AUS6819 v2 TANZANIA MAINLAND POVERTY ASSESSMENT www.worldbank.org/tanzania 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 Di- rectors 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 accep- tance 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 encour- ages 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 Copy- right 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. Contents Acknowledgements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix Acronyms and Abbreviations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi Executive Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxvii CHAPTER 1  Poverty and Inequality Trends. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Decline in Poverty and Extreme Poverty Since 2007. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 I.  Improvements in Households’ Living Conditions and Human Development Outcomes . . . . . . . . . . . . . . . 5 II.  Moderate and Fairly Stable Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 III.  CHAPTER 2  Poverty Profile. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Still Too Many Poor and Too Many People Clustered Around the Poverty Line . . . . . . . . . . . . . . . . . . . . . . . . 19 I.  The Characteristics of the Poor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 II.  III.  Migration and Poverty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 CHAPTER 3  Economic Growth and Poverty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 A Brief Review of Recent Economic Growth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 I.  The Growth Elasticity of Poverty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 II.  The Distributional Pattern of Growth. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 III.  CHAPTER 4  Uneven Geographic Decline in Poverty. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 Poverty Trends by Geographic Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 I.  Growth and Distributional Changes by Geographic Domains. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 II.  CHAPTER 5  Increasing Inequality between Geographic Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 The Sources of Urban-Rural Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 I.  Determinants of Inequality between Dar es Salaam and the Other Regions. . . . . . . . . . . . . . . . . . . . . . . . . . 57 II.  iii CHAPTER 6  Inequality of Opportunity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 Inequality of Opportunity in Household Consumption. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 I.  Inequality of Opportunity in Household Income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 II.  CHAPTER 7  Demographic Pressures Pose a Challenge to Poverty Reduction . . . . . . . . . . . . . . . . . . . . . 73 I.  Macro Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 II. The Demographic Transition in Tanzania . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 Patterns and Determinants of Fertility in Tanzania. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 III.  Main Findings and Directions for Further Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 IV.  References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 Appendix 1.A:  Poverty Estimation in the HBS 2007 and 2011/12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Appendix 1.B: Prediction Methods to Establish Comparability between the 2007 and 2011/12 Data . . . . . . . . . 97 Appendix 1.C: Welfare Dynamics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Appendix 1.D: Static Decomposition of Inequality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Appendix 2.A: Characteristics of the Poor and Poverty Correlates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Appendix 2.B: Multivariate Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 Appendix 2.C: Migration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .110 Appendix 3: Comparison of Poverty Trends Using NPS and HBS Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Appendix 4: The Unconditional Quantile Regression Model & Analysis of Spatial Inequality . . . . . . . . . . . . . . . . . 119 Appendix 5: Inequality of Opportunity: The Parametric Decomposition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 Appendix 6: Demography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 LIST OF TABLES Table I.1  Adjusted Poverty Rates for 2007 Using Prediction Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Changes in Economic Status across Quartiles, Wave 1 (2008/09) to Wave 3 (2012/13). . . . . . . . . . . . . . . 12 Table I.2  Decomposition of Inequality by Household Attributes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 Table I.3  Poverty Headcount for Alternative Poverty Lines, 2011/12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Table II.1  Households’ Demographic Structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 Table II.2  Migration by Gender and Period. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Table II.3  Table II.4  Characteristics of the Migrants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 Table II.5  Reasons for Migrating. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Relationship to the Head of the Household. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 Table II.6  Differences between Migrants and Nonmigrants before and after Migration. . . . . . . . . . . . . . . . . . . . . . . 30 Table II.7  Asset Differences between Migrants and Nonmigrants before and after Migration. . . . . . . . . . . . . . . . . 30 Table II.8  iv Tanzania Mainland Poverty Assessment Table II.9  Migrant Occupations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Amount of Domestic Remittances Received by Households . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Table II.10  Real GDP Growth in Tanzania by Sector, 2008–2013. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Table III.1  Adjusted Poverty Rates for 2007 by Geographic Domain Using Prediction Methods . . . . . . . . . . . . . . 48 Table IV.1  Endowments and Returns Effects of Some Specific Attributes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Table IV.2  LIST OF FIGURES Poverty and Extreme Poverty Incidence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .xiv Figure ES.1  Share of Households with Improved Housing Conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv Figure ES.2  Net Education Enrolment Rates. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv Figure ES.3  Annual Growth in GDP and GDP per Capita. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv Figure ES.4   Income Inequity in Tanzania by Gini Coefficient, 2001–2011/12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii Figure ES.5  Gini Coefficients in Sub-Saharan Africa. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii Figure ES.6  Figure ES.7  Distribution of the Poor Population by Geographic Area. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xviii Poverty Headcount by Geographic Domain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xx Figure ES.8  Sources of Urban-Rural Inequality: The Contribution of the Differences Figure ES.9  in Endowments and Returns to the Consumption Gap. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxi Contribution of Circumstance Variables to Consumption Inequality. . . . . . . . . . . . . . . . . . . . . . . . . . . . xxii Figure ES.10  Poverty Reduction by Number of Children (0–14 years), 2007–2011/12 . . . . . . . . . . . . . . . . . . . . . . . . xxii Figure ES.11  Poverty and Extreme Poverty Trends in Tanzania Mainland, 2007–2011/12 . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Figure I.1  Trends in Depth and Severity of Poverty in Tanzania Mainland, 2007–2011/12. . . . . . . . . . . . . . . . . . . . . . . 3 Figure I.2  Figure I.3  Adjusted Poverty Rates for 2007 Using Prediction Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Figure I.4  Trends in Dwelling Materials, 2007–2011/12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Figure I.5  Trends in Assets Ownership . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Trends in Agricultural Land Ownership . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Figure I.6  Gross Enrollment Rates in Tanzania and International Comparison. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Figure I.7  Primary and Secondary Net Enrollment Rates, 2001–2011/12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Figure I.8  Share of Children Enrolled in Primary School, by Age. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Figure I.9  Gross Enrollments by Gender and Gender Parity Index, 2011/12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Figure I.10  Educational Attainment Is Improving Slowly. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Figure I.11  Continued Reductions in Child Mortality, 2004/05–2010. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Figure I.12  Recent Improvements in Maternal Mortality, 2004/05–2010. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Figure I.13  Uneven Progress in Child Nutrition, 2004/05–2010/11. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Figure I.14  Figure I.15  Lorenz Curve and Inequality Coefficients. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Mean Consumption by Percentile. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Figure II.1  Contents v Poverty Estimates in Tanzania and Other Developing Countries by Percentage. . . . . . . . . . . . . . . . . . . 21 Figure II.2  Proportion of the Poor by Geographic Domain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 Figure II.3  Poverty by Demographic Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Figure II.4  Poverty Goes in Hand with Large Family Sizes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Figure II.5  Poverty Reduction by Number of Children (0–14 years), 2007–2011/12. . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Figure II.6  Poverty by Education Level of the Household Head. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Figure II.7  Poverty by Sector of Work of Head and Sources of Income of the Household. . . . . . . . . . . . . . . . . . . . . 26 Figure II.8  Figure II.9  Poverty by Migration Status. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Figure II.10  Access to Public Infrastructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 Major Sources of Remittances Received by Households. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Figure II.11  Channel and Primary Use of Remittances. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 Figure II.12  Tanzanian GDP Growth Rate. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Figure III.1  Comparison of Growth Rates across Countries. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Figure III.2  Sectoral Real Growth Rates in Tanzania. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Figure III.3  Sectoral Composition of Growth in Tanzania. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 Figure III.4  Growth Incidence Curves, 2001–2007 and 2007–2011/12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 Figure III.5  Growth and Redistribution Effects on Poverty Reduction (in percentage points) . . . . . . . . . . . . . . . . . 41 Figure III.6  Returns Effect and Endowments Effects over Time, Tanzania 2007–2011 . . . . . . . . . . . . . . . . . . . . . . . . . 42 Figure III.7  Basic Needs and Extreme Poverty Headcounts by Geographic Domain. . . . . . . . . . . . . . . . . . . . . . . . . . 46 Figure IV.1  Distribution of Poor Population by Geographic Area. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 Figure IV.2  Population Distribution by Consumption Quintiles and Area of Residence. . . . . . . . . . . . . . . . . . . . . . . 47 Figure IV.3  Depth and Severity of Poverty by Geographic Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 Figure IV.4  Adjusted Poverty Rates for 2007 by Geographic Domain Using Prediction Methods . . . . . . . . . . . . . 48 Figure IV.5  Growth and Redistribution Components of Changes in Poverty at the Regional Level. . . . . . . . . . . . 49 Figure IV.6  Growth Incidence Curves by Geographic Domain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Figure IV.7  Sources of Households’ Consumption Growth by Geographic Domain . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Figure IV.8  Decomposing Inequality by Regions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 Figures V.1  Unconditional Quantile Decomposition of Urban-Rural Inequality of Figure V.2  Real Monthly per Capita Consumption. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Unconditional Quantile Decomposition of Metropolitan-Nonmetropolitan Figure V.3  Inequality in Real Monthly per Capita Consumption. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 Consumption Inequality and Inequality of Opportunity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Figure VI.1  Share of Inequality of Opportunity in Tanzania Mainland and by Region . . . . . . . . . . . . . . . . . . . . . . . . . 63 Figure VI.2  Contributions of Family Background and Community Characteristics to Inequality. . . . . . . . . . . . . . . 65 Figure VI.3  The Contribution of Individual Circumstances to Inequality of Opportunity. . . . . . . . . . . . . . . . . . . . . . 66 Figure VI.4  vi Tanzania Mainland Poverty Assessment Inequality of Opportunity in Income . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 Figure VI.5  Contributions of Family Background and Community Characteristics to Income Inequality. . . . . . . 69 Figure VI.6  The Contribution of Individual Circumstances to Income Inequality of Opportunity. . . . . . . . . . . . . . 70 Figure VI.7  Tanzania’s Population is Projected to Reach 100 Million around 2040. . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Figure VII.1  Population Density will be Similar to China’s by 2050 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Figure VII.2  Mortality Has Fallen Rapidly but Fertility Remains High. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Figure VII.3  Age Structure and Dependency Ratio, 1950–2050 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Figure VII.4  The Decline in the Dependency Ratio Will Be Smaller than in East Asia . