46.1 48.1 47.8 48.1 47.8 Gender of household head, %  Men 71.8 73.2 69.1 71.6 72.3 77.6 71.0 74.7  Women 28.2 26.8 30.9 28.4 27.7 22.4 29.0 25.3 Highest level of education completed by household head, %   No education 19.9 26.0 8.7 17.8 28.5 34.2 19.2 29.4   Less than completed primary 13.9 17.1 8.0 12.8 18.5 18.7 15.6 19.5   Completed primary 49.6 48.1 52.3 49.6 49.6 45.9 59.1 49.2   Lower secondary 11.4 6.6 20.2 13.6 2.5 1.2 5.2 1.7   Upper secondary 3.4 1.6 6.6 4.0 0.8 0.1 0.6 0.0  University 1.8 0.5 4.2 2.2 0.1 0.0 0.4 0.1 Source: HBS 2017/18. younger than 15 and 65 and older) to the nondependent population (15–64 years old). Note: The dependency ratio is calculated as the ratio of dependents (­ 1 Appendix E describes regression analyses used to examine the main factors affecting household consumption and poverty status. Although the direction of causality is sometimes difficult to establish, the results allow for identifying variables closely related to levels of consumption and the likelihood of poverty. Results are in tables E.1 and E.2. 82 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T poor, four times as high as the poverty rate for h ­ ouseholds FIGURE 3.2: Poverty Headcount According to Gender of with no children (estimated at 10.6 percent). It is also Household Head, 2018, Percent percentage points (pp) higher than the national average 18 ­ 40 poverty rate and 28 pp higher than the poverty rate for house- holds with one or two children (Figure 3.1). The interaction 31.331.5 31.3 30 27.4 27.3 between family size and poverty is bidirectional; the more chil- 26.1 26.6 26.2 22.6 dren and dependents in a household, the less the household 20.3 20.7 can cover basic consumption needs and move out of poverty, 20 15.3 14.0 but poor households tend to have more children to compen- 9.2 sate for their inability to invest in other forms of human capital 10 and as insurance against infant mortality, trapping them in a vicious circle of poverty. 0 Tanzania Rural Urban Single Married Divorced Widowed Household size and number of dependents are in part Area Marital Status functions of the rural-urban poverty split. On average, rural Male Female families have 2.3 children younger than 15, compared with Source: HBS 2017/18. 1.3 in Dar es Salaam and 1.6 in other urban centers (Table 3.1). Thus, the dependency ratio of rural households is 50 percent higher than that of urban households; poor and rural house- in urban areas and in divorced households, where it exceeds holds must support more unproductive members than urban 22 pp. Urban widows are also approximately 14 pp poorer households. than urban widowers. Ownership of assets, especially trans- portation and communication equipment, is also significantly Poverty seems to be more prevalent among women. lower in women-headed households, which illustrates the It is difficult to study poverty according to gender because limited access of women to productive assets. household surveys assume equal distribution of consumption between members of a household and because of the status How much schooling the head has is closely related to the of women who head households, but there are indications incidence of poverty. This suggests that education is closely that poverty is more prevalent among women, particularly in linked to income generation. Households whose head has urban areas, where the poverty rate reaches 20.3 percent of completed lower secondary or more are less likely to be poor. women-headed households, compared with 14 percent of The poverty rate for households whose heads have no edu- male-headed households (Figure 3.2). Some types of wom- cation or did not complete primary school is approximately en-headed households are particularly vulnerable to poverty; 35 percent, compared with 26 percent for those whose head single and divorced women are more than 11 pp poorer than completed primary school and 6 percent for those whose men. The gap is high in rural and urban areas but particularly head has lower secondary education and above (Figure 3.3). Poor households tend to have less-educated heads and members, especially women-headed households. Approx- FIGURE 3.1: Poverty Headcount According Number of Children, 2018 , Percent imately 29 percent of heads of poor households have no education, and 19 percent did not complete primary school; 50 for women-headed households, these rates are 45 percent 44.2 and 14 percent, respectively (Table 3.1). Only 3.4 percent of 40 heads of poor households (2.2 percent of women-headed 30.7 ones) have more than a primary education, compared with 30 20 percent of nonpoor households. Similarly, only a few poor household members have completed more than primary 20 16.5 school; 43 percent have less than primary education, and another 47 percent have completed primary school only, 10 compared with 28 percent and 47 percent, respectively, in nonpoor households (Figure 3.4). Almost half of the mem- 0 bers of poor women-headed households have no education, Two kids or less 3-4 kids Five kids or more compared with approximately one-fourth of members of poor Source: HBS 2017/18. male-headed households. Controlling for sociodemographic Chapter 3 Profile of the Poor 83 FIGURE 3.3: Poverty Headcount According to Household more educated the head is. Education affects living standards Head Education, 2018, Percent and poverty reduction not only directly, but also through its effect on such things as health, productivity, and social 40 35.5 35.0 integration. The association between more education and higher living standards is highly significant in rural and urban 30 areas, accelerating with education beyond primary school, yet 26.0 the results indicate that, although education is the best shield 20 against poverty, primary education seems no longer sufficient to open up opportunities. 10 7.9 6.3 It may appear at first that households with younger heads 1.1 0 fare much better than those with older heads—poverty is lower when the household head is 30 or younger. However, n y y ry ry ity ar ar io da da rs im im at ve n n this is largely because younger heads are generally better- uc pr pr co co ni ed se se ed ed U educated, have just started the household, and so have few o et et er er N pl pl w pp Lo m om children. Controlling for other household sociodemographic U co C n characteristics in a multivariate model, the effect of the head’s a th ss age on poverty vanishes (Appendix E, tables E-1 and E-2)—it Le does not significantly affect living standards and poverty status. Source: HBS 2017/18. FIGURE 3.4: Educational Level of Household Members Children from poor households are at a ­ disadvantage— and Poverty Status, 2018, Percent another illustration of the rural-urban and gender divide. Enrollment in both primary and lower secondary 1.6 2.0 0.1 0.6 100 0.5 2.0 0.1 education is consistently lower for poor than nonpoor children 3.1 3.9 9.5 0.6 10.2 (Figure 3.5A). Nearly 20 percent of children from poor house- 17.1 19.4 80 39.1 holds aged 7–13 are not in school, compared to 13 percent of children in wealthier households (Figure 3.5B). The primary net 46.9 47.6 60 enrollment rate is more than 8 pp higher in urban than in rural 47.1 areas. It is also significantly higher in men-headed households 47.1 14.0 than in women-headed ones, with the gap being highest 40 17.0 17.2 among poor households. Similarly, only 25 percent of poor 12.8 11.6 44.7 children aged 14 to 17 years old are enrolled in lower second- 20 26.1 ary school, compared with 38 percent of other children. Overall, 24.3 18.4 16.0 Tanzania has systematically lower enrollment of poor than 0 of nonpoor children and low enrollment in lower secondary generally—34 percent for the whole country. For upper second- a s ds d d ld ni de de l ho ho za a a ary school, enrollment drops to 2.4 percent generally and to 0.6 n he he se se Ta ou ou - - en en percent for poor young people aged 18 to 19 years. rh rh M om oo o W Po -p on Households supported by agriculture are more likely to N Poor households be poor. Nearly 30 percent of households whose head works No Education Less than completed primary in agriculture are poor, highlighting its subsistence character. Completed primary Lower secondary In comparison, the poverty rate drops to 12 percent when Upper secondary University the household head works in trade or services and 7 percent Source: HBS 2017/18. when the head works in public administration (Figure 3.6A). The labor market profile of the poor is heavily skewed effects in the regression model, education appears to have a toward agriculture: 76 percent of poor household heads significantly positive relationship with consumption, and the work in agriculture, compared to 50 percent of non-poor returns to education rise meaningfully the more schooling ones (­Figure 3.6B). Households whose head works in trade the head has (Appendix E, tables E-1 and E-2). Similarly, the or services also have significantly higher consumption likelihood that a household will be poor drops significantly the (see tables E-1 and E-2). The relationship is particularly 84 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 3.5: Gross and Net Rates of Enrollment in School, 2018, Percent A. Gross enrollment rate B. Net enrollment rate 120 120 106.4 100.9 103.0 100 98.9 96.4 100 91.3 85.1 86.8 82.9 81.5 80 80 67.4 60 60 50.7 51.1 45.5 40 34.9 40 38.0 33.7 34.0 25.7 24.8 20 20 0 0 a n on n r or l r or a l oo a l n r or a l n r or ni ra ra ba ba ra ra oo oo oo ni ni ni Po ba ba Po Po Po za Ru Ru Ru Ru -p za Ur Ur za za -p -p -p Ur Ur n n n n on Ta on on Ta Ta Ta N N N N Primary Lower secondary Primary Lower secondary Source: HBS 2017/18. Note: Age categories for the denominators of the gross and net enrollment ratios are defined as follow:  - Primary schooling: 7 to 13 years old.  - Lower secondary schooling: 14 to 17 years old. FIGURE 3.6: Sector of Employment and Poverty, 2018, Percent A. Poverty Rates by HH Sector B. Sector of Employement by Poverty Status 100 5.5 1.3 4.8 7.4 Agriculture 29.5 16.3 8.5 14.8 80 7.3 18.7 17.0 Manufacturing 19.3 60 9.1 9.5 Services 12.1 40 75.5 50.0 54.3 20 Public administration 7.3 0 0 10 20 30 Non-poor Poor Tanzania Agriculture Industry Trade Other services Public administration Source: HBS 2017/18. strong in rural areas. Nearly 80 percent of individuals in A higher-status job for the household head is associated households identified as poor work in agriculture, compared with higher income and lower incidence and likelihood to 15 percent in the services sector and 5 percent in of poverty. The poverty rate for households operating their manufacturing, mining and construction. Only 1 percent own farm or working as unpaid family helpers is more than of working members of poor households are employed in double that of those self-employed in activities other than public administration. farming and almost quadruple the poverty rate of those in Chapter 3 Profile of the Poor 85 wage employment (Figure 3.7A). Seventy-eight percent of Overall, human capital and access to productive jobs poor household heads and 84 percent of household mem- are low for the poor and women. Nationally, 23 percent bers work on their own farms or as unpaid family helpers of women have no education, and 19 percent more did (Figure 3.7B). In contrast, approximately 30 percent of heads not complete primary school, compared with 13 percent of nonpoor households (and 25 percent of their members) of men with no education and 24 percent who completed are self-employed, and 17 percent (15 percent of their less than primary. The gender gap is larger in poor house- members) are wage employees. Regression results show that holds, in which 32 percent of women and 19 percent of men rural households whose heads own and manage nonfarm have no education. As a result, more women than men are businesses and urban households whose heads work as in unpaid household work and low-paying jobs. The gaps paid employees are far less likely to be poor (tables E-1 and are particularly large in poor households, exceeding 7 pp, E-2), although rural household heads managing their own although the gender gaps in education and employment businesses may be less productive than urban heads who are significantly lower in the younger generation, suggest- work with others, probably because they tend to hire family ing that gender differentials are starting to shrink and that members as a form of social support rather than based on policies to enhance girls’ education and empower women their skills. are beginning to bear fruit. FIGURE 3.7: Status of Employment and Poverty, 2018, Percent A. Poverty by Employment of HH B. Status of Employement by Poverty Status 100 Unpaid family helper/ 33.6 own farmer 80 53.0 Self employed without 61.2 18.5 employees 60 77.8 83.9 Self employed with 12.7 40 employees 23.9 19.5 20 6.4 Employee 8.9 14.7 4.7 16.7 10.7 2.9 14.6 1.7 0 4.7 3.8 0 10 20 30 40 Non-poor Poor Non-poor Poor Household Heads Individuals Employee Self-employed without employees Self-employed with employees Unpaid family helper/own farmer Source: HBS 2017/18. 86 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T II. Community-Based Infrastructure and Services Lack of access to infrastructure and community services leave poor households with fewer environment-based opportunities. Beyond household-specific characteristics, access to although the poor continued to be at a disadvantage in 2018 community services and to shared infrastructure constitute (Figure 3.8). For instance, only 13 percent of poor households critical dimensions of the profile of the poor. Community were served by tarmac roads, compared with 22 percent of services and assets like health facilities, roads, markets, and others. Conversely, 32 percent of wealthier households and communication networks are the backbone of household 43 percent of poor households lack access to any roads. development because they structure the household envi- Developing the road network and ensuring that all households ronment and promote emergence of new opportunities. can access at least the most basic trunk roads will be critical Moreover, provision of essential services and infrastructure given that, particularly in rural areas, households with road often reveal serious shortfalls for poor households, which are access are much less likely to be poor (tables E-1 and E-2). more likely to live in underserved communities that per- petuate their dire monetary situation and lack of access to Many Tanzanians lack access to health care centers. Approx- opportunities. imately 40 percent of Tanzanian households have no access to any type of health facility (e.g., health center, dispensary, Poor households tend to have less access to the road hospital). That includes nearly 50 percent of urban households network than others, although here Tanzania has made (Figure 3.9), even though the coverage of hospitals, both public progress since 2012. Between 2012 and 2018, the share and private, is higher for urban than rural households. Access to of communities served by the road network rose from 51 to 66 percent, driven by a larger increase in trunk roads. Access FIGURE 3.9: Access to Health Facilities, 2018, Percent of poor households to roads went up from 47 to 57 percent, 100 8.4 6.6 12.9 14.5 21.2 FIGURE 3.8: Access to Roads, 2018, Percent 80 51.9 100 46.5 55.2 30.6 45.1 9.2 12.7 60 20.2 22.1 80 40.4 40 50.4 43.9 60 48.2 41.5 45.3 45.7 20 40.6 36.4 40.3 40 36.2 0 Tanznia Rural Urban Non-poor Poor 20 40.5 43.5 34.4 32.2 23.4 Area Poverty 0 No health facility Tanznia Rural Urban Non-poor Poor Only health center/dispensary Area Poverty Only public/private hospital No road Trunk road Tarmac road Source: HBS 2017/18. Chapter 3 Profile of the Poor 87 public and private hospitals is also greater for nonpoor house- Poor households also suffer from less access to holds than poor ones, who mostly have access to small health ­ community-based infrastructure and other community centers and dispensaries. The situation has not changed much services. The share of poor households benefiting from since 2012, although access to health dispensaries increased public transportation, banks, or daily markets is less than for slightly for poorer households in urban areas outside Dar es better-off households (Figure 3.10). This is probably another Salaam. Controlling for other sociodemographic factors, access example of the rural-urban split, rural areas being historically to health facilities, especially hospitals, is significantly and posi- underserved. Access to primary schools may seem slightly tively correlated with higher consumption and less likelihood of higher for poor households, but only 38 percent of them have being poor. The significance and extent of the relationship are access to a secondary school, compared with 47 percent of particularly strong for urban households, from both consump- other households. Access to community-based services is also tion and poverty perspectives. systematically lower for poor households. ­ FIGURE 3.10: Access to Community Services and Infrastructure, 2018, Percent of Households 100 86 82.2 81 80 70.5 71 67 60 56.9 57 55 44.9 47 47.7 49 44 38 40 26.1 26 26 29 22.3 21 23.9 25 19 20 7.4 9 5.4 6 5 2 0 Public Daily Weekly Bank Primary Secondary Mobile Internet Informal SACCOs transportation market market school school phone signal connection nancial services Tanzania Non poor Poor Source: HBS 2017/18. 88 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T III.  Perception of Poverty Household self-assessments of their economic situation reveal widespread subjective poverty. Self-perception of the household situation reveals Self-assessment of their financial situation emphasizes that household financial and economic wellbeing was that households tend to consider themselves poorer severely degraded in 2018. Nationally, about half of than their neighbors. Nearly 47 percent of Tanzania’s households believed that their economic situation had households believed that they were worse off financially worsened since the year before, and only 26 percent than other households in the same community, and only reported improvement. The trend was consistent in all 26 percent felt that they were better off (Figure 3.11B). areas and for all poverty statuses (Figure 3.11A), and Forty-five percent of nonpoor households considered was particularly acute in urban households, of which themselves worse off than other households in their only 18.5 percent thought that their situation had community, and only 25 percent thought that their situation improved since 2017. More than half of poor and nonpoor was better, reflecting large unmet aspirations. Self- households thought that their economic condition had assessment of economic conditions tends to be worse than worsened; 31 percent of poor households assessed their monetary-based estimates because people tend generally economic situation as much worse, and another 22 percent to have higher monetary expectations than their estimated considered it a little worse. real basic needs. FIGURE 3.11: Subjective Poverty and Self-Assessment, 2018, Percent A. How do you compare the overall economic situation of B. How does this household economically and financially compare the household with one year ago? with others in this community this year? 100 2.5 3.0 1.4 2.5 2.2 100 2.1 2.6 1.0 2.2 1.4 14.1 15.7 19.1 23.8 23.6 24.5 22.2 25.7 23.0 80 29.1 80 28.0 33.2 26.9 60 22.8 23.3 20.9 60 28.9 29.4 20.0 26.6 24.5 21.6 40 40 22.9 23.2 26.4 26.0 21.9 25.1 24.4 24.9 20 20 28.1 25.9 32.0 27.4 30.9 23.7 26.6 21.7 20.7 20.5 0 0 Tanzania Rural Urban Non poor Poor Tanzania Rural Urban Non poor Poor Area Poverty Area Poverty Much worse A little worse Same Much worse A little worse Same A little better Much better A little better Much better Source: HBS 2017/18. Chapter 3 Profile of the Poor 89 FIGURE 3.12: Food Stress According to Poverty Status, FIGURE 3.13: Food Shortfalls According to Poverty 2018, Percent Status, 2018, Percent 100 100 4 10 13 11 11 7 7 13 9 18 15 20 19 15 28 24 14 80 80 20 28 25 25 24 25 26 26 11 29 37 35 12 60 31 34 33 12 9 27 10 10 60 10 13 19 18 40 14 11 11 12 40 13 15 71 52 51 53 55 61 49 52 47 20 32 34 34 34 20 34 35 40 0 0 r or r or r or r or r or r or r or r or oo oo oo oo oo oo oo oo Po Po Po Po Po Po Po Po -p -p -p -p -p -p -p -p on on on on on on on on N N N N N N N N Unable to eat Went without Worried to Ate only a few Ate less than Hungry but healthy and Skip a meal Ran out of food eating for a run out of food kinds of foods necessary cound not eat nutritious food whole day No issue Only once or twice Some months Almost every month No issue Only once or twice Some months Almost every month Source: HBS 2017/18. Source: HBS 2017/18 Approximately 20 percent of Tanzania’s households experience severe systematic difficulties in feeding their members. A significant number of Tanzanians, especially those in Nineteen percent of poor households reported having eaten poor households, faced high food stress, characterized less than necessary, and 15 percent ran out of food almost by fear of not having enough food. Almost every month, every month (Figure 3.13). Half of poor households ran out of 18 percent of poor households feared that they would run out food at least some months and sometimes every month, com- of food, compared with 10 percent of nonpoor households, pared with 32 percent of nonpoor households. Furthermore, and another 37 percent of poor households estimated that 13 percent of poor households reported facing situations this situation might occur during many months in the year every month in which their members were hungry because (Figure  3.12). Lack of healthy food, as well as of adequate there was nothing to eat, and 9 percent of them went with- food also characterizes food stress; 28 percent of poor house- out eating for a whole day almost every month. Altogether, holds believed they could not access healthy and nutritious the share of households that experienced no food shortages food almost every month, and 24 percent had access to ranged from one-third to one-half. In sum, for at least 20 limited types of food. percent of Tanzania’s households, there is a systemic risk of not having food. For some households, high food stress translates into difficul- ties in feeding themselves and ultimately to food shortfalls. The Productive Social Safety Nets program contributes to support the poor but its impact could be increased through wider coverage and better targeting. For many, access to the Productive Social Safety Nets reports that the program covers 1.2 million households, of (PSSN) program is essential for meeting basic consump- which 250,000 benefit from public works program (Box 3.1). tion needs, but its coverage is limited. The program is HBS 2018 found that about 1 million households and 4.9 managed by Tanzania Social Action Fund (TASAF), which million people benefit from the PSSN cash transfer program, 90 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T BOX 3.1 Tanzania Productive Social Safety Nets Program Implemented by the Tanzania Social Fund (TASAF), the School enrollment increased by more than 10 percent, •  Productive Social Safety Nets (PSSN) Program is a flag- particularly for primary school age children. Higher ship national social protection program that has been enrollment rates also helped increase the literacy rate, operational since 2012. The objective of PSSN is to enable particularly for primary school. poor households to increase incomes, opportunities, and consumption. It targets 15 percent of the total population of Beneficiary households were 8 percent more likely to •  the United Republic of Tanzania, including all households living visit a health provider, even when not sick. Use of health under the food poverty line (~650,000 households) plus those services was 21 percent higher for beneficiary households at risk of falling under the food poverty line if they experienced than for n­ onbeneficiaries; the benefit was even greater for a shock affecting their income (~350,000 households). under-fives. After a massive expansion process to introduce the Beneficiary households are 8 percent more likely to •  program nationwide, PSSN registered its intended cultivate farm plots and 18.6 percent more likely to beneficiary population of more than one million ­ own or raise livestock. The incidence of input use, households by September 2015, ahead of target. Cash which tends to be very low, rose between 18 percent transfers have been provided for the past four years to more and 38 percent for agricultural inputs that are linked than one million households in close to 10,000 villages in to higher productivity. PSSN beneficiaries shifted all 161 project area authorities. These include a basic trans- away from casual work into self-employment (which fer for all targeted households, an additional unconditional increased by 12.6 percent). Moreover, engagement transfer for households with children, and cash transfers with of nonfarm enterprises owned by beneficiaries in co-­ responsibility related to the uptake of health services for productive sectors increased, 19 percent more in trade the youngest children and to school attendance for school-age and 22 percent less in production. children. In addition, public works have reached nearly 300,000 Resistance to current and future shocks improved households in 44 project area authorities, completing 6,000 through increased savings, asset accumulation, improve- subprojects. Piloting of the productive inclusion and livelihoods ment of housing conditions, and take-up of health component has started in eight project area authorities; more insurance. Participating in the program reduced use of than 11,769 savings groups with 151,821 members (74 percent negative coping strategies (measured using a coping women) have been formed; and initial training on group orga- strategy index) by 19 percent. Beneficiaries improved nization, preparation of constitutions, savings mobilization, loan their housing and living conditions by using better roof management, and record keeping has been provided. materials (3 pp more likely to use higher quality build- The PSSN midline impact evaluation indicates that the ing materials) and improving drinking water sources project has had substantial human development and live- (4.4 pp less likely to use unimproved sources). The lihood outcomes. Early results show that PSSN, through cash likelihood of having any savings grew by 23 percent in transfers, has had positive and statistically significant results: treated households, which were 5.2 pp more likely to have transportation assets, 6.4 pp more likely to have com- Households receiving cash transfers experienced an addi- •  munication assets including mobile phones and radios, tional 10 percent reduction in poverty, accompanied by a and 6 pp more likely to have furniture and whose health 20 percent boost in monthly consumption. insurance registration tripled. of whom 291,000 households (1.4 million people) also benefit coverage reaches 11 percent of all rural and 5 percent of all from the public work program. PSSN reached 15 percent of urban households. extremely poor households, 14 percent of poor households, and 8 percent of nonpoor ones (Figure 3.14A). In line with Both poor and non-poor households report using PSSN the national distribution of the poor, the beneficiaries are income support mainly to buy food, though less poor mostly rural, constituting 78 percent of households benefiting; households invest more in productive assets. Poor Chapter 3 Profile of the Poor 91 beneficiary households appear to receive slightly larger sustainably above the poverty line. Approximately 23 percent transfers than their non-poor counterparts, but both mainly of current beneficiary households are in the two upper con- use PSSN benefits to purchase food, underscoring the sumption quintiles (15 percent in the fourth quintile, 8 percent high food stress in Tanzania (figures 3.14 B and C). About in the fifth), which makes them 7 percent of all households in 66 percent of beneficiaries reported using PSSN income the fourth quintile and 3 percent of all in the fifth. Of PSSN support to cover their food needs⸺70 percent among poor beneficiaries, 10 percent of those in the fourth quintile and households and 65 percent among non-poor ones, 13 percent 18 percent in the fifth invested the PSSN cash in productive to cover education and health expenses, and 21 percent assets, compared with only 4 percent of households in the to invest in productive assets, improve housing and cover poorest groups. Investing in productive assets may have other expenses. Of the poorest beneficiary households, only helped these households improve their living standards faster 4 percent invest in productive assets. While most non-poor and move to a higher income group. More than six years beneficiaries also use the funds mainly for food, about 10 since the program was designed, its targeting needs to be percent invest in productive assets. reviewed, but the process needs to be managed very carefully because some nonpoor beneficiaries may fall back into pov- PSSN is intended to target the poorest Tanzanians and erty if they are dropped from the program. Even households may have helped many beneficiaries escape poverty. in the highest quintiles are at risk because their productive According to HBS 2018, after the program had been in investments could be affected, something that would be even operation for three years, approximately 57 percent of PSSN more problematic if these investments are used to support beneficiaries were in the two poorest consumption quin- other poor households. Program targeting and recertification tiles, although approximately 69 percent of the households of beneficiary households need to be thoroughly analyzed to interviewed that reported benefiting from PSSN were above identify appropriate candidates, supported by measures to the poverty line, even though they were still receiving PSSN build the resilience of those who may no longer qualify so that support. For some this could be a temporary change of they do not fall back into poverty. Objective, standardized status; that is, for households that were close to the poverty processes must guide the analysis. line, the income support from the PSSN cash transfer may have allowed them to afford consumption above the poverty Without PSSN, basic needs and extreme poverty would line, but it was likely that they would fall back into poverty have been higher. Without PSNN income support, poverty if PSSN support were removed. For about one-quarter of would have been about 2 pp higher, which translates to an them, average consumption was only 20 percent higher than additional 1 million poor people, and extreme poverty would the poverty line; these households are at high risk of falling have risen from 8 to 9.2 percent, equivalent to 700,000 more back into poverty if income support is taken away. In other people. Expanding coverage of the program and better tar- cases, PSSN may have allowed some households to move geting it would help to accelerate poverty reduction. 92 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 3.14: PSSN Scheme, 2018, Percent A. PSSN Coverage Rate B. Use of Funds 15 100 4.9 5.6 13.9 9.1 7.1 3.8 80 9.5 3.5 10.7 4.4 10.2 10 7.5 8.8 60 7.6 5.4 40 5 64.7 69.8 20 0 0 Tanzania Rural Urban Non-poor Poor Non-poor Poor Area Poverty Food purchase School fees Medical expenses Productive assets purchase Construction/improving house Other C. Amount Received 100 16.9 18.0 12.8 15.3 20.3 80 14.9 14.7 16.7 17.2 21.0 60 42.0 47.5 46.1 60.8 40 43.0 20 20.3 22.8 22.5 11.5 15.7 0 Tanzania Rural Urban Non-poor Poor Area Poverty Less than 20,000 TSH 21,000 to 40,000 TSH 41,000 to 60,000 TSH More than 61,000 TSH Source: HBS 2017/18. Note: Figures 3.14-A and B are calculated on the basis of the shares of households that benefited from TASAF transfers. Chapter 3 Profile of the Poor 93 CHAPTER 4 The Multiple Facets of Poverty This chapter documents the progress that Tanzania has Despite progress in many well-being dimensions, levels made in many aspects of well-being and the deprivations achieved in some areas are still low. In particular, what is that the population continues to face. Because poverty is clear from anthropometric measures is the persistence of not solely about consumption deficits, this chapter exam- chronic malnutrition throughout the country, particularly ines whether improvements in nonmonetary dimensions in rural communities. Similarly, although many Tanzanians of well-being, such as housing conditions, assets, commu- have access to better sanitation, more than 70 percent of nity infrastructure, and human capital, have accompanied households have only unimproved sanitation facilities to improvements in living standards. Tanzanian households rely on, with many still practicing open defecation. Many have achieved better dwelling conditions, greater access to still lack efficient lighting and cooking energy sources, electricity, better sanitation facilities, and greater access to particularly in rural areas, where conditions in many house- water. Similarly, they have greater access to education, with holds are still grim. net enrollment in lower secondary rising and slowly improving the educational profile of the population. 96 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T I.  Living Conditions and Assets Ownership Housing conditions and access to basic services have improved over the past decade in both urban and rural areas but remain low. Improvements in housing conditions since 2007 are which continued, though more slowly, through 2018 (World evidence that living standards have been rising even for Bank, 2015). Nationally, the share of households with better rural and the poorest households. For instance, in 2007–12, wall, roof, and floor material went up respectively by 6, 16, there was considerable improvement in dwelling material, and 11 pp (Figure 4.1). The share of poor households with FIGURE 4.1: Housing Conditions, 2012 and 2018, Percent      A. Improved wall material          B. Improved roof material        C. Improved floor material 100 91 94 100 94 98 100 86 79 84 79 80 73 71 80 77 80 64 68 60 60 55 60 51 40 40 40 40 32 21 20 20 20 0 0 0 Tanzania Rural Urban Tanzania Rural Urban Tanzania Rural Urban 2012 2018          D. Number of bedrooms                       E. Dwelling ownership 3 100 5 5 6 8 8 6 8 2.5 7 2.4 2.2 2.4 18 17 2.1 2.1 2.1 2.2 2.3 2.2 80 2.0 44 36 2 1.8 60 89 85 40 76 75 1 50 57 20 0 0 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 Tanzania Rural Urban Tanzania Rural Urban Number of bedrooms Ratio person/bedrooms Owned Rented Provided free Source: HBS 2011/12 and 2017/18. Note: Improved wall material: stone, cement bricks, sun-dried bricks, and baked bricks. Improved roof material: iron sheets, tiles, concrete, and asbestos. Improved floor material: cement, ceramic tile, wood, and palm or bamboo. C h a p t e r 4 T h e M u lt i p l e Fa c e t s o f P o v e r t y 97 improved roofs went up 24 pp, from 50 percent in 2012 to electrification increased by 11 pp nationally—not enough to 74 percent in 2018. Nevertheless, 20 percent of Tanzanian ensure adequate electrification of the whole country, partic- households still suffer from pitiable wall and roof conditions, ularly in rural areas, where it increased only by 7 pp. Most of and 50 percent do not have adequate flooring. the progress was in secondary cities, where electrification went up by more than 20 pp, compared to just 12 pp in Dar Housing conditions improved faster in rural than in urban es Salaam. Energy sources for cooking illustrate the shortage areas, but rural dwellings are smaller and made of low- of electrical access: most households still rely on firewood er-quality material than their urban counterparts. Eighty-five and animal residuals (Figure 4.2C); 85 percent of rural and 83 percent of rural households own their dwellings, compared percent of poor households use these inefficient sources of with approximately half of urban ones. Home ownership is par- energy for cooking. ticularly high among the poor, although their dwellings are, on average, much smaller and made of lower-quality material than Tanzania’s plan to diversify toward solar energy to increase urban ones. The floors are made of earth or sand in 66 percent access to electricity is starting to bear fruit, particularly of rural houses, compared with 14 percent of urban houses; for in rural areas. The share of households using solar energy poor households, 80 percent of rural houses and 42 percent of as a source of lighting jumped from 2 percent in 2012 to urban ones have dirt floors. Similarly, roofs are made of grass 26 percent in 2018 (Figure 4.2B). The increase was particularly and leaves in 20 percent (25 percent for poor households) of marked in rural areas, where only 10.4 percent of households rural dwellings and 2 percent (6 percent for the poor) of urban are connected to the grid: In 2018, one-third of rural house- ones. Many houses that rural and poor families own are self- holds used solar power for light up from 2 percent in 2012. built and do not have property titles and thus cannot be used to alleviate the household’s poverty, such as for collateral to Many households lack adequate access to safe drinking obtain loans, investment against inflation, or intergenerational water. Although this has decreased from 35 percent in 2012, transfer of assets. approximately 26 percent of Tanzanian households lacked access to safe drinking water in 2018, with 8 percent having Access to electricity is problematic, particularly in rural access only to unprotected surface water (Figure 4.3A). Most areas and for poor households. Only about 29 percent of of the improvement was in rural areas, where the share of Tanzania’s households have access to the electric grid, a share households using safe water sources, whether basic or lim- that drops to 10 percent for rural and 7 percent for poor house- ited, increased from 55 percent in 2012 to 66 percent in 2018. holds (Figure 4.2A). Progress since 2012 has been modest; The Sustainable Development Goal (SDG)-based definition 98 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 4.2: Access to Electricity, 2012 and 2018, Percent A. Access to the electric grid 100 79.8 68.1 55.3 50 34.7 34.5 29.0 22.1 18.2 10.4 7.2 3.8 3.5 0 Tanzania Rural Other urban Dar es Salaam Non-poor Poor Area Poverty 2018 2012 B. Lightning source of energy                   C. Cooking source of energy 100 100 0 8 12 1 1 1 4 2 19 24 17 31 20 10 80 41 80 43 14 66 61 60 14 60 89 85 61 16 61 1 40 26 70 40 68 64 33 48 29 20 2 20 28 29 18 12 20 2 10 9 9 9 0 4 0 4 2 2 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 Tanzania Rural Urban Tanzania Rural Urban Electricity (grid and generator) Solar Electricity, kerosene and gas Coal/charcoal Kerosene/paraf n Other Firewood/animal residual Other Sources: HBS 2011/12 and 2017/18. of safe or basic water access includes several types of access. The distance to the main source of water has declined, Disaggregating the data based on the World Health Orga- contributing to the slight improvement in access to water. nization (WHO) classification gives a more detailed picture. Installation of piped water systems and connection of house- For instance, access to private piped water is rare: nationally, holds, primarily urban, to the water network has increased the 22 percent of households have access to piped water, ranging share of households accessing water directly on site; between from 51 percent in urban areas to 7 percent in rural areas 2012 and 2018, this share rose 5 pp nationally and 10 pp in (Figure 4.3B). Most SDG-based safe and basic water access urban areas (Figure 4.3C). The proportion of households therefore consists of protected community-dug wells and that are more than 1 kilometer away from the nearest source public taps located within the community or at a neighbor’s. of water fell from 29 percent to 23 percent, although the C h a p t e r 4 T h e M u lt i p l e Fa c e t s o f P o v e r t y 99 FIGURE 4.3: Access to Water and Sanitation, 2012 and 2018, Percent A. Access to water – SDG definition                B. Access to water – Disaggregated data 100 7 2 100 8 12 10 11 12 15 19 16 11 5 26 25 18 1 34 31 80 22 20 80 40 44 21 48 10 26 38 26 4 11 60 12 60 4 40 5 51 40 81 83 40 15 45 59 60 60 64 47 49 54 54 58 18 50 20 20 7 35 8 16 2 2 17 3 4 8 3 5 4 6 0 0 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 Tanzania Rural Urban Poor Tanzania Rural Urban Poor Basic water Limited water Piped water inside dwelling Piped water outside dwelling Improved drinking water Unimproved drinking water Unimproved water Surface water C. Time to source of water                    D. Access to sanitation – SDG definition 100 1 50 100 3 1 10 7 11 7 10 10 20 15 13 18 18 27 25 80 40 80 40 36 31 59 43 32 29 29 61 60 30 60 65 72 23 80 72 79 81 78 25 64 66 78 79 40 20 40 57 49 19 20 39 10 20 11 8 7 33 18 23 8 17 3 2 2 17 2 7 9 9 9 7 3 8 2 7 0 0 0 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 2012 2018 Tanzania Rural Urban Poor Tanzania Rural Urban Poor On site < 30 minutes > 30 minutes Basic sanitation Limited sanitation Distance > 1km (dry season) Unimproved sanitation Open defecation Sources: HBS 2011/12 and 2017/18. Notes: Figure 4.3A: Basic water: Drinking water from an improved water source located on the premises or that can be reached in no more than a 30-minute round trip. Limited water: Drinking water from an improved water source that requires longer than a 30-minute round trip to reach. Unimproved water: Drinking water from an unprotected spring or dug well. Surface water is from a river, dam, lake, pond, stream, canal, or irrigation canal. Figure 4.3B: Improved drinking water: From tube wells, boreholes, protected dug wells, protected springs, and rainwater collection. Unimproved drinking water: From unprotected springs or dug wells, surface water, and alternative methods, such as carts with small tanks or tanker-trucks. Classification is based on World Health Organization definition. Figure 4.3D: Basic sanitation: Nonshared flush toilets or ventilated improved pit latrines. Limited sanitation: Flush toilets or ventilated improved pit latrines shared with at least one other household. Unimproved sanitation: unimproved pit latrines, with or without slabs, open, etc. proportion of households located more than 30 minutes away sanitation systems of urban households led to most of the from their source of water increased in rural areas. progress between 2012 and 2018; the share of these with SDG-based basic sanitation (ventilated pit latrine and flush Meanwhile, access to sanitation is still very problem- toilets) increased nearly 17 pp, and the share of house- atic, particularly in rural areas. Improvements in the holds relying on limited sanitation rose 6 pp (Figure 4.3D). 