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Figure VII.5  Fertility Levels and Trends Differ across Geographic Zones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 Figure VII.6  LIST OF BOXES Box 1.1  Measuring Poverty in the HBS, 2011/12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Box 1.2  Inequality Decomposition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Box 2.1 Subjective Indicators of Deprivation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 Contents vii viii Tanzania Mainland Poverty Assessment Acknowledgements T he members of the core team that prepared background analysis on Tanzania’s demographic transition this report are Nadia Belhaj Hassine Belghith was led by Anton Dobronogov (GMFDR) with support from (GPVDR and TTL) and Isis Gaddis (GCGDR). Ad- Artemisa Flores Martinez (Consultant). The background ditional research and writing support was pro- analysis on migration in Tanzania was prepared by Gaia Nar- vided by Victoria Cunningham (GMFDR); Rose ciso (Trinity College Dublin) and Ani Rudra Silwal (University Mungai (GPVDR); Mahjabeen N. Haji (GMFDR); Ayago Esmu- of Sussex). bancha Wambile (GPVDR); and Shinya Takamatsu (AFRCE). Throughout the development of this report, the team ben- Critical support has been generously provided by the Den- efitted from fruitful discussions with and guidance from An- mark’s International Development Agency (DANIDA), Minis- drew Dabalen (GPVDR), Kathleen Beegle (AFCW1), Gabriela try of Foreign Affairs of Denmark, to some of the core work Inchauste (GPVDR), Nobuo Yoshida (GPVDR), Sergio Olivieri and to the background papers which underpinned the re- (GPVDR), and Yutaka Yoshino (GMFDR). port. We particularly acknowledge the help and support of Mette Brix Voetmann Melson and Steen Sonne Andersen. The team would like to extend its sincere thanks to the Na- tional Bureau of Statistics of Tanzania and the Ministry of The background papers and analysis reflects the work of Finance for the important support and critical feedback pro- more than 10 researchers, including World Bank staff, aca- vided throughout the preparation of the report. The team demics, and consultants. The core team worked closely with acknowledges with thanks the constructive comments each of the teams to ensure the consistency of the analyti- provided by a Technical Team formed by senior officials cal methods, findings, and main messages across the differ- from the Ministry of Finance, President’s Offices, Planning ent teams. Commission, Bank of Tanzania, REPOA and National Bureau of Statistics and including Anna Mwasha (Director for Pov- The data cleaning and construction of the consumption ag- erty Eradication Department), Omar Juma (Principal Econo- gregate using HBS 2001 was prepared by Remidius Ruhin- mist Planning Commission), Donald Mmari (REPOA), Sango duka (Researcher, University of Dar es Salaam and University Simba (2011/12 HBS Manager NBS), and Ekingo Magembe of Gothenburg). The section on poverty dynamics benefited (Principal Economist Poverty Eradication Department Min- from a background paper prepared by Dean Jolliffe (DECPI), istry of Finance). Ilana Seff (DECPI), Kashi Kafle (DECPI), and Jonathan Kastelic (DECPI). The background paper linking poverty and growth The major part of this report was prepared under the for- was produced by Lionel Demery (Consultant, Development mer World Bank unit, AFTP5, under the direction of Albert Economist) and Andy McKay (University of Sussex). The G. Zeufack (Sector Manager PREM) and overall guidance of ix Marcelo Giugale (Sector Director PREM). The team grate- Finally, the team offers its sincere thanks to Lydie Ahodehou fully acknowledges guidance from Albert G. Zeufack (Prac- (GMFDR), Justina Kajange (AFCE1), Grace A. Mayala (AFCE1), tice Manager GMFDR), Jacques Morisset (Program Leader Maude Jean Baptiste (GMFDR), Karima Laouali (GMFDR), AFCE1), Pablo Fajnzylber (Practice Manager GPVDR) and Senait Yifru (GPVDR) and Martin Buchara (GPVDR) for their Philippe Dongier (Country Director AFCE1). valuable assistance in arranging meetings and workshops with key government counterparts and stakeholders and in editing the document. x Tanzania Mainland Poverty Assessment Acronyms and Abbreviations DHS Demographic and Health Survey NPS National Panel Survey FDI Foreign Direct Investment PPP Purchasing Power Parity GDP Gross Domestic Product SSA Sub-Saharan Africa HBS Household Budget Survey T Sh Tanzanian shilling HIV Human Immunodeficiency Virus UN United Nations MDG Millennium Development Goal WDI World Development Indicators NBS National Bureau of Statistics WHO World Health Organization NGO Nongovernmental Organization Vice President: Makhtar Diop Country Director: Philippe Dongier Sector Director: Ana Revenga Sector Manager: Pablo Fajnzylber Task Team Leader: Nadia Belhaj Hassine Belghith xi Executive Summary S ince the early 2000s, Tanzania has seen remark- in Tanzania on how best to eradicate poverty. Such task re- able economic growth and strong resilience to quires a rigorous analysis of the evolution of poverty and external shocks. Yet these achievements were of the linkages between poverty, inequality, and economic overshadowed by the slow response of poverty growth. This report uses the availability of the new Tanzani- to the growing economy. Until 2007, the pov- an Household Budget Survey (HBS) for 2011/12, as well as erty rate in Tanzania remained stagnant at around 34 per- the new rebased GDP figures released in December 2014, cent despite a robust growth at an annualized rate of ap- as an opportunity to address these issues. More specifically, proximately 7 percent. This apparent disconnect between the report examines the recent trends in poverty and in- growth and poverty reduction has raised concerns among equality and their determinants and explores how respon- policy makers and researchers, leading to a consensus that sive poverty reduction was to economic growth and the this mismatch needed to be addressed with a sense of obstacles to achieving it. urgency. Over the past few years, the National Strategy for Growth Poverty and Extreme Poverty and Reduction of Poverty (MKUKUTA) in Tanzania has given Have Declined since 2007 high priority to eradicating extreme poverty and promot- ing broad-based growth. Achieving pro-poor growth has Basic needs poverty declined from 34.4 percent to 28.2 also been widely recognized by the World Bank as a criti- percent between 2007 and 2011/12 and extreme pov- cal strategy for accelerating progress toward its twin goals erty declined from 11.7 percent to 9.7 percent. The fig- of eliminating extreme poverty at the global level by 2030 ures come from the HBS’s consumption-based headcount and boosting shared prosperity by fostering income growth index, which measures the proportion of the population among the bottom 40 percent in every country. with a consumption level below the poverty line—28.2 per- cent of Tanzanians could not meet their basic consumption The official poverty figures announced by the government needs. The 9.7 percent of the population that is extreme- in November 2013 have revealed that the national strategy ly poor cannot afford to buy basic foodstuffs to meet their against poverty has begun to facilitate reductions. The basic minimum nutritional requirements of 2,200 kilocalories needs poverty rate has declined from around 34 percent to (Kcal) per adult per day. 28.2 percent between 2007 and 2012—the first significant decline in the last 20 years. Identifying the policy mecha- These poverty figures are estimated using, respectively, the nisms that have helped to increase the participation of the national basic needs poverty line of T Sh 36,482 per adult poor in the growth process and to speed pro-poor growth is per month and the national food poverty line of T Sh 26,085 therefore important for present and future decision-making per adult per month. xiii Poverty and Extreme Poverty Figure ES.1   approximately 1 percentage point per year between 2007 Incidence and 2011/12. 40% 35% Household Living Conditions and 34.4% 30% Human Development Outcomes 25% 28.2% Have Improved from Low Levels 20% The reduction in poverty was coupled with improve- 15% ments in living conditions, though from low levels. All 10% 11.7% households saw large improvements, between 2007 and 9.7% 5% 2011/12 in their housing conditions and modern ameni- ties such as television sets and mobile phones (Figure ES.2). 0% 2007 2011/12 2007 2011/12 Ownership of agricultural land improved as well, but posses- Poverty headcount Extreme poverty headcount sion of productive assets such as mechanized equipment (Basic needs poverty line) (Food poverty line) and big livestock is still limited. While these improvements were experienced mainly by less well-off households, mem- Source: HBS 2007 and 2011/12. bers of those households continue to suffer from different forms of deprivations. More than half of the poor and ru- ral dwellers still live in pitiable housing conditions and lack The depth and severity of poverty declined even important assets. Access to basic infrastructure (electricity, more strongly. Depth and severity capture the gaps be- piped water) also remains limited. tween poor households’ consumption level and the pov- erty line. They declined by 35 and 48 percent, respectively. Human development outcomes such as education, In other words, in addition to a decline in the share of the health, and nutrition improved as well, but overall lev- population living in poverty, Tanzania also witnessed a re- els remain low. Enrolments in primary education increased duction in the level of deprivation of those who remained markedly in 2001–7 but declined slightly in 2007–11/12 (Fig- in poverty. This suggests that poor households were able ure ES.3). However, there has been a remarkable expansion to reduce their consumption shortfall relative to the pov- erty line and that gains were larger amongst the poorest groups. Share of Households with Figure ES.2   Improved Housing Conditions The analysis of the poverty trend is challenged by 66% changes in the HBS design, but the adjustments made 60% 56% to counter the change in design support the decline of 46% poverty. Assessing the changes in poverty levels over time 40% 39% is subject to issues of comparability stemming from chang- 34% 33% es in the survey design and methodological improvements 20% implemented during the 2011/12 HBS. These issues were addressed using different methods, including the reevalu- 0% 2007 2011/12 2007 2011/12 2007 2011/12 ation of the consumption aggregates for HBS 2007 using the same approach as in 2011/12, as well as nonparamet- ric and parametric imputation procedures. The different Improved roof Improved oor Improved wall material material material adjustment methods support the decline of poverty and extreme poverty and show that poverty has dropped by Source: HBS 2007 and 2011/12. xiv Tanzania Mainland Poverty Assessment Figure ES.3   Net Education Enrolment Rates Poverty Has Become More 80% Responsive to Growth 84% 78% The poverty headcount appears to have declined just 60% as economic growth has continued to expand since 59% 2007. In December 2014, Tanzania released revised gross 40% domestic product (GDP) figures with a base year of 2007 (Figure ES.4). GDP growth averaged 6.3 percent from 2008 30% 20% to 2013, with a marked increase in volatility compared to 5% 15% the previous series of numbers. The new figures suggest a 0% stronger impact of economic growth on poverty reduction 2000/11 2007 2011/12 2000/11 2007 2011/12 than previously observed. Primary Lower secondary The magnitude of the poverty reduction response (Standard 1–7) (Form I–IV) to economic growth, however, depends on how eco- Source: HBS 2001, 2007, and 2011/12. nomic growth is defined. When growth is measured by changes in GDP per capita, the growth elasticity of poverty is –1.02 during 2007–2011/12—in other words, a 10 per- in lower secondary education, albeit from very low levels. cent increase in GDP growth per capita can be expected There are also growing quality concerns since education to produce a 10.2 percent decrease in the proportion of outcomes remain weak across all levels. the poor. When economic growth is defined using chang- es in mean household consumption calculated from HBS, Infant mortality (per 1,000 live births) declined from however, the growth elasticity of poverty is –4.0 during 68 in 2004/05 to 51 in 2010, and mortality of children the same period, indicating that an increase in household less than five years old declined from 112 to 81 during the same period. Improvements in maternal mortality Annual Growth in GDP and GDP Figure ES.4    have not been as significant, reflecting to some extent per Capita the lack of efficacy of the (public) health system and (base year 2007) financial constraints of the poorest households. 9% The welfare improvements did not hold across all 8% household groups. Despite the decline of poverty and 7% general improvements observed in households’ living con- 6% ditions, only 30 percent of the population has been able to 5% significantly improve their economic status and move to higher welfare classes. Around 12 percent of those at the 4% bottom of the consumption distribution remained trapped 3% in chronic poverty. Around 13 percent of the population has 2% moved down to the lowest quartile (bottom 25 percent) of 1% the consumption distribution. The movement across the welfare classes occurred mainly among the households 0% 2008 2009 2010 2011 2012 2013 in the middle economic classes, with those lacking assets GDP growth GDP per capita growth experiencing a worsening of their welfare and moving to lower economic status. Source: NBS, World Bank 2014. Executive Summary xv mean consumption would have a higher impact on pov- communications) that have limited capacity to create erty reduction than would changes in GDP per capita. jobs. Agriculture, which represents the main source of The Tanzania growth elasticity of poverty is higher than livelihood for the vast majority of the poor, grew by only the available estimates of about –3.0 suggested by pre- 4.2 percent per year in 2008–13, a lower rate than the vious studies (using survey mean figures) on developing overall economy of 6.3 percent. With growth mainly cen- countries. tered in national sectors where poorer Tanzanians are not particularly involved, the pro-poor growth would not be The difference between the estimates of the growth elas- expected. ticity of poverty found with the different measures of eco- nomic growth is quite common in developing countries, Pro-poor growth is actually the result of improvements but it seems to be larger in Tanzania. This is due to the in endowments and returns for poor households. Chang- discrepancy between the price deflators used to convert es in peoples income and consumption over time can be nominal GDP and household consumption values into broken down into changes in their personal characteristics real terms. The first measure uses the GDP deflator, which or endowments (for example, increased education levels, implies a much slower rate of inflation than price indices ownership of land and other assets, and access to employ- based on survey unit values and consequently a higher ment opportunities and basic services) and the returns that growth rate of real GDP per capita than of survey real mean they get for those endowments (for example, the returns household consumption. While there is no clear consensus to education, land productivity, and so forth). Households on which of these measures of economic growth is more in the 30 percent poorest groups experienced marked im- accurate, it seems that survey based data better reflect the provements in their endowments in assets, mainly trans- spending behavior of the poor and regional differences in portation and communication means, and in education. the cost of living. The improvements in endowments were coupled with an increase of the returns to their economic activity—essen- There are emerging signs of pro-poor growth in Tan- tially nonagricultural businesses—as well as to community zania. The poor are found to have benefitted dispropor- infrastructure, mainly local markets and roads, which have tionately from economic growth during the period 2007– had a positive influence on needy households’ living stan- 2011/12, in sharp contrast to the period 2001–07, during dards in recent years. which growth benefitted mainly the country’s richer groups. The relationship between growth and poverty involves Consumption Inequality Remains changes both in mean consumption and in the distri- Moderate and Fairly Stable bution of consumption across households. The decline of poverty at the national level is due to an increase in mean The Gini coefficient measures inequality in income household consumption as well as a reduction of inequali- or consumption expenditures across a nation’s pop- ty in the distribution of consumption between households, ulation; based on consumption per capita, it declined with the effect of inequality reduction being marginally modestly in Tanzania during the last decade. The Gini more important than the effect of consumption growth. coefficient of real per capita monthly consumption indi- Household consumption growth contributes by 40 percent cates that the level of inequality for Tanzania is approxi- to poverty reduction, while the reduction of inequality con- mately 36 in 2011/12, declining from around 39 in 2001– tributes by 60 percent. 07 (Figure ES.5). The improvements in the distribution of consumption seem to be driven by an increase of the The emerging signs of pro-poor growth contrast with consumption share accruing to the 20 percent poorest the nature of Tanzania’s economic growth. The latter segment of the population; this share grew by more than was driven mainly by fast-growing and relatively capi- 16 percent between 2007 and 2011/12. The population tal-intensive sectors (for example, finance, transport, and groups in the second income quintile of the population xvi Tanzania Mainland Poverty Assessment Figure ES.5   Income Inequity in Tanzania by There Are Still Too Many Poor Gini Coefficient, 2001–2011/12 and Too Many People Clustered 38.8 38.5 Around the Poverty Line 35.8 Around 12 million Tanzanian people are still below the poverty line. While the poverty headcount declined by around 18 percent, the absolute number of poor people only declined by 10 percent from 13.2 million to 11.9 million from 2007 to 2011/12, due to population growth. Likewise, the absolute number of extreme poor decreased by only 7 percent, declining from 4.5 million to 4.2 million. 2001 2007 2011/12 Poverty is particularly pervasive in the rural areas, where around 70 percent of the Tanzanian population Source: HBS 2001, 2007, and 2011/12. lives. About 10 million people in the rural population live in poverty, and 3.4 million live in extreme poverty, compared experienced an increase in their consumption share by 5 to less than 1.9 million living in poverty and 750,000 people percent, while those in top quintiles experienced a loss of in extreme poverty in the urban sector (Figure ES.7). around 4 percent. A large share of the population hovers around the pov- Tanzania’s inequality level compares favorably with erty line, likely to escape poverty but also prone to fall Sub-Saharan Africa and less developed countries. Tan- into it. Small changes in the national poverty line yield sig- zania’s Gini coefficient is below the Sub-Saharan Africa aver- nificant differences in estimated poverty levels, indicating age of 45.1 (Figure ES.6) and the low-income countries aver- a high concentration of individuals around the basic needs age of 40. It is on par with levels of inequality in South and threshold. For instance, a variation of the poverty line by 10 East Asia, which range around 38.4, and significantly lower percent (T Sh 120 per adult per day) would lead to a change than inequality levels in South America. of poverty rate by more than 20 percent. The significant Gini Coefficients in Sub-Saharan Africa Figure ES.6   0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Niger (2011) Ethiopia (2011) Sao Tome & P. (2010) Sierra Leone (2011) Tanzania (2012) Mali (2010) Congo, Rep. (2011) Mauritania (2008) Cameroon (2007) Angola (2009) Nigeria (2010) Chad(2011) Cabo Verde (2008) Uganda (2013) Togo (2011) Kenya (2005) Swaziland (2010) Cantral African Rep. (2008) Botswana (2009) Comoros (2004) Seychelles (2007) Source: HBS 2011/12 and WDI 2015. Executive Summary xvii Figure ES.7  Distribution of the Poor Population by Geographic Area (millions) 0.43 7 0 .1 1.52 1.7 Rural Rural Other Urban Other Urban Dar es Salaam Dar es Salaam 11.2 10.04 Source: HBS 2007 and 2011/12. number of people clustering around the poverty line sug- Poverty Is Associated with Rural gests that an important proportion of moderately poor peo- ple are positioned to move out of poverty, but also that an Status, Larger Families, Lower important proportion of nonpoor people are vulnerable to Education, and Low Access to falling into poverty. This fact is quite common in SSA coun- Infrastructure tries with poverty levels around 30 percent and requires a combination of policies to alleviate poverty and prevent Over 80 percent of the poor and the extreme poor in people from falling into it. Tanzania live in the rural areas. More than half of the rural poor depend on subsistence agriculture for their livelihoods. The incidence of poverty in Tanzania is about 15 per- centage points higher when using the international Poor households are larger in size and have more de- poverty line of $1.25 per person per day. The national pendents than nonpoor households. Households with poverty line reflects the country’s specific costs of basic five children and more have the highest poverty rates, fol- consumption needs but does not allow comparisons across lowed by elderly families whose head is 65 years old or old- countries. The international poverty line of $1.25 per person er. The interaction between family size and poverty is bidi- per day in 2005 purchasing power parity (PPP) exchange rectional. On the one hand, the large number of children rates is often used to evaluate a country’s poverty record and dependents affects the ability of the poor to cover vis-à-vis other low income countries or developing regions. their basic food needs and to move out of poverty. On the other hand, poor households tend to have more children Tanzania’s national poverty line is slightly lower than the to compensate for their inability to rise from poverty by international poverty line. Using the international pover- investing in the human capital of their children and hav- ty line shows that around 43.5 percent of the population ing many as an insurance strategy against infant mortality, lives in poverty in 2011/12. This increase of around 15 per- trapping them in a vicious circle of poverty. centage points, compared to the national poverty rate of 28.2 percent, is explained by the clustering of the popula- Poverty is negatively correlated with higher levels of ed- tion around the poverty line—the international standard ucation of the household head. Higher education levels of includes people considered just above the line using the the household’s head, particularly secondary and upper edu- national standard. cation, seem to be associated with better income-generating xviii Tanzania Mainland Poverty Assessment opportunities and significantly lower poverty levels. Educa- the household’s community of a daily market and mobile tion positively affects living standards and poverty reduction phone signal have a positive impact on consumption lev- both directly and indirectly through its impact on health els and reduce the probability of poverty. Access to these gains, productivity, social integration, and so forth. services is still quite limited in rural areas, hampering local opportunities to reduce poverty. Although primary education continues to be of crucial im- portance for fighting poverty, it alone seems no longer suf- Internal migration is related to lower poverty. Poverty ficient to increase poor people’s opportunities for economic levels appear to be much lower among migrant house- mobility and for moving out of poverty. Moreover, the re- holds. Migration is found to have a positive impact on wel- turns to education that have increased meaningfully in con- fare not only for migrants but also for their family left be- junction with higher levels of the head’s schooling appear hind, improving their living standards as well as the school to have declined in recent years. The expansion of educa- attendance of their children. tion and the increase of the general population’s education level might have induced changes in the requirements of Migrants are generally more educated, younger, and more the labor market and generated a decline of the rewards for prosperous than others. They tend to move towards big ur- years of schooling under a certain level. ban cities such as Dar es Salaam, Mwanza, and Zanzibar to seek better employment opportunities and living conditions. Wage employment and nonfarm businesses are as- The superiority of the characteristics of the migrants (for ex- sociated with lower poverty. Poverty rates are lowest ample, better education, higher living standards, and so forth) among households headed by government employees or may partly explain the improvement of their economic situ- employees in the private sector and NGOs. Interestingly, ation, but the positive effects of migration can easily be gen- households relying on nonagricultural businesses as a main eralized to less well-endowed people. While migration seems source of income appear to be experiencing a remarkable to be associated with lower poverty, it may prove to be less decline in poverty, suggesting that the development of beneficial in the long run as excessive migration might cause nonfarm employment can offer a pathway out of poverty. a displacement of poverty to the destination areas. This effect remains strong and very statistically significant even after controlling for—or holding constant—various other factors related to household well-being. The Decline in Poverty Is Uneven Geographically There has been a movement out of agriculture during the recent years, as the proportion of Tanzanian households Most of the improvements in the poverty indicators oc- whose main source of income is agricultural activity de- curred in Dar es Salaam. Poverty declined by over 70 per- clined from around 53 percent in 2007 to 39 percent in cent in Dar es Salaam but only by around 15 percent in the 2011/12. This seems to have reduced the negative influence rural sector, while it remained almost unchanged in the sec- of working in agriculture on living standards and poverty, ondary cities and towns, declining by only 5 percent (Figure probably due the fact that part of those who remained in ES.8). Although Dar es Salaam experienced the greatest pro- the sector are more productive and engaged more in cash portionate decline in poverty, the absolute number of poor crop production. people declined more in the rural areas, as 1.2 million rural people moved out of poverty as opposed to fewer than 300, Access to public infrastructure is also linked with low- 000 in the metropolitan city. er poverty. Poor households tend to have much lower access to private piped water, electricity, and tarmac roads. The uneven spatial decline of poverty is related to the Obstacles to infrastructure and services, particularly elec- pattern of economic growth, which was almost entirely tricity and roads, seriously limit the possibilities of the poor centered in Dar es Salaam, where most of the expanding to improve their living standards. Likewise, the presence in and flourishing sectors are concentrated. These include Executive Summary xix Poverty Headcount by Figure ES.8   their consumption by around 20 percent between 2007 and Geographic Domain 2011/12. This increase was driven mainly by the improve- ment of their endowments in assets (for example, increased 40% ownership of communication and transportation means 35% 39.4% and higher land ownership) as well as the improved access 33.4% 30% to community infrastructure (mainly roads). The returns to their endowments also increased, but to a lesser extent. In 25% particular, there has been an expansion of returns to both 22.7% 20% 21.5% nonfarm and household agricultural businesses followed 15% by a slight increase of returns lo land. Poor households in 14.1% 10% the secondary cities also experienced an increase of their consumption levels, by about 15 percent. This increase was 5% driven mainly by the increase of their endowments in assets 4.0% 0% 2007 2011/12 2007 2011/12 2007 2011/12 and the improvement of the returns to nonfarm activities and wage employment. Likewise, consumption of poor Rural Other urban Dar es Salaam households in Dar es Salaam increased by over 40 percent, Source: HBS 2007 and 2011/12. due mainly to the expansion of the returns to employment in public and private sectors followed by a slight increase of telecommunications, finance, and to a lesser extent con- the returns to nonfarm businesses. struction and manufacturing. Poorer households outside Dar es Salaam seem to have Increasing Inequality between experienced an increase of their consumption, despite Geographic Domains the limited growth in these regions. There were con- sumption gains among households in the poorest quintiles Inequality is increasing between urban and rural areas, not only in Dar es Salaam but also in regions where there as well as between Dar es Salaam and the other regions. was almost no growth (rural areas and secondary cities). Economic growth has benefitted most Tanzanians and started trickling down to the neediest, but the nature and Poverty reduction outside Dar es Salaam is driven main- composition of this growth induced an uneven increase ly by a reduction in inequality. The decline of poverty in of welfare at the regional level. Household consumption Dar es Salaam was driven by both an increase in mean con- grew faster in the metropolitan and urban zones than in sumption and an improvement in consumption distribu- rural areas, inducing an increase of inequality between the tion, while poverty reductions in rural and other urban areas geographic regions. The increase of interregional inequali- are due entirely to improvements in consumption distribu- ty was observed for all welfare groups but was much more tion (reduction of inequality). In these areas, the better-off pronounced among the richest groups. experienced declines in their consumption levels whereas the poorest quintiles appear to have experienced an in- Better off households in Dar es Salaam and urban zones crease in their consumption levels, albeit from low levels. have become richer due to expanding employment op- portunities and improving returns. Interregional inequal- The increase of the consumption of the poorest groups ity among better-off households is much higher (approxi- is driven essentially by the improvement of house- mately two times larger) and increasing faster than among holds’ endowments in rural areas and secondary cities, poorer households. This is mainly driven by the expanding while the increase in Dar es Salaam is explained mainly employment opportunities and the increase of returns to by the improvement of returns. Rural households in the wage work in the public and private sectors in Dar es Sa- 30 percent of poorest groups experienced an increase of laam and some urban zones. xx Tanzania Mainland