100 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Nevertheless, approximately 7 percent of Tanzanian percent of households have no sanitation facilities, and households lack any sanitation facility, and 65 percent another 79 percent rely on unimproved facilities. These rates have access only to unimproved traditional pit latrines. increase to 10 percent and 81 percent, respectively, for poor The problem is particularly acute in rural areas, where 10 households. Ownership of modern assets has risen; ownership of more traditional assets is falling. Tanzanian households have experienced some improve- of motorcycles has more than doubled, going up 7 pp in both ments in ownership of mobile phones, televisions, rural and urban areas outside Dar es Salaam, and 8 pp in non- motorcycles, cars, and other modern assets. In 78 percent poor households. In general, non-poor and Dar es Salaam of households, up from 57 in 2012, at least one member owns households recorded the highest increases in modern assets a mobile phone. For poor households, ownership of a mobile like televisions (+8 pp in Dar es Salaam), fridges (+4 pp), phone has gone up by 28 pp and for rural households by decoders (+39 pp), and laptops (+9 pp). Meanwhile, owner- 30 pp. In Dar es Salaam, more than 90 percent of households ship of such traditional items as radio sets or tape recorders are so equipped, as are 85 percent in other urban areas. has declined (Figure 4.4 and Table 4.1). Both the rising and Similarly, though still very low for poor households, ownership FIGURE 4.4: Ownership of Assets, 2012 and 2018, Percent A. Households’ asset ownership 100 25 78 80 20 63 58 57 60 55 15 43 40 33 34 10 24 26 24 20 16 15 5 10 11 10 8 7 5 7 10 1 5 1 5 3 1 0 1 1 2 4 1 4 4 3 0 0 –20 –5 –40 –10 –60 –15 op o o on ve ne e le n ge n e r er r er r t Ca de te se ot in Iro io on di yc o od ay hi rc pt id l vis ea Ra M nd St or ph c ic ac Ai Fr pl La Bi ec H us le c La m re ile VD D Te M ng ob e D p wi M Ta Se 2012 2018 Change (percentage points - right axis) Sources: HBS 2011/12 and 2017/18. C h a p t e r 4 T h e M u lt i p l e Fa c e t s o f P o v e r t y 101 FIGURE 4.4B: Livestock ownership 100 80 66 67 63 59 60 51 52 50 48 40 26 22 20 0 Tanzania Rural Urban Non poor Poor Area Poverty 2012 2018 Sources: HBS 2011/12 and 2017/18. TABLE 4.1: Change in Asset Ownership, 2012 and 2018, Percentage Points TANZANIA AREA POVERTY RURAL URBAN NON-POOR POOR Refrigerator 2.2 0.6 3.7 2.7 -0.3 Heater -1.9 -2.0 -2.3 -2.0 -1.7 Stove -5.0 -9.1 0.0 -4.1 -9.0 Large appliances Aircon -9.3 -2.7 -22.9 -10.9 -3.1 Mobile phone 20.9 27.6 6.0 18.5 29.6 Decoder 12.5 4.2 27.2 15.0 2.4 Television 8.3 5.3 11.0 9.4 3.1 Laptop 2.2 0.5 4.9 2.8 -0.2 Iron 2.1 0.1 3.5 2.7 -1.3 Music set 0.3 -0.1 0.8 0.3 0.0 Landline -0.3 0.0 -0.9 -0.4 0.0 Sewing machine -1.2 -0.8 -2.4 -1.4 -0.7 DVD player -3.0 -0.6 -9.2 -3.8 -0.4 Tape recorder -3.3 -2.6 -5.0 -3.9 -1.2 Small appliances Radio -12.2 -12.0 -13.3 -11.6 -14.9 Moto 6.6 7.1 5.6 7.9 1.5 Car 2.3 1.1 4.3 2.9 0.0 Transportation assets Bicycle 0.7 0.5 2.0 0.1 3.2 Sources: HBS 2011/12 and 2017/18. 102 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T falling trends are similar to the asset ownership trends seen in slightly for poor households, from 63 to 59 percent. Mean- 2007–12 (World Bank, 2015). while, ownership of land plots seems to have plunged dramatically. However, the changes observed may be attrib- Ownership of livestock rose slightly. Between 2012 and utable to a change in data collection methods that makes it 2018, the share of households owning livestock rose by 1 pp difficult to usefully compare 2012 and 2018 data. nationally, and by 4 pp in urban areas. It also decreased C h a p t e r 4 T h e M u lt i p l e Fa c e t s o f P o v e r t y 103 II.  Human Capital Tanzania’s Human Capital indicators are quite low. Tanzania FIGURE 4.5: Human Capital Index was ranked 128 of 157 countries in the 2018 World Bank Human Capital Project with a low Human Capital Index (HCI) Human Capital Index score of 0.4. Low expected years of schooling is among the 100 main limiting factors to the HCI. The World Economic Forum Global Human Capital Report of 2017 also ranked Tanzania 80 106 of 130 countries in HCI.1 Tanzania trails countries with 60 similar income levels, underperforming particularly in the 40 know-how sub-index (109th) due to the very low share of high- Know-How Capacity skilled employment, limited availability of skilled employees, 20 and low economic complexity (Figure 4.5). 0 Development Deployment Tanzania Kenya Uganda Rwanda South Africa Source: World Economic Forum Global Human Capital Report, 2017. Rising secondary education enrollment rates are slowly transforming the educational profile of Tanzanians. Gross and net enrollment rates in primary and lower sec- secondary. As a result, most students complete only half of ondary education rose slightly between 2012 and 2018, lower secondary and do not move on to upper secondary or but enrollment in lower secondary is still low. Between 2012 higher education. Net enrollment in upper secondary school and 2018, net primary enrollment increased by 14 pp (Figure is 2 percent, with a peak of 5 percent in urban areas. 4.6B) (15 pp in rural areas; 9 pp in urban areas). The combina- tion of stagnant or slipping gross enrollment rates in primary Because class repetition compounds itself over the course and lower secondary education and rising net enrollment of the curriculum, many students are older than would be rates (Figure 4.6A and B) suggests improvement in education expected for their placement. The discrepancies between quality and a drop in the number of children repeating classes gross and net enrollment rates suggest that class repetition and therefore being enrolled in classes not for their age. might be a cause. About 15 percent of all students report that they have failed a class, 16 percent for primary pupils, 10 Gross and net enrollment in upper secondary classes percent for lower secondary, and 9 percent for upper second- and beyond are very low. In Tanzania, education is free ary (Figure 4.7A). In primary school, 12 percent of enrolled until age 15, which corresponds to the second class in lower pupils are already older than 13 years and in upper secondary, 1 The components of the HCI in the World Bank Human Capital project are different from those of the HCI in the World Economic Forum Global Human Capital Report. 104 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 4.6: Gross and Net Enrollment Rates, 2012 and 2018, Percent A. Gross enrollment rate 120 106 106 101 99 100 95 92 83 80 67 60 55 46 42 40 35 21 19 20 9 8 4 3 0 Tanzania Rural Urban Tanzania Rural Urban Tanzania Rural Urban Primary Lower secondary Upper secondary B. Net enrollment rate 120 100 91 85 83 82 80 71 68 60 50 51 40 31 34 22 26 20 2 2 5 5 1 1 0 Tanzania Rural Urban Tanzania Rural Urban Tanzania Rural Urban Primary Lower secondary Upper secondary 2012 2018 Sources: HBS 2011/12 and 2017/18. Note: Age categories for the denominators of the gross and net enrollment ratios: primary, 7–13; lower secondary, 14–17; upper secondary, 18–19. more than 55 percent are older than 19, suggesting that class decreased 13 pp and the proportion with some lower second- repetition persists over the course of education (Figure 4.7B). ary rose 11 pp. Moreover, the educational profile is improving across generations; 11 percent of Tanzanians aged 15 to 29, More adults are now attaining lower secondary education. 17 percent of those aged 40 to 49, 23 percent of those aged Nationally, between 2012 and 2018, the share of adults aged 50 to 59, and 48 percent of those 60 and older have no formal 15 and older with no education dropped 10 pp and the share education. Similarly, 29 percent of individuals aged 15–29, with some lower secondary increased almost 8 pp (Figure 4.8). 7 percent of those aged 40 to 59, and 5 percent of those older The improvement was particularly remarkable in urban areas, than 60 have lower secondary education. where the proportion of adults without any formal education C h a p t e r 4 T h e M u lt i p l e Fa c e t s o f P o v e r t y 105 FIGURE 4.7: Repetition of Classes, 2018, Percent A. Repetition rates                       B. Enrollment by class of age 15.8 Tanzania 10.3 9.0 0.5 0.7 3.8 26.8 17.1 Rural 10.3 0.1 3.5 51.4 58.5 3.1 Area 0.4 13.2 29.8 11.6 12.3 Urban 10.4 11.1 11.3 74.8 15.0 3.1 16.0 Non poor 9.6 87.7 7.8 9.9 Poverty 15.3 Poor 12.9 0.0 Lower Upper Primary Superior secondary secondary 0 5 10 15 20 0 1 2 3 4 5 Primary Lower secondary Upper secondary 7–13 yo 14–17 yo 18–19 yo 20–24 yo > 24 yo Source: HBS 2017/18. FIGURE 4.8: Educational Achievements of Adults 15+, 2012 and 2018, Percent 100 0.6 0.1 0.4 3.6 1.5 5.0 1.7 1.4 0.7 1.2 3.0 8.9 3.0 10.5 6.2 16.7 17.0 80 27.9 45.8 46.6 46.7 60 46.0 48.3 40 14.1 44.7 11.7 15.4 12.5 6.7 20 30.8 34.4 25.8 7.3 20.3 23.3 10.3 0 2012 2018 2012 2018 2012 2018 Tanzania Rural Urban No education Less than completed primary Completed primary Lower secondary Upper secondary University Sources: HBS 2011/12 and 2017/18. 106 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T While there seems to have been progress in adult and maternal mortality, many under-5 children still suffer from chronic malnutrition. Over the last decade the total fertility rate (TFR) has Between 2010 and 2016, adult mortality fell from 5.1 to declined as a result of public family planning policies and 4.6 deaths per 1,000 population in women and from 5.0 to 4.3 increased education of women. The TFR has gone down by in men. Between 2015 and 2016, the probability of a 15-year- nearly one child over the past two decades (Figure 4.9A). For old dying before age 50 also declined: for women from rural women the TFR has followed the same declining trend but 23.6 percent to 18.1 percent (Figure 4.9D) and for men from for urban women it has gone up slightly, from 3.6 to 3.8 children, 24.2 percent to 17.4 percent. since 2005. In general, it seems that more education for women helps to bring down the TFR. For instance, women with no Anthropometric indicators for under-five children have education have a TFR of 6.9, compared with 5.3 for women improved over the last decade, but chronic undernutri- with complete primary education and 3.6 for women with lower tion remains severe, particularly in rural areas. For the last secondary education and beyond. 20 years, the share of under-five children with nutritional defi- ciencies has been decreasing (Figure 4.9C), although the fact The maternal mortality ratio has not changed significantly that too many under-five children are still stunted indicates over the last decade. In the 2016 Demographic and Health that chronic malnutrition is a structural problem. Nationally, Survey, there were 556 maternal deaths per 100,000 live births 35 percent of under-five children are stunted, 12 percent for the 10-year period before the survey (Figure 4.9B). The severely, which means that they are considered short for their confidence interval is large and overlaps with those of previ- age, highlighting problems with cumulative growth deficits ous years, indicating no significant differences. (Figure 4.9E). The problem is particularly acute in rural areas, where nearly 40 percent of under-five children are stunted. Meanwhile, over the last decade, adult mortality, including the probability of dying before age 50, has decreased. C h a p t e r 4 T h e M u lt i p l e Fa c e t s o f P o v e r t y 107 FIGURE 4.9: Health and Anthropometric Indicators, 2016      A. Trend in total fertility rate                B. Maternal mortality rate with confidence interval Of children per woman aged 15-49 Per 100,000 live births / women 15-49 8 800 6.6 6.3 6.5 6.5 6.1 6.0 690 666 6 6.2 600 5.8 5.7 578 556 556 5.6 5.4 4 5.1 5.2 466 454 446 4.1 3.8 3.6 3.7 400 2 3.2 353 0 200 1992 1996 1999 2005 2010 2016 2005 2010 2016 Tanzania Rural Urban C. Trend in anthropometric indicators                D. Adult mortality (15+) Percentage of under-5 children in situation of de cit Per 1,000 individuals at age 15 60 300 50 50 48 44 236 242 42 35 196 195 40 200 181 174 24 26 24 20 16 16 15 100 7 8 5 4 5 5 0 0 1992 1996 1999 2005 2010 2016 2005 2010 2016 Stunting Wasting Underweight Female Male E. Anthropometric indicators Percentage of under-5 children in situation of de cit 50 38.1 40 34.8 13 30 12 25.0 20 8 15.2 13.6 25 3 3 9.1 23 10 17 4.4 4.6 3.8 1 11 12 1 1 1 8 3 3 3 0 Tanzania Rural Urban Tanzania Rural Urban Tanzania Rural Urban Stunting Wasting Underweight Moderate Severe Total Source: DHS 2015/16. Note: Deficiencies in anthropometric indicators are measured for each indicator for which the Z-score is more than two standard deviations below (–2 SD) the median of the World Health Organization reference population. Stunting (height-for-age ratio) is a measure of linear growth retardation and cumulative growth deficits, identifying children who are short for their age (stunted) or chronically undernourished. Wasting (weight-for-height ratio) measures body mass in relation to body height or length and describes current nutritional status, identifying children who are thin (wasted) or acutely undernourished. Undernourishment (weight-for-age ratio) is a composite index of height-for- age and weight-for-height that takes into account acute and chronic undernutrition. 108 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T III.  Multitude of Deprivations in Well-Being The Tanzania National Bureau of Statistics is currently One-third of Tanzanians are multidimensionally deprived, developing a National Multidimensional Poverty Index. with the rate being much higher in rural areas. Approx- The approach is based on the principle that not only severe imately 34 percent of the population is deprived in at least shortfalls in consumption and income, but also deficits in one-third of the well-being indicators mentioned above. many other dimensions of living conditions can jeopardize the Deprivations are distributed unequally in terms of location; well-being of the population. This is part of the government’s 46 percent of the rural population and 10 percent of the urban effort to sustainably address poverty by going beyond the population is multidimensionally deprived (Table 4.2). proximate causes of deficits in consumption and understand- ing the different forms of deprivation that the population Deprivation tends to be greatest in the living standards faces to address the numerous underlying causes of poverty dimension. More than 30 percent of the population and vulnerability. Determining the multitude of dimensions in is deprived in sanitation, electricity, and cooking fuel which people are deprived and the complicated ways these (Figure 4.11); 29 percent in housing; 25 percent in drinking dimensions are enmeshed involved several consultations with water; and 21 percent in assets. Only 2.7 percent of the national stakeholders and international experts to define the population is deprived in the health dimension. National Multidimensional Poverty Index estimation method- ology. (See appendix F for a brief overview of the technical TABLE 4.2: Multidimensional Deprivations, 2018, Percent underlying model). The process has been completed, and HEADCOUNT the National Bureau of Statistics will publish the National Tanzania 34.2 Multidimensional Poverty Index soon. Rural 45.6 Urban 9.9 This chapter assesses the extent of deprivation in three Source: HBS 2017/18, Tanzania National Bureau of Statistics, and OPHI (2019). main dimensions of well-being, considering nine indicators of health, education and living standards (Figure 4.10). We considered individuals who were deprived in at least one-third of these indicators to be multidimensionally deprived. FIGURE 4.11: Deprivation Levels According to Welfare Dimension, 2018, Percent FIGURE 4.10: Well-being Dimensions to Assess the 40 Multitude of Deprivations 33.9 32.5 33.4 29.2 30 25.1 School attendance Years of schooling Child mortality Drinking water 20.9 Cooking fuel Sanitation Electricity 20 Housing 17.2 Assets 10.7 10 2.7 Health Education 0 Living standards (1/3) (1/3) (1/3) ce r g ts g n y ty el e lit in io sin se at an ci fu ta ol t tri w ta As ou nd g or ho ec ni ng n H m te sc ki Sa El ki at oo ld of rin Three welfare dimensions hi ol C rs D C ho a Ye Sc Source: Tanzania National Bureau of Statistics and OPHI 2019, preliminary steps to Living standards Education Health estimate the National Multidimensional Poverty Index. Note: Indicators within each dimension were equally weighted. Deprivation criteria are defined in appendix F, table F.1 Source: HBS 2017/18, Tanzania National Bureau of Statistics, and OPHI (2019). C h a p t e r 4 T h e M u lt i p l e Fa c e t s o f P o v e r t y 109 CHAPTER 5 Geographic Dimensions of Poverty This chapter examines geographic disparities in pov- village-level indicators that, with one exception, were erty in mainland Tanzania. It is divided into two parts: The derived from publicly available satellite imagery and first section presents new poverty estimates for 2018 at the remote-sensing data.1 These new estimates are not compa- regional and district levels, and the second section exam- rable with either of the two sets of district-level estimates ines the extent to which geographic disparities in poverty generated previously based on the 2012 Population Census are related to measures of urbanization, market access, (World Bank 2017; Kilama and Windeboom 2016).2 This is human capital, public services, and agro-climatic condi- primarily because of differences in survey methodology tions. The district level poverty estimates are generated between the 2012 and 2018 HBSs, which led to distorted based on the 2018 HBS, combined with a battery of estimates of poverty trends in many districts.3 1 The one exception is building counts, which were obtained from a development partner and are not publicly available. According to administrative boundaries defined in the 2012 Population Census, Tanzania contains 169 districts (of which 159 are on the mainland), which are subdivided into approximately 16,000 villages on the mainland. Combining the survey data with estimates derived from satellite indicators for villages that the survey does not cover increases the precision of the estimates sufficiently to make reporting district-level estimates feasible. See appendix G for methodological details 2 The methodology used to derive the World Bank estimates are described in Rascon and Audy (2016). 3 In particular, the 2011/12 HBS was based on the 2002 census sample frame. The change in the sample frame between 2002 and 2012 has large effects on estimated poverty rate in certain districts; for example, estimated poverty rose significantly in Longido district because of greater inclusion of settled Maasai in the 2012 sample frame. In several other districts, estimated poverty rates rose 20 to 30 pp between 2012 and 2018, which we believe stems more from methodological differences than deteriorating economic conditions. 112 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T I.  Geographic Disparities in Poverty Geographic inequalities in poverty are large. There are important disparities in poverty between agriculture on their own farms or as unpaid family helpers. ­geographic zones. The poverty rate varies from 14.5 p ­ ercent Maternal and child mortality, fertility rate, and prevalence of in the eastern zone to 34.5 percent in the western zone malaria are among the highest in this zone (DHS 2015/16). The (Figure 5.1). Poverty is also prevalent in the lake and southern southern and western zones also have high poverty rates and zones (>30 percent). Because the distribution of the popula- a high prevalence of health problems, but because popula- tion is uneven across geographic zones, zones where poverty tion density is low in these zones, there are fewer poor people. is highest do not necessarily contain the most poor people. The eastern and northern zones have the lowest poverty rates, Approximately one-third of the poor are located in the lake but because a substantial fraction of the population lives there zone, where approximately one-fourth of Tanzania’s mainland (18 percent and 12 percent, respectively), they contain a large population lives and the poverty rate is the second highest share of poor people. (Figure 5.2). Most households in this zone work in subsistence The analysis also reveals major geographic inequalities in FIGURE 5.1: Poverty Headcount by Geographic Zone, poverty across regions and districts. The standard deviation 2018, Percent FIGURE 5.2: Distribution of the Poor Population by 40 Geographic Zone, 2018, Percent 34.5 35 32.7 33.0 30 28.6 27.7 10.0 11.5 24.4 25 Western 19.3 9.0 20 Northern 14.5 Central 15 Southern Highlands 10 33.1 Southern 11.6 5 South West Highlands Lake 0 6.5 Eastern n n an rn rn an st ke n l ra er er er hl e hl e he La ds ds nt ig W t th ig uth st es ut Ce 6.7 Ea or H uth W So So N 11.5 So H Source: HBS 2017/18. Source: HBS 2017/18. Note: To estimate geographic differentials in poverty, Tanzania mainland was Note: To estimate geographic differentials in poverty, Tanzania mainland was divided into eight geographic zones. These are not official administrative areas, divided into eight geographic zones. These are not official administrative areas, they are based on DHS 2015/16 classification. They group administrative regions as they are based on DHS 2015/16 classification. They group administrative regions as follows: Western zone: Tabora, Kigoma; Northern zone: Kilimanjaro, Tanga, Arusha; follows: Western zone: Tabora, Kigoma; Northern zone: Kilimanjaro, Tanga, Arusha; Central zone: Dodoma, Singida, Manyara; Southern Highlands zone: Iringa, Njombe, Central zone: Dodoma, Singida, Manyara; Southern Highlands zone: Iringa, Njombe, Ruvuma; Southern zone: Lindi, Mtwara; South West Highlands zone: Mbeya, Rukwa, Ruvuma; Southern zone: Lindi, Mtwara; South West Highlands zone: Mbeya, Rukwa, Katavi, Songwe; Lake zone: Kagera, Mwanza, Geita, Mara, Simiyu, Shinyanga; and Katavi, Songwe; Lake zone: Kagera, Mwanza, Geita, Mara, Simiyu, Shinyanga; and Eastern zone: Dar es Salaam, Pwani, Morogoro. Eastern zone: Dar es Salaam, Pwani, Morogoro. C h a p t e r 5 G e o g r a p h i c D i m e n s i o n s o f P ov e r t y 113 of the head count poverty rate across districts is 13 percentage Patterns of night-time lights show expanding urbanization. points (pp). More than one-fifth of total inequality in economic Nighttime lights, which have long been used as a proxy for welfare, proxied by household consumption per adult, is due urbanization and economic development, can give a rough to geographic inequalities across districts, although it is likely indication of geographic patterns in economic development that this figure underestimates the true share of economic and urbanization (Elvidge et al. 2001; Henderson et al. 2012). inequality between districts, to the extent that household Judging from the growth in nighttime lights between 2013 consumption is measured with error. and 2018 (figures 5.3 and 5.4), urbanization has significantly expanded the economic footprint of Dar-es-Salaam and sec- Tanzania’s recent growth and poverty reduction may ondary cities and towns such as Dodoma, Arusha, Mwanza, have disproportionately benefited certain locations. For and Tarime. There has also been scattered growth in night- example, anecdotal evidence suggests that the adminis- time lights in the southern portion of the country and along trative capital of Dodoma has developed rapidly in recent the main north-south road corridors in the center of the coun- years, in anticipation of the long-awaited move of the try. The darkest areas, and those with modest growth in light, national government there, which was officially announced are in the west of the country near the border with Burundi, in April 2019. Furthermore, Tanzanian tourist arrivals have the Democratic Republic of the Congo, and Zambia, near doubled since 2006, raising the question of whether this Lake Tanganyika, indicating that economic development and has boosted poverty reduction in areas near tourist attrac- urbanization have stagnated in those areas, although patterns tions such as Mt. Kilimanjaro and the Serengeti National of nighttime lights tend to reflect settlement patterns more Park in northern Tanzania. strongly than poverty. FIGURE 5.3: Nighttime Lights, 2018 FIGURE 5.4: Changes in Nighttime Lights, 2013–18 Source: World Bank calculation based on Visible Infrared Imaging Radiometer Suite (VIIRS), Version 1; ngdc.noaa.gov. Notes: Nighttime light data from June 2018 are not available and are thus excluded Source: World Bank calculation based on VIIRS. from the analysis. Notes: Inserts show areas around Dar-es-Salaam and Dodoma. 114 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Pockets of poverty are concentrated in the northern and western parts of the country, although some are also found in the southern parts. Poverty is highest in regions in the west and ­ northwest. The district-level poverty map reveals four broad pockets Figure 5.5A shows region-level poverty estimates derived from of poverty. The first is in the north and northeast part of the the 2018 HBS. The poorest regions are Rukwa (45 ­ percent), country. Poverty is most pervasive in Longido district, which Simiyu (39 percent), Lindi (38 percent), Geita (37 percent), is just north of Arusha and just west of Mt. Kilimanjaro near and Mwanza (35 percent). The least impoverished regions are the Kenyan border and is estimated to have a poverty rate of Dar es Salaam (8 percent), Kilimanjaro (11 ­percent), Njombe greater than 60 percent. Longido is home to a large commu- percent), and Morogoro (19 percent). (13 ­ 4 nity of Maasai pastoralists, and a recent “ruby rush” at and around Mundarara mine has yet to create meaningful pov- Although these regional disparities are revealing, they fail erty reduction for residents. On the west side of the Serengeti to capture significant variations in living standards and is Itilima district and the neighboring Kwimba district, which poverty within each region. Figure 5.5B shows estimated are also among the most impoverished areas in the c ­ ountry. district poverty rates, which are derived based on the 2018 In sum, many of the country’s poorest districts lie roughly HBS and a variety of geospatial information that helps predict in a horizontal line from Longido in the east to Kwimba in the poverty rate in each district.5 the west. FIGURE 5.5: Estimated Poverty Rate A. According to Region B. According to District Source: World Bank estimates based on 2017/18 HBS and auxiliary variables. 4 For visualization in figure 5.5A, we used a shapefile constructed based on district and regional boundaries at the time of the 2012 Population Census. Therefore, Songwe, a new region that split from Mbeya in 2016, is not mapped, and the poverty rates for Mbeya (20.7 percent) and Songwe (21.4 percent) are similar. 5 See appendix G for details on the methodology used to derive district-level poverty estimates. C h a p t e r 5 G e o g r a p h i c D i m e n s i o n s o f P ov e r t y 115 TABLE 5.1: Poverty District Estimates, 2012 and 2018 are Tunduru and Mbinga in the Ruvuma region, Nanyumbu and Newala in the Mtwara region, and Lindi Rural and Liwarle SPEARMAN RANK CORRELATION 2018 2012 (WORLD BANK) 2018 poverty estimate 1.00 0.20 districts in the Lindi region. 2012 poverty estimate (World Bank) 0.20 1.00 These geographic patterns of poverty in 2018 should Average district household size in 2012 0.53 0.03 Average district literacy rate in 2012 -0.61 -0.14 not be compared with the 2012 district estimates to ana- District share of children aged 12 to 15 attending school -0.59 -0.08 lyze changes in district poverty rates. The rank correlation Source: 2012 census and World Bank staff estimates. between the 2018 district estimates and the 2012 estimates Note: The 2018 district estimates are more strongly correlated with census-based that the World Bank (2016) published is only 0.20 (Table 5.1). welfare indicators than the 2012 estimates. This correlation is too low for local economic shocks or growth patterns to explain, as two concrete examples illustrate. In A second pocket of poverty lies slightly further to the 2018, Kaliua district was very poor, with an estimated poverty west in the northern region, in parts of the Kigoma, rate of greater than 50 percent. This contrasts starkly with the Kagera, and Geita regions near the Kigosi and 2012 estimate of 3 percent. Second, Longido district was esti- Moyowosi game reserves. Poverty in this area is most mated to have a poverty rate higher than 60 percent in 2018, prevalent in Kibondo district, which is on the Burundi bor- compared with 32 percent in 2012. Large increases in poverty der and hosts large numbers of refugees, many of whom in these districts cannot explain these differences, which must arrived during unrest in neighboring countries in the mid- be due to changes in the methodology and source data that 1990s. Just north and east of Kibondo are four other dis- make the 2012 and 2018 estimates incomparable. tricts with high poverty rates: Chato, Biharamulo, Bukombe The new district poverty estimates for 2018 appear to be and Ngara. Buhigwe (south of Kibondo) and Kaliua and more consistent with indicators from the 2012 census than Urambo, approximately 300 km east of Buhigwe in the the previous 2012 d ­ stimates. For example, according ­ istrict e Tabora region, are also quite poor, with a poverty rate to the census, Kaliua had the fifth-lowest literacy rate in 2012, greater than 40 percent. which is more consistent with the high estimated poverty rate A third pocket of poverty lies in the western part of the in 2018 than the low poverty rate reported for 2012. Similarly, country, near the borders of the Democratic Republic of only 46 percent of 12- to 15-year-old children in Kaliua reported Congo and Zambia. Sumbawanga Rural is a small, particularly attending school in 2012, the lowest ratio of all districts in poor district in the Rukwa region, between Lake Tanganyika the country, further suggesting that it is a very poor district. and Lake Rukwa. Bordering Sumbawanga Rural to the north is Table 5.1 shows that the 2018 poverty estimates are much more Nkasi district, which also has a high rate of poverty. strongly correlated with measures of school attendance, liter- acy, and household size available in the census than the 2012 The other main pocket of poverty is in the southeast, in estimates. This confirms that the 2018 estimates are reasonably parts of the Ruvuma, Mtwara and Lindi regions bordering consistent with indicators from the 2012 census and that the Mozambique. Among the poorest districts in these regions 2012 and 2018 estimates are not comparable. Districts with high poverty rates tend to have many poor people. Although poor people tend to reside in districts with high people, p ­ articularly the area around Lake Victoria and Simiyu poverty rates, there are also populous districts with lower region in the north and around the Kigosi and Moyowosi poverty rates where many poor people live. Policymakers game reserves in the west. Districts such as Kwinba and may wish to target provision of public goods to areas where Kaliua host a large population of poor people because they large numbers of poor people live. Figure 5.6 shows the have large populations and high rates of poverty, although share of the country’s poor people according to district, there are also a significant number of poor people in Lushoto and ­ figure 5.7 shows the relationship between poverty rate district in Tanga and Ilala district near Dar es Salam, which and number of poor for 159 districts. The number of poor is have lower poverty rates but large populations. Conversely, lower in areas with lower rates of poverty, and the regions Longido and Nanyumbu districts, despite high poverty rates, with the highest poverty rates also have the most poor are sparsely populated and therefore have fewer poor people. 116 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 5.6: Estimated Share of Poor by District FIGURE 5.8: Multidimensional Deprivation Rate by Region, 2018, Percent Source: World Bank estimates based on 2017/18 HBS and auxiliary variables. The number of poor is derived based on the 2012 Population Census. Source: Estimates based on OPHI and NBS (2019) and 2017/18 HBS. FIGURE 5.7: Number of Poor and Poverty Rates in Districts and attainment, child mortality, and several indicators of asset ownership and housing quality into three components, 300 0.8 which are then combined into a single index. Because the Multidimensional Deprivation Index (MDI) is derived entirely from the 2018 HBS, it is available only at the regional level.6 Poverty incidence (2018) # of poor (in thousands) 0.6 Regional patterns in the MDI provide further evidence of large 200 geographic disparities in poverty (figures 5.8 and 5.9). 0.4 Although the geographic pattern of multidimensional 100 deprivation is broadly consistent with that of monetary 0.2 poverty, there are important differences. Multidimensional deprivation appears to be pronounced in the western part of the country, particularly in Kagera, Katavi, Tabora, Rukwa, and 0 0 Simiyu regions, which are also among the most impoverished # of poor (in thousands) Poverty incidence (2018) in terms of monetary poverty, but the detailed geographic patterns of multidimensional deprivation differ considerably Source: World Bank estimates based on 2017/18 HBS and auxiliary variables. The number of poor is derived based on the 2012 Population Census. from those of monetary poverty. For example, the pockets of monetary poverty in the south (e.g., Ruvuma region) are not Monetary poverty as poor in a multidimensional sense. The MDI may be failing to fully capture recent growth and development in particu- accompanies lack of larly areas. For example, Dodoma region is the seventh poor- human capital and access est of the 26 regions, according to the MDI, but ranks 18th in monetary poverty. In this case, the MDI may not be capturing to basic public services. Multidimensional deprivation is a measure of poverty 6 The MDI is similar to the Multidimensional Poverty Index (MPI). It was estimated in collaboration with the National Bureau of Statistics and the Oxford Poverty based on an index of access to services and human capi- and Human Development Initiative (OPHI) during the preliminary steps for the tal indicators. The index combines educational attendance calculation of the National Multidimensional Poverty Index. C h a p t e r 5 G e o g r a p h i c D i m e n s i o n s o f P ov e r t y 117 FIGURE 5.9: Number of Multidimensional Deprived by the rapid recent growth in Dodoma, visible in the nighttime Region, 2018, Thousand lights shown in figure 5.4, due to the anticipated movement of the national government administration. Similarly, the west- ern region of Katavi is the second poorest region according to the multidimensional index but only the 14th poorest of the 26 regions according to the monetary measure. Areas with a higher level of poverty also tend to have sig- nificantly less human capital and poorer access to pub- lic services such as water and sanitation. Figure 5.10 shows district-level geographic disparities in education attainment and access to basic services (e.g., improved water and sani- tation), which are derived from the 2012 Population Census. Figure 5.11 depicts how these nonmonetary measures of wel- fare are correlated with poverty incidence. There is a strong correlation between poverty incidence, education, and access to improved water and sanitation. Urban-rural inequalities are partially the source of these geographic disparities in human capital and public services, because urban residents on average tend to have more education and a denser net- work of public services. Education and access to basic public services are two of the most critical drivers of poverty reduc- tion in Tanzania, and addressing such geographic disparities Source: Estimates based on OPHI and NBS (2019) and 2017/18 HBS. ­ rucial step toward reducing geographic inequalities in is a c ­monetary poverty. FIGURE 5.10: Access to Education and Public Services, 2018 Sources: 2017/18 HBS and 2012 Population Census. 118 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 5.11: Relationship Between Poverty, Education and Public Services 0.8 0.8 0.8 Poverty incidence Poverty incidence Poverty incidence 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0 0.2 0.4 0.6 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 Access to improved sanitation % of no schooling Access to improved water Sources: World Bank estimates based on the 2017/18 HBS for poverty incidence and the 2012 Census for district-level share of people with no schooling and share of households with access to improved water/sanitation. C h a p t e r 5 G e o g r a p h i c D i m e n s i o n s o f P ov e r t y 119 II.  Drivers of Geographic Disparities in Poverty Urbanization and structural transformation provide are available, and these productivity gains come partly from Tanzania with an opportunity to make further strides the benefits of agglomeration economies, such as resource toward poverty reduction. Although Tanzania’s popu- sharing; quick, accurate matching; and knowledge spillovers lation is growing slightly faster than 3 percent a year, for (Duranton 2013). the last decade, the urban population has grown an aver- age of 5.5 percent a year. In 2012, approximately 30 per- As in many other developing countries, urbanization cent of the population was living in urban areas; by 2045, this has been a critical driver of geographic disparities in share is expected to reach 50 percent. Dar es Salaam is the poverty in Tanzania. Poverty rates are lower in larger and third-fastest-growing city in Africa (United Nations 2014). more densely populated districts (Figure 5.12). Poverty incidence is also less pronounced in districts with a greater Successful urbanization typically translates into level of nightlights, which reflects their greater levels of poverty reduction through a structural transformation urbanization and economic activity (Figure 5.13). Among from low-productivity agricultural employment into the least impoverished areas are the districts that house more-productive urban jobs in manufacturing and major urban cities, such as Dar es Salaam, Arusha, and services (Lewis 1954). In large cities, more-productive jobs Tanga. FIGURE 5.12: Poverty According to Population Density FIGURE 5.13: Poverty According to Nighttime Lights 4000 0.8 0.8 Population density (per sq. km) Poverty incidence (2018) 0.6 3000 0.6 Poverty incidence 0.4 2000 0.4 0.2 1000 0.2 0 0 0 0 0.5 1.0 1.5 2.0 Population density (per sq. km) Poverty incidence (2018) Nighttime light Sources: Population from 2012 Census; poverty from 2017/18 HBS. Sources: 2017/18 HBS and VIIRS. Note: Intensity of nighttime light is based on the natural log of maximum nighttime luminosity (2015) at the district level. 120 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Lack of access to urban centers limits the benefits of urbanization to a few urban centers and their immediate vicinity. The benefits of agglomeration economies can trickle down within a certain travel time. (See appendix H for more details). to other semi-urban and rural areas with good access The analysis also reveals that areas with limited market access to markets for products, labor, and ideas (Mayer 2008). are more likely to have a higher rate of poverty (Figure 5.15A); Those regions or districts with greater accessibility to major FIGURE 5.14: Market Access to Major Urban Centers in cities will attract more economic activities, investments, and Tanzania labor, which in turn will allow them to reinvest in market access and reinforce their agglomeration advantage. Research con- firms such beneficial spillover effects from urban to neighbor- ing rural areas, as seen in the cases of India and Nepal (Cali and Menon 2013; Fafchamps and Shilpi 2005). Conversely, poor areas with limited access to urban centers are more likely to remain trapped in poverty, worsening geographic inequali- ties (World Bank 2009; Page and Pande 2018). In Tanzania, a large swath in the north, northwestern, and southeastern parts of the country have particularly lim- ited access to markets, and these areas typically have higher levels of poverty. Areas surrounding Dar es Salaam— the commercial and economic center of the country—tend to have greater access to markets because they have a wide, dense road network system, which allows people living near the city to benefit from its agglomeration economies. This is Sources: World Bank’s estimates based on OpenStreetMap reflected in the varying levels of market accessibility displayed Notes: The Open Source Routing Machine (OSRM) algorithm is used to compute in figure 5.14, which is measured using the weighted sum of travel time between each village and major cities. “High” corresponds to areas in the top quartile of the market accessibility index; “Moderately high” the second top population in major cities that are accessible from each village quartile; “Moderate” the 2nd lowest quartile; and “Low” the lowest quartile. FIGURE 5.15: Poverty by Market Access and Distance to Dar es Salaam A. Poverty by Market Access B. Poverty by Distance to Dar es Salaam 0.8 0.8 0.6 0.6 Poverty incidence Poverty incidence 0.4 0.4 0.2 0.2 0 6 8 10 12 14 16 0 500 1000 1500 Market access Travel distance to Dar (in km) Sources: World Bank’s estimates based on OpenStreetMap. Note: The Open Source Routing Machine (OSRM) algorithm is used to compute travel time between each village and major cities. In the scatterplot, market access refers to the natural log of the market access index. C h a p t e r 5 G e o g r a p h i c D i m e n s i o n s o f P ov e r t y 121 travel distance to Dar es Salaam is also positively correlated of higher costs of transportation and fewer options of prod- with poverty incidence (Figure 5.15B), which means that the ucts available. These farmers also have limited access to out- farther away an area is from the country’s commercial hub, the put markets, so they receive lower prices for their harvest greater the incidence of poverty it has. (Aggarwal et al. 2017). The lack of market access traps these rural farmers in poverty while exacerbating existing geo- The positive economic dividends of access to jobs and graphic inequalities between rural and urban centers. trade opportunities in Dar es Salaam are hard to over- state. Although the Dar es Salaam administrative region Regional disparities and income divergences are widening accounts for a small fraction of mainland Tanzania’s land area because of a lack of connective infrastructure that facili- (0.16 percent) and total population (~10 percent), the region tates the flows of goods, services, ideas, and people. Lack accounts for approximately 40 percent of its manufacturing of market accessibility stems partly from the poor quality of employment and 53 percent of its manufacturing value. The roads, which undermines connectivity between rural areas region also contains 55 percent of the country’s manufactur- and urban markets (World Bank 2017). Insufficient roads limit ing establishments (66 percent excluding food, beverages, farmers’ access to input and output markets. Only 5 percent and tobacco). One study indicated that firms located near of rural households live less than 2 kilometers from a paved Dar es Salaam tend to be larger, offer better wages, and have road in poorer districts, and most trunk and tertiary roads are higher value added (Petracco et al. 2018). Dar es Salaam also in poor condition and are inaccessible during the rainy season accounts for a disproportionate share of the country’s port (World Bank 2012). In addition, because these roads are unre- traffic (~95 percent), dwarfing the importance of other cities as liable and inadequate, in many remote parts of the country, sources of imported manufactured goods (Storeygard 2016). post-harvest losses are high—an estimated 35 percent of total production. Limited and poorly maintained rural roads place The lack of market access significantly constrains rural a severe constraint on development of commercial agriculture farmers in Tanzania. For instance, farmers in rural areas with (World Bank 2017). limited market access pay higher prices for fertilizers because Poverty is more pronounced in tropical savannah zones, where agricultural productivity is low and tropical disease prevails. Climatic factors play an important role in determining geo- factor share of crop production and is deemed a critical ­ graphic inequalities in welfare and poverty. In Tanzania, for poverty reduction in Tanzania—is concentrated in the poverty is more pronounced in the tropical savannah zones, southern highlands (e.g., Iringa and Mbeya), the southwestern ­ which are dominant features of the northwest and southeast region (e.g., Shinyanga, Rukwa), and Arusha, which all have of Tanzania (Figure 5.16A). Although it is difficult to disen- country’s a largely nontropical climate and are home to the ­ tangle factors that explain geographic inequalities in pov- most fertile land (Luhunga 2017). erty between the tropical and non-tropical zones, researchers have shown that the tropical climate zones are unfavorable to Tropical zones are also prone to various diseases, with growth and poverty reduction for several reasons.7 important implications for health and labor productivity. Although there has been a rapid decrease in the prevalence of Tropical zones tend to have lower agricultural productiv- malaria in Tanzania, primarily due to sustained high immunization ity than temperate zones. For instance, major crops such as coverage and malaria prevention initiatives, it remains one of the maize, rice, and wheat tend to grow better in temperate and leading causes of death in children and mothers, with an esti- subtropical climates than in tropical zones. High tempera- mated 10 million cases reported in 2010 (World Bank 2017). tures, which mineralize organic materials, and heavy rainfall, which leaches them out of the soil, reduce soil quality (Sachs Malaria, which decreases human capital and productiv- et al. 2001). These unfavorable factors in tropical zones partly ity, is particularly common in the tropical savannah zones. explain why maize production—which accounts for the largest Figure 5.16B shows geographic disparities in the rate of 7 Hausmann (2001) shows that, on average, annual economic growth rates in tropical nations are 0.5 percent to 1 percent lower than in temperate countries. 122 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 5.16: Poverty vs. Tropical Climate and Diseases A. Köppen Climate Classification B. Plasmodium falciparum parasite C. Poverty and Tropical Savannah Incidence D. Malaria Incidence and Tropical Savannah Incidence 0.8 0.4 Malaria prevalence 0.6 0.3 Poverty incidence 0.4 0.2 0.2 0.1 0 0 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 % of Tropical Savannah area % of Tropical Savannah area E. Poverty and Malaria Incidence 0.8 0.6 Poverty incidence 0.4 0.2 0 0 0.1 0.2 0.3 0.4 Malaria prevalence Sources: Köppen Climate Classification data from Kottek et al. (2006); poverty estimates from the 2018 HBS; the Malaria Atlas Project. Notes: Figure 5.16B shows the proportion of the population aged 2 to 10 found to carry asexual blood-stage parasites calculated for the period 2000 to 2015. In figure 5.16E, malaria prevalence in the scatterplot corresponds to the maximum grid value of Plasmodium falciparum parasite rate in each district. C h a p t e r 5 G e o g r a p h i c D i m e n s i o n s o f P ov e r t y 123 infection with the Plasmodium falciparum parasite—a com- and welfare effect of malaria is substantial in Tanzania. Malaria monly used index of malaria transmission intensity that refers decreases learning capacity of individuals aged 5 to 25 and to the proportion of the population aged 2 to 10 found to undermines economic productivity of those aged 15 to 55 carry asexual blood-stage parasites. Tropical areas have (Mboera et al. 2007). More than 1 percent of GDP is devoted higher rates of malaria (Figure 5.16D), which is positively cor- to the disease (US$ 2.20 per capita), accounting for 39 percent related with poverty incidence (Figure 5.16E). The economic of total national health expenditures (Sicuri et al. 2013). The most impoverished areas are also prone to significant natural disaster risks such as drought. The climate has important implications for the poor’s vulner- settlements and economic activity in Dar es Salaam, Dodoma, ability to natural disaster risk. Although this analysis does not and other secondary cities since 2012, but poverty rates directly examine the correlation between poverty and natural across districts remain highly unequal, and when examining disaster risk, many of the pockets of poverty in the country are district-level estimates, four major pockets of poverty remain. vulnerable to such risk. Among the most frequent disasters that Urbanization is the primary driver of differences in welfare hit these poverty-stricken areas is drought, which has significantly across districts, although climatic conditions, natural disasters, contributed to food insecurity, livestock and agricultural loss, and and travel infrastructure play significant roles. Educational infectious disease transmission in Tanzania. attainment is also strongly correlated with poverty across districts. Overall, efforts to build human capital and increase Drought conditions are frequently observed in the north- access to urban areas would help the poorest districts reap ern (Arusha, Tanga, Manyara, Kilimanjaro, Mara), central more benefits from the country’s growth. (Dodoma, Morogoro), and southeastern (Mtwara, Lindi) regions. These geographic disparities in drought risks are FIGURE 5.17: Drought Hotspots depicted in figure 5.17, which shows the average level of drought severity measured using a composite index of drought length and dryness (with higher numbers indicating greater drought severity). Among the most drought-stricken areas is Longido dis- trict in Arusha, which is also estimated to have the highest pov- erty rate in the country. For instance, drought that hit the district in 2008/09 resulted in severe food insecurity and a significant loss of livestock (Bowen et al. 2010). Climate change exacerbates damage due to drought. Some of the previously highly productive areas of Tanzania such as the southern and northern highlands are expected to experience declining rainfall, frequent droughts, and sig- nificant increases in geographic and temporal variability of rainfall. These climatic risks will directly affect agricultural pro- ductivity because greater volatility in rainfall and prolonged droughts make cropping patterns unpredictable and induce ecological changes that encourage pests and diseases. Climate change will also result in shifts in agro-ecological zones, whereby particular crops will become unsuitable. Prolonged dry episodes may result in food shortages, which Source: Aqueduct Global Maps 2.1 (Gassert et al. 2014). damage people’s health and productivity (Irish Aid 2017). Note: Drought severity is the average length of drought times the dryness of the droughts from 1901 to 2008. Drought is defined as a continuous period when soil moisture remains below the 20th percentile. Length is measured in months, and Geographic inequalities in Tanzania remain a major dryness is the average number of percentage points by which soil moisture drops ­challenge. Images of lights at night suggest expansions in below the 20th percentile (Sheffield and Wood 2007). 124 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T CHAPTER 6 Agricultural Households and Nonfarm Enterprises I. Overview More people are engaged in nonfarm enterprises (NFEs). Despite an overall increase in the percentage of work- Between 2012 and 2018, the percentage of people who ing-age individuals running businesses, the proportion of reported having run a business in the previous 12 months women involved in farm enterprises and NFEs decreased increased from 17.5 ­ percent, and the per- percent to 19.7 ­ over time. In 2012, 54.4 ­percent of employed working-age centage of individuals engaged only in NFEs increased from individuals who engaged only in farming were women. This percent in 2012 to 10.5 ­ 6.2 ­ percent in 2018.1 This suggests that percent by 2018 (Table 6.2). The proportion of fell to 48.6 ­ working-age Tanzanians are shifting away from agriculture as women engaged in NFEs also fell across the country, except their sole economic activity and increasingly diversifying into in rural areas, where the percentage of women engaged in NFEs (Table 6.1). NFEs increased from 37.2 ­ percent in percent in 2012 to 43.6 ­ 2018. In contrast, in urban areas, female participation in NFEs decreased from 56.6 ­ percent. percent to 50.7 ­ Nonfarm households are smaller and have fewer dependents, higher education, and higher living standards. Households that operated NFEs had fewer children and of 11 ­percent for all household heads) (Figure 6.1D). In con- fewer household members. The average size of a household trast, households that participated primarily in farming activ- that operated an NFE in 2018 was 4.9, considerably smaller ities while also operating businesses had mostly completed than the average household size for those involved in NFEs only primary school; their primary school completion rate and farming (5.4 members) and those that rely on farming only was 57 ­percent in 2018 (Figure 6.1C). This suggests that bet- (6.3 members) (Figure 6.1A). NFE-only households had fewer ter-educated individuals are more likely to operate NFEs than children on average than other households, an average of 2, to engage in agricultural activities, underscoring the role of compared with 2.5 in households primarily engaged in farm- secondary education in accessing productive employment ing supported by NFE and 2.9 in households that engaged opportunities and escaping poverty in Tanzania. only in farming (Figure 6.1B). Nevertheless, because farm- ing is a labor-intensive activity (especially in areas with lim- Households that operated NFEs in urban areas were the ited resources) that relies heavily on unpaid household labor, best off. They consumed 2.5 times as much as households households with fewer children and thus less access to a pool with the lowest average monthly consumption—households of unpaid labor may find it feasible to operate only NFEs in rural areas that engaged only in farming (Figure 6.1F). as their main economic activity. That said, it seems like, in In rural and urban areas, households that participated Tanzania, operating NFEs is more a function of higher living primarily in farming but also operated NFEs had higher levels standards than of scarcity of cheap labor. of consumption than households that participated only in farming, suggesting a positive correlation between income Households that solely operated NFEs had more-educated and likelihood of operating an NFE, although it is not clear heads. In 2018, heads of households that only operated busi- whether NFEs help boost household incomes or those with nesses had the highest lower secondary school comple- higher income and consumption level are better able to tion rates (17 ­ percent, compared with the national average engage in NFEs. NFE ownership is defined as operation of an NFE involved in the provision of nonagricultural services; processing, production, and sale of agricultural by-products; trade; 1 professional services; mining; and agricultural services. An individual is considered to be operating an NFE if they report operating a business at any point in the 12 months before the survey. 126 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE 6.1: Participation in Farm and Nonfarm Activities, TABLE 6.2: Participation in Farm and Nonfarm Activities, 2012–18, ­percent 2012–18, ­percent 2012 2018 2012 2018 FARM FARM WITH NONFARM FARM FARM WITH NONFARM FARM FARM WITH NONFARM FARM FARM WITH NONFARM ONLY NONFARM ONLY ONLY NONFARM ONLY ONLY NONFARM ONLY ONLY NONFARM ONLY COMPLEMENT COMPLEMENT COMPLEMENT COMPLEMENT Tanzania mainland Tanzania mainland Households 46.4 11 9.4 37.3 9.6 14.4 Women 54.4 50.1 51.9 48.6 48.9 48.1 Individuals 57.7 8 6.2 37.8 7.3 10.5 Men 45.6 49.9 48.1 51.4 51.8 51.9 Rural Rural Households 60.3 14.6 2.4 52.7 13 8.7 Women 54.1 49.6 37.2 48.7 46.6 43.6 Individuals 69 10 1.6 51.9 9.6 6 Men 45.9 50.4 62.8 51.3 53.4 56.4 Urban Urban Households 25 4 22.6 12.4 4.1 23.6 Women 56.4 55.1 56.6 48.3 56.3 50.7 Individuals 36.5 3.1 16.8 12.2 3.3 18.7 Men 43.6 44.9 43.4 51.7 43.7 49.3 Sources: HBS 2011/12 and 2017/18. Sources: HBS 2011/12 and 2017/18. FIGURE 6.1: Socio-economic Status of Farm and Nonfarm Households, 2018, ­ percent  A. Average Household Size of Farm and NFE Households B.  Average Number of Children Younger than 15 in Farm and NFE Households 8 3.5 6.5 6.3 3 2.9 5.5 5.2 5.8 3.0 2.6 5.2 5 5.4 2.5 6 4.9 2.5 2.3 2 2 2 2.0 1.7 4 1.5 2 1.0 0.5 0 0 Rural Urban National Rural Urban National Farm only Farm with NFE complement NFE only Farm only Farm with NFE complement NFE only C. Household Heads Completing Primary Education D. Household Heads Completing Lower Secondary Education 80 25 70 65 20.6 57.3 57 56.5 55.9 20 17 60 54.9 54.2 48.3 49.4 15.5 50 15 11.5 11.1 40 9 30 10 7.6 20 4 4.9 5 10 0 0 Rural Urban National Rural Urban National Farm only Farm with NFE complement NFE only Farm only Farm with NFE complement NFE only C h a p t e r 6 A g r i c u lt u r a l H o u s e h o l d s a n d N o n fa r m E n t e r p r i s e s 127 FIGURE 6.1E Literate Household Heads FIGURE 6.1 F Average Monthly Household Consumption (TZS) 100 91.7 91 87.2 150,336.20 81.5 81.2 78.5 National 98,944.30 76 80 67.9 74,154.70 66.1 177,742.50 60 Urban 143,822.30 93,269.94 40 109,630.30 20 Rural 90,284.40 71,623.90 0 n 0 0 0 0 0 0 0 0 0 20 0 l 0 l na ra ba 0 0 0 0 00 00 00 00 00 00 ,0 ,0 ,0 ,0 Ru io 0, 0, 0, 0, 0, 0, Ur 20 40 60 80 at 10 12 14 16 18 N Farm only Farm with NFE complement NFE only Farm only Farm with NFE complement NFE only Source: HBS 2017/18. II.  Farm Households Overview of cash and staple crops. Between 2012 and 2018, the overall sale of cash crops top five cash crops were beans, paddy rice, sesame, cot- fell approximately 10 percentage points, and the propor- ton, and coffee (Table 6.3). By 2018, cassava had become the tion of farmers growing any of the top five cash crops fell most prevalent cash crop, followed by paddy rice, ground- 3.3 percentage points. A crop is defined as a “cash crop” nuts, sunflowers, and bananas. Of staples, maize was the most when more than half of the farmers who grow it sell it; other- grown crop in 2012 and 2018. Although some farmers sold wise, the crop is considered a staple (Guirkinger et al. 2015). maize, it was considered a staple crop because fewer than In 2012, farming households in Tanzania were selling an aver- 40 ­percent of farms were selling it in 2012 and 2018. In 2012, age of 72 ­percent of their total production of cash crops; by approximately 80 ­ percent of farms were growing maize, and 2018, that had fallen to 62 ­percent. In 2012, approximately 37 ­percent were selling it, these proportions decreased to 56 ­ percent of farming households were growing one or more 61 ­percent and 30 ­percent, respectively, in 2018. The reduc- of the top five cash crops. By 2018, this fell to 52.6 ­ percent. tion in the percentage of farmers selling maize dropped it Together these suggest a contraction in the larger market for from the second most sold to the least sold crop of the top cash crops, with some farming households exiting the market five staples sold in the market. and those who continued to operate selling a smaller share of their output. As the proportion of cash crop sales declined, the over- all value of the crops sold also declined. In 2012, the value The constitution of top five cash crops changed signifi- of the top five cash crops sold was almost six times that of cantly between 2012 and 2018, except for paddy rice, the crops consumed, indicating that farming households con- which retained its position as the second-most-important sidered it more profitable to sell those crops than to con- cash crop over these years. The top five cash and staple sume them; by 2018, the value of the top five cash crops sold crops for each year are determined according to the pro- was only about twice as much as the value of cash crops con- portion of farmers growing and selling them. In 2012, the sumed (Table 6.4). This suggests that, although they were still TABLE 6.3: Top Five Cash and Staple Crops, 2012 and 2018, ­ percent 2012 2018 PERCENTAGE OF FARMERS PERCENTAGE OF GROWING PERCENTAGE OF FARMERS PERCENTAGE OF GROWING CROP GROWING CROP FARMERS SELLING CROP CROP GROWING CROP FARMERS SELLING CROP Staple crops Maize 79.6 36.9 Maize 61 30.2 Cassava 20.7 24.9 Beans 27.9 39.3 Sweet potatoes 12.4 14.0 Sweet potatoes 17.7 45.9 Groundnuts 13.4 43.9 Sorghum 8.9 37.4 Sorghum 9.4 20.8 Cowpeas 6.5 37.8 Total (all five crops) 135 Total (all five crops) 115 Cash crops Beans 30.6 50.3 Cassava 18 60.1 Paddy rice 20.1 57.0 Paddy rice 15.5 62.3 Sesame 6.8 90.1 Groundnuts 14.3 50.1 Cotton 6.7 90.5 Sunflowers 9.7 66.2 Coffee 5.3 87.9 Bananas 6.9 61.1 Total (all five crops) 67.8 Total (all five crops) 49.7 Sources: HBS 2011/12 and 2017/18. percent because some farmers farm multiple crops and may have multiple farms. Note: percentage may be greater than 100 ­ C h a p t e r 6 A g r i c u lt u r a l H o u s e h o l d s a n d N o n fa r m E n t e r p r i s e s 129 selling the cash crops, farming households had begun to con- In 2012, most of the labor for cash and staple crops came sume a larger share. Farmers may have been selling more of from household members, with hired labor brought on during lower-value crops in 2018 than in 2012, but they were essen- the harvest season and other times of intense activity on the tially consuming a larger share (and selling a smaller share in farm.2 Sorghum is the most labor-intensive crop overall, and the market) of their cash crops. sweet potatoes are the least (Figure 6.2, Table 6.5), although not all cash crops are more labor intensive than all staple Cash crops in Tanzania are on average more labor intensive crops (Table 6.5); on average, the top five cash crops require than staple crops, requiring more paid and unpaid labor. 71 total days of labor per year, and the staple crops require 66 total days of labor. TABLE 6.4: Average Value Sold and Consumed of Top Five Cash Crops, 2012 and 2018, TZS TABLE 6.5: Mean Days of Labor According to Crop, 2012 AVERAGE VALUE AVERAGE VALUE VALUE SOLD AS SHARE OF TOTAL MEAN DAYS PER FARM PER YEAR SOLD CONSUMED CROP HARVESTED (%) CROPS PAID LABOR HOUSEHOLD LABOR TOTAL LABOR 2012 Cash crops Beans 96,686.9 63,519.1 58.9 Beans 2.6 44.7 47.3 Paddy rice 333,207.5 200,253.9 57.8 Paddy rice 7.6 76.8 84.4 Sesame 353,693.2 12,560.7 96.4 Sesame 3.1 77.0 80.1 Cotton 622,179.6 48,805.4 98.5 Cotton 3.1 58.7 61.8 Coffee 603,611.9 15,760.5 96.1 Coffee 3.0 79.1 82.0 Total 2,009,379.1 340,899.5 62.1 Total (all five crops) 3.88 67.26 71.12 2018 Staple crops Cassava 283,406.3 254,172.3 59.4 Maize 4.1 71.9 76.1 Paddy rice 759,522.8 251,860.6 51 Cassava 3.0 64.4 67.3 Groundnuts 103,453.9 58,745.4 59.9 Sweet potatoes 0.6 36.8 37.4 Sunflowers 147,663.5 77,898.2 81.5 Groundnuts 1.5 46.4 47.8 Bananas 258,570.7 174,168.5 54 Sorghum 1.8 97.6 99.5 Total 1,552,617.2 816,845.0 31.3 Total (all five crops) 2.2 63.42 65.62 Sources: HBS 2011/12 and 2017/18. Source: HBS 2011/12. FIGURE 6.2: Labor Intensity According to Crop, 2012 Labor Intensity Per Crop 98 99 100 90 84 77 77 80 79 82 76 80 72 70 64 67 59 62 60 50 45 47 46 48 40 37 37 30 20 8 10 3 3 3 3 4 3 1 2 1 0 Beans Paddy Rice Sesame Cotton Coffee Maize Cassava Sweet Groundnut Sorghum Potatoes Mean Days of Paid Labor Per Farm Per Year Mean Days of Household Labor Per Farm Per Year Mean Total Days of Labor Per Farm Per Year Source: HBS 201/12. 2 Data on labor according to crop were not available in HBS 2017/18. 130 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Characteristics of farming households Households that farm cash crops tend to be slightly more more cash crops. Similarly, in 2018, 82 ­percent of cash crop– educated than farming households that farm staple crops. farming households were classified as nonpoor, compared with Approximately 68 ­ percent of cash crop–farming household 77 ­percent of staple crop–farming households. Nevertheless, it heads had completed up to primary education, compared with is difficult to conclude whether households became less poor 66 ­ percent of staple crop–farming household heads. This sug- as a result of growing cash crops, or if they choose to grow cash gests that the level of education of the household head does crops because they were better off. not play a significant role in whether a household chooses to farm cash crops. Households that farmed only cash crops had higher aver- age monthly consumption rates than those who grow Poor households tend to produce more staple crops and are staple crops. The average monthly consumption level of less likely to sell their crops in the market. In 2018, 70 ­ percent households that farmed only cash crops was nearly 30 ­ percent of poor households farmed staple crops, and poor farmers were higher than that of staple crop–farming households selling on average only 22 ­ percent of their crops. This is a reduc- (Figure 6.3). This supports the idea that households that are tion from 2012, when 95 ­ percent of poor farmers grew staple growing and selling the majority of their crops are better off crops, suggesting that, over time, poor farmers started to grow than households that farm for subsistence. Access to infrastructure and farming households. Access to weekly or daily markets is limited for cash– Similarly, access to a road does not seem to significantly and staple crop–farming households. In 2018, 52 ­ percent increase the proportion of farming households that grow of farming households with access to a daily market, and cash crops. In 2018, 51 ­ percent of farming households with 53 ­percent of farming households with no access to a daily access to a road and 54 ­ percent of those with no access market produced cash crops. The similarity of these figures to a road produced cash crops. Likewise, access to a road suggests that proximity to a market is not the main factor in seems not to strongly affect decisions to farm staple crops the decision of farming households to produce cash crops. percent of farming households with (Table 6.6). In 2018, 89 ­ Access to a daily market does not seem to affect the deci- access to a road and 92 ­ percent of those with no access to a sion of farming households to produce staple crops either road produced staple crops. percent of farming households with (Table 6.6). In 2018, 89 ­ access to a daily market and 90 ­ percent of farming house- This does not mean that access to infrastructure does holds with no access to a daily market produced staple crops. not matter. The figures need to be interpreted with ­ caution, because households could grow more-profitable crops or increase their productivity and revenues if they had better FIGURE 6.3: Average Monthly Household Consumption access to markets and roads. Other factors (e.g., experience, According to Grown Crop Type, 2018, TZS knowledge, production) may have influenced the decision to 120,000 grow cash crops in situations of overall limited access to infra- 108,611.70 structure, but profitability and productivity would have been 98,471.21 100,000 higher with better infrastructure. 85,702.02 80,000 60,000 TABLE 6.6: Prevalent Types of Farming Households in percent Areas with Markets and Roads, 2018, ­ 40,000 CROPS NO DAILY DAILY NO WEEKLY WEEKLY NO TARMAC TARMAC 20,000 MARKET MARKET MARKET MARKET OR TRUNK OR TRUNK ROAD ROAD 0 Cash only 52.9 51.5 51.4 55.9 54.4 51.4 Cash and Staple Staple crops Cash crops only Staple only 90.5 89.1 89.8 90.9 91.9 88.9 crops only Source: HBS 2017/18. Source: HBS 2017/18. C h a p t e r 6 A g r i c u lt u r a l H o u s e h o l d s a n d N o n fa r m E n t e r p r i s e s 131 III.  Nonfarm Enterprise Households NFE sectors Because of a change in the survey questionnaire and processing of grains, tubers, and other products, indicating methodology, comparison of NFE sectors in 2012 and that the agro-processing industry is active in Tanzania. 2018 is not feasible. The International Standard Industrial Classification shows that, in 2012, the most prevalent NFEs The most common NFE sectors varied considerably specialized in retail sale of nonfood items that were not depending on whether the NFE owners were in rural or available in stores, stalls, or markets (Table 6.7). The Central urban areas. Although NFEs related to the production, pro- Product Classification shows that, in 2018, the most wide- cessing, and sale of agricultural products were most prevalent spread NFEs were engaged in the sale of agricultural, horti- in both urban and rural areas, retail trade services and the sale cultural, and market gardening products. This includes the of meat and fish products were two of the top five NFE sec- sale of cereals, vegetables, fruits and nuts, tubers, and pulses. tors in urban areas (Figure 6.4). A smaller share of urban NFEs The second most widespread NFE sector in 2018 was the were engaged in the sale of agricultural, horticultural, and market gardening products. Characteristics of NFEs The majority of NFEs in 2018 were mobile, with no fixed percent of NFEs that had no fixed location, 17 ­ 24 ­ percent pro- location, or were set up in the owner’s house with a des- vided business and production services. Of the 23 ­ percent of ignated space for conducting business activities. Of the NFEs whose activities took place at homes with a separate percent provided agricultural, forestry, and fishing space, 32 ­ products. TABLE 6.7: Top Five Sectors of Nonfarm Enterprises, 2012 and 2018, ­percent In 2018, savings from agricultural sales were the most SECTOR % popular source of start-up funds for NFEs. There was a con- 2012 Other retail sale not in stores, stalls, or markets 13.3 siderable difference in the sources of start-up funds for NFEs Other retail sale in nonspecialized stores 11.7 in rural and urban areas. NFEs in rural areas relied on pro- Retail sale of food in specialized stores 9.9 ceeds from agricultural activities for start-up funds, whereas Retail sale in nonspecialized stores with food, beverages, or tobacco predominating 9.7 most of those in urban areas were initially funded from own- Retail sale via stalls and markets of food, beverages, and tobacco products 6.8 er’s savings. The share of NFEs funded by owner’s savings fell Total 51.5 between 2012 and 2018, and the share of NFEs funded by 2018 SACCOS loans, bank loans, and loans from family and friends Agricultural, horticultural, and market gardening products 20.3 grew. The proportion of funds from loans from financial insti- Grain mill products, starches and starch products, other food products 10.0 tutions in total NFE start-up funds increased from 2.2 ­percent Meat, fish, fruits, vegetables, oils, and fats 5.9 None 5.8 percent in 2018 (Table 6.8). Although the share in 2012 to 3.2 ­ Retail trade services 5.0 remains marginal, it suggests that investors are relying less on Total 47.1 their own income and savings and increasingly using available Source: HBS 2011/12 and 2017/18. sources of credit. 