Poverty Assessment Despite the increasing disparities in returns, urban-ru- inequality is parental education, the partial effect of which ral inequality remains mostly due to large differences is around 20 percent, indicating a quite high persistence be- in households’ endowments. Urban households have tween parents’ and children’s socioeconomic attainments. higher living standards essentially because they have supe- rior endowments in terms of family size and composition, Family background contributes more to inequality education, assets, and access to services and employment than community characteristics. Family background opportunities (Figure ES.9). Rural households have been seems to have a greater influence on the disparity of living able to catch up somewhat with their urban counterparts standards than the characteristics of the local community, in education levels and asset ownership, but this has been such as access to basic services and infrastructure, connec- partly offset by increasing differences in family structure tion to markets and population centers, and so forth (Fig- and access to services and job opportunities. ure ES.10). This indicates significant problems of intergen- erational poverty and inequality persistence. Addressing the influence of parental education and background on Inequality Can Be Explained Partly children’s opportunities is a long-term mission that is of- by Family Background ten complex. But without additional policy actions, there are limited chances for the generations disadvantaged by The disparities of households’ endowments and living circumstances to spring out of the poverty and inequality standards are, to a large extent, the results of intergen- also endured by their parents. erational transmission of family background. Around one-fourth of total inequality in consumption in Tanzania Policy actions need to focus on developing endow- is due to circumstances that are outside individuals’ control, ments, especially those inherited from parents or re- such as age, gender, parents’ education, orphan status, and lated to community characteristics. Strategies for pro- region of birth. This is a quite sizeable share compared to oth- moting access to basic infrastructure and services need to er SSA countries, where the contribution of an individual’s be coupled with policy interventions to reduce disparities circumstances to inequality is less than one-fifth. The most in the distribution of circumstances and equalize opportu- important circumstance variables in accounting for overall nities. Education and labor market policies as well as fiscal Sources of Urban-Rural Inequality: The Contribution of the Differences in Figure ES.9   Endowments and Returns to the Consumption Gap Returns e ects and endowment e ects Returns e ects and endowment e ects by Area for Tanzania 2007 by Area for Tanzania 2012 1 1 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 Quantities Quantities Con dence interval/ Con dence interval/ Con dence interval/ Con dence interval/ endowment e ect returns e ect endowment e ect returns e ect Endowment e ect Returns e ect Endowment e ect Returns e ect Source: HBS 2007 and 2011/12. Executive Summary xxi Contribution of Circumstance Figure ES.10   Poverty Reduction by Number Figure ES.11   Variables to Consumption of Children (0–14 years), Inequality 2007–2011/12 0.25 40 Mean logarithmic deviation 0.20 20 0.15 0 0.10 –20 0.05 –40 2008 2010 2012 0 1 2 3–4 5+ Round Poverty headcount Poverty headcount 2007 (%) 2011/12 (%) Opportunity Inequality Family Background Decline in poverty headcount (%) Share Share Tanzania Community Characteristics Share Source: HBS 2007 and 2011/12. Source: NPS 2008, 2010, and 2012. future growth and its capacity to reduce poverty. At the household level, families with large number of children have limited capacity to reduce poverty (Figure ES.11). system reforms could contribute to reducing inequality of opportunity. Also, better targeted policies to expand the ac- At the national level, demographic pressures pose challenges cess to basic goods and services for people in vulnerable for public service provision, labor markets, land and resourc- circumstance groups may be instrumental for reducing the es, and so forth and can put a break on growth in per capi- disparity of opportunities and breaking the cycle of inter- ta incomes. The best way to reduce population growth and generational persistence of poverty. accelerate demographic change is by slowing down fertility. Empowering women through education and employment support, as well as with family planning services, would help Demographic Pressures Pose a to reduce fertility and stimulate per capita economic growth. Challenge to Poverty Reduction Tanzania is in the early stages of the demographic tran- Implications for Research and sition. With high fertility of around five births per wom- Policy en and the decline of mortality, the momentum of high population growth is expected to continue in the coming The focus of this report has been on the two recent waves years. The country could gain from a demographic divi- of HBS data and the information they provide on living stan- dend—meaning a large working-age population—start- dards and poverty in Tanzania. It reveals improvements in ing in 2020–30, but the dependency ratio (the proportion the poverty and inequality indicators and shows emerging of children below 14 years old and elderly above 65 years signs of pro-poor economic growth since 2007. The report in the household) will remain much higher than the levels identifies significant changes in the way economic growth achieved in East Asia 30 years ago. has been distributed across households in Tanzania and has found these to be associated with quite different experienc- High fertility may slow poverty reduction and under- es across the country. Urban households have experienced mine pro-poor growth prospects. The rapid population quite significant consumption gains, mainly in Dar es Sa- growth will continue to weigh heavily on the country’s laam, where most of the growth has taken place. There have xxii Tanzania Mainland Poverty Assessment also been gains among the rural poorer groups, essentially design of priority interventions should take into account the due to a reduction in inequality. Urban households in the diverse nature of poverty. three poorest deciles have benefitted from better access to employment opportunities, reflecting what has happened For the extreme poor who lack basic necessities and as- in Dar es Salaam. Those in rural areas, while also benefiting sets, priority could be given to safety nets and cash trans- from an increase of the returns to their agricultural and non- fer programs to help them enhance their livelihoods and farm activities, have seen much higher improvements of productivity. Such programs increase the levels and quali- their assets ownership. There seems to be a move away from ty of consumption of the poor, offer some security against agriculture, whilst at the same time those who remain in the unforeseen shocks, facilitate access to basic goods and ser- sector are likely to have benefitted from an increase of their vices, and advance the inclusiveness of the most vulnerable cultivation areas and improvements of their returns. There is population groups in the growth process. also some evidence that farming households have a greater commercial orientation in 2011/12 then they did in 2007. Moderate poor and vulnerable nonpoor households should benefit from a combination of prevention and promotion The report shows that despite the positive changes, the strategies that enable them to diversify their activity into number of poor, particularly in rural areas, is still disconcert- higher-return and more productive businesses. The follow- ingly high and the welfare disparity between the geograph- ing could be particularly beneficial in this effort: ic regions is widening. Households with a large number of children, whose heads have less education than coun-  The development of rural economy and agriculture terparts, who are engaged in subsistence agriculture and will be instrumental for an effective poverty-reduc- living in communities lacking infrastructure are likely to be tion strategy. The disadvantage of being engaged in the most poor. Many of them will pass on their poverty to agriculture seems to have diminished during recent their offspring. Even though the results point to the positive years, but what seems to matter to farmers is access to effects of economic growth on the poorest segments of the cash crops and to markets, indicating the importance of population outside Dar es Salaam, an important proportion encouraging a more commercial agriculture. Connec- of the population has not been able to fully benefit from tivity of farmers and rural poor people to infrastructure the economic prosperity of the country and remains vulner- using modern communication and transport means is able to poverty. Households that are located outside Dar es also vital for expanding their living standards. Although, Salaam have not been able to reach the levels of access to there is little evidence of the increase of access to local basic services and employment opportunities prevailing in markets in the rural communities, the increase of their the city. The levels of endowment in education and assets remain lower outside Dar es Salaam and other urban zones. Even households who could improve their endowment Mean consumption per percentile (TZS), 2011/12 base have not been able to find the returns in the local mar- 15,000 kets corresponding to those offered in the city. Improved productivity Growing productive in traditional sectors, modern sectors 10,000 access to markets Upgrade labor The analysis in this report provides policy pointers for pov- and o -farm skills and education Targeted diversi cation erty reduction. While poverty cannot be effectively tack- CCTs 5,000 led through stand-alone policy approaches and requires a cohesive multisectorial strategy, the findings may help 0 prioritize policy interventions tailored to enhance poverty 20 40 60 80 100 reduction. The basic tenets of conventional poverty reduc- Percentile tion strategies such as investment in human capital and Mean consumption Poverty line infrastructure, income and employment generation, and per day (1200 Tsh per day) control over fertility and family sizes largely remain, but the Poverty line –25% Poverty line +25% Executive Summary xxiii returns suggest that they play a positive role in influenc- On the basis of the analysis in the report, the following key ing households’ welfare. issues call for further investigation:  There are significant returns to undertaking busi-  The report reveals emerging signs of “pro-poor” growth ness activities in rural areas but also in some sec- with a changing structure of the economy and declin- ondary urban towns, strongly supporting the case ing contribution of agriculture. In recent years this sec- for diversification. Nonfarm business seems more tor’s growth has lagged other sectors in the economy, rewarding than agricultural activities, and households but agriculture continues to be the most important engaged in such businesses appear to have been more sector in rural areas and the main source of livelihoods successful than others in reducing their poverty. While for the poor. Stimulating agriculture through improving agriculture will remain the largest source of employ- farmers’ access to modern assets, enhancing their con- ment in Tanzania and there is no escaping the need to nectivity to infrastructure and markets and encourag- galvanize this sector, the role of nonfarm diversification ing a more commercial agriculture will be instrumental in absorbing the underemployed workforce, improving for an effective poverty-reduction strategy. However, poor households’ living standards and reinvigorating with the economic transformation, agriculture might the local economy needs to be recognized and promot- not be able to absorb the expanding rural labor force ed. Efforts by the government to accelerate the process and generate jobs commensurate with the aspirations of diversification could yield important returns in terms and education of youthful workforce; and agricultural of reducing poverty and increased income mobility. But advances alone will not meet the rural poverty chal- more work is needed to better understand how diversi- lenge. Higher diversification towards nonfarm activities fication to nonfarm activities can be enhanced in sec- can play an important role in boosting the local econ- ondary cities and rural areas. omy, promoting job creation and alleviating poverty. As it seems to be indicated by the results, efforts to ac- The road to inclusive growth is yet to be paved and the work celerate the process of diversification could yield quite is challenged by the widening urban-rural gap in living stan- significant benefits in terms of increased employment dards. Policy actions should focus on developing the endow- opportunities and reduced poverty not only in the rural ments of rural households, with special attention to improv- sector but also in secondary cities in urban areas. These ing the opportunities of new generations. There have been effects can be further investigated using the upcoming commendable efforts to promote basic education and ac- Integrated Labor Force Survey (ILFS) and the National cess to assets. These efforts need to spread more widely and Panel Surveys (NPS) to examine the incentives for agri- more evenly and need to be oriented toward the provision of cultural productivity and nonfarm diversification and to secondary and higher education, particularly in less favored better understand their growth and poverty alleviation regions. The report points to secondary schooling as being potentials in order to better inform the growth and pov- particularly important for escaping poverty, even among the erty reduction strategy. The factors driving the changes rural, farming population. One way in which this effect could in distribution pattern between 2007 and 2011/12 also be channeled would be in enabling farmers to use improved call for further investigation in the subsequent studies. inputs and technology. The promotion of education would help as well to equalize opportunities and contribute to  The relative decline of rural poverty appears to be driven breaking the cycle of intergenerational persistence of pover- by improvements of the endowments of poor house- ty. More efforts should be made to achieve broader coverage holds in assets and an expansion of their cultivation and better targeting of access to basic goods and services. areas. These improvements seem to be coupled with a Policies to reduce spatial disparities in endowments need to slight increase in the returns to land denoting potential be coupled with strategies to enable households to find the increase of land productivity, particularly for the poor. appropriate returns to their improved attributes in the local As most poor farmers are smallholders with low pro- markets by supporting increased labor and land productivity. ductivity and yields, there is need to further investigate xxiv Tanzania Mainland Poverty Assessment the underlying causes of the observed