132 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE 6.4: Top Five Nonfarm Enterprises Sectors, 2018, ­ percent A. Rural areas B. Urban areas Products of agriculture, horticulture Products of agriculture, horticulture 24.2 15.3 and market gardening and market gardening Grain mill products, starches and Grain mill products, starches and 9.2 11.1 starch products; other food products starch products; other food products Professional, technical and business services (except research, development, 5.6 Retail trade services 8.8 legal and accounting services) Beverages 5.5 Meat, sh, fruits, 7.5 vegetables, oils and fats Forestry and logging products 4.9 Unknown 7.4 0 5 10 15 20 25 30 0 5 10 15 20 Source: HBS 2017/18. TABLE 6.8: Most Predominant Sources of Start-Up Funds for Nonfarm Enterprises, 2012 and 2018, percentage of Total Funding 2012 2018 TANZANIA RURAL URBAN TANZANIA RURAL URBAN Income from agricultural production 32.3 46.5 9.5 33.2 48.5 12.9 Own savings 36.3 23.7 56.4 33.1 22.7 47.0 Other 8.7 10.9 5.1 9.9 9.4 10.5 Loan from family of friends 5.4 4.8 6.4 9.5 6.6 13.2 Gift from family or friends 10.3 7.7 14.3 5.1 5.5 4.6 Income from nonagricultural production 1.9 2.7 0.7 2.9 3.6 1.9 Inheritance 2.0 2.1 1.9 2.6 1.8 3.6 Loan from Savings and Credit Co-Operative Society 1.2 0.5 2.3 2.0 1.1 3.3 Sale of assets owned 0.8 0.7 0.9 0.5 0.6 0.5 Bank loan 1.0 0.3 2.2 1.2 0.3 2.5 Sources: HBS 2011/12 and 2017/18. The share of NFEs funded by loans from SACCOS market does not significantly affect the incidence of NFEs. increased slightly. Despite limited overall use, the share of Areas with weekly markets also showed similar trends, and funds from SACCOS loans in NFE start-up funds increased there was no appreciable difference in the types of NFEs in from 1.2 ­ percent in 2012 to 2 ­ percent in 2018. The use of areas with and without weekly markets. SACCOS loans was slightly higher for female NFE owners in urban areas (3.2 ­ percent in 2012, 2.8 ­ percent in 2018). NFEs have little effect on other households because only a few NFEs employ workers that are not members of their The presence of a daily or weekly market did not have a own households. NFE workers tended to be the owner of strong influence on type of NFE in 2018. Table 6.9 high- the business themselves or the owner’s household members. lights the effect of daily markets, weekly markets, and roads percent of NFEs were working proprietorships, In 2018, 26.5 ­ on the incidence of NFEs in rural and urban areas. In areas meaning that the business owners and partners were actively with no daily markets, 29 ­ percent of NFEs were engaged in engaged in the work of the business, and 25.2 ­percent of production of agricultural, forestry, and fishing products, com- NFEs reported having unpaid household members working in pared with 32 ­percent of NFEs in areas with a daily market. percent of NFEs hired paid labor in the their business. Only 7 ­ This difference of 3 pp suggests that the presence of a daily form of non-household members. C h a p t e r 6 A g r i c u lt u r a l H o u s e h o l d s a n d N o n fa r m E n t e r p r i s e s 133 TABLE 6.9: Incidence of Nonfarm Enterprises in Communities with and without Markets and Roads, 2018, ­ percent NO DAILY MARKET DAILY MARKET NO WEEKLY MARKET WEEKLY MARKET NO TARMAC OR TRUNK ROAD TARMAC OR TRUNK ROAD Agriculture, Forestry and Fishery Products 29.4 32.1 29.1 34.1 34.9 28.1 Ores and Minerals; Electricity, Gas and Water 1.3 1.6 1.4 1.3 1.5 1.3 Food Products, Beverages and Tobacco… 22.2 22.4 21.9 23.4 17.9 24.2 Other Transportable Goods, Except Metal Products, Machinery and Equipment 1.7 1.3 1.8 1.0 1.5 1.7 Metal Products, Machinery and Equipment 1.1 1.6 1.0 2.0 1.4 1.1 Constructions and Construction Services 2.7 1.8 2.6 1.8 2.1 2.6 Distributive Trade Services; Accommodation 9.5 6.1 10.1 3.0 6.8 9.3 Financial and Related Services… 0.9 0.2 0.8 0.1 0.6 0.7 Business and Production Services 14.8 19.2 15.4 18.6 18.4 15.0 Community, Social and Personal Services 7.3 7.4 7.1 8.0 6.2 7.8 None 6.2 4.3 6.1 3.9 5.0 5.9 Source: HBS 2018. FIGURE 6.5: Age of Nonfarm Enterprises, 2012 and 2018, ­ percent A. 2012 B. 2018 0.6 3.7 5.9 6.8 25.2 Less than a year Less than a year 19.7 30.3 1 to 2 years 1 to 2 years 3 to 5 years 3 to 5 years 6 to 10 years 6 to 10 years 38.8 More than 10 years More than 10 years Non responses 11.4 Non responses 17.1 23.1 17.4 Source: HBS 2011/12 and 2017/18. In 2018, only 11 ­percent of NFEs were registered with share of NFEs that had been established for six years or the Business Registrations and Licensing Agency in percent to 31 ­ longer before the survey fell from 42 ­ percent Tanzania—a slight decrease from 2012, when 11.7 ­ percent (Figure 6.5). Although this change may be because of of NFEs were registered with the agency. This may suggest the increase in businesses established between 2016 the persistence of informality among businesses. and 2018, it is also possible that many of the businesses established in 2012 were not functional in 2018. This raises In 2012 and 2018, the majority of NFEs were approxi- concerns regarding the sustainability and longevity of mately one to two years old. Between 2012 and 2018, the businesses in Tanzania. 134 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Characteristics of NFE households and owners Heads of NFE households were on average younger, more Gender, education, and location of the NFE owner likely to be married, and more educated than heads of affected average monthly revenue from NFEs consider- farming households. In 2018, heads of households that oper- ably.3 Female NFE owners earned on average less than half ated NFEs were three years younger on average than those of what male NFE owners earned, and NFE owners who had not operating NFEs (45 vs 48). In addition, the proportion completed lower secondary school had average monthly rev- of married heads in NFE households was 7 ­ percent higher enues more than two and a half times those of NFE owners than in households that did not operate an NFE. Moreover, who had completed only primary school (Figure 6.7A). NFEs in a higher proportion of NFE-operating household heads had rural areas earned the lowest average monthly revenue, high- completed secondary education, whereas farming household lighting the effect of location on NFE income. heads tended to have less education (Figure 6.6). FIGURE 6.6: Highest Level of Education Completed by Household Head, 2018, ­ percent 60 55.5 50 46.7 40 30 20 14.9 13.4 12.1 10.5 10 3.3 3.7 0.6 0.3 1.7 1.8 0 No education Less than Completed Completed lower Completed upper Completed completed primary primary secondary secondary university Non NFE household NFE household Source: HBS 2017/18. FIGURE 6.7: Average Monthly Nonfarm Enterprise Revenue According to Owner Characteristics, 2018, TZS A. Demographic characteristics 1,000,000 898,128 900,000 800,000 692,501 700,000 600,000 562,432 500,000 409,298 400,000 378,364 300,000 236,250 218,202 200,000 100,000 0 Average Female Male Rural Urban Primary Lower secondary All Owner's gender Location Owner's education Continues next page 3 Monthly NFE revenue includes revenue from the sale of products and goods, construction work, and services. C h a p t e r 6 A g r i c u lt u r a l H o u s e h o l d s a n d N o n fa r m E n t e r p r i s e s 135 FIGURE 6.7B: NFE age 602,110 600,000 500,000 480,276 434,961 400,000 300,000 257,686 206,897 200,000 117,525 100,000 0 Less than a year 1 to 2 years 3 to 5 years 6 to 10 years More than 10 years Non responses Source: HBS 2017/18. NFE monthly revenues generally increased with age. (Figure 6.7B). This suggests that NFEs that survive become NFEs that were 6 to 10 years old recorded the highest more profitable up to a certain age after which profitability average monthly revenue, although revenues were lower for starts to decline. NFEs that had been in operation for longer than 10 years 136 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T APPENDIX A Survey Description Official estimates of poverty in (mainland) Tanzania are food and non-food consumption transactions that occurred based on the Household Budget Surveys (HBS), which go over the course of 28 days for the first two and 14 days for back to the early 1990s. The HBSs are a series of repeated the latter, including consumption of self-produced items. cross-sectional surveys conducted by the Tanzania National Bureau of Statistics (NBS). As shown in Table 1.A-1 there Enumerators visited the households regularly to check have been four HBS rounds so far - 1991/92, 2000/01, 2007, and code the individual records. The HBSs further 2011/12 and 20117/18. All HBS collect data on household included a recall module for non-food expenditures, par- consumption, demographics (including education and ticularly (semi-)durables and other irregularly purchased health), asset ownership, housing, etc. The most recent items. 2011/12 and 2017/18 HBSs also contained a detailed labor The HBS instrument has evolved over time and there force and agricultural module. were significant changes between the HBS 2007, HBS There exists a second survey series suitable for poverty 2011/12 and HBS 2017/18. The first two used paper analysis, the National Panel Survey, which has had four method for data collection and the latter used Computer rounds so far (2008/09, 2010/11, 2012/13 and 2014/15). Assisted Personal Interviewing (CAPI) method, expect for The NPS is a longitudinal survey (tracking individuals) con- diary which was based on paper. While the 2007 HBS recall ducted every two years by the NBS and has a smaller sample module for non-food consumption was designed mainly size than the HBS. However, the panel nature of the data to capture expenditures on semi-durable and durable makes it a particularly attractive survey for studying poverty goods and only probed for a limited number of item cat- dynamics and transitions. Due to differences in the way the egories, the 2011/12 HBS included a much more detailed HBS and NPS surveys capture consumption we follow the and broader recall module. The recall module was further NBSs approach in that we draw (mainly) on the HBS data to expanded for HBS 2017/18. There is some evidence of measure poverty trends over time, though we make use of better supervision over time, which could have affected the the NPS to analyze poverty movements and dynamics. capture of food consumption in the diary. Finally, the 2007 and 2011/12 were representative at only Dar es Salaam, The 2007, 2011/12 and 20117/18 HBSs were imple- other urban, and rural levels while the 2017/18 HBS was mented during 12 months (HBS 2017-18 collected data representation at the regional (26 regions) level despite from December 2017 to November 2018). They use a maintaining the sample size at around 10,000 households. diary approach to collect consumption, where every indi- This was the result of significant improvement of the sam- vidual in a household is asked to record (on a daily basis) all pling procedure. Poverty line It was decided in consultations with the Tanzania (CAPI for nonfood recall and nonmonetary dimensions National Bureau of Statistics (NBS) that a new pov- and paper (PAPI) for diary food, 14 days diary). Other erty line (food and basic needs) will be estimated changes include: coverage of food and nonfood basket for HBS 2017/18 due to major changes in the survey items, which are larger in 2017/18 HBS; and changes in the methodology compared to HBS 2011/12 in terms COICOP coding of food basket items, using a more disag- of sample design, methodology of data collection gregated and detailed coding. Price changes over time Two methods can be used to adjust for price changes i) Unit values: fisher and Paasche price indices were esti- over time and estimate the real changes in consumption. mated respectively in HBS 2012 and 2018 based on unit The first is based on unit values and the second on official values (value/quantity) of a food basket of over 160 Consumer Price Index (CPI); items. These items represent over 95 percent of total food 138 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T consumption. Survey based deflators generally better reflect This can be due to the fact that inflation is higher in urban the spending behavior, particularly of the poor, as well as centers than rural areas. It may also not reflect the spend- temporal and regional changes/differences in the cost of ing behaviors at the national level and in particular in rural living. Based on this deflator, price increased by about 26 areas- where food consumption share and poverty are the percent between 2012 and 2018. highest. ii) Food CPI. This deflator has two major shortcomings. iii) Official CPI. This deflator includes food and non- First, it only includes food prices collected in main urban food items. While nonfood CPI suffers less shortcomings markets. Second it is based on a different coding of food than food one, given that nonfood urban prices are less items than those in the HBS. This indicator shows a higher biased, the coverage of main urban markets only remain inflation (around 55 percent) than survey food unit values. problematic. Robustness check In order to check the robustness of the methodology for one based on CPI food price, it reflects better the evolution estimating poverty levels and changes of welfare over of food prices as it has a better coverage of food items con- time, two methods were used: sumed in both urban and rural areas. i) Estimate the ratio of the food poverty lines of 2018 ii) Estimate the ratio of basic needs poverty lines of 2018 and and 2012 and compare the changes with food prices 2012 and compare the changes with inflation rate based on changes. The indicator shows an increase of the food line CPI. Both indicators show an increase of prices of respectively of 29 percent, which corresponds to the inflation rate based 35 percent and 38 percent during this period, which supports on food unit values. While this inflation rate is lower than the the robustness of the new estimates of the poverty line. Post-stratification Population and household weights in the survey were weights for all other strata (including regional strata) were post-stratified based on the 2018 population projec- adjusted by a factor of 1.010189. Household weights tion. Population weights for Dar es Salaam were adjusted were post-stratified by dividing the population by house- by a factor of 1.2 (=5147070/4273781). Population hold size. Appendix A 139 APPENDIX B Drivers of Poverty Reduction FIGURE B.1: Drivers of Poverty Reduction in Rural Areas FIGURE B.2: Drivers of Poverty Reduction in Urban Areas From 2012 to 2018 From 2012 to 2018 0.2 0.2 Difference in log real per Difference in log real per adult consumption adult consumption 0 0 –0.2 –0.2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Quantiles Quantiles CI/endowment effect Endowment effect CI/endowment effect Endowment effect CI/returns effect Returns effect CI/returns effect Returns effect   EXTREME POOR POOR MIDDLE CLASS RICHEST   RURAL URBAN RURAL URBAN RURAL URBAN RURAL URBAN Total 0.009 0.093*** 0.04*** 0.103*** 0.026** 0.131*** 0.118*** 0.278*** Endowments 0.118*** 0.224*** 0.095*** 0.226*** 0.07*** 0.147*** 0.153*** 0.123** Access to basic services 0.092*** 0.187*** 0.076*** 0.125*** 0.05*** 0.017 0.058** -0.073 Education 0.002 0.003* 0.005*** 0.011** 0.005*** 0.017*** 0.011*** 0.028*** Assets 0.017 0.079*** 0.012*** 0.095*** 0.016*** 0.113*** 0.041** 0.157*** Head nonfarm employment 0.001* −0.004* 0.004** 0.005* 0.004*** 0.005** 0.009*** 0.015* Demographic Structure 0.002 0.016** 0.001 0.001 0 0.001 0.01*** 0.001 Returns −0.109*** −0.131*** −0.054*** −0.124*** −0.044** −0.016 −0.035 0.154** Access to basic services −0.015 0.004 0.024** 0.036** 0.036* 0.041 0.019 0.06 Education −0.049* −0.186*** −0.064*** −0.095** −0.013 0.004** 0.007* 0.015*** Assets 0.001 0.016 0.006* −0.038* 0.008* −0.004 0.007* 0.072** Head nonfarm employment −0.003* 0.018** −0.012** 0.015* 0.007*** 0.005* 0.007* 0.035** Demographic Structure 0.016 0.081*** 0.044** 0.021* 0.001 0.051** −0.005 0.066** Source: HBS 2007, HBS 2011/12 and HBS 2017/18. Note: Extreme poor are population groups in the bottom 10 percent of the distribution; the poor are in the third decile; middle class are in the fifth decile, and the richest are in the top decile. * Significant at the 10 percent level; ** significant at the 5 percent level; *** significant at the 1 percent level. Numbers in parentheses are bootstrap standard deviations based on 100 replications. Appendix B 141 FIGURE B.3: Drivers of Poverty Reduction in Dar es FIGURE B.4: Drivers of Poverty Reduction in Other Urban Salaam From 2012 to 2018 Areas From 2012 to 2018 0.6 0.6 0.4 0.4 Difference in log real per Difference in log real per adult consumption adult consumption 0.2 0.2 0 0 –0.2 –0.2 –0.4 –0.4 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Quantiles Quantiles CI/endowment effect Endowment effect CI/endowment effect Endowment effect CI/returns effect Returns effect CI/returns effect Returns effect EXTREME POOR POOR MIDDLE CLASS RICHEST DAR ES SALAAM OTHER URBAN DAR ES SALAAM OTHER URBAN DAR ES SALAAM OTHER URBAN DAR ES SALAAM OTHER URBAN Total −0.120** 0.176*** 0.017 0.167*** 0.059* 0.188*** 0.383*** 0.309*** Endowments 0.304*** 0.188*** 0.208*** 0.200*** 0.079 0.152*** −0.041 0.130* Access to basic services 0.191** 0.130** 0.093* 0.106** -0.036 0.032 −0.023 0.028* Education −0.008 0.008* 0.007 0.011** 0.10* 0.038** 0.052*** 0.015** Assets 0.011* 0.103*** 0.134*** 0.102*** 0.169*** 0.136*** 0.277*** 0.076** Head nonfarm employment −0.029* 0.017** 0.002 0.006* −0.001** 0.013** 0.003** 0.012* Demographic Structure -0.01 0.022** −0.001 0.001 −0.022* −0.003 0.026 0.002 Returns −0.423*** −0.012 −0.191*** −0.033 −0.020 0.036 0.425** 0.179** Access to basic services 0.14 0.082*** 0.055 0.089*** 0.035 0.077** −0.032 0.09 Education −0.860*** −0.172** −0.019** −0.018* 0.063** 0.07* 0.044** 0.015*** Assets 0.007 0.003* −0.002* 0.019** 0.101 −0.049* 0.124* 0.098 Head nonfarm employment −0.025* 0.034** 0.021* 0.013* 0.172** 0.011* 0.262** 0.023*** Demographic Structure 0.113* 0.089* 0.134** 0.018* 0.0033 0.004 0.046 0.072* Source: HBS 2007, HBS 2011/12 and HBS 2017/18. Note: * Significant at the 10 percent level; ** significant at the 5 percent level; *** significant at the 1 percent level. Numbers in parentheses are bootstrap standard deviations based on 100 replications. 142 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T APPENDIX C Distributional pattern of Growth in Urban and Rural Areas FIGURE C.1: Growth Incidence Curves in Rural Areas, FIGURE C.2: Growth Incidence Curves in Rural Areas, 2007–2012 2012–2018 0.8 8 0.6 6 0.4 4 0.2 2 0 0 –0.2 –2 –0.4 –4 0 20 40 60 80 100 0 20 40 60 80 100 Growth rate by percentile Growth rate in mean Growth rate by percentile Growth rate in mean FIGURE C.3: Growth Incidence Curves in Urban Areas, FIGURE C.4: Growth Incidence Curves in Urban Areas, 2007–2012 2012–2018 8 8 6 6 4 4 2 2 0 0 0 20 40 60 80 100 0 20 40 60 80 100 Growth rate by percentile Growth rate in mean Growth rate by percentile Growth rate in mean Sources: HBS 2007, HBS 2011/12 and HBS 2017/18. 144 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T APPENDIX D Structure of Inequality Decomposition of Inequality The static decomposition of inequality enables one to   µ  IBetw =  Σk f j log    IWithin = Σ k f j GE 0j explore how the differences in households’ characteristics  j =1  µj   j =1   affect the level of inequality and provide important clues for understanding the underlying and changing structure   µj   µj   of real per capita consumption distribution in Tanzania. IBetw =  Σk j =1f j   log    IWithin = Σ k j =1υ j GE1j   µ    µ   The decomposition follows the approach of Cowell and Jenkins (1995) and consists of separating total inequality with fj the population share, uj the consumption share, in the distribution of consumption into inequality between and mj the mean consumption of subgroup j; m total mean the different household groups in each partition, IBetw, and ­consumption, GE 0 k Theil_L index, and GE1k Theil_T index of the remaining within-group inequality, IWithin. As the most subgroup j. commonly decomposed measures in the inequality literature come from the General Entropy class, mean log deviation with: (Theil_L) and the Theil_T indices in real per capita monthly consumption expenditure are used to identify the contribution y  yi  y  Theil _ L = 1 n Σn i =1 log  and Theil _ T = 1 n Σn log  i  .   yi  i =1  y  y of between-group differentials to total inequality. The General Entropy inequality measures allow total inequality to be equal to IBetw + IWithin and the amount of inequality explained by house- yi: is real monthly per capita consumption expenditure for holds attributes (or group of attributes) is measured by IBetw/ household i and y is mean real monthly per capita consump- Itotal, where between and within group inequalities are defined, tion expenditure. respectively, for Theil_L and Theil_T indices as Inequality of opportunity The approach to estimate the degree of opportunity therefore be determined by the extent to which the con- inequality associated with the distribution of both con- ditional distribution of outcomes on circumstances, F(y|C), sumption and income is based on the framework of differs from F(y). Bourguignon et al. (2007). The method is based on the separation of the determinants of household’s outcome Inequality of opportunity can be estimated as the dif- (­ consumption or income), yi, into a set of circumstances vari- ference between the observed total inequality in the ables, denoted by the vector Ci ; efforts variables, denoted by distribution of consumption or income and inequality the vector Ei and unobserved factors, represented by ui. The that would prevail if there were no differences in cir- cumstances. Let F (y  ) be the counterfactual distribution of outcomes function can be specified as: outcomes when circumstances are identical for all individuals. yi = f(Ci,Ei,ui ) i:1…N (1) The opportunity share of inequality can be defined as: The circumstances variables are economically exogenous ΘrP = 1− ( (y I F ) ) (2) since they are outside the individual’s control but effort factors I (F ( y )) may be endogenous to circumstances as an individual’s actions may be influenced by its gender, parental background etc. The first step for computing ΘP consists on estimating a specific model of (1), which can be expressed in the following Equality of opportunity occurs, in the Roemer’s (1998) log-linear form: sense, when outcomes are independently distributed from circumstances. This independence implies that circumstances ln(yi)=Cia+Ei b + ui(3) have no direct causal effect on outcomes and no causal impact on efforts. The degree of opportunity inequality can Ei = ACi + ei 146 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T where a and b are two vectors of coefficients, A is a matrix The subscripts d and r, in ΘP, denote respectively of coefficients specifying the effects of the circumstance that inequality of opportunity is estimated directly variables on effort and ei is an error term. Model (3) can be or residually by eliminating the contribution of effort expressed in reduced from as: or circumstances to outcomes. The direct and resid- ual methods can yield different figures of opportunity ln(yi) = Cid + hI  (4) inequality and the only inequality measure for which where d = a + b  A and hi = ui + ei b. the two methods give the same results is the mean log deviation (Theil_L), which has a path-independent Inequality of opportunity can be measured using equation (2) decomposition when the arithmetic mean is used as the where the counterfactual distribution is obtained by replacing reference income or consumption (Foster and Shney- yi with its estimated value, from equation (4), and which can erov, 2000). By using the mean log deviation inequality be expressed as: y  i = exp(Cδˆ + η ˆ i ). In this decomposition, the index the residual and direct methods give the same variation in y i can be interpreted as the influence of effort opportunity inequality measures. because circumstances are set to be equal for all households, and inequality of opportunity is measured as a residual. The parametric approach allows the estimation of the partial effects of one or some circumstance variables on outcomes, Inequality of opportunity can also be measured directly by while controlling for the others, by simulating distributions eliminating the contribution of effort to outcomes, using the (y such as: y ij = exp(C jδˆ j + C h≠ jδˆ h≠ j + η ˆ i ), where F  j ) is the smoothed distribution, obtained from the predicted values counterfactual outcomes distribution obtained by keeping of outcomes based on circumstances in equation (4) while circumstance C j constant. ignoring the remaining variation in the residuals: i = exp(Ciδˆ ) (5) z The inequality share specific to circumstance j can be com- puted residually by: The share of inequality of opportunity can thus be measured by: ( ) (6)  (z I F ) ΘP j = 1− ( ( ))  y I F j Θ = I (F ( y )) d I (F ( y )) P Social mobility: Father vs Son and Mother vs Daughter FIGURE D.1: Intergenerational Mobility Poor Population, Father vs Son and Mother vs Daughter, Percent A. Education 100 3 0 3 1 100 2 3 0 3 3 6 0 4 7 Education level of the daughter (%) 13 12 Education level of the son (%) 90 90 22 25 27 80 34 80 32 16 36 25 45 10 70 48 5 70 47 57 60 60 26 11 15 17 50 25 37 50 27 58 40 40 29 20 25 30 30 56 23 49 20 47 20 46 37 26 30 29 23 0 10 16 10 22 14 0 0 n y y y ry n y y ry y ry ry io ar ar ar ar ar ar tio a a a a at im im nd nd im im nd nd nd nd uc ca pr pr co co pr pr co co co co u ed se ed e ed se se se se se e ed m m o et e e So o N et e d e d So m ov N pl m ov te te pl So Ab m So e e Ab m pl pl Co Co m m Co Co Education level of the father (%) Education level of the mother (%) No education Some primary Completed primary Some secondary Above secondary Completed secondary continued Appendix D 147 FIGURE D.1B. Employment Status 100 100 Employment status daughter(%) 90 90 Employment status son (%) 80 80 70 65 70 60 80 85 82 84 60 82 82 87 88 87 88 50 50 100 40 40 30 30 20 25 20 12 6 12 14 11 10 3 3 5 3 14 0 7 2 10 8 9 7 2 2 4 11 0 0 6 4 0 ee ee ed d ed rk ld d ed rk ld d ed d e e e e oy oy wo ho wo ho oy rk oy rk er tir tir oy oy er pl pl wo wo e e Re Re pl pl pl pl us us Em Em em em em em er ho er ho ev ev Un Un lf lf id id Se Se N N pa pa Un Un Employment status of the father (%) Employment status of the mother (%) Employee Self employed with others Self employed alone Unpaid family helper/own farm FIGURE D.1C. Industry 100 2 5 100 7 8 8 90 3 90 18 5 12 6 6 3 29 80 36 80 39 46 8 Industry daughter (%) 8 34 70 70 Industry son (%) 9 60 10 60 10 50 14 50 100 24 40 80 40 82 72 74 30 58 30 53 52 20 41 20 37 10 10 0 0 re es es n n re ry rt ry rt tio tio tu po po vic tu st st ic ul rv du ra du ra ul er ns ns ric se st ric ist In In s tra tra i Ag in in er Ag er m m d d th th an an ad ad O O de de lic ic bl b a a Tr Tr Pu Pu Industry father (%) Industry mother (%) Agriculture Industry Trade and transport Other services Public administration Source: HBS 2018. 148 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T APPENDIX E Multivariate Regressions and Determinant of Consumption and Poverty We perform a regression analysis to examine the main method and the second using the probit model. The estima- factors affecting households’ consumption and poverty. tion results are reported respectively in Tables E-1 and E-2. This allows us to identify the main correlates of poverty. We use two regression models. The first examines the impact of the It is worth mentioning that the direction of causality is household socioeconomic characteristics on the logarithm of sometimes difficult to establish in these kinds of analysis. real per capita household consumption, and the second inves- The results below allow the identification of variables closely tigates the determinants of the probability of being poor. The related with poverty, but the direction of causation will neces- first model is estimated using the Ordinary Least Square (OLS) sitate analysis that is more sophisticated. TABLE E.1: Correlates of Consumption, 2018 TANZANIA RURAL URBAN Household socio-demographic characteristics Household size 0.113*** 0.107*** 0.129*** (0.00) (0.00) (0.01) Share of members aged 0–14 years –0.207*** –0.154*** –0.372*** (0.03) (0.04) (0.07) Share of members aged 65+ years –0.347*** –0.403*** –0.179 (0.04) (0.05) (0.09) Age of household head 0.004*** 0.003*** 0.006*** (0.00) (0.00) (0.00) Gender of household head –0.048** –0.047** –0.039 (0.02) (0.02) (0.03) Education of the head (omitted: no education)   Incomplete primary 0.043* 0.039 0.086 (0.02) (0.02) (0.06)   Primary completed 0.167*** 0.145*** 0.249*** (0.02) (0.02) (0.05)   Lower secondary 0.372*** 0.358*** 0.457*** (0.03) (0.04) (0.05)   Upper secondary 0.559*** 0.394*** 0.685*** (0.05) (0.05) (0.08)  University 0.842*** 0.788*** 0.910*** (0.07) (0.11) (0.09) Head of household migrant 0.026 –0.001 0.109*** (0.02) (0.02) (0.03) Household economic activity Sector of employment of the head (omitted: agriculture)  Manufacturing 0.065* 0.059 0.082 (0.03) (0.04) (0.05)  Services 0.108*** 0.172*** 0.094** (0.02) (0.03) (0.03)   Public administration 0.098 0.129* 0.045 (0.05) (0.05) (0.08) Status of employment of the head (omitted: unpaid family helper/own farm)  Employee 0.169*** 0.159*** 0.220*** (0.03) (0.04) (0.04)   Self-employed with employees 0.311*** 0.195*** 0.432*** (0.04) (0.05) (0.06)   Self-employed without employees 0.092*** 0.092** 0.137*** (0.02) (0.03) (0.04) (Table Continued on next page) 150 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE E.1: Correlates of Consumption, 2018 (Continued) TANZANIA RURAL URBAN Has any agricultural plot 0.103*** 0.092*** 0.121*** (0.01) (0.02) (0.03) Has any livestock 0.169*** 0.205*** 0.046 (0.02) (0.02) (0.03) Housing characteristics Access to sanitation (omitted: open defecation)   Basic sanitation 0.329*** 0.315*** 0.413*** (0.03) (0.04) (0.12)   Limited sanitation 0.174*** 0.297*** 0.256* (0.04) (0.05) (0.12)   Unimproved sanitation –0.007 –0.006 0.100 (0.03) (0.03) (0.12)   Other sanitation –0.082 –0.048 –0.163 (0.06) (0.06) (0.18) Access to water (omitted: unimproved water)   Basic water 0.021 0.046* –0.218*** (0.02) (0.02) (0.07)   Limited water –0.034 0.002 –0.389*** (0.02) (0.02) (0.08)   Other water 0.031 0.059* –0.213* (0.02) (0.02) (0.08) Daily market –0.025 –0.030 –0.016 (0.02) (0.02) (0.03) Access to road (omitted: no road)   Trunk road 0.049*** 0.053*** 0.049 (0.01) (0.02) (0.03)   Tarmac road 0.059** 0.020 0.081* (0.02) (0.03) (0.04) Mobile phone signal 0.037** 0.022 0.083** (0.01) (0.01) (0.03) Access to health facility (omitted: no health facility)   Only health center/dispensary –0.030* –0.015 –0.027 (0.01) (0.02) (0.03)   Only public/private hospital 0.209*** 0.097 0.359*** (0.04) (0.06) (0.05)   Health center/dispensary and public/private hospital 0.058* 0.071* 0.133*** (0.03) (0.03) (0.04) Geographic location (omitted: rural)   Other urban centers 0.023 (0.02)   Dar es Salaam 0.315*** (0.03) Constant 11.432*** 11.446*** 11.372*** (0.05) (0.05) (0.15) Observations (0.03) 6,675 2,788 R-squared 0.476 0.430 0.467 Source: HBS 2017/18. Appendix E 151 TABLE E.2: Correlates of Poverty, 2018 TANZANIA RURAL URBAN Household socio-demographic characteristics Household size 0.120*** 0.111*** 0.162*** (0.01) (0.01) (0.02) Share of members aged 0–14 years 0.486*** 0.612*** 0.153 (0.10) (0.11) (0.21) Share of members aged 65+ years –0.357** –0.395** –0.236 (0.13) (0.15) (0.33) Age of household head 0.001 0.003 –0.005 (0.00) (0.00) (0.00) Gender of household head 0.023 –0.044 0.147 (0.05) (0.05) (0.12) Education of the head (omitted: no education)   Incomplete primary –0.007 –0.054 0.191 (0.06) (0.06) (0.18)   Primary completed –0.177*** –0.179*** –0.143 (0.05) (0.05) (0.15)   Lower secondary –0.763*** –0.859*** –0.592** (0.11) (0.12) (0.20)   Upper secondary –0.470* –0.746** –0.288 (0.21) (0.25) (0.29)  University –1.556*** –1.050* –1.755*** (0.33) (0.46) (0.41) Head of household migrant –0.034 0.038 –0.255** (0.05) (0.05) (0.10) Household economic activity Sector of employment of the head (omitted: agriculture)  Manufacturing –0.093 –0.136 –0.061 (0.10) (0.11) (0.17)  Services –0.270*** –0.363*** –0.220 (0.08) (0.09) (0.13)   Public administration –0.227 –0.545* –0.008 (0.27) (0.25) (0.38) Status of employment of the head (omitted: unpaid family helper/own farm)  Employee –0.219* –0.052 –0.362* (0.11) (0.12) (0.18)   Self-employed with employees –0.235 –0.189 –0.326 (0.12) (0.16) (0.21)   Self-employed without employees –0.146 –0.233** –0.078 (0.08) (0.08) (0.14) Has any agricultural plot –0.167*** –0.193*** –0.104 (0.04) (0.04) (0.11) Has any livestock –0.211*** –0.261*** –0.024 (0.05) (0.05) (0.11) (Table Continued on next page) 152 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE E.2: Correlates of Poverty, 2018 (Continued) TANZANIA RURAL URBAN Housing characteristics Access to sanitation (omitted: open defecation)   Basic sanitation –0.490*** –0.661*** –0.530 (0.11) (0.12) (0.35)   Limited sanitation –0.746*** –0.877*** –0.800* (0.11) (0.19) (0.34)   Unimproved sanitation –0.002 0.019 –0.136 (0.07) (0.07) (0.33)   Other sanitation 0.081 0.007 0.311 (0.19) (0.17) (0.65) Access to water (omitted: unimproved water)   Basic water –0.006 –0.014 0.351 (0.05) (0.05) (0.20)   Limited water 0.015 0.022 0.337 (0.06) (0.06) (0.27)   Other water –0.052 –0.116 0.469 (0.07) (0.07) (0.26) Daily market 0.098* 0.148** –0.028 (0.05) (0.05) (0.10) Access to road (omitted: no road)   Trunk road –0.125** –0.138** –0.168 (0.04) (0.04) (0.11)   Tarmac road –0.058 0.109 –0.224 (0.07) (0.08) (0.12) Mobile phone signal –0.007 0.028 –0.093 (0.04) (0.04) (0.09) Access to health facility (omitted: no health facility) Only health center/dispensary 0.045 0.010 0.126 (0.04) (0.04) (0.11) Only public/private hospital –0.376* –0.208 –0.867** (0.15) (0.19) (0.31) Health center/dispensary and public/private hospital –0.288*** –0.318** –0.332* (0.08) (0.10) (0.14) Geographic location (omitted: rural)   Other urban centers 0.122 (0.07)   Dar es Salaam –0.151 (0.11) Constant –1.091*** –1.070*** –1.183* (0.14) (0.14) (0.46) Observations 9,463 6,675 2,788 Source: HBS 2017/18. Appendix E 153 APPENDIX F Multidimensional Deprivation and Multidimensional Poverty Index (MPI) 154 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Alkire and Foster (2011) propose a simple methodol- The first step is to determine a threshold or depriva- ogy for the measurement of multidimensional poverty, tion cutoff, zj > 0, for each dimension, according to which employs a generalization of the conventional which individuals can be considered as deprived in that Foster-Greer-Thorbecke (FGT) poverty measures to ­dimension. Then, construct the n × d matrix of deprivations account for multidimensionality. The approach builds g0 = [g0ij], where g0ij = 1 when yij < zj (deprived) and g0ij = 0 if on the work on multidimensional poverty and deprivation yij ≥ zj (non-deprived). A vector C of deprivation scores is con- developed by the Oxford Poverty & Human Development structed from the matrix g0, where the deprivation score for Initiative (OPHI) and introduces an intuitive approach to each individual i is defined by the following weighted sum: identify the poor using two forms of cutoff: one within each of the relevant dimensions of the welfare to deter- c i = ∑ j w j gij 0 mine whether a person suffers shortfalls in that dimension, where wj is the weight associated with each dimension j, and and a second across dimensions that delineates how summing to d. widely deprived a person must be in order to be con- sidered poor and identifies the poor by ‘counting’ the The second step consists in identifying the poor, and is dimensions in which an individual is deprived. They pro- based on the selection of a cutoff level for the depriva- pose an adjusted FGT measure that is particularly suitable tion scores and a definition of an identification function. for use with ordinal data and informs on the breadth of Let k ≤ d is the poverty cutoff and ρk(yi ; z) is the identification multiple deprivations of the poor. function defined as follows: Consider a number of relevant dimensions of well-being, ρk (yi ; z) = 1 if ci >k    (i is poor) d ≥ 2, for a population of n individuals. The well-being and dimensions might relate to education, living standards, ρk (yi ; z) = 0 if ci < k   (i is nonpoor) or access to basic services, etc. The individuals achievements are denoted by the n × d matrix y = [yij], where yij ≥ 0 is the ρk(yi ; z) identifies individual i as poor when the number of achievement of individual i in dimension j. dimensions in which he/she is deprived is at least k. Incidence or headcount ratio Based on qk, the headcount ratio, which measures the This is analogous to the conventional income headcount ratio proportion of people identified as multidimensional poor, ­ which measures the incidence of poverty, but in a multidimen- can be defined as: sional setting. ∑ ρk ( y i , z ) n q The headcount ratio has two main shortcomings: first, it H (y ,z ) = i =1 = remains unchanged if a poor individual becomes deprived n n in a new dimension. Second, it does not allow the evaluation of the contribution of each dimension to poverty. Intensity of multidimensional poor’s deprivation To address these shortcomings, Alkire and Foster (2011) where A measures the average proportion of deprivations in suggest an additional measure that assesses the breadth which the poor are deprived, through calculating the percent- of deprivation experienced by the poor: age of total deprivations each poor person has (ci(k)/d) and calculating the average of those percentages across the poor ∑ c (k ) n A= i =1 i (dividing by the number of poor only, q). dq Appendix F 155 Construction of the MPI The Multidimensional Poverty Index (MPI) is then defined The contribution of each dimension to poverty, CDj, can be as a combination of the headcount and the average calculated using MPI as: proportion of deprivation to inform on the prevalence of poverty and the average extent of a poor individu- w j  n  ∑ i =1 j ij w g 0 (k ) al’s deprivation. It is given by the simple product of H and   d CD j = A: MPI = HA. MPI represents the proportion of weighted w jnMPI deprivations experienced by the poor relative to the maxi- mum potential deprivations that could be experienced by the The multidimensional deprivation index (MDI) used in chap- whole population. ters 4 and 5 is very similar to the MPI. FIGURE F.1: Welfare Dimensions and Deprivations Criteria DIMENSIONS INDICATORS WEIGHT DEPRIVATION CRITERIA Education Years of schooling 1/9 No household member age 10 years or older has completed at least five years of schooling. (1/3) School attendance 1/9 Any school-age child (6 to 15 years old) is not attending school. Health Child mortality 1/9 Any child has died in the family in the five-year period preceding the survey. (1/3) Living Standards Housing 1/9 The household has unimproved wall, floor and/or roof housing material (1/3) Electricity 1/9 The household has no electricity. Sanitation 1/9 The household has no improved sanitation facility. Drinking water 1/9 The household does not have access to improved drinking water. Cooking fuel 1/9 The household cooks with dung, wood, or charcoal. Assets 1/9 The household does not own at least two of the following assets: radio, TV, telephone, bicycle, motorbike or refrigerator. Source: Authors definitions based on Tanzania National Bureau of Statistics and OPHI (2019). 156 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T APPENDIX G Methodology for Generating Small-Area Poverty Estimates Obtaining accurate and reliable estimates of local pov- to produce reliable estimates below a certain geographical erty is difficult due to the high costs of collecting welfare level, such as provinces or districts. The 2018 HBS is no excep- data that allows for such analysis. Household surveys – from tion, and is not considered to be representative below the which poverty estimates are derived – are typically too small provincial level. Methodology Traditionally, poverty mapping methods estimate a FGT0 are obtained by estimating the nested-error model ­ random effect regression model using survey data (Equation 1); generating district-level effects ud* ~ N 0, σ 2 ˆu ( ) containing per capita income or consumption data and ­ * and unit-level effects εid ~ N 0, σ 2 ( ) ˆ ε ; and then calculating use the estimated parameters to simulate welfare in population welfare values through micro-simulation based a contemporaneous census. Because there is no recent on the sample and non-sample values of explanatory vari- ­ census in Tanzania, however, applying this method would ables as specified in Equation 1. To estimate the Means produce biased estimates. (Lange, Pape, and Putz 2018). Squared Errors (MSE) of FGT0, we follow Molina and Rao To overcome this challenge, we combine the household (2010), which proposes a parametric boostrap MSE estima- sample data with remote sensing indicators at the village tor following the bootstrap method for finite populations of level, linking the source and auxiliary data geographi- González-Manteiga et al. (2008).1 Alternative estimates gen- cally through village identifiers. To combine these two data erated using a village level model, a variant of the sub-area sources, we employ the Empirical Best Prediction (EBP) estimator proposed by Torabi and Rao (2014), yielded simi- Method (Molina and Rao 2010, Battese, Harter, and Fuller lar results. 1988). The EBP modifies the traditional ELL method (Elbers, Lanjouw, and Lanjouw, 2003) in two main ways. First, random The specification of the consumption model in Equation 1 effects are introduced at the level of the district instead of affects the district-level poverty rates. The set of variables the enumeration area. Second, these district random effects included in each specification was chosen from the list of can- are conditioned on the sample data. This method therefore didate variables using the stepwise selection process (with a efficiently combines household level information on per cap- mitigates significance level threshold of 0.01 for removal). This ­ ita consumption, which are only available in sample villages, ­ ample, which the potential for over-fitting the model to the s with an exhaustive set of village level prediction based on would make the resulting poverty estimates less precise. indicators derived from satellite data. The stepwise selection process yielded a model with 40 right hand side variables (including regional dummies) as reported In the EBP method, the two-level nested error regression in Table G.1. The model achieved a reasonably high R2 of model is first estimated: 0.23 despite the fact that the explanatory variables are at the village level whereas consumption is measured at the house- ­ y id = xid β + ud + εid , i = 1, ... .,N d = 1, ... . ,D hold level. ( 2  ud ~ N 0, σ u ) ( 2 ,  εid  ~ N 0, σ ε ) (1) Finally, we apply a rescale factor for each region to ensure that where yid corresponds to log per capita consumption for our regional averages of EBP estimates match the regional household i in district d. ud and eid are district-specific and rates of poverty based solely on the HBS 2018, which is household-specific random errors. The EBP estimates of regionally representative.2 1 The R “sae” package has been used to implement this estimation. 2 This is the simple ratio or pro-rata benchmarking considered in Pfefferman, et al (2014). That analysis evaluates several benchmarking procedures in cross-sectional simula- tions. In the simulations considered, the simple ratio benchmarking performs at least as well as several more complex benchmarking procedures proposed in the literature. 158 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Data Geospatial data on the right-hand side of equation are the minimum, maximum, mean, sum, and standard drawn from a number of different sources: which are catego- deviation of those statistics for each village. rized into four different groups – urbanization, market access, agro-­climate, and natural disaster shocks and summarized as Agro-climate: below: – Rainfall and temperature data from Willmott and Matsuura (2018) capture the monthly estimates of pre- Urbanization: cipitation and temperature at 0.5° resolution. Values – Night-time light data from the Visible Infrared are interpolated for each grid node from an average of Imaging Radiometer Suite (VIIRS) at a spatial resolu- 20 different weather stations, with corrections for ele- tion of 15 arc-seconds. We used the annual compos- vation. The annual average of precipitation and tem- ite measure of nighttime light from 2015 and monthly perature for each village between 2010 and 2017 are composite measures from December 2017 through included. December 2018. – Elevation data from AidData’s GeoQuery (Goodman – Global human settlement layer (GHSL) data contain et al., 2019) provides gridded data on elevation (in information on built-up areas on a global scale for the meters) at a 500m resolution, which are based on following years: 1975, 1990, 2000, 2014. It is published Jarvis et al. (2008). by the Joint Research Center (JRC) of the European – Climate data from Kottek et al. (2006) offer a world Commission and is derived from data collected by map of the Köppen-Geiger climate classification, Landsat satellites. We compute the percentage which divides climates into five main climate groups of total built-up area observed in 2014 that was (tropical, dry, temperate, continental, and polar), with constructed prior to 1975 or during 1975-1990, each group being divided based on seasonal precipi- 1975-1990, and 2000-2014. tation and temperature patterns. The raster version of – Population data from WorldPop Africa provide 2015 the map is available at 0.5° resolution. estimates of numbers of people per pixel available at – Crop yield data from IFPRI HarvestChoice Dataverse a spatial resolution of 0.000833333 decimal degrees (Wood-Sichra et al. 2016) capture the estimated (or approx. 100m at the equator); and from Global yield of various crops (ton/km) at 0.5° resolution. In Human Settlement dataset, which also provides the our analysis, three major crops in Tanzania – maize, grid-level population estimate of 2015, informed by ­ sorghum and rice – are considered. the distribution and density of built-up as mapped in the Global Human Settlement Layer (GHSL) global – The net primary productivity (NPP) and the layer per corresponding epoch. Normalized Difference Vegetation Index (NDVI), which are commonly used as indicators to – Agglomeration Index provided by the World Bank’s characterize vegetation health and vigor. These data Geospatial Operations Support Team (GOST) based are available at a resolution of 0.1 degree. on Uchida and Nelson (2009) is an composite index of urbanization, which takes into account population Market Access: density, the population of a “large” urban center, and travel time to that large urban center. – See Appendix H for more details – Building footprints provided by GOST based Natural Disaster Risks: on data shared by Ecopia and Maxar offers various indicators of building areas, counts, and – Flood and drought data from UNEP/DEWA/ densities using different measurement scales. GRID-Europe measure the estimated size of The characteristics of building densities are computed GDP (economic) and population (physical) based on the sizes of or distance to 5 or 25 nearest that are exposed to flood and drought events buildings around each building point and compute per 10km grid. Appendix G 159 Results This section discusses the reliability and accuracy of the TABLE G.1: Beta Model Results (Variables Selected poverty map estimates. We use the average coefficient of through Stepwise Process) variation across districts as the main criteria to judge how the VARIABLES COEF. STD. ERROR T-STATISTICS precision and reliability of our estimates compare with the Agglomeration index (sum) –0.036 0.007 –5.413 direct estimates obtained from the 2018 HBS. The coefficient Air temperature in 2013 –0.038 0.004 –8.717 of variation for each district is defined as the standard error Annual nighttime light 2015 (sum) 0.075 0.01 7.729 of the district poverty estimate divided by the estimated dis- Building density (mean of building areas for 5 nearest neighbors) 0.067 0.014 4.913 trict poverty rate. The consumption model – the model used Building density (mean of distance to 25 nearest neighbors) –0.275 0.047 –5.806 to predict consumption as a function of spatial covariates Building density (mean of distance to 25 nearest neighbors) 0.184 0.043 4.267 (Equation 1) – has a reasonably high R2 of 0.23, which means Building density (standard deviation of distance to 5 nearest that the geospatial variables explain 23 percent of variation neighbors) 0.066 0.011 5.802 in household consumption in the first stage. This is impres- Elevation (district-level) –0.119 0.019 –6.134 sive given that the geospatial variables are only capable of Flood economic shocks (2018) 0.012 0.002 5.512 explaining variation in household consumption across vil- Flood economic shocks (2018) –0.022 0.005 –4.364 lages. The results from the consumption model are presented NDVI max (2018/11) (1km resolution) 0.624 0.133 4.705 in Table G.1. NDVI mean (2018/07) (1km resolution) 1.668 0.322 5.18 NDVI mean (2018/07) (1km resolution) –1.834 0.341 –5.376 Figure G.1 also presents a comparison of the Coefficients NDVI standard deviation (2017/1-2018/3) (70m resolution) 0.041 0.006 7.003 of Variation (CVs) of direct estimates of FGT0 at the dis- NDVI standard deviation (2017/1-2018/3) (30m resolution) 0.016 0.002 7.107 trict level with comparable small area estimates. We see NDVI standard deviation (2017/1-2018/3) (50m resolution) –0.055 0.008 –7.088 that the average CV of the small area estimates (18%) achieves NDVI sum (2017/1-2018/3) (50m resolution) 0 0 5.228 about a 50% reduction in the CV compared to direct estimates NDVI sum (2018/07) (1km resolution) –0.011 0.002 –6.817 (35% or 38% depending on the method used).3 Coincidentally, NDVI sum (2018/07) (1km resolution) 0.011 0.002 6.132 the average CV of the small area district estimates is approxi- NDVI sum (2018/07) (1km resolution) 0 0 –6.974 mately equal to the average CV of the direct estimates of the Nighttime light mean (2018/07) 0.18 0.027 6.618 routinely published provincial poverty rates, which is also 18%. Nighttime light mean (2018/09) –0.151 0.024 –6.206 Nighttime light min (2018/08) –0.05 0.013 –4.025 This method was validated using data from the 2012 Nighttime light sum (2018/03) 0 0 5.719 ­population census. To do the validation, we constructed a Population estimate (mean) –0.098 0.017 –5.9 “welfare index” comprised of the first principal component Precipitation in 2013 –0.005 0.001 –4.213 of several household welfare indicators present in the Precipitation in 2014 –0.008 0.002 –4.678 census. These welfare indicators include the household’s Precipitation in 2017 0.011 0.002 7.127 size and dependency ratio, whether the household is a N = 9465, RMSE = 0.547, Adjusted-R2 = 0.235. Notes: Original raster data are available at the pixel level and aggregated by village beneficiary of the main social protection program, whether and district levels before being used for the consumption model. The minimum, the household contains a disabled member, the education maximum, mean, and total values of those pixels at the village or district levels are computed and used to explain variation in consumption. Region dummies are and age of the head, and whether the head was engaged in excluded from the table for brevity. agricultural work. The correlation between the mean value of this welfare index and mean consumption is 0.6 across districts. We then added a constant of five to the first principal HBS, and drew a sample of households to mimic the structure component index and defined a welfare deprivation line of of the 2018 survey. Finally, we used the same remote sensing approximately 4, which classified approximately 20 percent indicators listed above, pertaining in some cases to earlier of the population as deprived. We then drew a “synthetic years, to predict welfare deprivation rates at the district level sample” consisting of the same villages sampled in the 2018 using the EBP method. The CV of 35% is based on the standard method of clustering standard errors by enumeration areas (PSUs), which accounts for correlation between households in the same 3 enumeration area. This overstates the precision of the survey estimate by ignoring correlation between hosueholds in different enumeration areas within the same district. Partly for this reason, Molina and Marhuenda (2015) recommend using a Horwitz-Thompson estimator to estimate standard errors from survey data. Using this method gives an estimated average CV of 38 percent. This is also a fairer comparison to the 18 percent average CV yielded by the small area estimation model, since both allow for welfare to be correlated for all households within a district. 160 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE G.1: CV of Direct vs. Small Area Estimates of The results of this validation test were encouraging. The Poverty Rates at the District level correlation between estimated and actual census depriva- tion rates was 0.86. The average standard error and coef- 100 ficient of variation fell by more than half compared to the survey alone. Furthermore, the estimated 95% confidence 80 intervals from the EBP procedure contained the actual cen- sus deprivation rates in 72 percent of the districts. This is the same percentage of districts contained within the 95% con- 60 fidence intervals obtained from taking district means in the synthetic survey, which underestimates standard errors by 40 assuming that poverty rates are independent across enu- meration areas. Overall, the validation procedure provides 20 evidence that, in this case, combining survey data with an exhaustive set of village-level remote sensing indicators 0 generates district estimates that are sufficiently precise to CV from EBP estimates CV from HBS direct estimates publish. Notes: This figure shows the value of coefficients of variation (CV) for each district- level poverty estimates from SAE method and direct estimates based on the 2018 HBS only. Appendix G 161 APPENDIX H Market Accessibility 162 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Definition In this report, we define market access as a measure of A priori, there is no agreement on what the value of accessibility from one origin to all destinations based l should be. For the purpose of this study (and for simple on travel distance (or travel time). More formally, market interpretation), we set it such that the market potential access for a given location (or origin) i can be expressed as (or attractiveness) of a location j to a location i decays in follows: half for every additional 60 min of travel time. Travel time is computed based on how long people need to travel by car to MA i = ∑P e j j − lρij get from an origin i to a destination j given the existing road network system in Tanzania, as recorded in OpenStreetMap. Travel time is computed based on the OSRM algorithm.1 where Pj refers to the population of a location (or destina- In this study, destinations are defined as all major cities in tion) j, rij is travel time between locations i and j, and l is a Tanzania and origins are points on the road network that are trade elasticity or decay parameter. In other words, market closest to the centroid of each village.2 The market potential access is the weighted sum of population in all the destina- of those destinations are measured based on the size of city tions, which are weighted by travel time/distance. population, which is derived based on the 2012 Census. Data The calculation of market access requires the following and 2) road network data. We use the census population and information: 1) census population and geographical geographical locations of the cities from Brinkoff (2010). For coordinates (e.g., longitude and latitude) of all major cities; road network information, we rely on OpenStreetMap. 1 The Stata command “osrmtime” is used to compute travel time between the centroid of each village and major cities. 2 The cities that are taken into account for the calculation of market access include: Dar es Salaam, Mwanza, Arusha, Mbeya, Morogoro, Tanga, Kigoma, Dodoma, and Songea. Appendix H 163 APPENDIX I Poverty Dynamics 164 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T A Cautionary Note on Comparing HBS Poverty Estimates with Non-Harmonized NPS Poverty Estimates Tanzania has two surveys from which it solicits consump- FIGURE I.1: Levels and Trends of Poverty in Tanzania tion data suitable for poverty analysis; the Household Mainland using HBS and NPS (Non-Harmonized), Percent Budget Survey and the National Panel Survey. Both surveys 40 34.4 are integrated household surveys which, in addition to con- sumption, also solicit information on a broad range of top- 28.2 Poverty incidence 30 (% of population) ics that allow for determinants of outcomes to be assessed 20.8 and linkages to be made across sectors. The National Panel 20 18.1 16.6 14.6 Survey is a series of nationally representative household panel surveys with four rounds (2008/09, 2010/11, 2012/13, and 10 2014/15). The data are collected every two years from smaller sample sizes whose representation is at the national level 0 (Mainland and Zanzibar) and four sub-national domains—Dar 07 2 9 1 3 5 /1 /0 /1 /1 /1 20 11 08 10 12 14 es Salaam, other urban areas, rural areas, and Zanzibar. 20 20 20 20 20 HBS NPS While both the HBS and NPS provide the data to track the evolution of aggregate poverty numbers, the NPS Sources: HBS and NPS. Note: NPS estimates in graph are based on a non-harmonized methodology. also provide data to analyze the poverty dynamics.1 NPS also provide data to analyze micro-level determinants of pov- erty reduction at the household level. Though the key pur- surveys that make direct comparisons of consumption unad- pose of NPS is enable detailed study of poverty dynamics, few visable. First, while HBS requests households to maintain a researchers have exploited the data for such purposes. diary to collect data on food consumption data, NPS requests a key household member to recall consumption of the past The non-harmonized estimates of poverty using the NPS seven days. Second, to measure food consumption of home-­ reveal relatively lower levels of poverty and an opposite produced food, HBS solicits self-reported monetary val- poverty trend to the estimates produced using the HBS. ues from home producers while NPS uses the unit values of Since HBS and NPS are nationally representative surveys col- households that purchased those food items in the same lecting consumption data over relatively similar periods, in locality. Third, the level of commodity detail on food and non- principle they ought to provide a similar—although not an food consumption solicited through HBS is greater than that identical—picture of the evolution of poverty. Figure I.1 shows of NPS due to a more comprehensive and disaggregated list. the non-harmonized poverty trend for Mainland using the These methodological differences affect the level and distri- National Panel Survey and the poverty trend using the HBS. bution of the consumption aggregate, even if the information Poverty trends using HBS 2007 and HBS 2011/12 revealed a is solicited from the same sample. decline in poverty headcount of six percentage points from 34.4 percent to 28.2 percent, while poverty estimates from the The construction of the poverty line differed in several first three rounds of the NPS revealed lower levels of poverty areas making direct comparisons also potentially mislead- and an increasing poverty trend. ing. A common approach for estimating poverty lines is the cost of basic needs approach, which measures the cost of Three main ingredients underlie the differences in lev- acquiring enough food to provide adequate daily nutrition els and trends: (i) consumption aggregate, (ii) poverty per person (food line) plus the cost of some non-food essen- line, and (iii) intertemporal deflators. Regarding consump- tials (non-food component). This is the method used for both tion, there are key methodological differences between the the HBS and the NPS with some variation. In addition to the 1 The main purpose of NPS is to provide data to monitor national development objectives, evaluate specific policies and programs, and better understand the determinants of poverty reduction in Tanzania. Appendix I 165 impact differences in the consumption measurement meth- no significant differences in the intra-year and spatial defla- ods have on the poverty line, differences in the reference tors between the surveys, there were important differences in period, reference population and cost per calorie method the inter-year deflators. For instance, HBS revealed a higher played a role in the mismatch. HBS derives the cost of buying rate of inflation between 2007 and 2011/12 (over 100 percent 2,200 calories per adult per day based on the food consump- for food and 90 percent overall) compared to NPS which esti- tion patterns prevailing in a population whose per adult nom- mated food inflation to be 21 percent between 2008 and 2010 inal consumption fell between the 2nd and 5th decile (reference and 34 percent between 2010 and 2012. population) during a period of 30.4 days (reference period) valued at national median prices. In contrast, NPS derived the In conclusion, differences in the poverty levels between cost of buying 2,200 calories per adult per day based on the the two surveys were mainly attributed to differences in food consumption patterns prevailing in a population whose the methods for constructing the poverty line and differ- per adult real consumption is below the median during a ences in trends were mainly attributed to differences in period of 28 days valued at prices faced by the reference pop- the temporal deflator. ulation. The non-food component of the basic needs poverty line uses the average food consumption share of the popu- lation whose total consumption is close to the food line. For FIGURE I.2: Comparison of Basic Needs Poverty Rates HBS, these people are those whose total consumption per Across Survey Rounds, Mainland, Percent adult is between the food line and 1.2 times the food line. For NPS, they are those whose total consumption per adult is 40 34.4 in the bottom 25 percent. Poverty incidence 28.2 27.9 (% of population) 30 26.8 26.4 27.4 Finally, adjustments in differences cost of living compari- 20 sons across survey years differed in important ways. Both HBS and NPS use the Fisher price index calculated by geo- 10 graphic stratum and survey quarter to adjust for w­ ithin-year and spatial price differences. Both surveys rely on household 0 expenditure data for information about price levels rather 07 12 16 18 3 5 /1 /1 20 20 20 20 than market prices. While HBS price index is a weighted aver- 12 14 20 20 age of food and non-food indices, the NPS price index uses HBS DHS/ HBS NPS a food index only. A Fisher price index is also used to adjust HBS for inter-year differences, with the HBS index based on the Sources: Authors’ calculations based on HBS 2007, 2011/12 and 2018, NPS 2012/13 weighted average of food and non-food indices and the and 2014/15, DHS 2015/16. NPS index based on the food index only. While there were Note: The NPS consumption aggregate includes clothing and footwear. Progress in Poverty, Inequality and Shared Prosperity The evolution of poverty in Tanzania since 2007 is addressed most of the mismatches, some variations in esti- explored using all available nationally representative sur- mates between the two surveys remain due to differences in veys in Tanzania. We complement the poverty estimates the survey design, survey instrument, and other idiosyncrasies from the HBS using nationally-representative surveys – that are more difficult to adjust. whether they are consumption or non-consumption surveys— to successfully monitor trends in poverty since 2007. Where The purpose of this section is to evaluate the poverty non-consumption surveys are available – as was with the trends in Tanzania using alternative nationally representa- DHS—survey-to-survey imputations were used to derive esti- tive surveys. The Household Budget Survey, which is used to mates of consumption, and thus poverty. Where consumption produce the official poverty estimates, is conducted approxi- surveys were available – as was with the NPS—poverty esti- mately once every 6 years and is very expensive to implement. mation methodologies were harmonized with the methodol- In the interim, there are other surveys that are less costly and ogy used with the HBS. While methodological harmonization conducted more frequently that can be exploited for more 166 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T up to date poverty estimates. Together with medium term FIGURE I.3: Comparison of Basic Needs Poverty Rates forecasting, we show how poverty can be monitored in the Across Survey Rounds by Geographic Domain, Percent absence of a recent HBS. 40 35.4 34.7 Poverty in Tanzania Mainland has been declining in the last Poverty incidence (% of population) 30 decade, resulting in 1 in 4 people living in poverty in 2018 21.0 compared to 1 in 3 a decade earlier. Poverty has declined 20 18.3 11.8 13.6 about 8 pp in one decade from 34.4 percent in 2007 to 26.4 10 percent in 2018. The harmonized NPS poverty estimates also 1.4 2.2 reveal similar levels and a declining trend of similar magnitude 0 to the HBS.2 3 5 3 5 3 5 3 5 /1 /1 /1 /1 /1 /1 /1 /1 12 14 12 14 12 14 12 14 20 20 20 20 20 20 20 20 Spatial disparities in poverty remain high but progress Dar es Other urban Rural Zanzibar in poverty reduction and shared prosperity were made Salaam in rural areas and Zanzibar, where poverty is highest. These areas recorded declines in poverty from a high base, Sources: Authors’ calculations based on HBS 2007, 2011/12 and 2018, NPS 2012/13 and 2014/15, DHS 2015/16. while Dar es Salaam and other urban areas recorded slight Note: The NPS consumption aggregate includes clothing and footwear. increases (Figure I.3). The decline in national poverty was cou- pled by some progress towards shared prosperity between 2012 and 2014, much of which was concentrated in rural areas FIGURE I.4: Projected HBS Poverty Rates with NPS and Zanzibar. In Dar es Salaam and secondary cities, the mid- (Harmonized) and DHS (Imputed) Rates dle class seem to have benefitted the most from growth. The spatial distribution of food poverty also mirrors that of basic 30 27.9 27.4 needs poverty, with rural areas having the most elevated lev- 26.8 Poverty rate (%) 26.4 els of food poverty, followed by Zanzibar, other urban areas 25 and Dar es Salaam. 20 The positive growth outlook for Tanzania is expected to keep the country on a downward poverty trend, 15 2012 2013 2014 2015 2016 2017 2018 2019 2020 but Tanzania may still not reach the twin goals. Assuming sustained economic growth in Tanzania, the propor- Distribution-neutral, pass-though = 0.29 tion of people living in poverty is expected to fall (Figure I.4). NPS 2012/13 DHS 2015/16 (no cell) NPS 2014/15 HBS 2018 Using the Tanzania HBS as the underlying micro-data for pov- erty projection calculations, projections based on the neu- Sources: Authors’ calculations using HBS 2011/12, HBS 2017/18, DHS 2010, DHS tral-distribution approach with the empirical pass-through of 2015/16, NPS 2012/13, NPS 2014/15. Note: Actual poverty rate in 2012. Projection is from 2013 to 2020. Projections use a neutral 0.29 reveal a decline in poverty. The empirical pass-through distribution with empirical pass-through = 0.29 based on GDP per capita in constant LCU. rate of 0.29 applied to the projections suggests that GDP per GDP per capita is drawn from MFMod and adjusted by population growth. capita growth is transmitted into welfare growth at a 3:1 ratio. At this pace of poverty reduction, the poverty will be about 9 percentage points above the goal in 2030 and the absolute number of poor is expected to either remain constant or to increase as the population continues to grow rapidly.3 2 The harmonized methodology constructed the poverty lines using the same methodology as the HBS. The harmonized methodology also used the CPI to adjust for price differences across time. Expenditures on clothing and footwear are also included in the NPS consumption aggregate beginning with the 2012/13 survey, which was the first year the information was solicited in the NPS. 3 The limitations of tools and assumptions used to generate poverty projections must be recognized. The distribution-neutral approach can predict poverty rates relatively well if the distribution of the welfare aggregate (or inequality) remains the same over time. However, assuming growth with no distributional changes can be a strong assumption since widening (narrowing) distributions tend to attenuate (hasten) poverty reduction. Appendix I 167 Harmonizing the NPS Poverty Measurement Methodology The harmonized methodology constructed poverty lines generated from 2012/13 onward account for clothing and using the same methodology as that used for the HBS and footwear, unless otherwise indicated. deflated consumption temporally using the CPI. The CPI rather than the NPS inflation figures is used for adjusting con- To perform sensitivity analyses on the poverty estimates sumption variation over time. Because the harmonization pro- and offer a longer-term perspective on the poverty trends, cess is detailed in Belghith, Lopera et al. (2018), the focus here clothing and footwear is excluded for estimates starting is on the more recent adjustments that have been made for in 2008/09 to 2014/15 and the base year for the poverty the purposes of this study. threshold is changed to the year 2010 so that the harmo- nized estimates can be compared to the non-harmonized Zanzibar is incorporated into the analysis. Because estimates. Figure I.5 is a visual of the data and assumptions the motivation of Belghith, Lopera et al. (2018) was to for the proceeding analysis. compare Mainland poverty estimates for HBS and NPS, estimate for Zanzibar were not included. The inclusion Finally, to evaluate poverty trends, the panel data are of Zanzibar in this analysis implies that spatial price index, treated as a series of cross-sectional data. To do so, consumption aggregate and poverty line will be affected, cross-sectional weights are used instead of panel weights thus estimates will vary slightly from those reported in to estimate the trends. No adjustments were made to the Belghith, Lopera et al. (2018). NPS 2008/09 weights as this was the baseline survey and no adjustments were made to the NPS 2010/11 weights as there Expenditures on clothing and footwear are included in was very little attrition between the first round and second the consumption aggregate beginning with the 2012/13 round. A cross-sectional weight for round 3 was constructed. survey. The HBS measurement methodology has tradition- ally included expenditures on clothing and footwear but this The final step in the cross-sectional weight calculation information was not solicited in the first two rounds of NPS. process is to apply a post-stratification adjustment. The NPS began soliciting information of expenditures on cloth- post-stratification (or calibration) adjustment weights the ing and footwear for men, women and children using a sev- total population generated from the survey weights to a en-day recall in NPS 2012/13. Because information on clothing known population total. In general, this adjustment should and footwear was not solicited in the first two rounds of NPS, be made at the lowest level of administrative disaggrega- the consumption aggregate and consequently the poverty tion for which population totals are available and for which line used in Belghith, Lopera et al. (2018) excluded this com- at least one cluster has been selected into the survey. In ponent. The construction of the harmonized poverty line Tanzania, this level would normally be the region. In the case also considers clothing and footwear. Thus, poverty trends of the 2012, however, it is not possible to use the region FIGURE I.5: Data and Assumptions Used for Analyzing NPS Poverty Trends NPS Cross-section NPS Cross-section NPS Cross-section NPS (Aggregate excludes 2008/09 2010/11 2012/13 2012/13 clothing and footwear) NPS Cross-section NPS (Aggregate excludes 2012/13 2014/15 clothing and footwear) NPS Cross-section NPS (Aggregate includes 2012/13 2014/15 clothing and footwear) Note: Darker boxes reflect the base year under consideration. 168 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE I.1: New Administrative Divisions TABLE I.2: Total Population Before and After Stratification NEW REGION ORIGINATED FROM TOTAL NUMBER OF HOUSEHOLDS TOTAL NUMBER OF INDIVIDUALS Geita The region was created from Bukombe district in Shinyanga region, PROBABILITY POST-STRATIFIED PROBABILITY POST-STRATIFIED Chato from Kagera region and Geita from Mwanza region WEIGHTS (FINAL) WEIGHTS WEIGHTS (FINAL) WEIGHTS Katavi Mpanda district from Rukwa region Dar es Salaam 1,062,462 1,083,381 4,181,107 4,299,458 Njombe Ludewa, Makete, and Njombe districts from Iringa region Other urban – Mainland 1,706,152 1,888,763 7,080,161 7,095,990 Simiyu Formed by taking Bariadi, Meatu, and Maswa districts from Shinyan- Rural – Mainland 6,179,499 6,054,641 31,872,998 32,432,820 ga region and Busega district from Mwanza Zanzibar 239,212 250,212 1,279,012 1,334,143 Source: www.statoids.com/utz.html because the o ­ riginal sample was selected prior to the cre- Based on the above issues with using the region as the ation of four new districts (Table I.1). Additionally, the sam- level for post-stratification, instead the country was ple was selected using the 2002 population census as a divided into four “areas”: Dar es Salaam, other main- sampling frame and the information for post-stratification is land urban areas, rural mainland areas, and Zanzibar. Post- sourced from the newer 2012 population census. This is an stratification was then performed at this level. See Table I.2 issue because certain clusters, particularly those near urban for total number of households and individuals following the areas, where reclassified from rural to urban, or sub-divided post-stratification adjustments. and reclassified. Harmonized NPS Trends in Food and Basic Needs Poverty i.  