improvements  The exercise of linking observed poverty outcomes to in land endowments and returns and whether these data in economic growth identified the choice of price latter are resulting from increased yields, more produc- deflator as critical. This issue needs to be explored in tive use of resources, or higher diversification and com- more depth in future work. plementarities between households’ activities. There is also need to explore whether there are real increased  The 2011/12 HBS provides an excellent basis for small efforts towards cash crops or food crops will remain the area poverty mapping, since it was concurrent with the largest source of cash income to rural households. HBS 2012 Population and Housing Census. This would pro- and NPS with agricultural surveys provide an invaluable vide a significant data base for geographically disaggre- opportunity to examine these issues and to investigate gated policy advice and development planning and for the constraints poor farmers face in raising productivity, a better understanding of the characteristics of the poor. accessing markets, and diversifying (both within farm- ing and into nonfarm activity). This would help a better understanding of the patterns of poverty dynamics and the underlying causes of poverty persistence. Executive Summary xxv Introduction Poverty in Tanzania remains a persistent problem. In order accurate estimates of poverty and the opportunity to ex- for the government, development partners, and other key plore inherent nuances. stakeholders to reduce poverty and boost shared prosper- ity in Tanzania, it is essential to understand poverty trends, The report starts by sketching the positive changes in wel- inequality, and their dynamics. Specifically, it is necessary fare, poverty, and inequality. It then presents the challeng- to determine whether growth has become more pro-poor. es that remain to be addressed and the main obstacles to The last poverty assessment by the World Bank (2007b) for poverty reduction. Chapter 1 examines the trends in poverty Tanzania, and more recent World Bank reports (World Bank and inequality in mainland Tanzania during recent years and 2011, 2012a, 2013d) flagged the sluggish response of pov- explores the evolution of the nonmonetary dimensions of erty to growth as a concern. well-being. Chapter 2 reviews in detail the characteristics of the poor, including an analysis of the economic effects of mi- This report is based primarily on the new Household Bud- gration. Chapter 3 examines the response of poverty to eco- get Survey (HBS) for 2011/12, which provides up-to-date nomic growth and investigates the distributional issues at information and an opportunity to explore the latest trends the national level. Chapter 4 analyzes the pattern of poverty on economic growth at the household level. The availability trend by geographic domain. Chapter 5 explores the sourc- of the 2011/12 HBS allows an updated and more detailed es of spatial inequalities. Chapter 6 analyses inequality of analysis of the poverty situation in Tanzania and provides opportunity in consumption and income and explores the opportunity to relate recent poverty trends to the perfor- effects of family background on the persistence of poverty mance of the economy at large. The quality of data is higher and inequality. Finally, chapter 7 examines the demographic than in previous surveys, and this ensures more reliable and transition and determinants of fertility. xxvii Chapter  1 Poverty and Inequality Trends Key Messages ➤➤ Poverty has declined by around 1 percentage point per year since 2007. ➤➤ Living conditions and human capital outcomes have improved over time, but the achievements are low com- pared to neighboring countries. ➤➤ Inequality is moderate and compares favorably with other SSA countries. The poverty assessment for Tanzania (World Bank 2008) re- different approaches to address the comparability issues vealed a stagnant level of poverty at around 33–36 percent and discusses their effects on the estimation of the poverty between 2001 and 2007, raising concerns that the country trend. Tanzanian poor people are identified as those facing may be off-track in meeting the Millennium Development consumption shortfalls, but poverty is not a single econom- Goal (MDG) target of poverty reduction by 2015 as well as ic condition and it goes beyond consumption deficits. Thus, the Bank goal of ending extreme poverty by 2030. the second section examines the evolution of the nonmon- etary dimensions of welfare and explores how these factors The availability of the Household Budget Survey (HBS) for have evolved over time for both the whole population and 2011/12 allows an updated and more detailed analysis of the most disadvantaged groups. It also investigates the dy- the poverty situation in Tanzania. These data are not only namics of well-being to identify the household groups fac- more timely but also of improved quality over previous ing chronic poverty and those switching between states of household surveys, thereby permitting more reliable and well-being and deprivation. The third section investigates accurate analysis of the latest trends in poverty and inequal- the evolution and structure of inequality. ity as well as in other nonmonetary dimensions of welfare. The first section of this chapter examines the poverty trend Decline in Poverty and Extreme I.  since 2007. Analyzing changes in poverty over time is chal- Poverty Since 2007 lenged by the changes in HBS design between 2007 and 2011/12. In particular, changes that occurred in the length The basic needs and extreme poverty headcount rates for of the reference period and degree of commodity detail Tanzania Mainland were, respectively, 28.2 percent and 9.7 for nonfood items affect the welfare trends and complicate percent in 2011/12. The headcount rates are based on the offi- comparisons of poverty levels over time. The section uses cial National Bureau of Statistics (NBS) definition of basic needs 1 and food poverty lines, estimated at, respectively, T Sh 36,482 Poverty and Extreme Poverty Figure I.1   per adult per month and T Sh 26,085.5 per adult per month. Trends in Tanzania Mainland, 2007–2011/12 According to the 2011/12 Tanzania HBS, 28.2 percent of the 40% population is poor, with monthly consumption per adult equivalent below the basic needs poverty line, and 9.7 35% 34.4% percent lives in extreme poverty, below the food poverty 30% line, and hence cannot afford to buy enough food to meet 28.2% 25% the minimum nutritional requirements of 2,200 kilocalories 20% (Kcal) per adult equivalent per day (see Box 1.1 for details). 15% The poverty rate has declined by around 6 percentage 10% 11.7% 9.7% points since 2007.1 The official national (basic needs) pov- 5% erty rate is estimated at 33.6 percent in 2007, but cannot 0% be compared to the new headcount rate for 2011/12 due 2007 2011/12 2007 2011/12 to significant changes in the survey design as well as im- Poverty headcount Extreme poverty headcount provements in the methodology for the measurement of (Basic needs poverty line) (Food poverty line) consumption aggregate and poverty line. This poverty as- Source: HBS 2007 and 2011/12. sessment tries to address these changes by reestimating the consumption aggregates for HBS 2007 using the same ap- proach as in 2011/12 and by adjusting the current poverty inequality among the poor. The estimate of the poverty gap line by the price changes between 2007 and 2011/12.2 This for 2011/12 indicates that the average consumption level of yielded a poverty estimate of 34.4 percent for 2007, sug- a poor Tanzanian is around 93 percent of the national pov- gesting a poverty reduction at the national level by around erty line, suggesting that many of the poor are very close 6 percentage points (or 18 percent).3 to the poverty line and that small income transfers would help a significant decline in poverty. Likewise, the severity Extreme poverty also declined, but by a lower degree. The of poverty is estimated at 2.3 percent, indicating a low level proportion of the population with consumption below the of inequality among the poor Tanzanian population groups. food poverty line declined from 11.7 percent in 2007 to 9.7 percent in 2011/12, falling by around 2 percentage points (or 16 percent) between 2007 and 2011/12 (Figure I.1). The food 1 The rest of the text uses poverty rate for basic needs headcount poverty line is updated using a food Fisher price deflator, cal- poverty rate and extreme poverty for extreme headcount poverty culated from unit values of the HBS 2007 and HBS 2011/12 rate. data, which shows higher inflation than the combined food 2 To estimate the poverty line for HBS 2007, we adjusted the pov- and nonfood price deflator used to update the basic needs erty line of HBS 2011/12 by a food and nonfood Fisher price index, poverty line. This leads to a stronger increase in the food calculated from unit values of the HBS 2007 and HBS 2011/12 data. poverty line than in the basic needs poverty line between 3 It should be noted that even though the 2007 poverty headcount 2007 and 2011/12 and thus to stronger variation of the basic ratio did not change much through the revision, both measured needs poverty rates than in extreme poverty figures. consumption and the poverty line were substantially increased. Consumption per adult rose by almost one-third. This is due partly to the fact that the revised aggregate includes education, health, The depth and severity of poverty declined more strongly and communication expenditures, which were previously exclud- than the poverty headcount. The depth of poverty (or poverty ed, and partly due to a different way of drawing on the diary and re- gap) measures the average consumption expenditure short- call data for nonfood spending (see Appendix 1.A). The 2007 basic fall of the poor as a share of the basic needs poverty line, needs poverty line has also been revised upwards, from T Sh 13,998 while the severity of poverty (or squared poverty gap) reflects (see URT 2009) to T Sh 19,201 (see URT 2014). 2 Tanzania Mainland Poverty Assessment Box 1.1  Measuring Poverty in the HBS, 2011/12 As it is typically the case in SSA, the HBS 2011/12 uses consump- consumption per “adult equivalent.” This requires equivalence tion as the key welfare measure to analyze poverty. This consump- scales to convert household members of different age and sex tion aggregate comprises food consumption, including food into a standardized adult based on assumptions about caloric produced by households themselves, as well as expenditures on requirements. The HBS 2011/12 poverty analysis follows in this a range of nonfood goods and services (e.g., clothing, utilities, tradition and uses consumption per adult equivalent as the key transportation, communication, health, education, etc.). However, welfare measure. Price deflators are used to adjust consumption the consumption aggregate does not include rent or other hous- per adult equivalent for differences in prices across geographic ing-related expenditures, nor does it include expenditures on domains and over the course of the HBS fieldwork. larger consumer durable items (such as cars, TVs, computers, etc.). The poverty lines are based on the cost-of-basic-needs To the extent that better-off households devote a larger propor- approach. The HBS 2011/12 food poverty line (T Sh 26,085.5 per tion of their total consumption to durable goods, this omission adult per month) is based on the cost of a food basket that creates certain biases and underestimates “true” consumption delivers 2,200 calories per adult per day (given consumption among wealthier families. This matters less for poverty analysis, patterns in a reference population). The basic needs poverty line where the focus lies on the bottom-end of the distribution, but it (T Sh 36,482 per adult per month) adds an allowance for basic can have a significant impact on estimated inequality. nonfood necessities to the food poverty line. Further technical The HBS 2011/12, as most household surveys, collects details on the construction of the HBS 2011/12 consumption consumption data at the level of households. For the purpose of aggregate, adult equivalence scale, price deflators, and poverty poverty and welfare analysis total household consumption needs line can be found in Appendix 1.A. to be adjusted for differences in household size and composition. The basic needs headcount poverty rate (or as used in the This is to account for the fact that, for instance, a single-person text, poverty rate) measures the proportion of the population household requires less consumption than a family of five. One whose monthly (price-adjusted) total household consumption possible approach is to compute consumption per capita, which per adult equivalent is below the basic needs poverty line, and implicitly assumes that all members of the household require the the extreme headcount poverty rate (used in the text as extreme same level of consumption. Another approach, which is wide- poverty rate) measures the proportion of the population whose spread in the context of Sub-Saharan Africa, where typically a monthly (price-adjusted) total household consumption per large share of consumption is spent on food items, is to compute adult equivalent is below the food poverty line. Figure I.2 shows a strong decline of the poverty gap and Trends in Depth and Severity of Figure I.2   severity index by, respectively, 35 and 48 percent, suggest- Poverty in Tanzania Mainland, ing that an important proportion of poor households have 2007–2011/12 been able to reduce significantly their consumption short- 12% fall relative to the poverty line and that the gains were par- ticularly large among the poorest groups. 10% 10.3% 8% These poverty trends still face comparability issues emanat- ing from the changes in the survey design, and these issues 6% 6.7% are further addressed using different prediction techniques. 4.5% 4% The reconstruction of the 2007 consumption aggregate and poverty line can account for changes in the methodology 2% 2.3% to estimate poverty, but they cannot correct for variations 0% in the survey’s design that occurred between 2007 and 2007 2011/12 2007 2011/12 2011/12, such as the changes in the reference period for Depth of Poverty (P1) Severity of Poverty (P2) which nonfood consumption are reported and the changes in the degree of commodity detail. Source: HBS 2007 and 2011/12. Poverty and Inequality Trends 3 We use a semiparametric method and two imputations meth- from a model of consumption estimated using 2011/12 ods, namely, the small area estimation poverty mapping ap- survey data. The explanatory variables used in the mod- plication and multiple imputations chained equations to ad- el are restricted to those that are comparable across the dress the remaining comparability problems and analyze the two surveys, and the relationship between consumption poverty trend between 2007 and 011/12. The three methods and its correlates is assumed to be stable over time in or- proceed as follows (more details are in Appendix 1.B): der to ensure the perfect comparability of consumption across the two surveys. This approach circumvents the  The semiparametric method. This approach, proposed by need for using price deflators and uses the poverty line Tarozzi (2002), is based on the assumption that HBS 2007 for 2011/12 to measure the predicted poverty for 2007. and 2011/12 share the same distribution of per-dult equivalent consumption conditional on a set of some  Multiple imputations chained equations (MI chained). Im- variables that have not been affected by the changes in plemented in STATA with the mi impute chained com- the questionnaire design. It is therefore possible to use mand, these are based on Rubin’s (1987) work to deal observations on these variables from HBS 2007, together with missing values generated by nonresponse in sur- with information on the structure of the conditional dis- vey-based research. The method is close in spirit to the tribution in HBS 2011/12, to recover the marginal distri- poverty mapping technique and consists in filling in miss- bution of consumption in HBS 2007. The approach then ing values for multiple variables using iterative methods uses the reweighting procedure of Dinardo et al. (1996) and chained equations. The approach accommodates to estimate the poverty counts for HBS 2007. arbitrary missing-value patterns and uses less restrictive assumptions than the poverty mapping method.  