Comparing consumption increases when the base year changes from 2010/13 to per adult equivalent 2012/13 and when clothing and footwear is included. The increase resulting from shifting the base year up is to be A consumption-based measure of welfare is used to obtain expected as prices in 2012/13 are higher than in 2010/12. poverty indicators. The consumption aggregate provides a Since 2010/11, real consumption per adult has been ris- summary measure of living standards based on households’ ing nationally and across geographic domain, with the expenditures on goods and services. Table I.3 compares the exception of Dar es Salaam. This is consistent regardless of average monthly per adult expenditures in Tanzania Shillings the temporal deflator used except in 2014/15 where there is (TZS) across NPS surveys. The expenditures are expressed in a more moderate change when the CPI is used. This is due (i) nominal terms, (ii) real terms adjusted spatially and season- to the relatively higher level of inflation between 2012/13 ally within the survey, (iii) real terms adjusted temporally using and 2014/15 when using the CPI (13 percent) than using survey-based deflators, and (iv) in real terms adjusted tempo- the s­ urvey-based deflator (5 percent). Despite the decline rally using the consumer price index (CPI). The survey-based between 2012 and 2014, Dar es Salaam fairs significantly bet- deflator is a Fisher price index constructed from the unit val- ter than other urban areas and rural areas on Mainland in ues in the NPS and is based on food values only. terms of higher levels of consumption. Residents in Dar es Three sets of estimates are provided to highlight the evo- Salaam consume 1.5 times than resident in other urban areas lution of the consumption aggregates as the base year and almost 3 times more than residents in rural areas. The changes and an additional consumption item is included to levels of consumption for residents in Zanzibar are closer in the aggregate. The first set of estimates are the consumption magnitude to those of rural residents on Mainland than to res- aggregates from all 4 rounds excluding clothing and footwear idents in urban areas on Mainland. (method 1). The base year for the temporal price adjust- Consumption regained a positive trajectory for all con- ments is NPS 2010/11. Similarly, the second set of consump- sumption groups, with the middle-income groups ben- tion aggregates excludes clothing and footwear but uses efitting most. The positive trajectory follows a decline in NPS 2012/13 as the base year for temporal adjustments. The consumption between 2008 and 2010 likely resulting from the third set of consumption aggregates includes clothing and global financial crisis. The middle-income groups (third and footwear (method 2) and uses NPS 2012/13 as the base year. fourth quintile) benefited most with their total consumption Overall, the temporally adjusted real consumption aggregate Appendix I 169 TABLE I.3: Comparison of Consumption Per Adult Per Month in Tanzania METHOD 1 METHOD 1 METHOD 2 2008/09 2010/11 2012/13 2014/15 2012/13 2014/15 2012/13 2014/15 NATIONAL Nominal consumption 48,140 55,817 78,012 85,389 78,012 85,389 80,781 89,015 Real consumption 49,032 56,512 80,033 87,722 80,033 87,722 82,867 91,441 Real consumption, temporally adjusted (survey-based) 59,887 56,512 58,751 62,151 80,033 84,698 82,867 86,950 Real consumption, temporally adjusted (CPI) 58,595 56,512 63,233 61,618 80,033 77,988 82,867 81,294 DAR ES SALAAM Nominal consumption 124,702 138,145 177,747 181,540 177,747 181,540 184,747 189,810 Real consumption 109,086 123,403 161,638 161,461 161,638 161,461 168,000 168,776 Real consumption, temporally adjusted (survey-based) 133,237 123,403 118,656 114,396 161,638 155,896 168,000 160,487 Real consumption, temporally adjusted (CPI) 130,362 123,403 127,709 113,413 161,638 143,544 168,000 150,048 OTHER URBAN Nominal consumption 69,006 74,314 103,889 118,069 103,889 118,069 107,487 122,848 Real consumption 66,957 72,978 106,927 120,404 106,927 120,404 110,638 125,283 Real consumption, temporally adjusted (survey-based) 81,781 72,978 78,493 85,307 106,927 116,254 110,638 119,130 Real consumption, temporally adjusted (CPI) 80,016 72,978 84,482 84,574 106,927 107,043 110,638 111,381 RURAL Nominal consumption 37,186 42,704 58,524 63,297 58,524 63,297 60,488 65,946 Real consumption 40,294 45,522 62,559 68,774 62,559 68,774 64,662 71,660 Real consumption, temporally adjusted (survey-based) 49,214 45,522 45,924 48,727 62,559 66,404 64,662 68,141 Real consumption, temporally adjusted (CPI) 48,152 45,522 49,428 48,308 62,559 61,143 64,662 63,708 ZANZIBAR Nominal consumption 44,103 54,171 65,876 74,433 65,876 74,433 69,202 77,932 Real consumption 41,805 54,304 72,755 81,474 72,755 81,474 76,412 85,293 Real consumption, temporally adjusted (survey-based) 51,061 54,304 53,409 57,725 72,755 78,666 76,412 81,104 Real consumption, temporally adjusted (CPI) 49,959 54,304 57,483 57,229 72,755 72,433 76,412 75,829 Source: Authors’ calculation based on NPS 2008/09, 2010/11, 2012/13 and 2014/15. Note: The highlighted year represents the base year. per adult varying by 10 to 15 percent between 2010/11 starting from round 3 using a more comprehensive consump- and 2014/15. Food and non-food consumption growth was tion aggregate. Table I.5 shows the food lines per month also strongest among this group at about 11 percent and for all four rounds of the NPS. The first major set of columns 14 ­ percent respectively. Between the same period, the rich- shows the food lines when consumption excludes clothing est quintile experienced an increase in total consumption and footwear and NPS 2010/11 is assumed as the base year. per adult of 8 percent, while the bottom 40 experienced The second major column shows the poverty lines when con- only a 4 percent increase. Table I.4 presents the average real sumption excludes clothing and footwear and NPS 2012/13 is consumption per adult per month across surveys for total con- ­ assumed as the base year. The final set of columns shows the sumption, food consumption and non-food consumption and poverty lines when consumption includes clothing and foot- their distribution by quintile. The disaggregation by quintile wear and NPS 2012/13 is assumed as the base year. Note the provides a snapshot of the welfare from the poorest to the inclusion or exclusion of an item to consumption, even a non- richest population groups and the evolution of their welfare. food one, will affect the food lines because the reference pop- ulation is chosen from the distribution of total consumption. ii.  Poverty lines To make food lines comparable across time, inter-year adjustments in cost of living are made. Rows (b) and The food line reflects the cost of acquiring a food basket (c) show the food line of the base year converted to the survey that delivers 2,200 calories per adult per day. A food line year prices using the survey-based price deflators and the CPI of TZS 23,371 per month is estimated should one wish to esti- respectively. In general, inflation using the CPI is lower for the mate poverty starting with round 1 using an aggregate that first three rounds of NPS, with prices increasing by 19 percent excludes clothing and footwear. A food line of TZS 32,339 and 26 percent, compared to the 22 ­ percent and 34 percent per month is estimated should one wish to estimate poverty increase using the Fisher food index. However, between NPS 170 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE I.4: Comparison of Total, Food and Non-Food Consumption Per Adult Per Month in Tanzania AGGREGATE EXCLUDES CLOTHING AND FOOTWEAR BASE AGGREGATE INCLUDES CLOTHING AND AGGREGATE EXCLUDES CLOTHING AND FOOTWEAR BASE YEAR = NPS 2010/11 YEAR = NPS 2012/13 FOOTWEAR BASE YEAR = NPS 2012/13 2008/09 2010/11 2012/13 2014/15 2012/13 2014/15 2012/13 2014/15 Mean total consumption per adult equivalent per month, adjusted by inter-year price variation (CPI) National 58,595 56,512 63,233 61,618 80,033 77,988 82,867 81,294 Poorest Quintile 21,443 20,310 20,800 20,846 26,326 26,385 27,278 27,625 2nd Quintile 33,471 32,071 33,808 33,777 42,789 42,750 44,317 44,588 3rd Quintile 45,193 43,718 47,644 48,248 60,302 61,066 62,425 63,707 4th Quintile 63,350 60,803 69,981 70,099 88,573 88,723 91,563 92,572 Richest Quintile 129,652 125,790 144,001 135,262 182,258 171,198 188,822 178,210 Mean food consumption per adult equivalent per month, adjusted by inter-year price variation (CPI) National 41,065 38,546 43,201 41,919 80,033 77,988 82,867 81,294 Poorest Quintile 16,183 14,952 15,796 15,366 26,326 26,385 27,278 27,625 2nd Quintile 25,621 23,767 25,791 25,325 42,789 42,750 44,317 44,588 3rd Quintile 34,321 32,136 35,177 35,154 60,302 61,066 62,425 63,707 4th Quintile 46,359 43,531 49,379 48,806 88,573 88,723 91,563 92,572 Richest Quintile 82,953 78,431 89,939 85,050 182,258 171,198 188,822 178,210 Mean non-food consumption per adult equivalent per month, adjusted by inter-year price variation (CPI) National 17,530 17,963 20,032 19,698 25,354 24,932 28,189 28,238 Poorest Quintile 2,835 2,796 2,662 2,857 3,369 3,616 4,301 4,873 2nd Quintile 5,438 5,813 5,638 5,946 7,136 7,525 8,736 9,385 3rd Quintile 9,099 9,758 9,920 10,914 12,555 13,813 14,708 16,333 4th Quintile 16,692 17,802 19,663 20,651 24,887 26,137 27,984 30,004 Richest Quintile 53,957 53,731 62,288 58,164 78,836 73,617 85,471 80,618 Source: Authors’ calculation based on NPS 2008/09, 2010/11, 2012/13 and 2014/15. Note: The highlighted year represents the base year. TABLE I.5: Comparison of Food Lines Across NPS Survey Rounds in Tanzania   Method 1 Method 1 Method2   2008/09 2010/11 2012/13 2014/15 2012/13 2014/15 2012/13 2014/15 (a) Survey-specific food line in survey year prices 18,027 23,371 32,339 33,729 32,339 33,729 32,286 33,689 Adjusted for inter-year variation in cost of living using survey-based deflators (b) Base year food line in survey year prices 19,135 23,371 31,373 32,986 32,339 34,009 32,286 33,438 Food price index 0.82 1 1.34 1.41 1 1.05 1 1.05 Adjusted for inter-year variation in cost of living using the CPI (c) Base year food line in survey year prices 19,557 23,371 29,580 33,272 32,339 36,376 32,286 36,316 CPI ratio 0.84 1 1.26 1.42 1 1.13 1 1.13 Source: Authors’ calculation based on NPS 2008/09, 2010/11, 2012/13 and 2014/15. Note: The highlighted year represents the base year. 2012/13 and NPS 2014/15, prices rose by 13 percent using the footwear will affect the poverty line more directly through the CPI compared to 5 percent using the Fisher food index. scaling factor. Table I.6 compares the basic needs poverty line across survey rounds. The basic needs poverty line reflects the food line and the cost of basic non-food essentials. The basic needs pov- erty line of TZS 30,053 in is estimated for 2010/11 when the iii.  Trends in food and basic aggregate excludes clothing and footwear and a poverty line needs poverty of TZS 40,008 is estimated for a more comprehensive aggre- gate starting in 2012/13. As with the food lines, the poverty Overall, food poverty has been declining slightly since lines will vary based on whether the consumption aggregate 2010/11. Figure I.6 shows the national trends for food includes/excludes clothing and footwear, the inter-year defla- ­ poverty rates using the harmonized methodology and var- tor used and the base year chosen. Inclusion of clothing and ious assumptions on the consumption aggregate and the Appendix I 171 TABLE I.6: Comparison of Basic Needs Poverty Lines Across NPS Survey Rounds in Tanzania   METHOD 1 METHOD 1 METHOD2   2008/09 2010/11 2012/13 2014/15 2012/13 2014/15 2012/13 2014/15 (a) Survey-specific food line in survey year prices 22,470 30,053 40,008 42,913 40,008 42,913 41,987 44,936 Adjusting for inter-year variation in cost of living using survey-based deflators (b) Base year poverty line in survey year prices 24,605 30,053 40,343 42,417 40,008 42,074 41,987 43,486 Food price index 0.82 1 1.34 1.41 1 1.05 1 1.05 Adjusting for inter-year variation in cost of living using the CPI (c) Base year poverty line in survey year prices (CPI) 25,148 30,053 38,037 42,784 40,008 45,002 41,987 47,228 CPI ratio 0.84 1 1.26 1.42 1 1.13 1 1.13 Source: Authors’ calculation based on NPS 2008/09, 2010/11, 2012/13 and 2014/15. Note: The highlighted year represents the base year. FIGURE I.6: Comparison of Food Poverty Rates Across FIGURE I.7: Comparison of Food Poverty Rates Across NPS Survey Rounds, Percent Survey Rounds by Geographic Domain, Percent 20 20 18.4 18.8 15.5 16.0 Food poverty incidence 14.0 14.3 Food poverty incidence 15 13.2 (% of population) 12.1 12.2 15 11.7 (% of population) 10 8.3 8.6 10 6.1 5 4.6 5 0.5 0.3 0 0 9 1 3 5 3 5 3 5 /0 /1 /1 /1 /1 /1 /1 /1 3 5 3 5 3 5 3 5 08 10 12 14 12 14 12 14 /1 /1 /1 /1 /1 /1 /1 /1 12 14 12 14 12 14 12 14 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 Method 1 Method 1 Method 2 Dar es Salaam Other urban Rural Zanzibar Source: Authors’ calculations based on NPS 2008/09, 2010/11, 2012/13 and 2014/15. Source: Authors’ calculations based on NPS 2008/09, 2010/11, 2012/13 and 2014/15. Note: Squares indicate the base year assumed for each method. Consumption Note: Squares indicate the base year assumed for each method. Consumption aggregate includes clothing and footwear. aggregate includes clothing and footwear. base year. While the different assumptions have little impli- FIGURE I.8: Comparison of Basic Needs Poverty Rates cation on the trends, the choices matter for the levels of food Across Survey Rounds, Percent poverty. Changing the base year for method 1 from NPS 2008/09 to NPS 2010/11 elevates the rates. 30 26.6 27.0 27.7 27.2 26.4 24.0 23.6 24.0 Poverty estimates reveal that about one in four people Poverty incidence (% of population) in Tanzania lack the minimum resources to afford a basic 20 standard of living. The harmonized methodology also sug- gests a decline in poverty since 2010/11 (Figure I.8). The 10 influence of the base year and type of aggregate on the lev- els of poverty is more muted than on the levels of food pov- 0 erty, offering estimates that are closer in range across the 9 1 3 5 3 5 3 5 /0 /1 /1 /1 /1 /1 /1 /1 08 10 12 14 12 14 12 14 methodologies. 20 20 20 20 20 20 20 20 Method 1 Method 1 Method 2 Source: Authors’ calculations based on NPS 2008/09, 2010/11, 2012/13 and 2014/15. Note: Squares indicate the base year assumed for each method. 172 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T BOX I.1 Poverty Lines The food poverty line is the cost of a food basket that deliv- The basic needs poverty line is based on the cost of attain- ers 2,200 calories per adult per day. The basket of foods is ing a basic standard of living. It consists of the cost of derived from the food consumption patterns of the reference acquiring food for adequate nutrition and the cost of non- population—the 2nd to 5th decile of the total per adult con- food essentials. The non-food component of the basic sumption distribution. Quantities consumed are converted needs poverty line reflects the average non-food con- into calories and valued at the median unit values prevailing in sumption of households with total consumption close to the country. The daily food line is estimated as follows: the poverty line. In this case, the reference households are those whose consumption lies between the food line zF = ∑q k k pk × 2,200 and 1.2 times the food line. The basic needs poverty line calculated by dividing the food line by the average is ­ ∑q k k ck proportion of total consumption devoted to food for the ­ reference group. where q is the total quantity of food item k consumed in the reference population, p is the national median unit value of item k and c is the caloric conversion factor for item k. iv.  Shared prosperity Zanzibar, those in the middle of the consumption distribution experienced much higher growth rates in per adult consump- Tanzanians experienced positive consumption growth tion than the rest of the population. after an initial period of negative growth, but richer Tanzanians gained more. The first three panels highlight Growth became more pro-poor after 2012. Beginning in the consumption growth starting from 2008 to each succes- 2012, the consumption growth of poorer Tanzanian’s began sive round (Figure I.9). Between 2008 and 2010, consumption to exceed that of richer Tanzanians (Figure I.11). To evalu- per adult shank for virtually all consumption groups, with sig- ate whether a more comprehensive consumption aggregate nificant declines for the poorest decile. Consumption fell by ­ provides a different picture of consumption growth, compar- about 2 percent each year. The decline is likely related to the isons of consumption growth are made when the aggregate effects of the global financial crisis, which started to manifest includes clothing and footwear versus when it excludes the in 2008 in the United States before spreading to the rest of aforementioned item. The inclusion or exclusion of clothing the world. Between 2008 and 2014, consumption growth was and footwear has little effect on the distributional pattern of positive though it showed signs of cooling off between 2012 growth. and 2014. However, the poorest decile, appear not to have Consumption inequality is moderate and stable except regained their 2008 levels of consumption. The fourth panel for the increase in 2012. The Gini index of real per ­ capita focuses on the period of growth between 2010 and 2014. ­ consumption per month – which measures the extent to which It reveals that all consumption groups made positive gains consumption is unevenly distributed in the p ­ opulation— in consumption, but that on average, those in the bottom is approximately 38 in round 1, 2 and 4. In 2012, ­inequality 40 experienced lower rates of growth. rose to about 40. Higher levels of inequality are observed on The 2010–2015 GICs reveals a period of pro-poor growth Mainland compared to Zanzibar. The relatively higher variabil- only for Dar es Salaam (Figure I.10). The slight rise in pov- ity of income on Mainland is driven by the welfare ­ differences erty and decline in inequality in Dar es Salaam thus sug- between urban and rural areas. Secondary cities, which have gests that while the city is becoming more equal, growth in the highest inequality, maintained a constant level of inequal- consumption of the poor was not sufficient to reduce pov- ity. However, inequality declined in Dar es Salaam while it erty. Rural areas, which experienced rising inequality amid grew in rural areas, leading to a switch in ranking over the declining poverty, saw a disproportionate positive growth years. Inequality on Zanzibar was more volatile over the years, in consumption for the top 60 percent. For urban areas and increasing in some and declining in others. Appendix I 173 FIGURE I.9: Growth Incidence Curves, 2008–2015 A. Growth incidence, Tanzania, 2008–2010 B. Growth incidence, Tanzania, 2008–2012 0 4 3 –1 Consumption growth (%) Consumption growth (%) 2 –2 1 –3 0 –4 –1 –5 –2 –6 –3 –7 –4 0 20 40 60 80 100 0 20 40 60 80 100 Consumption percentile Consumption percentile Growth rate by percentile Growth rate in mean Growth rate by percentile Growth rate in mean C. Growth incidence, Tanzania, 2008–2015 D. Growth incidence, Tanzania, 2010–2015 3 6 5 Consumption growth (%) 2 Consumption growth (%) 4 1 3 0 2 –1 1 –2 0 –3 –1 0 20 40 60 80 100 0 20 40 60 80 100 Consumption percentile Consumption percentile Growth rate by percentile Growth rate in mean Growth rate by percentile Growth rate in mean Source: Authors’ calculations based on NPS 2008/09, 2010/11, 2012/13 and 2014/15. Note: Method 1 consumption estimates are used. FIGURE I.10: Growth Incidence Curves by Geographic Domain, 2010–2015 A. Dar es Salaam, 2010–2015 B. Other urban areas, 2010–2015 6 8 5 7 Consumption growth (%) Consumption growth (%) 4 6 3 5 2 4 1 3 0 2 –1 1 –2 0 –3 –1 –4 –2 0 20 40 60 80 100 0 20 40 60 80 100 Consumption percentile Consumption percentile Growth rate by percentile Growth rate in mean Growth rate by percentile Growth rate in mean continued 174 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE I.10 C. Rural areas, 2010–2015 D. Zanzibar, 2010–2015 4 5 4 Consumption growth (%) 3 3 2 2 1 0 –1 1 –2 –3 0 –4 –5 0 20 40 60 80 100 0 20 40 60 80 100 Consumption percentile Consumption percentile Growth rate by percentile Growth rate in mean Growth rate by percentile Growth rate in mean Source: Authors’ calculations based on NPS 2010/11 and 2014/15. FIGURE I.11: Comparison of Growth Incidence Curves of Consumption Excluding and Including Clothing and Footwear, 2012–2015 A. Tanzania, 2012–2015 (consumption excludes clothing B. Tanzania, 2012–2015 (consumption includes clothing and footwear) and footwear) 7 7 6 6 5 Consumption growth (%) 5 Consumption growth (%) 4 4 3 3 2 2 1 0 1 –1 0 –2 –1 –3 –2 –4 –3 –5 –4 –5 0 20 40 60 80 100 0 20 40 60 80 100 Consumption percentile Consumption percentile Growth rate by percentile Growth rate in mean Growth rate by percentile Growth rate in mean Source: Authors’ calculations based on NPS 2012/13 and 2014/15. FIGURE I.12: Gini Index of Monthly Real Consumption Per Capita, Mainland 45 Gini index 40 40.4 38.8 35 2012/13 2014/15 Source: Authors’ calculations based on NPS 2012/13 and 2014/15. Appendix I 175 FIGURE I.13: Lorenz Curve and Gini Index by Geographic Domain Lorenz Curve by Area Lorenz Curve by Area Tanzania 2008/09 National Panel Survey Tanzania 2010/11 National Panel Survey 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Cumulative population proportion Cumulative population proportion Dar es Salaam Rural Line of perfect equality Other urban Zanzibar Lorenz Curve by Area Lorenz Curve by Area Tanzania 2012/13 National Panel Survey Tanzania 2014/15 National Panel Survey 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0 0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Cumulative population proportion Cumulative population proportion Dar es Salaam Rural Line of perfect equality Other urban Zanzibar 2008/09 2010/11 2012/13 2014/15 P90/ P90/ P90/ P90/   GINI INCOME SHARES GINI INCOME SHARES GINI INCOME SHARES GINI INCOME SHARES P10 P10 P10 P10 LOW TOP LOW TOP LOW TOP LOW TOP                   QUINTILE QUINTILE QUINTILE QUINTILE QUINTILE QUINTILE QUINTILE QUINTILE Mainland 38.2 5.2 6.9 45.6 38.6 5.4 6.8 46.1 40.3 6.1 6.2 47.0 38.8 5.7 6.4 45.5 Dar es Salaam 35.7 5.4 6.7 42.1 33.4 4.5 7.8 41.4 33.4 4.2 8.0 41.7 30.3 4.0 8.5 38.9 Other urban 36.4 5.3 6.6 43.8 36.2 4.9 7.3 43.5 36.1 5.4 6.9 43.3 35.6 5.5 6.7 41.8 Rural 31.9 4.1 8.2 40.5 32.7 4.2 8.0 41.0 35.1 4.7 7.3 42.8 33.7 4.5 7.7 41.6 Zanzibar 32.8 4.3 8.1 41.1 31.8 3.9 8.8 41.0 33.5 4.3 8.2 42.1 29.2 3.8 8.4 37.1 Source: Authors’ calculations based on NPS 2008/09, 2010/11, 2012/13 and 2014/15. Note: To ensure comparability across all survey rounds, the consumption expenditure excludes clothing and footwear. However, the estimates are robust across methodologies. The inclusion of clothing and footwear does little to affect the 2012/13 and 2014/15 Gini, Income shares, p90/p10 ratio and Lorenz curve. 176 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Poverty Projections Frequent and timely consumption and income data corroborate the forecasted poverty levels and trends with needed for designing programs and policies to eliminate estimates obtained from survey instruments collected later, poverty are not always available often due to the costly and to obtain prospective estimates where up-to-date pov- and time-consuming nature of primary data collection. erty estimates are not available. Poverty is projected using To facilitate poverty monitoring in situations of data scar- two approaches; the distribution-neutral approach and the city or between surveys rounds, alternative techniques can growth elasticity approach. The distribution neutral approach be employed to estimate poverty. One technique involves assumes everyone’s consumption (income) grows at the same projecting national poverty using a nationally representative rate while inequality essentially remains unchanged. Growth consumption data. Another approach involves estimating elasticity approach assumes both growth and distribution poverty using a nationally representative non-consumption effects influence poverty reduction. data into which consumption data is imputed. Both tech- niques avoid the associated costs and delays of collecting For both projection approaches, growth in GDP per cap- new data while allowing for more up-to-date and reliable ita in constant local currency unit is assumed to drive the poverty estimates. changes in consumption or poverty. Forecasts of GDP per capita growth are simulated through the World Bank’s mac- The latest available data from the Household Budget roeconomic modelling tool, the Macro-fiscal model (MFMod) Survey and macroeconomic forecasts is used to forecast and are adjusted by population growth. Growth’s relationship changes in national poverty headcount over the medium to poverty is more direct under the growth elasticity approach term. This allows us to derive a picture of future outlooks, as it measures the percent change in poverty resulting from BOX I.2 Distribution Neutral and Growth Elasticity Approach The distribution-neutral approach uses the consump- which is the ratio between annualized poverty growth tion distribution of HBS 2011/12 to project future pov- and the annualized GDP growth between periods A andB. erty rates assuming consumption grows over time without In this case, B is 2012 and A is 2007. P is the poverty rate inequality changing. The advantage of this approach is that and G is the GDP per capita in constant local currency ­ poverty rates will reflect the curvature of the consumption unit (LCU). The point-to-point method is calculated as ­ distribution since the actual consumption distribution is follows: used for the projections. The growth elasticity approach determines how much  PB  −1 ­ economic growth contributes to poverty reduction and   PA   E point −to −point = uses the poverty-to-growth elasticity to project future pov-  GB   −1 erty rates. The poverty-to-growth elasticity is the ratio of  GA   a percent change in poverty rates to a percent change in consumption (or income). It is estimated by two different methods, one being the annualized method and the other which is the ratio between poverty growth and the GDP being the point-to-point method. The annualized method is growth between periods A and B. calculated as follows: 1  PB  B− A −1   PA   Eannualized = 1  GB  B− A −1   GA   Appendix I 177 a percent change in growth. Under the neutral-distribution FIGURE I.14: Actual and Projected Poverty Rates Using approach, growth’s impact on poverty is channeled through the Distribution Neutral and Growth-Elasticity Approaches its effect on welfare. It is assumed there are factors that mit- 30 igate the effect of GDP per capita growth on welfare aggre- Poverty incidence (% of population) 28.2 gate growth from household surveys. The pass-through can be understood as this factor of adjustment. The assumptions 25 used for both projection approaches are highlighted in the 23.2 table below. 22.7 20 While poverty’s downward trend is expected to con- 19.4 tinue over the next years, Tanzania may not reach the 18.4 twin goals. Projections based on a distribution-neutral 15 approach shows more modest poverty reduction while 2012 2013 2014 2015 2016 2017 2018 2019 2020 national elasticities show a faster rate of decline suggest- Distribution-neutral, pass-though = 0.29 ing a more optimistic trajectory of welfare.1 The distribution Annualized elasticity = –1.40 neutral approach, which offers a more moderate outlook of Point-to-point elasticity = –1.23 Regional elasticity = –0.73 poverty reduction, best reflects Tanzania’s poverty reduc- tion trajectory. Source: Authors’ calculations using HBS 2007, HBS 2011/12, and MFMod forecasts of GDP per capita in constant LCU adjusted by population growth (March 2018 version). The growth elasticity approach is likely to underestimate Notes: Fig. I.14: Actual poverty rate in 2012. Projection is from 2013 to 2020. Projections use: poverty given the location of the poverty line relative to  a)  neutral distribution with empirical pass-through = 0.29; b) annualized mean consumption and initial inequality. Elasticity approach elasticity (2007–2012); c) point-to-point elasticity (2007–2012) and d) point- to-point elasticity at the regional level. provides optimistic results because it accounts for inequality  b)  Fig. I.15: Vertical line is the logged HBS 2011/12 poverty line. at a lower level.2 The curvature of the consumption distribu- tion implies that the relationship between poverty reduc- tion and growth is non-constant. This relationship changes FIGURE I.15: Kernel Density Function of Log Real with each successive shift in the consumption distribution Monthly Consumption Per Adult and with changes in the distribution. The closer the poverty 0.8 line is to the mean of the distribution, the higher the elas- ticity as is the case in Tanzania (see figure below). However, this relationship weakens with each successive upward shift 0.6 in the consumption distribution assuming the poverty line is located below the mean. This implies the use of elasticity for 0.4 0.2 TABLE I.7: Growth Elasticities and Pass-Through Rates GROWTH-POVERTY TYPE OF PROJECTION ELASTICITY TYPE ELASTICITY PASS-THROUGH 0 (2007–2012) 8 10 12 14 16 Distribution Neutral 0.29 ln(monthly expenditure per adult equivalent, real) Growth Elasticity Annualized —1.40 HBS 2007 HBS 2011/12 Growth Elasticity Point-to-point —1.23 Growth Elasticity Regional† —0.73 Source: Authors’ calculations using HBS 2007, HBS 2011/12, and MFMod forecasts of GDP per capita in constant LCU adjusted by population growth (March 2018 version). Source: Authors’ calculations based on the HBS 2007 and 2011/12 and MFMod Notes: forecasts (March 2018 version). Fig. I.13: Actual poverty rate in 2012. Projection is from 2013 to 2020. Projections use: † The regional elasticity reflects the growth-poverty elasticity for Sub-Saharan  c)  neutral distribution with empirical pass-through = 0.29; b) annualized Africa for the period circa 1990–2012 based on poverty rates using the elasticity (2007–2012); c) point-to-point elasticity (2007–2012) and d) point- international poverty line of US$ 1.90 (2011 PPP). It is estimated using the point-to- to-point elasticity at the regional level. point methodology.  d)  Fig. I.14: Vertical line is the logged HBS 2011/12 poverty line. 1 Since the regional elasticity is lower than the national elasticities, it provides a more moderate forecast of poverty reduction. The lower regional elasticity relative to the national elasticities is expected since the regional poverty rates are higher than the national rates, thus they will produce a lower percent change in poverty than would national poverty rates. 2 The survey-to-survey imputations suggest inequality increased slightly between 2012 and 2016. 178 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T poverty projections is likely to underestimate poverty rates. However, since the distribution of consumption appears to The distribution in consumption can also change. Between have increased since 2011, the poverty reduction impact from 2007 and 2011, the consumption distribution became more growth may have been attenuated, thus resulting in projec- equal, which is incorporated into the growth elasticity. tions that underestimate poverty. Poverty Imputations Using the Demographic and Health Survey Survey-to-survey imputation (SSI) is a method that aims to ii.  Comparability of HBS facilitate poverty monitoring in situations where the avail- and DHS surveys able nationally representative survey does not contain consumption data. SSI uses an imputation model developed The accuracy of survey-to-survey predictions relies on key on consumption data from a representative survey to impute assumptions: (i) the two survey questions have variables in into non-consumption data sets. The imputed consumption common that are solicited in a consistent ­manner, (ii) the vari- data is then used to estimate poverty. The surveys utilized in ables in common between the two surveys explain a large share this imputation exercise consist of a base survey on which a of the intertemporal change in household expenditure and consumption model is developed and two complementary poverty, and (iii) the sampling design between the two surveys surveys—one implemented prior to the base survey and the is relatively similar (Newhouse, Shivakumaran et al. 2014). other implemented after—on which imputed estimates of consumption are developed. Comparability in geographical coverage across the HBS and DHS surveys is ensured. First, the geographic scope of DHS 2010 and 2015/16 is restricted to Mainland. While i. Data HBS 2011/12 is based on a representative sample of Tanzania Mainland, the DHS 2010 and 2015/16 provides a represen- The 2011/12 Tanzania Household Budget Survey (HBS) tative sample for the United Republic of Tanzania (Mainland is used as the base survey to model consumption.3 and Zanzibar). Second, new regions in the DHS 2015 are Household expenditures and consumption were collected recoded to match the original regions in HBS 2011/12 since using a diary that recorded both household purchases and they did not exist during the implementation of HBS 2011/12 consumption over a 28 days period (National Bureau of (Table I.8). As of the 2012 census, four new administrative Statistics 2013). Using this expenditure and consumption data, regions—Geita, Katavi, Njombe and Simiyu—were created. the HBS reports a consumption-based poverty headcount No adjustment was made to DHS 2010 since the sampling index of 28.2 percent based on the national basic needs pov- frame was based on the 2002 Population and Hosing Census erty line is TZS 36,482 per adult equivalent per month (World as was the HBS 2011/12. Finally, a sample for Dar es Salaam Bank 2015). is identified for the DHS data. While HBS is representative of Dar es Salaam, other urban areas and rural areas on Mainland, The Tanzania Demographic and Health Surveys (DHS) 2010 DHS is representative of urban and rural areas on Mainland and 2014 is used to impute consumption and thus pov- and on Zanzibar. Since DHS is not representative at the level erty. The DHS is a nationally representative survey that solicits of Dar es Salaam, the DHS sample size for Dar es Salaam is information on fertility, family planning, childhood mortal- smaller than if a separate stratum had been created for it. The ity, nutrition, maternal and child health, domestic violence, smaller DHS sample size for Dar es Salaam implies relatively malaria, adult mortality, and HIV/AIDS-related knowledge less precise predictions for Dar es Salaam as compared to the and behavior. A nationally representative sample of women other strata. age 15–49 in all selected households and men age 15–49 in one-third of selected households were interviewed (National The time-period of the HBS and DHS should also match. Bureau of Statistics and ICF Macro 2011). The fieldwork of the HBS survey was conducted for one year 3 The survey collects data on a wide range of individual and household characteristics, including: education, employment and health status; ownership of consumer goods and assets; housing structure and building materials; household access to services and facilities; ownership of non-farm businesses; and agricultural activities. Appendix I 179 TABLE I.8: New Regions created in March 2012 and Correspondence to Original Regions and Districts NEW REGION ORIGINATED FROM Geita The region was created from Bukombe district in Shinyanga region, Chato from Kagera region and Geita from Mwanza region Katavi Mpanda district from Rukwa region Njombe Ludewa, Makete, and Njombe districts from Iringa region Simiyu Formed by taking Bariadi, Meatu, and Maswa districts from Shinyanga region and Busega district from Mwanza Source: www.statoids.com/utz.html during October 2011–October 2012. The DHS 2010 was car- utilized to select a subset of predictors from the set of com- ried out December 2009–May 2010 and the DHS 2015 carried mon variables in the HBS 2011/2012 and DHS 2010 and out August 2015–February 2016. Given the overlap between 2015/16. A final selection step involves the manual inspection the DHS and the HBS, there was no need to restrict the period of the sensibility of results. in the base survey where the consumption model is devel- oped. However, to control for the seasonal factors, monthly The second step is choosing a model to estimate the dummies are included in the imputation model. coefficients for each predictor. Two types of models are ­ tested—a linear regression and predictive mean ­ matching. Variables used in the consumption models need to be ­ ossible Sensitivity to the inclusion of cellphones as a p available in both the HBS and DHS. The common vari- ­ predictor is tested. The regression considers sampling able set used to develop the consumption model consists weights (pweight in STATA). Finally, to allow for differences of household demographics, dwelling characteristics, own- across domains, a separate consumption model is developed ership of durables, ownership of land, characteristics of the for each of the strata: Dar es Salaam, other urban areas, and household head and his/her spouse, location variables, and rural areas. The poverty estimates of the consumption mod- indicators of month of interview. The stability of the variables els are compared with the poverty estimates from actual con- ensures that the consumption model developed in the base sumption data in the consumption survey. survey can reasonably be used to impute consumption into the non-consumption survey. The third step involves producing multiple imputations expenditure estimates for each household based in the non-consumption-survey. Based on the resulting imputed iii. Sample consumption estimates, poverty headcount and their stan- dard errors are calculated. The sample sizes used for the analysis are reflected in Table I.9. The lasso (least absolute shrinkage and selection operator) is a shrinkage and selection method for linear regression. TABLE I.9: Survey Sample Sizes The method minimizes the sum of squared errors, but is sub-   HBS 2011/12 DHS 2010 DHS 2015 ject to an extra penalty term that sets a bound on the sum of Dar es Salaam 3,016 383 700 the absolute values of the coefficients, as can be seen in the Other Urban 3,040 1,856 2,934 following equation: Rural 4,130 7,384 8,929 Mainland 10,186 9,623 12,563 2 Sources: HBS 2011/12, DHS 2010 and 2015. n  p  p p ∑  y i − β 0 − ∑ β j x ij  + λ ∑ β j = RSS + λ ∑ β j .  i =1 j =1 j =1 j =1 iv. Methodology A model is fit using all predictors and the penalty term shrinks the coefficient estimates towards zero. If the tuning The imputation process consists of three stages: the first parameter λ preceding the penalty term is sufficiently large, stage consists of choosing predictors for developing a some of the estimates are forced to be equal to zero. In this model of log per capita consumption of a household using way, the lasso performs variable selection and yields models the HBS 2011/12 consumption survey. Two automated vari- that involve only a subset of the variables. Therefore, select- able selection methods—lasso and backwards stepwise—are ing a good value of λ for the lasso is crucial and depending 180 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T on the value of λ, the lasso can produce a model involving any the sign of their coefficients are different across the differ- number of variables. ent modeling strata since these situations can be an artifact of multi-collinearity and may affect the precision of the esti- An optimal λ is chosen through cross-validation. A grid of mated coefficients. λ values is created between the λ that results in a model with no parameters to one that results in the regular linear regres- Multiple imputation (MI) is a Monte Carlo simulation-based sion estimates and the cross-validation error for each value procedure that better reflects the uncertainty inherent in a of λ is computed. The tuning parameter value for which the given prediction model. Instead of predicting a single value cross-validation error is smallest is selected. Finally, the model for a target outcome variable, several cases (in this study 20) is re-fit using all of the available observations and the selected are predicted to better reflect sampling variability. The overall value of the tuning parameter. For this procedure, the glmnet prediction for a given observation is then simply the average and caret R packages are used (Kuhn 2008, Friedman, Hastie over the 20 estimates from the separate simulations while the et al. 2010). final variance estimate reflects variation within and between simulation rounds. Lasso estimates can increase prediction accuracy in cases when the least squares estimates have excessively high For a continuous target variable, two typical multiple variance. This reduction in variance comes at the expense imputation methods are linear regression and predictive of an increase in bias however (James, Witten et al. 2013). mean matching (PMM). The linear regression model relies This has popularized the practice of post-lasso estimation– on normality of the model and is superior to other methods that is, using lasso regression as a pre-processing step, where when the underlying normality holds, but can be more sen- the subset of non-zero variables selected is then subse- sitive than other methods, such as PMM, to violations of this quently used in an ordinary least squares (OLS) linear regres- assumption (StataCorp 2017). In the case of linear regres- sion to generate non-biased coefficient estimates (Belloni and sion, we use the complete-data observations (this is the HBS Chernozhukov 2013). 