Small area estimation poverty mapping application. This approach is based on Elbers et al. (2003) and Christi- The different prediction approaches support the decline of aensen et al. (2012). It replaces per-adult equivalent con- poverty between 2007 and 2011/12, but reveal a slightly sumption data in HBS 2007 by predicted consumption lower pace of poverty reduction. Depending on the meth- using both available information on household charac- od used, poverty appears to have declined by around 4–5 teristics (sociodemographic attributes and assets owner- percentage points (or 12–15 percent), which is slightly low- ship) in 2007 as well as the parameter estimates obtained er that the decline of 6 percentage points (or 18 percent) Figure I.3  Adjusted Poverty Rates for 2007 Using Prediction Methods 35% 32.4% 31.9% 33.2% 30% 28.2% 25% 20% 15% 13.1% 13.7% 13.6% 10.4% 9.7% 10% 8.6% 8.6% 6.7% 5% 3.4% 4.7% 3.4% 3.2% 3.0% 1.0% 2.7% 0.9% 1.8% 2.3% 1.3% 0.6% 0% Headcount Depth Severity Headcount Depth Severity Headcount Depth Severity Headcount Depth Severity Semi-parametric (Tarozzi) MI Chained Poverty Mapping 2011/12 Extreme Poverty Poverty Source: HBS 2007 and 2011/12. 4 Tanzania Mainland Poverty Assessment Table I.1  Adjusted Poverty Rates for 2007 Using Prediction Methods Semi-parametric MI chained (with cell MI chained (without Poverty mapping (with Poverty mapping (Tarozzi) phone) cell) cell phone) (without cell) Extreme Extreme Extreme Extreme Extreme Poverty Poverty Poverty Poverty Poverty Poverty Poverty Poverty Poverty Poverty Headcount 13.1% 32.4% 13.7% 31.9% 11.9% 28.6% 13.6% 33.2% 10.6% 28.6% Depth of poverty 3.4% 10.4% 3.0% 8.6% 2.6% 7.5% 2.7% 8.6% 2.2% 7.0% Severity of poverty 1.3% 4.7% 1.0% 3.4% 0.9% 2.9% 0.9% 3.2% 0.7% 2.6% Source: HBS 2007 and 2011/12. reported above. Interestingly, the decline in extreme pover- living standards have been accompanied by improvements ty appears to be quite higher using the prediction methods. in other nonmonetary dimensions of well-being such as It varies between 3 to 4 percentage points against a decline housing conditions, assets, and human capital. The section of only 2 percentage points observed above. This is due to also examines the dynamics of well-being in Tanzania and the fact the prediction methods attenuate the effects of investigates the population groups facing chronic lack of food prices inflation on extreme poverty (see Figure I.3). well-being and those “switching” between states of well-be- ing and deprivation. One problem faced with the prediction methods is related to the difficultly of selecting the consumption correlates A.  Housing Conditions and Assets that are comparable across the HBS 2007 and 2011/12 sur- Housing conditions have improved considerably between veys. These methods are quite sensitive to some household 2007 and 2011/12, providing evidence for rising living stan- characteristics, especially demographic structure and own- dards, including for rural and the poorest households. At the ership of assets. The ownership of certain assets, in partic- national level, the share of households with improved wall ular cell phones, vary a great deal over time. Including cell material went up by 12 percentage points, from 34 percent phones in the prediction models violates the assumption in 2007 to 46 percent in 2011/12. Likewise, improved roof of stability of the consumption correlates, while excluding material went up by 10 percentage points at the nation- them introduces an omission bias. The exclusion of the cell al level and improved floor material by over 5 percentage phones from the prediction models results in lower predict- points (Figure I.4). Interestingly, the rise in improved housing ed poverty measures for HBS 2007, suggesting a very low characteristics seems to have occurred mainly in the rural decline of poverty over time (Table I.1). While the prediction areas and for households in the poorest segments (Tables models including cell phone may bias upward poverty es- 1.C-1 and 1.C-2 in Appendix  1.C). Improved dwelling con- timates for 2007, excluding them would introduce an omis- ditions increased by over 40 percent for households in the sion bias as these devices contribute significantly to house- lowest quintiles, against less than 30 percent for the richest holds’ welfare. Further research will be needed to explore segments. Despite these improvements, more than half of alternative ways to address this problem. poor households and rural dwellers continue to suffer from pitiable housing conditions. Improvements in Households’ II.  Ownership of modern assets increased while ownership Living Conditions and Human of traditional goods deteriorated. There have been some Development Outcomes improvements in ownership of communication and trans- portation devices, mainly cell phones, TV and videos, mo- As poverty is not solely about consumption deficits, this torcycles, and mopeds. Ownership of other selected house- section examines whether the observed improvements of hold items, such as mosquito nets and cooking stoves, also Poverty and Inequality Trends 5 Figure I.4  Trends in Dwelling Materials, 2007–2011/12 67.8% Other Urban 50.6% 97.1% wall material Dar es Salaam 85.8% Improved 33.1% Rural 21.9% 46.1% National 34.1% 69.2% Other Urban 62.9% 96.8% oor material Dar es Salaam Improved 88.1% 22.3% Rural 17.0% 38.8% National 33.3% 90.5% Other Urban 84.6% 99.2% roof material Dar es Salaam Improved 95.3% 54.8% Rural 42.0% 66.2% National 55.8% 0% 20% 40% 60% 80% 100% 2011/12 2007 Source: HBS 2007 and 2011/12. Note: Improved roof means iron sheets, tiles, concrete, or asbestos sheets. Improved floor means cement, ceramic tiles, vinyl, or wood/ bamboo. Improved wall means stones, cement bricks, or baked bricks. improved, related partly to public interventions for the (NPS) data, show that households tend to replace traditional former. The ownership of these assets appears to have im- devices such as radio and bicycle by more upgraded goods, proved more markedly among the less well off. The propor- such as TVs or motorbikes. tion of poor households having a mobile phone has multi- plied by seven from 5 percent to around 39 percent and the Ownership of agricultural land, particularly large plots, im- proportion of poor families owning mosquito nets almost proved substantially for poor households. For the poor and doubled (Tables 1.C-3 and 1.C-4). Conversely, ownership of nonpoor alike, there has been an improvement of ownership more traditional assets such as basic furniture items, radios, of agricultural land with areas over 5 acres, but the improve- and bicycles has declined. It seems that households have ments are more marked for poor households (Figure  I.6). replaced these items by more modern ones, as can be seen The ownership of plots of marginal size also improved, while from the decline of bicycles and increase of motorcycles that of plots of small and medium sizes declined, particularly and mopeds or the decline of radio and increase of TVs in for the poor. The increase of land ownership seems to have Figure I.5 and Table 1.C-4. This is further confirmed by the resulted in a decline of plots provided for free. While this can analysis of Seff et al. (2014) who, using National Panel Survey be considered as a positive sign, the impact on small and 6 Tanzania Mainland Poverty Assessment Figure I.5  Trends in Assets Ownership Increases in modern assets ... ... but declines in more traditional items 100 35 100 16 90 90 14 30 80 80 12 70 25 70 Percetage points Percetage points 60 60 10 20 Percent Percent 50 50 8 40 15 40 6 30 10 30 4 20 20 5 10 2 10 0 0 0 0 Cell Phone Cooking Stove Mosquito net TV Video Fridge Computer Motor Cycle/ moped Car Sewing Machine Telephone (landline) Iron Chair Table Bed Bicycle Water Heater Radio Sofa 2007 2011/12 increase 2007 2011/12 decline (percentage points-right axis) (percentage points-right axis) Source: HBS 2007 and 2011/12. subsistence farmers who face liquidity constraints might be B.  Human Development negative in the short term. The increase of large plots for Human development outcomes have improved since the ear- the poor can contribute to the improvement of their pro- ly 2000s, but overall levels remain low, particularly in compar- ductivity and living standards, but additional support will be ison to other neighboring countries as well as developing re- necessary to help them better exploit these resources. gions. This section examines education and health outcomes. Trends in Agricultural Land Ownership (%) Figure I.6   a. Poor Households b. Nonpoor Households 80 80 69 71 70 67 70 65 65 60 60 56 60 60 53 48 50 50 37 40 40 36 32 31 35 31 30 30 26 26 24 20 20 19 19 19 18 20 18 20 18 17 17 16 16 15 14 14 12 11 10 10 8 Marginal 0 8 5 2 2 Marginal 2 Marginal 2 0 Marginal 1 0 1 0 0 Small Medium Large Marginal Small Medium Large Small Medium Large Small Medium Large Marginal Small Medium Large Small Medium Large Owned Land Rented Land Provided free Owned Land Rented Land Provided free 2007 2011/12 Source: HBS 2007 and 2011/12. Note: Marginal is less than 0.5 acres; small is between 0.5 and 2.5 acres; medium is between 2.5 and 5 acres, and large is 5 acres and over. Poverty and Inequality Trends 7 Education Gross Enrollment Rates in Figure I.7   Gross enrollment rates at all levels of schooling are con- Tanzania and International sistently lower than the average for SSA countries and Comparison (%) much below achievements in other developing regions 120 (Figure I.7). 100 80 Primary school enrollments increased sharply after the in- 60 40 ception of the Primary Education Development Program 20 in 2001, but some of these gains appear to be eroding. 0 Primary Secondary Tertiary According to HBS data the primary net enrolment rate in- creased from 59 percent in 2000/01 to 84 percent in 2007, Tanzania Sub-Saharan Africa East Asia and Pacici c Latin America and Caribbean but then fell back to 78 percent in 2011/12 (Figure I.8). Ad- ministrative data from the Education Management Infor- Source: World Development Indicators (WDI 2014). mation System generally show higher enrollment rates Note: Regional aggregates for year 2011 and Tanzania estimates than the household surveys but confirm the recent decline for 2012. in net and gross enrollment rates.4 The recent declines are quite disconcerting given the approaching MDG target Primary and Secondary Net Figure I.8   date for achieving universal primary education. Enrollment Rates, 2001–2011/12 There has been a remarkable expansion in access to low- Primary Education er secondary education under the Secondary Education 1.0 0.8 Development Program, although from very low levels. In 0.6 2000/01 only 5 percent of the population of lower second- 0.4 ary school age (14–17 years) was in school. This proportion 0.2 rose to 17 percent in 2007 and 31 percent in 2011/12. The 0.0 Mailand Rural Other Dar es surge was particularly pronounced in rural areas, where the Tanzania Urban Salaam net enrollment rate at the lower secondary level increased Lower Secondary Education from 2 to 22 percent from 2000/01 to 2011/12. However, 1.0 upper secondary enrollments remain negligible, at below 2 0.8 percent of the population ages 17–22 years old in 2011/12. 0.6 0.4 0.2 Increased enrollments have gone in hand with a reduction 0.0 Mailand Rural Other Dar es in late enrollments, particularly between 2001 and 2007, Tanzania Urban Salaam and more recently a reduction in overage enrollments. Be- tween 2001 and 2007 the share of children ages 7 years (the 2000/01 2007 2011/12 compulsory school age) enrolled in school increased from Source: HBS 2000/1, 2007 and 2011/12. 23 to 60 percent, though it then fell slightly to 57 percent in 2011/12. In addition, the share of overage children (14 years and over) enrolled in primary school declined substantive- percent of 15-year-old children were still enrolled at the pri- ly between 2007 and 2011/12 (Figure I.9). While in 2007, 60 mary level, this share had declined to 26 percent in 2011/12, partly a reflection of the increased (and earlier) transition to the secondary level. 4 See Gaddis and Hoogeveen (2013) for a discussion of discrep- ancies between survey-based and administrative enrollment rates. Also note that enrollment rates are proxied by attendance rates in While primary education is not marked by significant gen- the HBS. der inequality, girls continue to be less likely to attend upper 8 Tanzania Mainland Poverty Assessment Share of Children Enrolled in Figure I.9   percent in 2001 to 33 percent in 2011/12, while the share of Primary School, by Age the population with some or completed lower secondary education has become more prevalent, increasing by about 100% 10 percentage points.5 80% 60% While access to education has improved, education out- comes at the primary and secondary levels remain poor 40% and uneven. Weak learning outcomes are documented, for 20% 0% instance, by the 2012 Uwezo Learning Assessment, which 5 6 7 8 9 10 11 12 13 14 15 16 17– shows that only 26 percent of Standard 3 students can read 24 a Standard 2 level Kiswahili story. Even in Standard 7, the final 2000/01 2007 2011/12 year of the primary education cycle, almost one-quarter of Source: HBS 2000/1, 2007 and 2011/12. students do not meet Standard 2 level proficiency. Results are somewhat better for basic numeracy (where the results have improved since 2010), but even worse for English. The levels of education. The HBS 2011/12 shows that there were results, which are representative at the district level, also 3.8 million male and 3.7 million female students in primary reveal large geographic inequalities—with pass rates of 79 school, which results in a gender parity index of 98 percent. percent in the highest performing regions and of 27 percent Administrative enrollment data for 2011/12 even shows a in the lowest performing districts (Uwezo, 2013). slight advantage for girls. However, gender parity declines to 84 percent at the lower secondary level and 56 percent at the upper secondary level. Gender inequality is more pro- Health and Nutrition Infant mortality (which measures the probability of infants nounced for gross than for net enrollments, indicating that dying before their first birthday per 1,000 live birth) dropped the gap in enrollment probabilities between boys and girls from 68 in the 2004/05 to 51 in the 2010 (Figure I.12).6 Un- is larger for children outside the official school age. der-five mortality, which measures the probability of chil- dren dying between birth and the fifth birthday, declined Increased access to primary and secondary education is from 112 in the 2004/05 to 81 in the 2010. Since both indi- slowly transforming the educational structure of the labor cators were already on a declining trend during the 1990s, force. As shown in Figure I.11, the share of the population Tanzania stands good chances of achieving the MDG target ages 15-years-old and over, who have no education or less of reducing child mortality by two-thirds by 2015 (com- than completed primary education, has declined from 45 pared with 1990). Figure I.10  Gross Enrollments by Gender and There is also cautious evidence of recent progress in mater- Gender Parity Index, 2011/12 nal mortality. At 454 deaths per 100,000 live births—per the 5,000 100% 2010 Demographic and Health Survey (DHS)—the maternal mortality ratio remains high, though it has come down from Enrollment (‘000) 4,000 80% Gender Parity 3,000 60% 2,000 40% 5  However, as discussed in World Bank (2014) the educational 1,000 20% makeup and skill composition of the Tanzanian labor force today 0 0% still resembles that of Thailand in 1975 and the country has a long Primary Lower Upper Tertiary Secondary Secondary way to go to catch up with the emerging economies in Asia and Latin America. Male Female Gender Parity Index 6 Child mortality estimates in the 2004/05 DHS refers to the period Source: HBS 2000/1, 2007 and 2011/12. 2000–04/05, and in the 2010 DHS to the period 2006–10. Poverty and Inequality Trends 9 Educational Attainment is Figure I.11   Figure I.13  Recent Improvements in Maternal Improving Slowly Mortality, 2004/05–2010 Educational Attainment Population 15+. 2000/01 800 deaths per 100,000 live birth University 600 578 Upper Secondary 454 Lower Secondary 400 Completed Primary Some Primary 200 No Education 0 60 50 40 30 20 10 % 10 20 30 40 50 60 1995–2004 2001–2010 Educational Attainment Population 15+. 2011/12 Source: DHS data; NBS; and ICF Macro 2011. Note: The figures include data for Zanzibar. University Upper Secondary Lower Secondary Anthropometric indicators for young children show some Completed Primary improvement since 2004/05, but the trends are uneven and Some Primary malnutrition continues to be widespread. Stunting, defined No Education as reduced height for age and an indicator of chronic malnu- 60 50 40 30 20 10 % 10 20 30 40 50 60 trition, was consistently high between the 2004/05 and 2010 Males Females DHS (at 42–44 percent). It came down to 35 percent in the NPS 2010/11 but rebounded to 37 percent in the NPS 2012. Source: HBS 2001 and 2011/12. Underweight (low weight for age) fell slightly from 16 to 13 percent. Wasting, measured as low weight for height and an Continued Reductions in Child Figure I.12   indicator of acute food shortage or infectious disease (such Mortality, 2004/05–2010 as diarrhea), increased from 2.6 percent in 2004/05 to 6.6 per- cent in 2010, but declined to 4.2 percent in 2012 (Figure I.14). 120 112 death per 1000 live birth 100 91 Malnutrition appears even more widespread if one consid- 81 80 68 ers the risk a child faces of suffering undernourishment at 60 58 51 some point in time. Fifty-five percent of children less than 40 3 years old at the time of NPS 2008 fieldwork were stunt- 20 ed, and 22 percent were underweight, at least once before 0 reaching age five (based on at least two independent obser- 2000–2004 2004–2006 2006–2010 vations at different points in time). This risk falls slightly for Infant mortality Under-5 mortality children of the same age at the time of NPS 2010. Fifty-two Source: DHS data; NBS; and ICF Macro 2011. Note: The figures include data for Zanzibar. 7 The lack of statistical significance partly mirrors that maternal death is a rare event in the surveys, so that mortality ratios tend 578 in the 2004/05 DHS (Figure I.13). While this change is to come with large standard errors (NBS and ICF Macro 2011). In not statistically significant, it suggests a departure from the addition, maternal mortality rates are measured for the 10-year increase in maternal mortality observed between the 1996 period preceding the survey, which also implies that the indica- and 2004/05 DHS.7 However, Tanzania will not achieve the tor does not react immediately to changes in the socioeconomic MDG targets on maternal mortality. environment. 10 Tanzania Mainland Poverty Assessment Uneven Progress in Child Nutrition, 2004/05–2010/11 Figure I.14   50 44 43 42 40 37 35 Percent 30 20 16 16 16 14 13 10 5 7 4 3 4 0 DHS 2004/05 NPS 2008/09 DHS 2010 NPS 2010/11 NPS 2012/13 DHS 2004/05 NPS 2008/09 DHS 2010 NPS 2010/11 NPS 2012/13 DHS 2004/05 NPS 2008/09 DHS 2010 NPS 2010/11 NPS 2012/13 Stunting Wasting Underweight Source: DHS Statcompiler (2014); URT (2011); and Seff et al. (2014). Notes: Figures include data on Zanzibar. Children are below 5 years of age. Z-score below – 2 SDs. Based on the 2006 WHO child growth standards. percent of those children were at risk for ever being stunted economic status quartiles, where economic status is mea- and 19 percent for being ever underweight. sured by consumption and each quartile represents one- fourth of the population. It examines the main characteris- Infections such as Malaria and HIV continue to account for a tics of the households who experienced a decrease in their substantial burden of disease. With an estimated 10 million economic status or remained trapped in the poorest quartile. malaria cases in 2010, Tanzania continues to be one of the most affected country by the disease in the World.8 HIV/Aids There are substantial variations in households’ economic prevalence in 2012 was estimated at 5.1 percent of the pop- status, both positive and negative. Around 60 percent of the ulation ages 15–49 years, slightly above the SSA average population changed economic status, in the distribution of (4.7 percent) but somewhat below prevalence rates in other consumption, between 2008 and 2013.10 About half of them East African countries (for example, Uganda 7.2 percent and moved up in economic status, while the other half experi- Kenya 6.1 percent). enced a deterioration of their economic status. The poorest and richest population groups were less likely to change their C.  Dynamics of Well-being economic status than those in the middle classes. Living conditions and human capital appear to have im- proved over time, despite the persistence of important depri- Table I.2 shows movement between economic status quar- vations and gaps in many dimensions of human well-be- tiles for the first and last round of the NPS. Economic status ing. But these improvements are not homogenous for all household groups and may hide significant fluctuations in 8 The World Health Organization considers Tanzania to be one of the the well-being. Some households may have experienced four countries with the highest malaria burden in Africa, along with improvements in their economic status, while others may Nigeria, DRC, and Uganda (WHO 2012). The other data cited in this have fallen into a state of poverty. These dynamics cannot be section are based on the World Development Indicators (WDI 2014). tracked by cross-sectional HBS data but require panel data The analysis is based on the paper by Seff et al. (2014). The three 9  series. This study uses the three waves of the NPS—fielded in waves of NPS are for 2008/09; 2010/11, and 2012/13. 2008/09, 2010/11, and 2012/13—to explore more in depth 10  The change of economic status is related to the change of quar- the dynamics of well-being in Tanzania during the past five tile in the distribution of per adult equivalent consumption, where years.9 The analysis examines the movement in and out of each quartile represents 25 percent of the population. Poverty and Inequality Trends 11 quartiles are created for each wave of the NPS using real con- those better endowed with assets have been more able to sumption expenditure per adult equivalent, where the first improve or at least maintain their economic status. quartile reflects those at the bottom 25 percent of annual expenditure and the fourth quartile represents consumers at The urban residents were more likely than the rural ones to the 75th percentile of expenditure and above. The results are maintain their economic status, but the difference between presented through transition matrices, where the diagonal the two areas, in the likelihood of maintaining the economic moving from the top left to the bottom right reflects those status quartile, significantly declined over time. Between the individuals who have maintained their level of consumption first two waves of NPS data, the percentage of individuals expenditure between rounds, those in the bottom left trian- who maintained their economic status was 48 percent in ur- gle have fallen to a lower welfare quartile, and individuals in ban areas against 37 percent in rural sectors. These percent- the upper right triangle have improved their welfare quartile. ages dropped, respectively, to 43 percent and 39 percent between the last two waves. Many Tanzanians are trapped in poor well-being status.11 Of those who were in the poorest quartile in 2008, about half (12 percent) were still in the poorest quartile in 2013. Such Moderate and Fairly Stable III.  individuals are likely trapped in chronic poverty. Inequality Many Tanzanians have experienced a deterioration in their This section examines the extent and structure of inequal- living standards. Around 30 percent of the population has ity in the distribution of household consumption expendi- moved to lower economic status during the past five years. tures, using data from three rounds of HBS for 2001, 2007, Among them 13 percent have moved to the lowest quar- and 2011/12. It is now widely admitted that above a certain tile, falling into poverty. This reveals that many Tanzanians threshold, inequality undermines growth and poverty-al- are vulnerable to poverty, even among those that are not leviation efforts and affects the length of growth spells.12 currently poor. Those who became poor are generally those Reaching a better understanding of how pervasive and who lack assets, mainly agricultural land and livestock, while deep are inequalities in Tanzania, would help the design of policies to accelerate the reduction of poverty. Table I.2   Changes in Economic Status across The Level and Trend of Consumption A.  Quartiles, Wave 1 (2008/09) to Inequality Wave 3 (2012/13) Tanzania shows moderate levels of inequality in 2012. With the Gini coefficient estimated at less than 40, inequality in Wave 3 quartiles Wave 1 Tanzania is moderately high by international standards but quartiles 1st (poorest) 2nd 3rd 4th (top) Total lower than Sub-Saharan average inequality. The Gini coeffi- 1st 12% 7% 4% 2% 25% cient of real per capita monthly consumption indicates that (poorest) (1.0) (0.7) (0.5) (0.3) the level of inequality for Tanzania is approximately 36, be- 2nd 8% 7% 7% 3% 25% low the SSA average of 45.1 and the low income countries (0.7) (0.6) (0.7) (0.4) average of 40.13 Among East African countries, Tanzania’s 3rd 4% 7% 7% 7% 25% (0.4) (0.6) (0.6) (0.7) Gini coefficient is below that of Burundi, Kenya, Uganda, and 4th (top) 1% 4% 7% 13% 25% (0.2) (0.4) (0.6) (0.9) 11 Here we use a relative concept of “poverty,” which basically im- Total 25% 25% 25% 25% plies that the household falls into the poorest consumption quartile. Source: Seff et al. 2014. 12 See UNDP 2013; Chambers and Krause 2010; and Berg and Os- Notes: Point estimates are weighted to population of individuals try 2011, among others. in wave 1; Standard errors in parentheses are corrected for stratifi- cation and clustering. Total observations are: 3,082. 13  Africa’s Pulse (2013) and WDI Gini indicators. 12 Tanzania Mainland Poverty Assessment Rwanda and is only slightly higher than Ethiopia.14 It is on par areas, as can be seen from the changing shape of the Lo- with levels of inequality in South and East Asia, which range renz curves in Figure I.15. Much of the reduction in inequal- around 38.4, and significantly lower when compared to parts ity seems to be driven by an increase in the welfare share of South America, such as Mexico, Bolivia, and Brazil, where accruing to the poorest segment of the population, as the levels of inequality range from 47 to 55.15 consumption share of the poorest quintile grew by more than 16 percent between 2001 and 2011/12 and by over 20 It is worth mentioning that the levels of inequality in Tanzania percent during the past five years, except in the secondary are likely higher than the figures reported here, as the avail- cities, where it grew by only 11 percent over the past decade able surveys fail to sample the richest households and to cap- (bottom part of Figure I.15). Even though part of the increase ture the rising concentration of wealth among people at the in the share of consumption going to the bottom quintile top end of the distribution. Also, the consumption aggregate can be attributed to improvements in the survey design, the used to measure inequality excludes expenditures on housing adjusted inequality estimates using the reweighting proce- and durable goods. Expanding the food and nonfood expen- dure, as well as the small area estimation techniques, reveals diture aggregates to include these expenses would probably also positive changes over the past decade in the consump- increase inequality. Finally, expenditure-based measures of tion shares of the lowest quintile groups. inequality tend to underestimate income inequality because expenditure is closer to permanent income and is likely to be The Structure of Consumption B.  less dispersed than current income. Inequality The positive picture of equalization of consumption distri- Inequality in Tanzania shows a slightly decreasing trend over bution patterns in Tanzania may hide persisting inequali- time. The Gini coefficient decreased from 38.8 to 35.8 be- ties between groups. It is important, thus, to examine the tween 2001 and 2012 (see Figure I.15). The HBS and NPS data- structure of inequality and to investigate the extent to sets show slightly different levels and trends for inequality. which consumption inequality is attributable to variations This is possibly due to differences in measurement methods between population subgroups. This investigation can be of consumption expenditures between the two datasets. The carried out by the decomposition (or breakdown) of in- first uses the diary method and the second a seven-days-re- equality by population subgroups, which consists of sepa- call method for collection of food consumption data.16 Also, rating overall inequality in the distribution of consumption NPS data do not collect information on clothing expenditures, into inequality within population subgroups and inequality and there have been no changes in the survey design similar between them. (For more see Box 1.2.) to those introduced in HBS. But although the inequality es- timates from NPS did not confirm the declining trend of in- Table I.3 provides summary results of the shares of inequal- equality, it still provides evidence of moderate and fairly stable ity explained by the differences between population sub- inequality at a level below 40-as estimated by Gini index. groups partitioned according to eight household attributes (the gender, age, educational level, activity status, and sec- For the rest of the analysis in this section, the study uses tor of employment of the household head and the regional the HBS, as it is the nationally representative survey that is specifically designed for national measures of poverty and 14 The Gini coefficients in some East African countries are 46 for inequality. Burundi in 2012; 47.7 in Kenya in 2005; 44.3 in Uganda in 2009; 50.8 in Rwanda in 2011; and 33.6 in Ethiopia in 2011. Dar es Salaam and secondary cities display more unequal World Development Indicators database (WDI 2014). The GINI 15  distributions of consumption than rural areas. The Gini coef- coefficient for Latin American countries is based on income which ficients are respectively of 36, 38, and 30 for the capital city, generally shows higher variability than consumption. rest of urban, and rural areas in 2011/12. The distribution of 16 A study by Beegle et al. (2012) revealed that diary-based collec- consumption is equalizing over time in all the regions, with tion of food consumption data leads to lower inequality estimates the most substantial improvement occurring in the rural than recall-based collection. Poverty and Inequality Trends 13 Figure I.15  Lorenz Curve and Inequality Coefficients Lorenz Curve by Area Lorenz Curve by Area Lorenz Curve by Area Tanzania 2011–2012 Tanzania 2001 Household Budget Survey Tanzania 2007 Household Budget Survey Household Budget Survey 1.0 1.0 1.0 0.9 0.9 0.9 0.8 0.8 0.8 0.7 0.7 0.7 0.6 0.6 0.6 0.5 0.5 0.5 0.4 0.4 0.4 0.3 0.3 0.3 0.2 0.2 0.2 0.1 0.1 0.1 0.0 0.0 0.