2011/12 for which we have consumption information) to esti- mate a linear regression with an associated set of coefficients A further variable selection method one can use is b that relates the target variable y log per capita household b ackwards stepwise regression to iteratively drop ­ consumption with the rest of the predictors. The model is non-­s ignificant coefficients. The post-lasso OLS defined as: c oefficient e ­ ­ stimates, based on the subset of non-zero variables selected by the lasso pre-processing step, may b + eh(1) yh = a + X’h    ­ everal non-significant coefficients. Backwards results in s stepwise ­ v ariable selection begins with the full least where yh is log per capita expenditure, a is the intercept, X’ squares model containing all predictors (in this case the is the vector of explanatory variables for household h, b is non-zero lasso s ­ ubset), and while the least-significant the vector of regression coefficients, and eh is a stochastic term is “insignificant” at the specified significance level error term. We specify probability weights for each obser- of 5 ­ p ercent, removes it and re-estimates the model and vation where each observation is weighted by the inverse repeats the p ­ rocedure until all terms are significant at of its probability of being sampled and use a robust White the 5 percent level (StataCorp 2017). Note that in both sandwich estimator to compute the variance-covariance the lasso and stepwise routines, dummies of a categori- matrix. cal variables are treated as individual predictors and their significance are tested i ­ ndividually and not as a group, Multiple imputation by linear regression follows the which could lead to ­ d ifferent results. The final subset of ­following steps: significant predictors is then used in multiple imputation • Fit the weighted linear regression model (1) on the routine. observed data to obtain estimates of the model parameters β ˆ and σˆ 2. The predictors and their associated coefficients that results from the two automated procedures described 2 • Simulate new parameters b* and σ * from their joint above undergo one final manual check. Variables are posterior distribution under the conventional non-in- removed if their coefficients exhibit a different sign than suggested by their bi-variate correlation with income, or if ( ) formative improper prior Pr β ,  σ 2 ∝ 1 σ 2  . Appendix I 181 TABLE I.10: Comparison of Means of Variables by Geographic Domain OTHER URBAN Comparison of Means of Variables (linearized standard errors in parentheses) VARIABLE TYPE VARIABLE 2010 2012 2015 C Total number of members living in the HH 4.7 (0.1) 4.7 (0.2) 4.6 (0.1) C Number of children aged 0-5 years 0.8 (0.0) 0.8 (0.1) 0.8 (0.0) D hcomp==single parent with kids 0.2 (0.0) 0.2 (0.0) 0.1 (0.0) D hcomp==couple with kids 0.6 (0.0) 0.6 (0.0) 0.6 (0.0) D floor==Ceramic tiles/marumaru 0.0 (0.0) 0.0 (0.0) 0.1 (0.0) D floor==Earth/sand/Dung 0.3 (0.0) 0.3 (0.0) 0.2 (0.0) D cook==Firewood 0.3 (0.0) 0.3 (0.0) 0.3 (0.0) D ttoilet_fc==Flush toilet 0.3 (0.0) 0.2 (0.0) 0.4 (0.0) D Radio and Radio Cassette 0.7 (0.0) 0.6 (0.0) 0.6 (0.0) D Telephone (land line) 0.0 (0.0) 0.0 (0.0) 0.0 (0.0) D Television 0.3 (0.0) 0.3 (0.0) 0.4 (0.0) D Iron (Charcoal or electric) 0.5 (0.0) 0.4 (0.0) 0.4 (0.0) D Motor Vehicles 0.0 (0.0) 0.0 (0.0) 0.1 (0.0) D Motor cycle 0.1 (0.0) 0.1 (0.0) 0.1 (0.0) D (Head) Gender 0.7 (0.0) 0.7 (0.0) 0.7 (0.0) D attain==incomplete secondary 0.2 (0.0) 0.2 (0.0) 0.1 (0.0) D Household head has no spouse (1) or has a spouse (0) 0.4 (0.0) 0.4 (0.0) 0.4 (0.0) D attainsp==incomplete secondary 0.1 (0.0) 0.1 (0.0) 0.0 (0.0) D Livestock na. na. 0.3 (0.0) 0.3 (0.0) C for continuous variable and D for dummy RURAL Comparison of Means of Variables (linearized standard errors in parentheses) C Total number of members living in the HH 5.3 (0.1) 5.3 (0.1) 5.3 (0.1) C Number of children aged 0-5 years 1.1 (0.0) 1.1 (0.0) 1.0 (0.0) C Adult females over 18 years old 1.3 (0.0) 1.3 (0.0) 1.3 (0.0) C Adult males over 18 years old 1.2 (0.0) 1.2 (0.0) 1.1 (0.0) D hcomp==single parent with kids 0.1 (0.0) 0.1 (0.0) 0.1 (0.0) D hcomp==couple with kids 0.7 (0.0) 0.7 (0.0) 0.7 (0.0) D Number of sleeping rooms 2.3 (0.0) 2.1 (0.0) 2.2 (0.0) D floor==Earth/sand/Dung 0.8 (0.0) 0.8 (0.0) 0.8 (0.0) D wall==Timber 0.0 (0.0) 0.0 (0.0) 0.0 (0.0) D toilet==Shared toilet 0.2 (0.0) 0.2 (0.0) 0.2 (0.0) D cook==Firewood 0.9 (0.0) 0.9 (0.0) 0.9 (0.0) D cook==Kerosene 0.0 (0.0) 0.0 (0.0) 0.0 (0.0) D Radio and Radio Cassette 0.6 (0.0) 0.5 (0.0) 0.5 (0.0) D Iron (Charcoal or electric) 0.2 (0.0) 0.1 (0.0) 0.1 (0.0) D Motor Vehicles 0.0 (0.0) 0.0 (0.0) 0.0 (0.0) D Motor cycle 0.0 (0.0) 0.0 (0.0) 0.1 (0.0) D Bicycle 0.5 (0.0) 0.4 (0.0) 0.4 (0.0) C Head years of education 4.5 (0.1) 4.9 (0.1) 5.0 (0.1) D enrol==primary 0.0 (0.0) 0.0 (0.0) 0.0 (0.0) D attain==incomplete primary 0.2 (0.0) 0.2 (0.0) 0.2 (0.0) D attain==complete primary 0.4 (0.0) 0.5 (0.0) 0.5 (0.0) D attain==incomplete secondary 0.0 (0.0) 0.0 (0.0) 0.0 (0.0) C Number of spouses of household head 0.7 (0.0) 0.7 (0.0) 0.7 (0.0) D Household head has no spouse (1) or has a spouse (0) 0.3 (0.0) 0.3 (0.0) 0.3 (0.0) (Table Continiued on next page) 182 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE I.10: Comparison of Means of Variables by Geographic Domain (Continued) DAR ES SALAAM Comparison of Means of Variables (linearized standard errors in parentheses) VARIABLE TYPE VARIABLE 2010 2012 2015 C Total number of members living in the HH 4.1 (0.2) 4.0 (0.1) 4.3 (0.1) C Adult females over 18 years old 1.2 (0.1) 1.3 (0.0) 1.3 (0.0) D hcomp==single parent with kids 0.1 (0.0) 0.1 (0.0) 0.1 (0.0) D hcomp==couple with kids 0.5 (0.0) 0.5 (0.0) 0.6 (0.0) D hcomp==elderly HH 65& above 0.0 (0.0) 0.0 (0.0) 0.0 (0.0) D cook==Firewood 0.1 (0.0) 0.0 (0.0) 0.0 (0.0) D ttoilet_fc==Traditional pit latrine 0.6 (0.1) 0.6 (0.0) 0.4 (0.0) D Refrigerator or freezer 0.3 (0.0) 0.3 (0.0) 0.4 (0.0) D Television 0.6 (0.0) 0.6 (0.0) 0.6 (0.0) D Iron (Charcoal or electric) 0.6 (0.0) 0.6 (0.0) 0.6 (0.0) D (Head) Gender 0.8 (0.0) 0.8 (0.0) 0.8 (0.0) D agehd==agehd 40-49 0.2 (0.0) 0.2 (0.0) 0.2 (0.0) D agehd==agehd 50-59 0.2 (0.0) 0.1 (0.0) 0.1 (0.0) C Head years of education 8.3 (0.4) 8.7 (0.2) 8.8 (0.3) D attain==higher 0.1 (0.0) 0.1 (0.0) 0.1 (0.0) D Household head has no spouse (1) or has a spouse (0) 0.4 (0.0) 0.4 (0.0) 0.3 (0.0) D agesp==agesp 50-59 0.0 (0.0) 0.0 (0.0) 0.1 (0.0) C First spouse years of education 4.5 (0.2) 5.0 (0.2) 5.7 (0.3) D month==November 0.0 (0.0) 0.1 (0.0) 0.0 (0.0) D month==December 0.1 (0.1) 0.1 (0.0) 0.0 (0.0) D Livestock na. na. 0.1 (0.0) 0.2 (0.0) Source: DHS 2010, HBS 2011/12, DHS 2015/16. Note: Linearized standard errors in parentheses. D=dummy, C=continuous. • Obtain one set of imputed values y m by simulating 1 • Find the d donor candidates for which y ˆi − y ˆj 2 ­ inimal, and randomly sample one of them. is m from N ( X m β * ,  σ * Ih1×h1 ) where y = (y1,y2, …,yh)’. Usual values for d are 3, 5 and 10. We use d = 5. • Repeat steps 2 and 3 to obtain M sets of imputed These ­ constitute the cases’ “closest neighbors”. 1 m ­values y m  ,… ,  y m . 1 The imputed values y m will be the observed yi from the yj min donor candidate identified above. The predictive mean matching MI method combines the standard linear regression method above with a PMM is especially attractive for predictive variables that nearest-neighbor imputation approach. The imputed values ­ are not normally distributed since by drawing from the are generated by modifying step 3 above: observed data, PMM preserves the distribution of the observed values in the missing part of the data, which  denote the predicted value of the rows with an • Let y i makes it more robust than the fully parametric linear regres- observed yi where i = 1, …,hi. sion approach (StataCorp 2017). This procedure results in  20 imputed datasets which are then analyzed and combined • Likewise, let y j denote the predicted value of the rows using Rubin’s rules for valid statistical inferences that properly with missing yj where y = 1, …,h0. reflect the uncertainty due to missing values (Rubin 1987). Appendix I 183 v. Results Final backwards and forwards imputation results into the 2010 and 2015 respectively, as well as the observed (direct Table shows the mean values of the final selected vari- estimation) values for 2012 are shown in Table I.12. Our ables. The results show little signs of volatility between the imputation results suggest that the poverty rate decreased HBS 2011/12 and DHS survey rounds highlighting the consis- nationally approximately 1 percentage point between 2010– tency in which the questions were solicited and captured. 2015. This decrease was concentrated in Dar es Salaam - the capital saw a decrease of 2 percentage points, while Rural areas Two consumption models are compared; one with cell- had a decrease of half a percentage point. Other Urban areas phone ownership and one without cellphone ownership show an increase of approximately 1 percentage point. A note as a predictor variable. The validity of an imputation model of caution about interpreting these results is in order. Given the depends on the assumption of constant relative prices magnitude of these estimated changes, as well as the width across time. Stifel and Christiaensen (2006) recommend of their confidence intervals, it is perhaps more reasonable excluding from the imputation models variables whose rates to assume that poverty has remained relatively stable in the of return are likely to change markedly in the face of evolv- period under analysis in Tanzania. ing economic conditions. Likewise, Harttgen, Klasen et al. (2013) argue for removing certain consumer durables since The Gini coefficient shows more pronounced changes improvement in asset ownership that outpaces income during the same period. A national increase of three per- growth can create “asset drift”. In many developing coun- centage points is observed while both other urban areas tries, cellphones are among the assets for which ownership and rural areas had an increase of two percentage points. has expanded rapidly, and Tanzania is no exception (see Along with the decrease in poverty, Dar es Salaam also saw a Table I.11). This reflects the overall rapid reduction in cell decrease in inequality with the Gini coefficient decreasing by phone prices and service fees and not so much of income one percentage point. growth. As a result, owning a cell phone may have a very dif- ferent budgetary impact in 2015 than in 2010 (Newhouse, Shivakumaran et al. 2014Newhouse, Shivakumaran et al. TABLE I.12: Poverty Estimates and Imputations 2014). The final models excluded cellphones for the reasons POVERTY HEADCOUNT outlined above. All tested model specifications including (% OF POPULATION) DHS 2010 HBS 2011/12 DHS 2015/16 cellphones in the imputation model resulted in more opti- Mainland 28.1 28.2 26.8 (1.2) (1.8) (1) mistic poverty trends. Other Urban 18.9 21.5 20.1 (2.7) (4.5) (2.1) Comparing the sensitivity of the distribution of imputed Rural 31.9 33.4 31.5 values to observed values for each stratum led to the lin- (1.4) (2.3) (1.2) ear regression MI method for the Dar es Salaam and other Dar es Salaam 7.6 4.0 5.3 urban strata and PMM for the rural stratus to be chosen. (2.9) (.7) (1.7) Source: Authors’ calculations based on HBS 2011/12, DHS 2010 and 2015/16. In addition to variables with inconsistent coefficient signs, regional dummies were also dropped to decrease the effect of time-invariant variables. TABLE I.13: Gini Coefficient TABLE I.11: Share of Cellphone Ownership GINI COEFFICIENT DHS 2010 HBS 2011/12 DHS 2015/16 DHS 2010 HBS 2011/12 DHS 2015/16 Mainland 0.33 0.34 0.36 46 57 78 Other Urban 0.36 0.37 0.38 (1.1) (1.3) (0.7) Rural 0.30 0.29 0.32 Source: Authors’ calculations based on HBS 2011/12, DHS 2010 and 2015/16. Dar es Salaam 0.33 0.35 0.32 Note: Standard errors in parentheses. Source: Authors’ calculations based on HBS 2011/12, DHS 2010 and 2015/16. 184 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Using synthetic panel to explore movements in and out of poverty, 2010–2014. This section presents results for poverty dynamics in i.  Technical choices and Tanzania based on synthetic panel estimates for the 2nd motivation and 4th round of the National Panel Survey (NPS). The results for the overall population are presented contain infor- In the next lines, the choices made to construct the syn- mation on poverty profiles, i.e. movements in and out of thetic panel are detailed. The main criterion to decide poverty for different population subgroups, defined by char- between different options has been the successful prediction acteristics of the household head. These tables contain results of the overall poverty rate in period 2. The results presented both for joint probabilities (i.e. the absolute probability of below show that the combination selected for obtaining the showing a particular type of poverty dynamics) and condi- main results displayed does best at predicting poverty rates in tional probabilities (i.e. the probability of transitioning to or period 2. staying in a particular poverty status in 2014, given one´s pov- erty status in 2010). The main results presented in this doc- Income Model: The models to predict consumption expen- ument are obtained using the method outlined in Dang diture must consist exclusively of variables that do not and Lanjouw (2013) [henceforth DL]. These results are much vary over time. “Model 1” includes only information on sex, more accurate—in terms of getting the right poverty num- marital status, literacy, education and big region of birth. bers in each of the rounds—than DL results from previous “Model 2” additionally contains information on characteristics versions, which has been achieved by weighting the income that are invariant as long as the household does not move, regressions. such as whether the household lives in a rural or urban area, and whether the household head is indigenous to the area The work required synthetic panel estimation of mobil- she lives in, with particular dummies for those that migrated ity to be split into three parts, and in all of them it to the North or the East. The models include only significant is necessary to make choices. The first is the choice variables, Model 1 has an R2 of around 0.3 while Model 4 has of a good model for predicting consumption expendi- an R2 of around 0.4. ture, the second is the estimation of the intertemporal ­ correlation of residuals in those models, and the third is In addition to the income model, there is the additional to join the pieces (combined with assumptions on error choice whether to estimate the model with or without structure) to produce final estimates of transitions in and weights. This is mostly a practical question, as there is little out of poverty. The choices taken in all of these steps are theoretical guidance of why either option would be prefer- explained below. able. The Table below shows that the estimates are much better when population weights are used to estimate these income models. Then, results from Model 1 are consider- TABLE I.14: Poverty Rates by Age, 2010 and 2014 NPS 2 ably more accurate. This can be explained by the fact that AGE RANGE MEAN STD. ERROR CONFIDENCE INTERVAL households move from urban to rural areas and between All 26.51 1.26 24.03 28.99 regions, but Model 2 assumes that these characteristics are 25–70 26.56 1.28 24.05 29.07 fixed. Therefore, predicting with characteristics from period 20–65 24.53 1.32 21.93 27.14 1, it misses those improvements and predicts slightly higher NPS 4 poverty rates. AGE RANGE MEAN STD. ERROR CONFIDENCE INTERVAL All 23.93 1.36 21.25 26.60 Residual distribution: There are two main options in the liter- 25–70 24.03 1.40 21.27 26.78 ature: Dang and Lanjouw (2013)1 suggest to assume bivariate 20–65 24.57 1.50 21.62 27.51 normality, and Bourguignon and Moreno (2015)2 [hence- Source: Authors’ calculations based on NPS 2010/11 and 2014/15. Note: The estimates account for complex survey design. forth BM] put forward an alternative based on calibrating the parameters of a Gaussian mixture g(u) that best fits the distri- bution of innovations ui = e2 – re1, and then simulating from 1 http://documents.worldbank.org/curated/en/967911468330016425/pdf/WPS6504.pdf 2 https://editorialexpress.com/cgi-bin/conference/download.cgi?db_name=NEUDC2015&paper_id=301 Appendix I 185 that distribution in order to obtain synthetic expenditure for TABLE I.15: Poverty Dynamics based on Synthetic Panels, round 2. It is a priori not obvious to which extent the more Tanzania 2010–2014 involved and flexible BM procedure should be of advantage, JOINT PROBABILITIES CONDITIONAL PROBABILITIES it probably depends on the degree of violation of the bivari- POVERTY STATUS ESTIMATE POVERTY STATUS ESTIMATE ate normality assumption. Poor, Non-poor 15.91 Poor -> Non-poor 57.91 Non-poor, Poor 12.79 Non-poor -> Poor 16.68 Given that the DL procedure has been used more widely Non-poor, Non-poor 60.02 Non-poor -> Non-poor 83.32 until now and is therefore more consolidated, the team Poor, Poor 11.29 Poor -> Poor 42.09 decided to prioritize it in this application. A comparison of Source: Authors’ calculations based on NPS 2010/11 and 2014/15. Note: Predictions are obtained using population weights. N = 3529 for NPS2 and tables shows that, when income models are estimated with N =3066 for NPS4. The sample includes household heads that were between 20 weights, the fit obtained with the DL approach in terms of and 75 years old by the time of NPS2. Estimations are based on income Model 1 (weighted regressions) and DL methodology (ρy estimated from cohort means, predicting correctly the poverty rates in periods 1 and 2 is very and normality assumed for residuals). good, and comparable if not better to the performance of the BM* model that had emerged as the most accurate from the comparison of methodological choices with unweighted TABLE I.16: Characteristics Associated with Poverty regressions (see Tables below). Mobility, Synthetic Panel Estimates, Tanzania 2010–2015: Joint Probabilities POOR, NON-POOR, NON-POOR, POOR, An additional (maybe minor) degree of freedom when CHARACTERISTICS HHH NON-POOR POOR NON-POOR POOR applying the DL method is whether to produce the esti- 1. HOUSEHOLD CHARACTERISTICS mates based on the households of round 1 or on those SEX HHH   seen in round 2. Table 5 also shows that using round 1 char- Female 17.36 13.27 55.97 13.40 acteristics leads to better estimates. Male 15.52 12.65 61.12 10.71 Age HHH Age range: 20–75. The performance of synthetic panel Younger than 35 12.02 13.13 66.41 8.43 methods is only slightly better than for the other pre-selected Between 36 and 50 15.89 13.02 59.93 11.17 age range (25–60). Moreover, the evolution of the poverty rate Older than 50 18.75 12.24 55.52 13.50 for the broader range fits more closely the evolution of the Literacy HHH Non-literate 20.09 18.00 39.56 22.35 overall population, with a certain (non-significant) decline in Literate 14.57 11.11 66.58 7.74 poverty between 2010 and 2014 (see Table below). In order to Education HHH reflect that change in the synthetic panel estimates, the age Lower than complete primary 19.75 16.25 45.59 18.41 range 20–75 is chosen. complete primary 15.65 11.83 64.60 7.91 Higher than complete primary 4.12 5.03 89.78 1.06 Residual correlation ρ: This parameter, which measures Civil Status HHH the inter-temporal correlation of the part of income not Never married 7.83 6.61 82.68 2.88 explained by the prediction model for consumption expendi- Married - monogamous 15.10 12.06 62.90 9.94 ­ anels Dang ture, is essential in the construction of synthetic p Married - polygamous 17.09 14.04 55.38 13.49 and Lanjouw (2013) provide a formula to derive this inter-tem- Living together 17.33 14.05 55.40 13.22 poral correlation of errors in the ­models above from the Separated / Divorced 12.22 15.64 61.20 10.94 Widow 20.08 11.50 55.13 13.29 inter-temporal correlation of income ρy, so that we just need Education Father HHH to obtain the latter. Father never went to school 15.98 13.74 57.68 12.61 Father went to school 15.90 12.69 60.25 11.15 DL suggest using inter-temporal correlation of mean Education Mother HHH cohort expenditure (meaning average household expendi- Mother never went to school 16.00 12.77 59.93 11.29 ture for households with a head born in the same year(s)) Mother went to school 15.10 12.91 60.76 11.22 as a proxy for household inter-temporal correlation of Sector HHH expenditure. An alternative to this approach would be to All except agriculture/fishing 12.43 10.49 69.55 7.53 consider the household correlation coefficient as obtained Agriculture or fishing 17.43 13.79 55.86 12.92 from the panel for rounds 1 and 3 for the corresponding age Employment type HHH Farmers and fishers (all) 17.64 13.74 55.50 13.11 range (see Table below). Regular wage 12.24 10.88 68.82 8.06 Self employed 14.04 11.85 65.28 8.83 Others 16.59 12.37 59.58 11.46 186 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE I.17: Characteristics Associated with Poverty TABLE I.18: Characteristics Associated with Poverty Mobility, Synthetic Panel Estimates, Tanzania 2010–2015, Mobility, BM* Synthetic Panel Estimates, Tanzania 2010– Joint Probabilities 2015, Joint Probabilities CHARACTERISTICS HHH POOR, NON-POOR, NON-POOR, POOR, POOR, NON-POOR, NON-POOR, POOR, NON-POOR POOR NON-POOR POOR CHARACTERISTICS HHH NON-POOR POOR NON-POOR POOR 2. LOCATION, DEMOGRAPHIC CHARACTERISTICS AND DWELLING OWNERSHIP  1. Sex HHH Rural/Urban         Female 20.50 12.27 54.74 12.49 Male 15.21 13.87 60.93 9.98 Urban 12.68 10.21 69.81 7.30 2. Age Rural 17.03 13.68 56.63 12.67 Older than 45 18.77 13.06 56.31 11.86 Migration Younger than 45 14.18 13.94 62.56 9.32 3. Literacy Never migrated 16.85 13.56 57.13 12.46 Non-Literate 22.05 17.99 38.98 20.98 Migrated more than 5 Literate 14.52 12.10 66.22 7.17 years ago 16.68 11.93 60.42 10.97 4. Education Migrated less than 5 Lower than Compl. years ago 12.64 11.43 67.58 8.35 Primary 21.04 16.82 45.05 17.09 Compl. Primary 15.41 12.74 64.42 7.44 Region Higher than Compl. West 15.81 19.22 46.80 18.16 Primary 4.44 5.68 88.87 1.01 North 14.90 9.26 69.33 6.51 5. Civil Status Central 19.40 14.52 49.33 16.75 Never Married 9.64 5.47 81.64 3.25 Married - Monogamous 14.69 13.35 62.76 9.19 S Highlands 18.10 12.13 59.14 10.63 Married -Polygamous 17.96 15.91 53.97 12.16 South 15.24 15.25 56.85 12.66 Living together 17.06 14.96 55.37 12.60 SW Highlands 18.65 12.22 57.80 11.33 Separated / Divorced 14.87 12.69 63.23 9.21 Lake 16.33 14.31 56.89 12.46 Widow 23.30 11.05 52.29 13.36 East 10.57 7.59 77.52 4.32 Benchmark: Overall Population 16.35 13.53 59.61 10.52 Zanzibar 16.24 9.90 64.39 9.47 Source: Authors’ calculations based on NPS 2010/11 and 2014/15. Household Size Note: Predictions are obtained using population weights. N = 3529 for NPS2 and 4 or less HH members 16.42 12.96 58.79 11.83 N = 3066 for NPS4. The sample includes household heads that were between 20 and 75 years old by the time of NPS2. Estimations used Model 1, BM calibration 5 or more HH members 14.31 12.23 63.87 9.59 (adapted for residual weights) and simulation procedure to deal with residuals Female Share in HH (with 1000 simulations in the simulation step) and ρy from NPS1–3 panel Less than 50% women 15.88 12.59 60.40 11.13 50% women or more 15.95 13.05 59.51 11.50 TABLE I.19: Characteristics Associated with Poverty Dependency Ratio Mobility, Synthetic Panel Estimates, Tanzania 2010–2015, No dependents 13.35 11.15 67.19 8.31 Conditional Probabilities POOR -> NON-POOR NON-POOR-> POOR -> Below 25% 15.31 10.60 64.02 10.08 CHARACTERISTICS NON-POOR -> POOR NON-POOR POOR Between 25% and 50% 15.57 12.22 61.61 10.61 1. CHARACTERISTICS HOUSEHOLD HEAD Between 50% and 75% 16.69 14.05 56.70 12.56 Sex HHH Above 75% 18.07 14.52 52.85 14.56 Female 55.87 18.26 81.74 44.13 Male 58.57 16.28 83.72 41.43 People per room Age HHH Less than 2 people Younger than 35 57.93 15.72 84.28 42.07 per room 15.14 12.40 62.07 10.39 Between 36 and 50 58.12 16.95 83.05 41.88 2 people per room Older than 50 57.68 17.12 82.88 42.32 or more 16.57 13.12 58.25 12.06 Literacy HHH Non-Literate 46.73 30.37 69.63 53.27 Dwelling ownership Literate 64.99 13.45 86.55 35.01 Does not own 11.79 11.32 69.34 7.55 Education HHH Owner 16.81 13.11 57.98 12.10 Lower than Compl. Primary 51.19 25.33 74.67 48.81 Benchmark: Compl. Primary 66.20 14.55 85.45 33.80 Higher than Compl. Primary 79.09 4.76 95.24 20.91 Overall Population 15.91 12.79 60.02 11.29 Civil Status HHH Source: Authors’ calculations based on NPS 2010/11 and 2014/15. Never Married 72.63 6.80 93.20 27.37 Note: Predictions obtained using population weights. N = 3529 for NPS2 Married - Monogamous 59.70 15.23 84.77 40.30 and N = 3066 for NPS4. The sample includes household heads between 20 Married -Polygamous 55.26 19.30 80.70 44.74 and 75 years old by the time of NPS2. Estimations based on income Model 1 Living together 56.17 19.28 80.72 43.83 (weighted) and DL methodology. Separated / Divorced 51.73 19.56 80.44 48.27 Widow 59.83 16.27 83.73 40.17 (Table Continiued on next page) Appendix I 187 TABLE I.19: Characteristics Associated with Poverty TABLE I.20: Characteristics Associated with Poverty Mobility, Synthetic Panel Estimates, Tanzania 2010–2015, Mobility, BM* Synthetic Panel Estimates, Tanzania 2010– Conditional Probabilities (Continued) 2015. Conditional Probabilities POOR -> NON-POOR NON-POOR-> POOR -> POOR -> NON-POOR -> NON-POOR -> POOR -> CHARACTERISTICS NON-POOR -> POOR NON-POOR POOR CHARACTERISTICS HHH NON-POOR POOR NON-POOR POOR Education Father HHH 1. Sex HHH Father never went to School 55.23 18.37 81.63 44.77 Father went to School 58.20 16.52 83.48 41.80 Female 62.15 18.31 81.69 37.85 Education Mother HHH Male 60.38 18.54 81.46 39.62 Mother Never went to School 58.05 16.68 83.32 41.95 2. Age Mother went to School 56.67 16.67 83.33 43.33 Sector HHH Older than 45 61.27 18.83 81.17 38.73 All except agriculture/fishing 61.62 12.34 87.66 38.38 Younger than 45 60.35 18.23 81.77 39.65 Agriculture or fishing 56.87 18.85 81.15 43.13 Employment type HHH 3. Literacy Farmers and fishers (all) 56.82 18.90 81.10 43.18 Non-Literate 51.24 31.58 68.42 48.76 Regular Wage 59.53 12.92 87.08 40.47 Literate 66.95 15.44 84.56 33.05 Self employed 60.81 14.52 85.48 39.19 Others 58.64 16.32 83.68 41.36 4. Education 2. LOCATION, DEMOGRAPHIC CHARACTERISTICS AND DWELLING OWNERSHIP Lower than Compl. Primary 55.18 27.18 72.82 44.82 Rural/Urban         Compl. Primary 67.45 16.51 83.49 32.55 Urban 62.89 12.00 88.00 37.11 Higher than Compl. Primary 81.51 6.01 93.99 18.49 Rural 56.78 18.53 81.47 43.22 Migration 5. Civil Status Never migrated 56.90 18.26 81.74 43.10 Never Married 74.81 6.28 93.72 25.19 Migrated more than 5 years ago 59.83 15.60 84.40 40.17 Married - Monogamous 61.51 17.54 82.46 38.49 Migrated less than 5 years ago 59.52 13.70 86.30 40.48 Region Married -Polygamous 59.63 22.77 77.23 40.37 West 45.67 28.30 71.70 54.33 Living together 57.51 21.28 78.72 42.49 North 69.40 10.93 89.07 30.60 Central 53.11 21.75 78.25 46.89 Separated / Divorced 61.74 16.72 83.28 38.26 S Highlands 62.74 16.04 83.96 37.26 Widow 63.56 17.45 82.55 36.44 South 53.92 20.27 79.73 46.08 SW Highlands 61.95 16.46 83.54 38.05 Benchmark: Lake 56.16 19.18 80.82 43.84 Overall Population 60.84 18.50 81.50 39.16 East 70.58 8.22 91.78 29.42 Source: Authors’ calculations based on NPS 2010/11 and 2014/15. Zanzibar 62.75 12.44 87.56 37.25 Note: Predictions are obtained using population weights. N = 3529 for NPS2 and Household Size N = 3066 for NPS4. The sample includes household heads that were between 20 4 or less HH members 57.55 17.17 82.83 42.45 and 75 years old by the time of NPS2. Estimations used Model 1, BM calibration 5 or more HH members 59.25 15.23 84.77 40.75 (adapted for residual weights) and simulation procedure to deal with residuals Female Share in HH (with 1000 simulations in the simulation step) and ρy from NPS1–3 panel. Less than 50% women 58.22 16.37 83.63 41.78 50% women or more 57.51 17.10 82.90 42.49 Dependency Ratio No dependents 61.07 13.44 86.56 38.93 Below 25% 59.75 13.39 86.61 40.25 Between 25% and 50% 58.89 15.68 84.32 41.11 Between 50% and 75% 56.45 18.94 81.06 43.55 There are arguments in favor of each of the choices (com- Above 75% 54.82 20.62 79.38 45.18 piled in B: Individual or cohort rho?). The authors would People per room be inclined to use the individual 2008–2012 ρy, while others Less than 2 people per room 58.73 15.79 84.21 41.27 2 people per room or more 57.29 17.48 82.52 42.71 strongly favor the cohort 2010–2014 option. As long as we use Dwelling ownership the cohort ρy estimated with 1-year cohorts, which is the clos- Does not own 60.20 13.27 86.73 39.80 est to the individual panel ρy, practical differences are going Owner 57.58 17.53 82.47 42.42 Benchmark: to be from negligible to inexistent. As a compromise, The Overall Population 57.91 16.68 83.32 42.09 authors use that cohort rho (0.541), as long as it is mentioned Source: Authors’ calculations based on NPS 2010/11 and 2014/15. explicitly that it is very similar to the panel parameter from Note: Predictions obtained using population weights. N = 3529 for NPS2 and N = 3066 for NPS4. The sample includes household heads between 20 and 75 years old by the 2008–2012, which has been done in the DL estimates pre- time of NPS2. Estimations based on income Model 1 (weighted) and DL methodology. sented below. 188 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE I.21: Model Selection. DL, Weighted vs. Unweighted Regressions PRED. POVERTY ACTUAL POVER- ACTUAL POVER- MODEL AGE RANGE BASE ROUND WEIGHTS R1 TY R1 DIFF. R1 POVERTY R2 TY R2 DIFF. R2 DIFF. TREND 1 25-60 1 No 20.96 24.76 3.80 18.78 24.45 5.67 –1.87 1 25-60 1 Yes 26.04 24.76 –1.28 23.94 24.45 0.51 –1.79 1 25-60 2 No 20.51 24.76 4.25 19.04 24.45 5.41 –1.16 1 25-60 2 Yes 25.42 24.76 –0.66 24.00 24.45 0.45 –1.11 1 20-75 1 No 22.04 26.87 4.83 18.80 24.11 5.31 –0.48 1 20-75 1 Yes 27.20 26.87 –0.33 24.07 24.11 0.04 –0.37 1 20-75 2 No 20.73 26.87 6.14 18.82 24.11 5.29 0.85 1 20-75 2 Yes 25.45 26.87 1.42 23.81 24.11 0.30 1.12 2 25-60 1 No 22.68 24.76 2.08 21.74 24.45 2.71 –0.62 2 25-60 1 Yes 25.94 24.76 –1.18 26.23 24.45 –1.78 0.60 2 25-60 2 No 19.72 24.76 5.04 20.14 24.45 4.31 0.73 2 25-60 2 Yes 23.00 24.76 1.76 24.20 24.45 0.25 1.51 2 20-75 1 No 23.54 26.87 3.32 21.47 24.11 2.64 0.68 2 20-75 1 Yes 27.12 26.87 –0.25 26.17 24.11 –2.06 1.81 2 20-75 2 No 20.05 27.87 7.82 19.87 24.11 4.24 3.58 2 20-75 2 Yes 23.41 27.87 4.45 23.98 24.11 0.13 4.32 Source: Authors’ calculations based on NPS 2010/11 and 2014/15. Note: All estimates using population weights. ρ derived from ρy estimated from NPS 2-4 1-year cohorts for the corresponding age range. DL estimates can be obtained using NPS2 (Rd. 1) or NPS4 (Rd.2) characteristics. The “benchmark” differs minimally from the poverty rates in Table 1 due to a few dropped observations because of missing observations for the variables in the prediction model. Standard errors for the (slightly different) benchmark poverty rates can be found in Table 1. TABLE I.22: Model Selection. Unweighted Regressions, DL vs. BM MODEL AGE METHOD POVOUT POVIN NEVERPOOR BOTHPOOR POVERTY R2 BENCHMARK 1 25-60 DL (Rd. 1) 12.92 10.74 68.29 8.04 18.78 24.45 1 25-60 DL (Rd. 2) 12.37 10.90 68.59 8.14 19.04 24.45 1 25-60 BM 15.29 11.32 63.92 9.47 20.79 24.45 1 25-60 BM* 14.22 13.29 61.95 10.54 23.83 24.45 1 20-75 DL (Rd. 1) 14.29 11.05 66.91 7.75 18.80 24.11 1 20-75 DL (Rd. 2) 13.11 11.21 68.06 7.62 18.82 24.11 1 20-75 BM 17.65 11.23 61.90 9.22 20.45 24.11 1 20-75 BM* 16.35 13.53 59.61 10.52 24.05 24.11 2 25-60 DL (Rd. 1) 13.84 12.90 64.42 8.84 21.74 24.45 2 25-60 DL (Rd. 2) 12.12 12.54 67.74 7.60 20.14 24.45 2 25-60 BM 14.94 13.01 62.23 9.82 22.83 24.45 2 25-60 BM* 14.02 14.84 60.40 10.74 25.58 24.45 2 20-75 DL (Rd. 1) 15.14 13.06 63.39 8.41 21.47 24.11 2 20-75 DL (Rd. 2) 12.92 12.74 67.21 7.13 19.87 24.11 2 20-75 BM 17.31 12.76 60.37 9.56 22.32 24.11 2 20-75 BM* 16.16 14.92 58.21 10.71 25.63 24.11 Source: Authors’ calculations based on NPS 2010/11 and 2014/15. Note: All estimates using population weights. ρ derived from ρy from NPS1-3 panel for the corresponding age range. DL estimates can be obtained using NPS2 (Rd. 1) or NPS 4 (Rd.2) characteristics. BM estimates obtained based on Round 1 characteristics. BM* estimates differ from BM estimates by taking into account residual weights in the calibration step. The “benchmark” differs minimally from the poverty rates in Table 1 due to a few dropped observations because of missing observations for the variables in the prediction model. Standard errors for the (slightly different) benchmark poverty rates can be found in Table 1. TABLE I.23: ρy for Different Age Ranges, Panel estimates TABLE I.24: ρy for Different Cohort Definitions. 20–75, ρy HOUSEHOLDS, NPS 1–3 Pseudo-panel estimates AGE RANGE UNWEIGHTED WEIGHTED N ρy HOUSEHOLDS, NPS 2–4 All 0.5136 0.535 4870 COHORT DEFINITION WEIGHTED UNWEIGHTED COHORTS 20–75 0.5361 0.554 4491 Year of Birth 0.5451 0.5458 56 25–60 0.5596 0.5724 3482 2 Years of Birth 0.7851 0.7807 28 Source: Authors’ calculations based on NPS 2010/11 and 2014/15. 3 Years of Birth 0.7939 0.7767 19 Note: All coefficients are significantly different from 0 at the 1% level (verified with Source: Authors’ calculations based on NPS 2010/11 and 2014/15. pwcorr, sign). Note: All coefficients are significantly different from 0 at the 1% level (verified with pwcorr, sign). Appendix I 189 TABLE I.25: ρy for Different Cohort Definitions. 25–60, TABLE I.26: ρy over 2 years in Different Periods, Panel Pseudo-panel estimates estimates ρy HOUSEHOLDS, NPS 2–4 ρy HH, NPS 1–2 ρy HH, NPS 2–3 COHORT DEFINITION WEIGHTED UNWEIGHTED COHORTS AGE RANGE WEIGHTED UNWEIGHTED WEIGHTED UNWEIGHTED Year of Birth 0.6074 0.6398 36 All 0.593 0.6159 0.6256 0.6058 2 Years of Birth 0.6999 0.7219 18 20–75 0.6006 0.6222 0.6332 0.6141 3 Years of Birth 0.8279 0.8404 12 25–60 0.6113 0.6409 0.6489 0.6347 Source: Authors’ calculations based on NPS 2010/11 and 2014/15. Source: Authors’ calculations based on NPS 2010/11 and 2014/15. Note: All coefficients are significantly different from 0 at the 1% level Note: All coefficients are significantly different from 0 at the 1% level (verified with (verified with pwcorr, sign). pwcorr, sign). ii.  Detailed description of report for details about these geographic zones). Using “big models and some variables regions” is convenient not only from reasons related to esti- mation (higher significance, parsimony), it is as well not obvi- Model 1: sex, marital status, literacy, education and big ous what would be the best way to deal with new regions in region of birth of the household head. NPS4 if we would use the smaller ones. Model 2: Additionally, whether the household lives in a rural Education: Categories merged to “below primary”, “some or urban area, and whether she is indigenous to the area he/ primary” (up to 6th degree), completed ­ primary (“7th or she lives in, with particular dummies for those that migrated higher”), “some lower secondary”, “completed lower sec- to the North or the East. ondary” (Form IV as highest education), “upper second- ary” (if A) or “diploma” and “university”. The decision is Marital status: Same as “detailed marital status” with the based on checking that grades within a c ­ ategory had similar exception that separated and divorced households are coefficients in the regressions, and also on a superficial look merged into one category. on the structure of the education system (upper secondary is split into A-levels and diploma courses). Region of birth of the household head: Information obtained from dist_born is in individual file. This is pre- Migrant: Use information in res_dur on whether the house- ferred to “region where the household lives” in the income hold has always lived in the same “district”3. Authors add models because it’s more clearly time-invariant. Initial 2 dummies for households that have moved to either the East 25/30 regions merged to the 9 DHS Geographic zones (or or the North, because they are highly significant in prediction “big regions”- see chapter 5 of this volume and 2015/16 DHS of expenditures. Additional material for analysis of poverty dynamics. The first three rounds of the NPS are used to analyze pov- The number is slightly higher at 14,680 if the relevant restric- erty dynamics in Tanzania. Most of the analysis uses the bal- tion is that the respondent was in 2008 and 2012.5 anced panel of individuals which is comprised of those who were successfully interviewed in NPS1, NPS2 and NPS3.4 The Selective attrition is not a major concern when using the first three rounds of the NPS tracked eligible household mem- first three rounds of the NPS as a longitudinal dataset. bers between 2008 and 2012, and those who were successfully Normally, one of the biggest concerns when modelling pov- interviewed in multiple survey rounds comprise the “balanced erty dynamics using longitudinal data is selective attrition. If panel” used in the analysis. The full balanced panel dataset individuals and households exit the panel in a non-random comprises 14,464 individuals for whom there is consumption way, then there is the possibility that results could be unreli- expenditure data in each of the first three rounds of the NPS. able. Fortunately, the attrition rates in the NPS are very low 3 Looking at the variable “district born”, it seems to refer to what the household data file calls “region”. 4 This restriction is necessary for the analysis of three-round poverty status, and for the analysis of chronic versus transitory poverty. 5 See, for example, the round 1 to round 3 transition matrices, and the regression results for poverty exit and entry between NPS1 and NPS3. 190 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T FIGURE I.16: Data and Assumptions used for Analyzing Poverty Dynamics NPS Actual NPS Actual NPS (Aggregate excludes clothing and footwear) 2008/09 panel 2010/11 panel 2012/13 NPS Synthetic Synthetic NPS (Aggregate excludes clothing and footwear) 2010/11 panel panel 2014/15 Note: Dark boxes reflect the base year under consideration. and do not significantly bias estimates of the poverty head- Four rounds of NPS are used to analyze poverty dynam- count or poverty trends. 6 ics in Tanzania. The number of rounds of data being used, and the kind of analysis being done (for example, longitudinal or NPS4, which was implemented in 2014/15, is a cross-­ cross-sectional) depends on the particular research question sectional survey of households and individuals based on being addressed. In 2008 the proportion of missing data on the a new redrawn sample, and therefore cannot be used as occupational sector of the household head is so high as to be part of a panel in the same way that 2008 to 2012 can. unusable. As such, all analysis of occupational sectors uses a Figure I.16 presents a graphical description of the panel combination of NPS2, NPS3 and NPS4. In contrast, there is reli- data used for the panel analysis (top graphic) and con- able data on the main source of income in the household for all trasts it to the structure of the data used in the NPS2 to four rounds of the NPS, and trends and transitions of this vari- NPS4 synthetic panel analysis (bottom graphic). There is, able are presented over the longest possible time period. however, value in including NPS4 in the analysis because it can be used to provide the most recent poverty and sector cross-sectional decompositions for Tanzania. This chapter A description of the poverty rates uses data following the outline in the top part of the ­ figure – and population shares that is, households and individuals are directly linked through time. Poverty rates increased significantly in rural areas, even though they were flat or fell in all other areas between 2008 and 2012. This was the case for both the full sample in each round of data, and for the 14,709 panel members who FIGURE I.17: Kernel Density Distributions of Real were present in the first and third rounds. Even though the Consumption Expenditure share of the population living in rural areas decreased slightly from 2008 to 2012, the rise in the poverty rate in this group 0.8 from 28.1 percent to 30.3 percent drove the overall aggregate increase in poverty over the period. As shown in Table I.27, 0.6 the overall increase in the poverty rate was slightly higher for Density balanced panel members than it was for the full sample.7 0.4 0.2 The geographical differences in the extent of poverty were stark in 2008, and they had increased further by 0 2012. The poverty rate for panel members living in rural areas 8 10 12 14 was 27.9 percent in 2008 and grew to 32.1 percent in 2012. Log of real 2010 adult equivalent expenditure This stands in contrast to a drop from 3.1 percent to 0.5 per- Wave 1 Wave 2 Wave 3 cent in Dar es Salaam, and from 29.5 percent to 21.8 percent in Zanzibar. Source: Authors’ calculations based on NPS 2008/09, 2010/11, 2012/13. 6 The attrition rate between NPS1 and NPS2 was 3%, while the attrition rate between NPS2 and NPS3 was 4%. 7 Balanced panel members are those respondents who were successfully interviewed in each of the first three rounds of the NPS. Appendix I 191 TABLE I.27: Poverty Rates and Population Shares FULL SAMPLE BALANCED PANEL MEMBERS ONLY ROUND 1 ROUND 3 ROUND 1 ROUND 3 POV. RATE POP. SHARE POV. RATE POP. SHARE POV. RATE POP. SHARE POV. RATE POP. SHARE Dar es Salaam 3.3% 1.0% 3.1% 0.5% 6.9% 8.4% 6.4% 9.0% (0.003) (0.001) (0.003) (0.002) Other Urban 11.6% 11.8% 11.8% 11.1% 14.7% 14.1% 14.6% 14.0% (0.007) (0.006) (0.008) (0.008) Rural 28.1% 30.3% 27.9% 32.1% 75.4% 74.6% 76.1% 73.8% (0.005) (0.004) (0.005) (0.006) Zanzibar 28.8% 20.7% 29.5% 21.8% 2.9% 2.9% 2.9% 3.2% (0.009) (0.010) (0.010) (0.011) Overall 24.0%   24.9%   24.0%   26.0%   (0.004) (0.003)   (0.004) (0.004) N 16 709 24 727   14 680 14 680 Source: Authors’ calculations based on NPS. TABLE I.28: Proportion of Panel Members Poor and TABLE I.30: Marginal Effects of Probit for Poverty Exit Non-Poor, 2008 to 2010 TRANSITIONS OUT OF POVERTY ROUND 2 ROUND 1 VARIABLES W1 TO W2 W1 TO W3 NON-POOR POOR Non-poor 60.8 12.1 72.9 Individual  Age –0.000 0.001 Poor 14.6 12.5 27.1 (0.001) (0.001) Round 1 75.4 24.6 100  Male –0.031 0.005 Source: Authors’ calculations based on NPS.   (0.019) (0.023)  Married 0.028 0.032 (0.030) (0.032)   Primary education 0.017 0.072** TABLE I.29: Transition Matrix: 2010 Poverty Status (0.021) (0.029) Conditional On 2008 Poverty Status   Secondary education 0.145** 0.083 (0.057) (0.061) ROUND 2   Other education 0.008 –0.024 NON-POOR POOR (0.059) (0.078) Non-poor 83.4 16.6 100   Moved district (NPS3) 0.329*** (0.037) Round 1 Poor 53.7 46.3 100 Household Head Source: Authors’ calculations based on NPS.   Male household head 0.057** -0.028 (0.027) (0.031)   Age of household head –0.001 0.000 (0.001) (0.001) FIGURE I.18: Share of Chronic and Transitory Poverty by   Primary education 0.050** –0.090*** Place of Residence (0.024) (0.029)   Secondary education 0.398*** 0.240*** (0.032) (0.041)   Other education 0.166*** -0.064 Chronic poor (0.063) (0.075) Household   Number of children –0.027*** –0.001 Transient poor (0.005) (0.005)   Number of adults –0.000 0.016** (0.006) (0.007) Never poor   Number of elders –0.067*** –0.014 (0.021) (0.026) 0 10 20 30 40 50 60 70 80 90 100   Other urban –0.222*** –0.050 Percentage (0.061) (0.069)  Rural –0.282*** –0.214*** Zanzibar Rural Other urban Dar es Salaam (0.051) (0.060)  Zanzibar –0.279*** –0.293*** Source: Authors’ calculations based on NPS 2008/09, 2010/11, 2012/13. (0.056) (0.066) (Table Continiued on next page) 192 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE I.30: Marginal Effects of Probit for Poverty Exit TABLE I.31: Marginal Effects of Probit for Poverty Entry (Continued) (Continued) TRANSITIONS OUT OF POVERTY TRANSITIONS INTO POVERTY ROUND 1 VARIABLES W1 TO W2 W1 TO W3 ROUND 1 VARIABLES W1 TO W2 W1 TO W3 Main income source   Number of adults 0.005** -0.011***   Cash crops –0.018 –0.021 (0.002) (0.003) (0.031) (0.036)   Number of elders –0.021** –0.013  Livestock 0.303*** 0.078 (0.011) (0.012) (0.070) (0.083)   Other urban 0.062*** 0.080***  Business 0.135*** -0.170*** (0.010) (0.011) (0.043) (0.052)  Rural 0.194*** 0.176***  Wages -0.056 0.006 (0.009) (0.008) (0.036) (0.045)  Zanzibar 0.125*** 0.160***  Remittances 0.088* -0.119* (0.013) (0.016) (0.053) (0.062)  Other –0.143*** 0.005 Main income source (0.050) (0.059)   Cash crops –0.034** –0.043*** Observations 3,535 2,454 (0.013) (0.015) Standard errors in parentheses  Livestock –0.041** –0.043** *** p<0.01, ** p<0.05, * p<0.1 (0.020) (0.021) Source: Authors’ calculations based on NPS 2010/11,2012/13 and 2014/15.  Business –0.094*** –0.103*** (0.014) (0.017)  Wages –0.103*** –0.060*** (0.015) (0.017) TABLE I.31: Marginal Effects of Probit for Poverty Entry  Remittances –0.106*** –0.059** TRANSITIONS INTO POVERTY (0.024) (0.025) ROUND 1 VARIABLES W1 TO W2 W1 TO W3  Other –0.037 –0.027 (0.028) (0.032) Individual Observations 11,952 8,462  Age 0.000 –0.001* Standard errors in parentheses (0.000) (0.000) *** p<0.01, ** p<0.05, * p<0.1  Male 0.002 0.008 Source: Authors’ calculations based on NPS 2010/11, 2012/13 and 2014/15. (0.008) (0.010)  Married –0.023* 0.003 (0.012) (0.012)   Primary education –0.033*** –0.043*** (0.010) (0.014)   Secondary education –0.071*** –0.113*** (0.018) (0.020)   Other education –0.086*** –0.089*** (0.020) (0.029)   Moved district (NPS3) –0.039*** (0.015) Household Head   Male household head –0.037*** –0.010 (0.011) (0.013)   Age of household head 0.001** -0.001 (0.000) (0.000)   Primary education –0.037*** –0.052*** (0.011) (0.014)   Secondary education –0.074*** –0.147*** (0.012) (0.018)   Other education –0.076*** –0.084*** (0.021) (0.024) Household   Number of children 0.012*** 0.007*** (0.002) (0.002) (Table continued on next page) Appendix I 193 Additional Material for Analysis of Poverty, Sectoral Mobility and Shocks TABLE I.32: Inter-Temporal Household Income Source Composition and Transitions: 2008–2012 INITIALLY POOR   2012/13   FOOD CROP SALES CASH CROP SALES LIVESTOCK SALES BUSINESS INCOME WAGES REMITTANCES OTHER 2008/9 Food crop sales 41.1% 6.3% 1.0% 8.1% 8.3% 1.4% 2.5% Cash crop sales 2.7% 4.7% 0.2% 1.0% 0.9% 0.0% 0.3% Livestock sales 0.7% 0.1% 0.5% 0.2% 0.0% 0.1% 0.0% Business income 1.4% 0.4% 0.1% 2.7% 0.8% 0.2% 0.4% Wages 2.5% 0.4% 0.4% 0.9% 2.5% 0.0% 0.1% Remittances 0.8% 0.2% 0.4% 0.5% 0.6% 0.2% 0.1% Other 1.0% 0.2% 0.0% 0.8% 0.7% 0.1% 1.6% INITIALLY NON-POOR   2012/13 2008/9 Food crop sales 27.1% 4.9% 1.5% 5.2% 5.5% 1.5% 1.1% Cash crop sales 4.0% 2.8% 0.6% 0.9% 0.4% 0.2% 0.3% Livestock sales 1.3% 0.2% 1.4% 0.5% 0.5% 0.2% 0.0% Business income 3.3% 1.3% 0.1% 8.9% 2.6% 0.5% 0.5% Wages 2.2% 0.4% 0.0% 4.0% 8.9% 0.3% 0.3% Remittances 1.2% 0.1% 0.1% 0.9% 0.7% 0.8% 0.1% Other 0.4% 0.2% 0.1% 0.7% 0.7% 0.1% 0.4% Source: Authors’ calculations based on NPS 2010/11 and 2012/13. TABLE I.33: Household Income Source Transition Matrices Conditional on Initial Poverty Status: 2008–2012 INITIALLY POOR     2012/13   FOOD CROP SALES CASH CROP SALES LIVESTOCK SALES BUSINESS INCOME WAGES REMIT. OTHER 2008/9 Food crop sales 59.8% 9.2% 1.4% 11.9% 12.1% 2.1% 3.6% 100 Cash crop sales 27.7% 47.3% 2.4% 10.4% 8.8% 0.0% 3.4% 100 Livestock sales 47.6% 3.3% 31.9% 9.8% 0.0% 7.4% 0.0% 100 Business income 22.9% 6.5% 2.3% 45.6% 13.8% 2.5% 6.4% 100 Wages 37.0% 5.6% 5.5% 13.0% 37.9% 0.0% 1.0% 100 Remittances 29.0% 6.5% 13.6% 17.8% 21.8% 7.5% 3.8% 100 Other 22.8% 4.4% 0.0% 18.6% 16.6% 1.1% 36.4% 100 INITIALLY NON-POOR     2012/13 2008/9 Food crop sales 57.9% 10.5% 3.3% 11.0% 11.7% 3.2% 2.4% 100 Cash crop sales 43.8% 31.0% 6.0% 9.3% 4.5% 1.9% 3.5% 100 Livestock sales 30.6% 4.4% 33.4% 12.7% 12.7% 5.6% 0.6% 100 Business income 19.3% 7.4% 0.6% 51.7% 15.3% 3.0% 2.8% 100 Wages 13.7% 2.6% 0.1% 24.7% 55.3% 2.0% 1.7% 100 Remittances 30.4% 1.3% 2.5% 23.1% 18.1% 21.2% 3.4% 100 Other 15.9% 8.2% 2.6% 26.1% 28.7% 4.0% 14.5% 100 Source: Authors’ calculations based on NPS. 194 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE I.34: Marginal Effects of Probit Regressions on TABLE I.35: Marginal Effects of A Probit Regression on Poverty Status, 2010, 2012 and 2015 Poverty Status In 2012 Conditional On 2010 Baseline VARIABLES (1) POOR NPS2 (2) POOR NPS3 (3) POOR NPS4 Variables Industry –0.138*** –0.167*** –0.116*** VARIABLES (1) ALL (2) INITIALLY POOR (3) INITIALLY NON-POOR (0.046) (0.032) (0.038) HHH changed sector –0.084*** –0.183** –0.033 Services –0.132*** –0.147*** –0.162*** (0.032) (0.078) (0.031) (0.027) (0.024) (0.026) HHH age –0.001 –0.003 –0.000 Other/unknown –0.111*** –0.050 –0.117*** (0.001) (0.002) (0.001) (0.030) (0.035) (0.036) HHH male 0.016 0.097 0.017 HHH age 0.000 –0.001 0.001 (0.040) (0.084) (0.042) (0.001) (0.001) (0.001) HHH married –0.047 –0.139 –0.027 HHH male –0.089** –0.036 –0.030 (0.041) (0.089) (0.043) (0.035) (0.032) (0.031) HHH primary –0.079*** 0.001 –0.081** HHH married 0.020 –0.016 0.008 (0.029) (0.054) (0.032) (0.036) (0.032) (0.033) HHH secondary –0.263*** –0.191 –0.205*** HHH primary –0.097*** –0.035 –0.132*** (0.032) (0.135) (0.033) (0.026) (0.023) (0.027) HHH other edu –0.188*** 0.035 –0.171*** HHH secondary –0.254*** –0.190*** –0.196*** (0.049) (0.182) (0.045) (0.033) (0.031) (0.041) Number children 0.021*** 0.021* 0.016*** Number children 0.019*** 0.024*** 0.026*** (0.006) (0.012) (0.005) (0.005) (0.005) (0.005) Number adults –0.017** –0.035** –0.015** Number adults 0.011* –0.001 –0.003 (0.007) (0.014) (0.007) (0.006) (0.007) (0.007) Number elders 0.016 0.017 0.003 Number elders 0.045* 0.018 –0.012 (0.030) (0.053) (0.027) (0.023) (0.022) (0.029) Other urban 0.095*** 0.016 0.063*** Other urban 0.116*** 0.150*** 0.115*** (0.024) (0.244) (0.016) (0.030) (0.026) (0.037) Rural 0.260*** 0.089 0.193*** Rural 0.232*** 0.231*** 0.165*** (0.017) (0.221) (0.014) (0.024) (0.016) (0.030) Zanzibar 0.280*** 0.247 0.191*** Zanzibar 0.231*** 0.269*** 0.111*** (0.039) (0.239) (0.035) (0.035) (0.035) (0.039) Moved district –0.055 –0.212** 0.001 Observations 3,828 4,687 3,276 (0.036) (0.092) (0.034) Standard errors in parentheses Observations 2,920 587 2,333 *** p<0.01, ** p<0.05, * p<0.1 Standard errors in parentheses Source: Authors’ calculations based on NPS 2010/11, 2012/13 and 2014/15. *** p<0.01, ** p<0.05, * p<0.1 Base categories: Agricultural household, household head has no education, Source: Authors’ calculations based on NPS 2010/11 and 2012/13. household is located in Dar es Salaam. Base categories: Agricultural household, household head has no education, household is located in Dar es Salaam. i.  Transitions in Employment Type TABLE I.36: Transitions in Employment Type Between W2 and W3 (Whole Sample) WAVE 3   PAID NON-AG NON-AG AG AG UNPAID WAVE 2 EMPLOYEE SELF EMPLOYED UNPAID WORKER SELF-EMPLOYED UNPAID WORKER APPRENTICE TOTAL Paid Employee 6.7 1.7 0.2 1.8 1.2 0.1 11.6 Non-Ag Self Employed 2.0 7.0 0.4 2.3 2.0 0.0 13.6 Non-Ag Unpaid Worker 1.1 0.9 0.9 0.8 4.2 0.1 8.1 Ag Self-Employed 3.7 3.0 0.4 18.8 9.5 0.1 35.4 Ag Unpaid Worker 4.0 2.1 1.0 7.7 16.4 0.1 31.3 Total 17.4 14.7 2.8 31.3 33.3 0.4 100.0 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. Appendix I 195 TABLE I.37: Transitions in Employment Type Between W2 and W3 (Escaped Poverty) WAVE 3   WAVE 2 PAID EMPLOYEE NON-AG SELF EMPLOYED NON-AG UNPAID WORKER AG SELF-EMPLOYED AG UNPAID WORKER UNPAID APPRENTICE TOTAL Paid Employee 1.7 0.6 0.2 1.9 0.9 0.0 5.3 Non-Ag Self Employed 0.7 2.8 0.1 1.3 2.3 0.0 7.0 Non-Ag Unpaid Worker 0.5 0.3 0.0 1.5 5.2 0.2 7.8 Ag Self-Employed 4.1 2.4 0.2 19.6 10.9 0.0 37.2 Ag Unpaid Worker 5.6 3.5 0.8 10.7 21.7 0.3 42.7 Total 12.7 9.6 1.3 35.1 40.9 0.5 100.0 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. TABLE I.38: Transitions in Employment Type Between W2 and W3 (Entered Poverty) WAVE 3 NON-AG SELF NON-AG UNPAID AG UNPAID WAVE 2 PAID EMPLOYEE EMPLOYED WORKER AG SELF-EMPLOYED WORKER UNPAID APPRENTICE TOTAL Paid Employee 2.5 0.8 0.0 2.2 2.1 0.0 7.6 Non-Ag Self Employed 0.6 2.7 0.6 1.7 2.8 0.0 8.3 Non-Ag Unpaid Worker 0.8 0.9 0.6 0.9 4.0 0.1 7.3 Ag Self-Employed 4.9 2.5 0.3 24.2 11.4 0.0 43.3 Ag Unpaid Worker 3.7 0.8 1.3 7.4 20.3 0.0 33.5 Total 12.4 7.7 2.8 36.4 40.6 0.1 100.0 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. TABLE I.39: Transitions in Employment Type Between W2 and W3 (Trapped in Poverty) WAVE 3   PAID NON-AG NON-AG AG AG UNPAID WAVE 2 EMPLOYEE SELF EMPLOYED UNPAID WORKER SELF-EMPLOYED UNPAID WORKER APPRENTICE TOTAL Paid Employee 1.5 0.1 0.2 0.8 1.6 0.0 4.1 Non-Ag Self Employed 0.8 1.1 0.1 1.3 0.5 0.0 3.7 Non-Ag Unpaid Worker 0.7 0.6 0.4 0.4 4.4 0.0 6.6 Ag Self-Employed 4.1 3.0 0.2 23.0 14.0 0.0 44.4 Ag Unpaid Worker 5.7 0.7 1.4 8.7 24.7 0.0 41.2 Total 12.8 5.5 2.4 34.1 45.3 0.0 100.0 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. TABLE I.40: Transitions in Employment Type Between W2 and W3 (Never Poor) WAVE 3   PAID NON-AG NON-AG AG AG UNPAID WAVE 2  EMPLOYEE SELF EMPLOYED UNPAID WORKER SELF-EMPLOYED UNPAID WORKER APPRENTICE TOTAL Paid Employee 9.7 2.4 0.2 1.8 1.0 0.1 15.2 Non-Ag Self Employed 2.8 10.1 0.5 2.9 2.0 0.0 18.3 Non-Ag Unpaid Worker 1.3 1.2 1.3 0.7 4.0 0.1 8.6 Ag Self-Employed 3.1 3.3 0.5 16.7 7.8 0.1 31.5 Ag Unpaid Worker 3.3 2.3 0.9 6.9 12.9 0.1 26.4 Total 20.3 19.2 3.4 28.9 27.8 0.5 100.0 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. 196 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T ii. Sector Transitions TABLE I.41: Transition Matrix for Sector of Employment Between W2 and W3 (Whole Sample) WAVE 2 WAVE 3 AGRIC. & MANUFAC- ELECT. AND TRANSP., FINANCE, INSUR., OTHER SECTOR OF EMPLOYMENT FISHING MINING TURING UTILITIES CONSTR. TRADE STORAGE, COM. REAL ESTATE SERVICES TOTAL Agriculture & fishing 66.3 0.4 1.0 0.0 0.6 4.9 1.0 0.0 0.8 74.9 Mining 0.3 0.4 0.0 0.0 0.0 0.1 0.0 0.1 0.0 0.9 Manufacturing 0.7 0.0 1.0 0.0 0.1 0.5 0.2 0.0 0.1 2.6 Electricity and utilities 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.0 0.2 Construction 0.4 0.0 0.1 0.0 0.2 0.1 0.2 0.0 0.1 1.0 Trade 4.1 0.3 0.5 0.0 0.2 4.7 1.0 0.0 0.6 11.5 Transport, storage, communication 0.6 0.0 0.1 0.0 0.1 0.8 1.8 0.1 0.2 3.7 Finance, insurance, real estate 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.0 0.2 Other Services 1.5 0.0 0.2 0.0 0.1 0.5 0.2 0.1 2.5 5.2 Total 73.9 1.1 2.9 0.1 1.4 11.7 4.3 0.3 4.4 100 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. TABLE I.42: Transition Matrix for Sector of Employment Between W2 and W3 (Escaped Poverty) WAVE 2 WAVE 3 AGRIC. & MANUFAC- ELECT. AND TRANSP., FINANCE, INSUR., OTHER SECTOR OF EMPLOYMENT FISHING MINING TURING UTILITIES CONSTR. TRADE STORAGE, COM. REAL ESTATE SERVICES TOTAL Agriculture & fishing 77.5 0.7 2.0 0.0 0.4 5.3 0.7 0.1 0.9 87.5 Mining 0.7 0.6 0.0 0.0 0.0 0.1 0.0 0.0 0.0 1.4 Manufacturing 0.4 0.0 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.6 Electricity and utilities 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Construction 0.3 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.3 Trade 4.7 0.2 0.0 0.0 0.1 1.3 0.2 0.0 0.0 6.5 Transport, storage, ­communication 0.5 0.0 0.1 0.0 0.0 0.3 0.3 0.0 0.0 1.1 Finance, insurance, real estate 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Other Services 1.9 0.1 0.2 0.0 0.1 0.3 0.0 0.0 0.2 2.7 Total 85.8 1.5 2.3 0.0 0.6 7.4 1.2 0.1 1.1 100 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. TABLE I.43: Transition Matrix for Sector of Employment Between W2 and W3 (Entered Poverty) WAVE 2 WAVE 3 AGRIC. & MANUFAC- ELECT. AND TRANSP., STOR- FINANCE, INSUR., OTHER SECTOR OF EMPLOYMENT FISHING MINING TURING UTILITIES CONSTR. TRADE AGE, COM. REAL ESTATE SERVICES TOTAL Agriculture & fishing 82.7 0.0 0.6 0.0 0.8 2.9 0.2 0.0 0.6 87.7 Mining 0.2 0.0 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.4 Manufacturing 0.8 0.0 0.4 0.0 0.0 0.0 0.3 0.0 0.0 1.4 Electricity and utilities 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Construction 0.9 0.0 0.1 0.0 0.2 0.0 0.0 0.0 0.0 1.1 Trade 3.4 0.0 0.1 0.0 0.2 3.3 0.0 0.0 0.0 6.9 Transport, storage, communication 0.2 0.0 0.0 0.0 0.0 0.0 0.1 0.0 0.2 0.6 Finance, insurance, real estate 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 Other Services 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.4 1.6 Total 89.2 0.0 1.3 0.0 1.1 6.4 0.6 0.2 1.3 100 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. Appendix I 197 TABLE I.44: Transition Matrix for Sector of Employment Between W2 and W3 (Trapped in Poverty) WAVE 2 WAVE 3 TRANSPORT, STORAGE, SECTOR OF EMPLOYMENT AGRICULTURE & FISHING MANUFACT. CONSTR. TRADE AND COM. OTHER SERVICES TOTAL Agriculture & fishing 87.7 0.3 0.7 4.0 0.6 0.2 93.6 Manufacturing 0.0 0.0 0.0 0.0 0.0 0.0 0.0 Construction 0.3 0.0 0.0 0.0 0.0 0.0 0.3 Commerce 2.7 0.2 0.0 1.6 0.4 0.2 5.0 Transport, storage, com. 0.4 0.1 0.0 0.1 0.0 0.0 0.6 Other Services 0.4 0.0 0.0 0.0 0.0 0.0 0.5 Total 91.6 0.6 0.7 5.7 1.0 0.4 100.0 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. TABLE I.45: Transition Matrix for Sector of Employment Between W2 and W3 (Never Poor) WAVE 2 WAVE 3 AGRIC. & MANU­ ELECT. AND TRANSP., FINANCE, INSUR., OTHER SECTOR OF EMPLOYMENT FISHING MINING FACTURING UTILITIES CONSTR. TRADE ­STORAGE, COM. REAL ESTATE SERVICES TOTAL Agriculture & fishing 56.1 0.4 1.0 0.0 0.5 5.3 1.2 0.0 1.0 65.6 Mining 0.3 0.4 0.0 0.0 0.0 0.2 0.0 0.1 0.0 1.1 Manufacturing 1.0 0.0 1.5 0.0 0.1 0.8 0.2 0.0 0.2 3.8 Electricity and utilities 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.0 0.2 Construction 0.3 0.0 0.1 0.0 0.4 0.1 0.3 0.0 0.1 1.2 Trade 4.3 0.5 0.7 0.0 0.3 6.5 1.5 0.0 1.0 14.9 Transport, storage, communication 0.8 0.0 0.1 0.0 0.1 1.3 2.8 0.1 0.3 5.6 Finance, insurance, real estate 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.0 0.2 Other Services 1.6 0.0 0.4 0.0 0.2 0.8 0.3 0.1 3.9 7.4 Total 64.5 1.4 3.7 0.2 1.8 15.0 6.4 0.4 6.5 100 Source: Authors’ calculations based on NPS 2010/11 and 2012/13. TABLE I.46: Poverty Transition and Change in Hours Worked by Sector WHOLE SAMPLE SECTOR WAVE 2 MEAN HOURS WAVE 3 MEAN HOURS DIFFERENCE Agriculture & fishing 557 692 135 *** Mining 1328 2032 704 *** Manufacturing 1591 2041 450 *** Electricity and Utilities 2699 1954 -745 *** Construction 1318 1461 142 Trade 1076 1857 781 ** Transport, storage, communications 1744 2512 768 Finance, Insurance, real estate 1422 2737 1315 ** Other Services 1818 2295 477 *** Total 739 1053 314 POOR-POOR SECTOR WAVE 2 MEAN HOURS WAVE 3 MEAN HOURS DIFFERENCE Agriculture & fishing 548 668 119 Manufacturing 6 2040 2033 Construction 523 605 82 Trade 1061 1433 371 Transport, storage 81 1860 1779 Services 603 1736 1133 Total 561 738 177 (Table Continiued on next Page) 198 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE I.46: Poverty Transition and Change in Hours Worked by Sector (Continued) NON-POOR - NON-POOR SECTOR WAVE 2 MEAN HOURS WAVE 3 MEAN HOURS CHANGE Agriculture & fishing 572 717 145 Mining 1594 2197 603 Manufacturing 1594 2091 497 Electricity & utilities 2702 1954 -748 Construction 1392 1689 297 Trade 1082 1970 888 Transport, storage 1717 2572 855 Finance, insurance 1580 2864 1285 Services 1902 2331 429 Total 844 1236 392 POOR-NON-POOR SECTOR WAVE 2 MEAN HOURS WAVE 3 MEAN HOURS CHANGE Agriculture & fishing 533 669 137 Mining 530 1449 919 Manufacturing 1123 1684 561 Electricity & utilities 1540 1920 380 Construction 2086 664 -1422 Trade 726 1442 716 Transport, storage 2191 2519 328 Finance, insurance 0 2806 2806 Services 1026 1827 801 Total 575 801 226 NON-POOR-POOR SECTOR WAVE 2 MEAN HOURS WAVE 3 MEAN HOURS CHANGE Agriculture & fishing 545 657 111 Mining 96 1689 1593 Manufacturing 997 577 -420 Construction 1103 1506 403 Trade 1474 2326 853 Transport, storage 1288 1440 152 Finance, insurance 224 224 Services 1305 2037 732 Total 608 747 139 iii.  Transitions in Types of Agricultural Work TABLE I.47: Transitions Between Types of Agricultural Work (Whole Sample) WAVE 2 WAVE 3 CROP LIVESTOCK MIXED OTHER TOTAL Crop 19.6 0.6 9.1 3.5 32.8 Livestock 0.8 0.8 1.2 0.4 3.2 Mixed 15.4 0.8 32.5 2.4 51.1 Other 6.2 0.7 3.0 3.0 12.9 Total 41.9 2.9 45.8 9.4 100.0 Appendix I 199 TABLE I.48: Transitions Between Types of Agricultural TABLE I.50: Transitions Between Types of Agricultural Work (Escaped Poverty) Work (Trapped in Poverty) WAVE 2 WAVE 3 WAVE 2 WAVE 3 Crop Livestock Mixed Other Total CROP LIVESTOCK MIXED OTHER TOTAL Crop 18.9 1.4 9.4 3.8 33.5 Crop 27.4 0.1 7.5 3.9 39.0 Livestock 0.6 1.6 1.8 0.2 4.2 Livestock 1.1 0.0 1.2 0.3 2.7 Mixed 14.9 0.6 30.2 1.9 47.5 Mixed 16.8 0.6 24.8 3.5 45.8 Other 6.9 0.7 3.2 4.1 14.8 Other 7.2 0.2 1.7 3.4 12.5 Total 41.3 4.2 44.6 9.9 100.0 Total 52.6 1.0 35.3 11.2 100.0 TABLE I.49: Transitions Between Types of Agricultural TABLE I.51: Transitions Between Types of Agricultural Work (Entered Poverty) Work (Non-poor Non-Poor) WAVE 2  WAVE 3 WAVE 2 WAVE 3 CROP LIVESTOCK MIXED OTHER TOTAL CROP LIVESTOCK MIXED OTHER TOTAL Crop 18.1 0.4 10.4 3.5 32.4 Crop 17.7 0.6 9.2 3.2 30.7 Livestock 0.6 0.6 0.4 0.4 2.0 Livestock 0.7 0.8 1.3 0.5 3.3 Mixed 16.9 0.8 33.2 2.2 53.0 Mixed 14.6 0.9 35.8 2.4 53.6 Other 6.9 0.5 3.0 2.3 12.7 Other 5.4 1.0 3.2 2.8 12.4 Total 42.5 2.2 47.0 8.4 100.0 Total 38.4 3.2 49.4 8.9 100.0 TABLE I.52: Shocks Experienced in Wave 2 TABLE I.53: Shocks Experienced in Wave 3  POVERTY STATES PERCENTAGE OF MEDIAN NUMBER OF MEAN NUMBER OF POVERTY STATES PERCENTAGE OF MEDIAN NUMBER OF MEAN NUMBER OF HOUSEHOLDS SHOCKS EXPERIENCED SHOCKS EXPERIENCED HOUSEHOLDS EXPERI- SHOCKS EXPERIENCED SHOCKS EXPERIENCED EXPERIENCING A SHOCK BY HOUSEHOLDS BY HOUSEHOLD ENCING A SHOCK BY HOUSEHOLDS BY HOUSEHOLD PP 86% 2.0 3.1 PP 84% 2 2.4 NN 85% 3.0 3.1 NN 79% 2 2.5 PN 84% 2.0 3.0 PN 83% 3 2.7 NP 87% 3.0 3.1 NP 79% 2 2.6 Total 85% 3.0 3.1 Total 81% 2 2.5 200 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T TABLE I.54: Poverty Transitions and Exposure to Shocks TABLE I.56: Transitions across consumption quintiles in Wave 2 WHOLE POPULATION SHOCK TYPE PP NN PN NP TOTAL WAVE 2 QUINTILE WAVE 3 QUINTILE   Drought or flood 38% 27% 33% 31% 30% 1 2 3 4 5 Total Crop disease or crop pests 33% 27% 33% 32% 29% 1 8.0 5.6 3.3 2.4 0.8 20.1 Livestock died or were stolen 27% 24% 26% 28% 25% 2 5.7 5.5 4.6 3.2 1.0 20.0 Household business failure, non-agricultural 1% 7% 4% 3% 5% 3 3.8 5.0 5.4 4.0 1.9 20.0 Loss of salaried employment or non-payment of salary 0% 4% 2% 2% 3% 4 1.7 3.3 5.4 5.8 3.9 20.0 Large fall in sale prices for crops 32% 29% 29% 30% 30% 5 0.6 0.9 2.0 4.7 11.8 20.0 Large rise in price of food 50% 53% 47% 51% 52% Total 19.7 20.2 20.7 20.1 19.3 100.0 Large rise in agricultural input prices 23% 28% 23% 32% 27% PP Severe water shortage 29% 33% 30% 34% 32% WAVE 2 QUINTILE WAVE 3 QUINTILE   Loss of land 5% 5% 4% 5% 5% Chronic/severe illness or accident of household member 11% 7% 6% 9% 8%   1 2 3 4 5 TOTAL Death of a member of household 16% 11% 15% 11% 12% 1 65.2 13.9 0.0 0.0 0.0 79.0 Death of other family member 28% 34% 32% 28% 32% 2 16.6 4.4 0.0 0.0 0.0 21.0 Break-up of household 5% 5% 5% 6% 5% 3 0.0 0.0 0.0 0.0 0.0 0.0 Jailed 0% 1% 0% 1% 1% 4 0.0 0.0 0.0 0.0 0.0 0.0 Fire 3% 2% 3% 2% 2% 5 0.0 0.0 0.0 0.0 0.0 0.0 Hijacking/robbery/burglary/assault 6% 10% 7% 5% 8% Total 81.7 18.3 0.0 0.0 0.0 100.0 Dwelling damaged, destroyed 0% 0% 0% 1% 0% NN Other 1% 3% 1% 1% 2% WAVE 2 QUINTILE WAVE 3 QUINTILE   N 898 5252 1163 904 8217   1 2 3 4 5 Total 1 0.0 0.0 0.0 0.0 0.0 0.0 2 0.0 4.6 4.3 3.6 1.3 13.9 TABLE I.55: Poverty Transitions and Exposure to Shocks 3 0.0 7.0 8.8 6.6 3.1 25.5 in Wave 3 4 0.0 4.4 8.8 9.5 6.3 29.0 SHOCK TYPE PP NN PN NP TOTAL 5 0.0 1.2 3.3 7.8 19.3 31.6 Drought or flood 35% 27% 37% 36% 31% Total 0.0 17.2 25.3 27.5 30.1 100.0 Crop disease or crop pests 20% 19% 22% 27% 21% PN Livestock died or were stolen 16% 17% 19% 18% 17% WAVE 2 QUINTILE WAVE 3 QUINTILE   Household business failure, non-agricultural 2% 5% 2% 3% 4% 1 2 3 4 5 TOTAL Loss of salaried employment or non-payment of salary 1% 3% 2% 1% 2% 1 0.0 25.7 22.0 15.6 5.4 68.7 Large fall in sale prices for crops 19% 20% 25% 23% 21% 2 0.0 10.0 13.3 6.6 1.5 31.3 Large rise in price of food 56% 47% 45% 50% 48% 3 0.0 0.0 0.0 0.0 0.0 0.0 Large rise in agricultural input prices 26% 21% 27% 24% 23% 4 0.0 0.0 0.0 0.0 0.0 0.0 Severe water shortage 18% 23% 27% 22% 23% 5 0.0 0.0 0.0 0.0 0.0 0.0 Loss of land 7% 3% 5% 4% 4% Total 0.0 35.7 35.2 22.2 6.9 100.0 Chronic/severe illness or accident of household member 6% 7% 7% 6% 7% NP Death of a member of household 12% 8% 11% 9% 9% Death of other family member 16% 31% 29% 27% 29% WAVE 2 QUINTILE WAVE 3 QUINTILE   Break-up of household 4% 5% 5% 3% 4%   1 2 3 4 5 TOTAL Jailed 1% 0% 1% 0% 0% 1 0.0 0.0 0.0 0.0 0.0 0.0 Fire 1% 1% 1% 2% 1% 2 30.9 5.1 0.0 0.0 0.0 35.9 Hijacking/robbery/burglary/assault 3% 8% 5% 3% 6% 3 31.7 6.2 0.0 0.0 0.0 37.9 Dwelling damaged, destroyed 0% 1% 1% 0% 1% 4 14.1 5.6 0.0 0.0 0.0 19.7 Other 3% 5% 2% 2% 4% 5 5.0 1.5 0.0 0.0 0.0 6.5 N 898 5,250 1,165 903 8,216 Total 81.7 18.3 0.0 0.0 0.0 100.0 Appendix I 201 TABLE I.57: Change in Consumption between Wave 2 and Wave 3 POVERTY STATE MEAN CHANGE IN CONSUMPTION MEDIAN CHANGE IN CONSUMPTION MINIMUM MAXIMUM STANDARD DEVIATION PP 5201 5227 –19223 29117 9328 NN 26812 16430 –452529 804593 62146 PN 41160 31582 8777 371165 33306 NP –17762 –14102 –250715 6533 18835 Total 21042 12782 -452529 804593 53465 202 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T References Aggarwal, Shilpa, Brian Giera, Dahyeon Jeong, Patrick Ebers, C., Lanjouw, P., & Leite, P. G. 2008. “Brazil within Brazil: Olobo, Jonathan Robinson, and Alan Spearot. 2017. Testing the Poverty Map Methodology in “Market linkages, trade costs, and technology adop- tion in rural Tanzania: Preliminary evidence from a pilot Minas Gerais.” The World Bank. in Kilimanjaro Region.” Innovations for Poverty Action Ellis, Peter; Roberts, Mark. 2016. Leveraging Urbanization in (Ref. #: F-89307-TZA-1). South Asia : Managing Spatial Transformation for Prosperity Arce, Carlos E., and Jorge Caballero. 2015. Tanzania: and Livability. Washington, DC: World Bank. © World Bank. Agricultural Sector Risk Assessment. Agriculture Global https://openknowledge.worldbank.org/handle/10986/22549 Practice Technical Assistance Paper. World Bank. License: CC BY 3.0 IGO. Washington, DC. © World Bank. Elvidge, Christopher D., et al. “Night-time lights of the world: Battese, George E., Rachel M. Harter, and Wayne A. Fuller. 1994–1995.” ISPRS Journal of Photogrammetry and Remote “An error-components model for prediction of county Sensing 56.2 (2001): 81–99. crop areas using survey and satellite data.” Journal of the Ebers, C., Lanjouw, P., & Leite, P. G. (2008). Brazil within Brazil: American Statistical Association 83.401 (1988): 28–36. Testing the poverty map methodology in Minas Gerais. Belghith, N. B. H., M. A. Lopera, A. E. Ndip and W. Karamba The World Bank. (2018). “Analysis of the mismatch between Tanzania Employment and Labour Relations (General) Regulations, Household Budget Survey and National Panel Survey Schedule 1. 2017. Government Notice No. 47. https://www. data in poverty and inequality levels and trends.” Policy ilo.org/dyn/natlex/docs/ELECTRONIC/104378/127290​/F1​ Research Working Paper, World Bank Group. 8361. 574762067/TZA104378.pdf. Belloni, Alexandre, and Victor Chernozhukov. “Least squares Fafchamps, M., and F. Shilpi. 2005. “Cities and specialisation: after model selection in high-dimensional sparse models.” Evidence from South Asia.” Economic Journal 115 (503): Bernoulli 19.2 (2013): 521–547. 477–504. Bowen PA, Hillyard J, Hartwig K, Langford S, Harvey M, James Friedman, J., T. Hastie and R. Tibshirani (2010). “Regularization S, 2010. The whole village project village reports for Elerai, paths for generalized linear models via coordinate descent.” Eworendeke, Kimoukuwa, Tingatinga, Kiserian, Sinya, and Journal of statistical software 33(1): 1. Kitendeni in Longido District. Available at: wholevillage.umn. edu/documents/Longido.pdf. Accessed August 5, 2010. Gassert, F., M. Landis, M. Luck, P. Reig, and T. Shiao. 2014. Cali, M., and C. Menon. 2013. “Does Urbanization Affect “Aqueduct Global Maps 2.1.” Washington, DC: World Rural Poverty? Evidence from Indian Districts.” World Bank Resources Institute. Available online at http://www.wri.org​ Economic Review 27 (2): 171–201. /publication/aqueduct-metadata-global. Dang, H.-A. and P. Lanjouw (2013). “Measuring poverty Guirkinger, Catherine, Jean-Philippe Platteau, and Tatiana dynamics with synthetic panels based on cross-sections.” Goetghebuer. “Productive inefficiency in extended agricultural households: Evidence from Mali.” Journal of Dang, H.-A. H., P. F. Lanjouw and R. Swinkels (2017). Who Development Economics 116 (2015): 17–27. doi:10.1016/j. remained in poverty, who moved up, and who fell down? jdeveco.2015.03.003. Poverty Reduction in the Course of African Development. M. Nissanke and M. Ndulo. Oxford, Oxford University Press. Goodman, Seth, Ariel BenYishay, Zhonghui Lv, and Daniel Runfola. 2019. “GeoQuery: Integrating HPC systems and Duranton, Gilles. 2013. “Agglomeration and jobs in devel- public web-based geospatial data tools.” Computers oping countries.” Background paper for the 2013 World and Geosciences 122: 103–112. Development Report. Harttgen, K., S. Klasen and S. Vollmer (2013). “An African Elbers, Chris, Jean O. Lanjouw, and Peter Lanjouw. “Micro–level growth miracle? Or: what do asset indices tell us about estimation of poverty and inequality.” Econometrica 71.1 trends in economic performance?” Review of Income and (2003): 355–364. Wealth 59: S37–S61. 204 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T Hausmann, Ricardo. 2001. “Prisoners of Geography.” Foreign Luo, Shan, and Zehua Chen. “Sequential Lasso cum EBIC Policy 122, 44–53. for feature selection with ultra-high dimensional feature space.” Journal of the American Statistical Association Henderson, C. R. 1953. “Estimation of variance and covariance 109.507 (2014): 1229–1240. components.” Biometrics 9 (2), 226–252. Luhunga, Philbert M. 2017. “Assessment of the impacts of Henderson, J. Vernon, Adam Storeygard, and David N. Weil. climate change on maize production in the Southern “Measuring economic growth from outer space.” American and Western Highlands sub-agro ecological zones of economic review 102.2 (2012): 994–1028. Tanzania.” Front. Environ. Sci. 5(51): 1–16. Huang, R. and Hidiroglou, M. (2003). “Design consistent Mayer, Thierry. 2008. “Market potential and development.” estimators for a mixed linear model on survey data. Joint Working Paper DP 6798, Center for Economic Policy Statistical Meetings.” Section on Survey Research Methods Research, Washington, DC. 1897–1904. Mboera, Leonard E. G., Emmanuel A. Makundi, and Andrew Hulme, D. and A. Shepherd (2003). “Conceptualizing chronic Y. Kitua. 2007. “Uncertainty in malaria control in Tanzania: poverty.” World Development 31(3): 403–423. Crossroads and challenges for future interventions.” Am. J. Trop. Med. Hyg. 77(6): 112–118. Irish Aid. 2017. TANZANIA CLIMATE ACTION REPORT FOR 2016. Irish Aid. McKay, A. and D. Lawson (2003). “Assessing the extent and nature of chronic poverty in low income countries: Issues Jalan, J. and M. Ravallion (1998). “Transient poverty in and evidence.” World Development 31(3): 425–439. post-reform rural China.” Journal of Comparative Economics 26(2): 338–357. Molina, I., & Marhuenda. 2015. “sae: an R package for small area estimation.” The R Journal 7(1): 81–98. James, G., D. Witten, T. Hastie and R. Tibshirani (2013). An introduction to statistical learning, Springer. National Bureau of Statistics (2013). Key Findings: 2011/12 Household Budget Survey Tanzania Mainland. Jarvis, A., H.I. Reuter, A. Nelson, E. Guevara. 2008, Hole- filled seamless SRTM data V4, International Centre for National Bureau of Statistics and ICF Macro (2011). “Tanzania Tropical Agriculture (CIAT), Available from http://srtm​ Demographic and Health Survey 2010.” .csi.cgiar.org. Newhouse, D., S. Shivakumaran, S. Takamatsu and N. Yoshida Kilama, Blandina, and Wietze Lindeboom, 2016, “Region and (2014). How survey-to-survey imputation can fail, The district poverty estimates for Tanzania, 2012”, mimeo. World Bank. Kottek, Markus, Jurgen Grieser, Christoph Beck, Bruno Nguyen, M. C., Corral, P., Azevedo, J. P., & Zhao, Q. 2018. Rudolf, and Franz Rubel. 2006. “World Map of the “Sae: A Stata package for unit level small area estimation.” Köppen-Geiger climate classification updated.” The World Bank. Meteorol. Z., 15, 259–263. Nguyen, Nga Thi Viet; Dizon, Felipe F. 2017. The Geography Kozbur, Damian. “Testing-based forward model selection.” of Welfare in Benin, Burkina Faso, Côte d’Ivoire, and Togo. American Economic Review 107.5 (2017): 266–69. World Bank, Washington, DC. World Bank. Kuhn, M. (2008). “Caret package.” Journal of statistical software Petracco, Carly Kathleen, Sanchez-Reaza, Javier. 2018. Jobs 28(5): 1–26. Diagnostic: Tanzania (English). Jobs Series. issue no. 16. Washington, DC: World Bank Group. Lewis, W. Arthur. “Economic development with unlimited sup- plies of labour.” The manchester school 22.2 (1954): 139–191. Rascon, Ericka, and Robin Audy, 2016, Poverty mapping in Tanzania, mimeo Lipton, M. and M. Ravallion. 1995. Poverty and policy. Handbook of Development Economics. J. Behrman and Ravallion, M. and M. Huppi (1991). “Measuring changes T. N. Srinivasan. Amsterdam, North Holland. III. in poverty: A methodological case study of Indonesia REFERENCES 205 during an adjustment period.” World Bank Economic Van der Weide, Roy. 2014. “GLS estimation and empirical Bayes Review 5(1): 57–82. prediction for linear mixed models with heteroskedasticity and sampling weights: a background study for the POVMAP Rubin, R. (1987). “Multiple imputation for nonresponse in Project.” The World Bank. surveys (J Wiley & Sons, New York, NY).” Valderrama, D. and M. Viveros. 2014. Sectoral gains and pop- Sachs, Jeffrey D., Andrew D. Mellinger and John L. Gallup. ulation shifts. Poverty and inequality measures in practice: 2001. “The Geography of Poverty and Wealth.” Scientific A basic reference guide with Stata examples, World Bank American March 2001. Group, Poverty and Equity Global Practice. Sheffield, Justin, & Eric F. Wood. 2008. “Projected changes Willmott, C. J. and K. Matsuura. 2018. “Terrestrial air tem- in drought occurrence under future global warming from perature and precipitation: Monthly and annual time series multi-model, multi-scenario, IPCC AR4 simulations.” (1900 - 2017)” (V 5.01). Online. Climate Dynamics 31(1): 79–85. Wood-Sichra, U., A.B. Joglekar and L. You. 2016. “Spatial Sicuri, Elisa, Ana Vieta, Leandro Lindner, Dagna Constenla, production allocation model (SPAM) 2005: Technical and Christophe Sauboin. 2013. “The economic costs of Documentation”. HarvestChoice Working Paper. malaria in children in three sub-Saharan countries: Ghana, Washington, D.C.: International Food Policy Research Tanzania, and Kenya.” Malaria Journal 12(307): 1–14. Institute (IFPRI) and St. Paul: International Science and Technology Practice and Policy (InSTePP) Center, University Silwal, A, D Newhouse, R Engstrom, and A Bedada, 2019, “The of Minnesota. spatial distribution of poverty in Sri Lanka in 2016”, mimeo. World Bank. 2009. World Development Report 2009: StataCorp (2017). Stata 15 Base Reference Manual. . College Reshaping Economic Geography. World Bank, Station, TX, Stata Press. Washington, DC. World Bank. https://openknowledge. worldbank.org/handle/10986/5991 License: CC BY 3.0 IGO. Stifel, D. and L. Christiaensen. 2006. Tracking poverty over time in the absence of comparable consumption data, The World Bank. 2012. Tanzania Economic Update, October World Bank. 2012: Spreading the Wings. World Bank, Washington, DC. World Bank. https://openknowledge.worldbank.org/han- Storeygard, Adam. 2016. “Farther on down the road: Transport dle/10986/26849 License: CC BY 3.0 IGO. costs, trade and urban growth in Sub-Saharan Africa.” Review of Economic Studies 83: 1263–1295. World Bank. 2015. Tanzania mainland poverty assessment. World Bank, Washington, DC. World Bank. Tibshirani, Robert. 1996. “Regression shrinkage and selection via the lasso.” Journal of the Royal Statistical Society: Series World Bank. 2017. Tanzania - Systematic country diagnostic B (Methodological) 58.1: 267–288. (English). World Bank, Washington, DC. World Bank. United Nations. 2014. World Urbanization Prospects: The 2014 World Bank. 2018. Tanzania Macro and Poverty Outlook. Revision. New York: United Nations. World Bank, Washington, DC. World Bank. 206 TA N Z A N I A M A I N L A N D P O V E R T Y A S S E S S M E N T