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 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 Cumulative population proportion Line of perfect equality Dar Essalam Other Urban Rural 2001 2007 2011/12 Income shares Income shares Income shares Low Top p90/ Low Top Low Top Gini p90/p10 quintile quintile Gini p10 quintile quintile Gini p90/p10 quintile quintile National 38.78 5.42 6.52 45.88 38.50 5.18 6.62 45.72 35.84 4.39 7.73 44.07 Rural 37.23 5.08 6.83 44.55 35.54 4.66 7.26 43.33 29.86 3.53 8.98 39.06 Other urban 38.80 5.69 6.27 45.49 39.96 5.96 5.98 46.58 38.14 4.92 6.96 45.65 Dar es Salaam 39.77 5.60 6.44 46.55 40.12 5.60 6.44 47.26 36.04 4.36 7.74 44.40 Source: HBS 2001, 2007, and 2011/12. location, the urban/rural status, and the demographic com- 2007 and more than doubled since 2001. This increase is position of the household).17 mainly driven by the widening disparities between house- hold groups whose head has not completed the primary Over 20 percent of total real per capita consumption in- education level and whose head is illiterate as well as by equality in 2011/12 can be explained by inequality between the more than proportionate expansion of the mean con- six groups of households sorted by the educational attain- sumption level of tertiary educated groups relative to the ment of the head. As expected, mean consumption levels other groups. Families headed by university graduates seem of the different educational groups increase with the edu- to have been able to benefit from economic growth more cation of household head, and more than double when the than the other households. education of the head is above completed primary. There are also substantial differences in average consumption lev- Inequality between geographic regions is increasing as els between household groups headed by university gradu- well. Even though consumption inequality remained rel- ates and those headed by secondary graduates. atively stable or slightly decreased over time, the wel- fare gaps between urban and rural areas and between Differences between education groups seems to be in- creasing over time—the share of inequality attributable to the household head’s education, in both Theil_L and Teil_T, 17 For details on the different household characteristics used in is around 6 percentage points higher in 2011/12 than in the decomposition, see appendix 1.D. 14 Tanzania Mainland Poverty Assessment Box 1.2  Inequality Decomposition The static decomposition of inequality enables one to explore how the differences in households’ characteristics affect the level of inequality and provide important clues for understanding the underlying and changing structure of real per capita consumption distribution in Tanzania. The decomposition follows the approach of Cowell and Jenkins (1995) and consists of separating total inequality in the distribu- tion of consumption into inequality between the different household groups in each partition, IBetw, and the remaining within-group inequality, IWithin. As the most commonly decomposed measures in the inequality literature come from the General Entropy class, mean log deviation (Theil_L) and the Theil_T indices in real per capita monthly consumption expenditure are used to identify the contribution 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 households attributes (or group of attributes) is measured by IBetw/Itotal , where between and within group inequalities are defined, respectively, for Theil_L and Theil_T indices as  k  µ  k IBetw = ∑ f j log     IWithin = ∑ f j GE0j  µ j   j =1 j =1    k µj   µj  k IBetw = ∑ f j   log   IWithin = ∑ v j GE1j  j=1  µ   µ  j =1 with fj the population share, νj the consumption share, and µj the mean consumption of subgroup j; µ total mean consumption, k GE 0k Theil_L index, and GE 1 Theil_T index of subgroup j. n y n y  y ∑i=1log y with: Theil _ L = 1/ n  and Theil _T = 1/ n∑i =1 i  log  i   i y y yi: is real monthly per capita consumption expenditure for household i and y is mean real monthly per capita consumption expenditure. geographic regions increased substantially. Differences be- These widening disparities can be explained mainly by the tween urban and rural areas as well as disparities between uneven growth of the average consumptions of household geographic locations account for more than 17 percent of groups across the different geographic locations, as con- overall inequality in the most recent survey. The differences sumption levels for households in the coastal and central in average consumption levels between household groups zones have increased proportionately more than for house- living in urban zones and those located in rural areas are holds in the other regions. quite substantial. The welfare gap between these groups has widened over time, increasing by over 9 percentage There are quite important welfare disparities between sec- points between 2007 and 2011/12 and more than tripling tors of employment groups. The share of total inequality at- since 2001. This increase is driven by the considerable ex- tributable to the differences in the mean consumptions of pansion of the average consumption level of households these sectors is around 13 percent. Household groups head- in Dar es Salaam, which grew proportionately much more ed by government employees and private sector employ- than average consumptions of household groups in the ees are much better off than groups with heads employed other locations. in the other sectors. Inequality between these groups slight- ly increased in 2011/12 due to a more than proportionate Interregional inequalities, already important in the begin- increase of the average consumption level of household ning of the decade, are gaining importance over time, in- groups headed by private sector employees relative to the creasing by more than 10 percentage points since 2001. other groups. Poverty and Inequality Trends 15 Decomposition of Inequality by Household Attributes Table I.3   2001 2007 2011/12 Share of inequality explained by (%) Share of inequality explained by (%) Share of inequality explained by (%) Theil-L Theil-T Theil-L Theil-T Theil-L Theil-T Education of head 9.94*** 10.20*** 14.70*** 15.40*** 20.80*** 21.10*** (0.021) (0.028) (0.010) (0.012) (0.015) (0.014) Gender of head 0.000 0.000 0.000 0.000 0.001 0.001 (0.000) (0.000) (0.000) (0.000) (0.001) (0.001) Age of head 1.99*** 1.73*** 1.19*** 1.04*** 1.32*** 1.08*** (0.006) (0.005) (0.003) (0.003) (0.003) (0.003) Activity stat. of head 0.751 0.700 0.48** 0.39** 0.32* 0.25* (0.014) (0.016) (0.002) (0.001) (0.001) (0.001) Empl. sector of head 9.87*** 9.13*** 12.60*** 12.10*** 13.70*** 12.60*** (0.013) (0.011) (0.010) (0.010) (0.010) (0.009) Family type 12.10*** 13.00*** 10.50*** 11.20*** 10.60*** 10.30*** (0.018) (0.021) (0.008) (0.009) (0.008) (0.009) Urban/rural status 5.76*** 5.39*** 8.69*** 8.27*** 19.10*** 17.40*** (0.009) (0.010) (0.007) (0.007) (0.012) (0.012) Regional location 6.79*** 6.03*** 11.50*** 10.50*** 18.40*** 16.60*** (0.012) (0.010) (0.010) (0.009) (0.011) (0.011) Source: HBS for 2001, 2007, and 2011/12. * 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. Differences in households’ demographic composition The gender, age, and activity status of the household head accounts for a quite significant share of total inequality, have marginal explanatory powers barely exceeding 1 per- amounting to around 11 percent. Households comprised cent. Total consumption inequality is overwhelmingly a of only adults all over 14 years old, whether single or in matter of inequality within these various household groups. couples, are much better off than the rest of household The low share of gender in these decompositions can be groups, while elderly households whose head is 65 years explained by the low proportion of woman-headed house- old or over seem to face severe hardships and have the holds in the sample, amounting to less than 20 percent, and lowest mean per capita consumption levels. The contribu- the particular status of women who head their own house- tion of family composition to inequality seems to slightly holds, as most are widowed, running their own agricultural decline over time. business, or benefitting from remittances from family abroad. 16 Tanzania Mainland Poverty Assessment Poverty and Inequality Trends 17 18 Tanzania Mainland Poverty Assessment Chapter  2 Poverty Profile Key Messages ➤➤ Despite the improvements in the poverty level, there are still nearly 12 million Tanzanians living in poverty. ➤➤ Poverty is associated with larger families, lower education, and low access to infrastructure. ➤➤ Nonfarm diversification and internal migration can make an impact on poverty alleviation ➤➤ Migration contributes to raising the welfare of migrant households as well as that of their families left behind. The previous chapter shows that poverty has started to Still Too Many Poor and People I.  decline and that improvements of households’ living stan- dards have been coupled with improvements in the other Clustered Around the Poverty dimensions of well-being. However, despite these positive Line changes, around one-third of the Tanzanian population continues to live in poverty, and an important proportion of Around 12 million Tanzanians continue to live below the the population in the poorest groups is likely to be trapped poverty line. The improvements of poverty over time have in persistent poverty. not resulted in a significant decline of total number of poor people. The rapidly growing population in Tanzania—which The “Tanzanian poor” are not a homogeneous group and increased from around 38 million in 2007 to 42 million in poverty is not a single problem that can be solved with a 2011/12—slowed the reduction of the absolute size of poor stand-alone or uniform package of policy measures. In order population.19 As a result, the total number of poor people for the government and other stakeholders to instigate ap- declined by only 10 percent, falling from 13.2 million in 2007 propriate pro-poor measures, it is necessary to understand to 11.9 million in 2011/12. The absolute number of extreme in detail the characteristics and profiles of the most disad- poor declined even more slowly, by 7 percent, falling from vantaged groups and the different constraints they face. 4.5 million to 4.2 million. This explores the correlates of poverty in Tanzania and who is most affected. As recent literature on Tanzania highlights the positive impact of migration on economic mobility and 18  See Beegle et al. (2011) and Christiaensen et al. (2013). poverty, the analysis will focus on the welfare payoffs of in- 19  The total population for mainland Tanzania in HBS 2011/12 (42.3 ternal migration and examine the potential for geographic million) is slightly lower than the total population captured in the mobility to improve poor households’ living standards.18 2012 population and housing census (43.6 million). 19 A large share of the Tanzania population is clustered around Mean Consumption by Percentile Figure II.1   the poverty line, vulnerable to fall into poverty but also like- (US$ at 2005 PPP) ly to escape from it. The poverty depth indicator shows that 15 Mean consumption per capita many of the poor are close to the poverty line, likely to escape per day -US$ 2005 PPP poverty through small improvements of their living stan- dards. Likewise, slight increases in the value of the poverty 10 line can lead to significant increases in the estimated poverty levels. For instance a rise of the poverty line by 25 percent—a 5 mere additional T Sh 300 per adult per day—increases the poverty headcount by more than 50 percent (Table II.1).20 0 0 20 40 60 80 100 The international poverty rate is 43.5 percent, around 15 Percentile percentage points higher than the national poverty rate. Mean consumption/ US$ 1.25 day/person Tanzania’s national poverty line reflects the country’s spe- Nat. poverty line US$ 2 cific costs of basic consumption needs, but is difficult to compare with other countries poverty thresholds. To over- Source: HBS 2011/12. come this issue the international poverty line of US$1.25 per capita per day (in 2005 PPP exchange rate) is often used to percentile, indicating a quite significant increase of the pov- evaluate a country’s poverty record vis-à-vis other develop- erty rate due to relatively small changes in the poverty line. ing countries or regions.21 Tanzania’s international poverty rate compares favorably with The US$1.25 international poverty line is slightly higher other SSA countries with similar income levels. Poverty is than the 2011/12 basic needs poverty line but yields a sig- slightly lower than the SSA average of 46.8 percent. As can nificantly greater poverty rate of 43.5 percent compared to be seen in Figure II.2, poverty is relatively less pervasive in the national poverty rate of 28.2 percent.22 This is due to the Tanzania than in the neighboring countries that have similar clustering of people around the poverty line, as can be seen also in Figure II.1, which plots average consumption (per capita per day) for each percentile of the consumption dis- 20  T Sh 120 corresponds to US$0.17 at current official exchange rates. tribution. At the bottom end of the distribution, the curve 21  Global poverty estimates are based on an international poverty appears relatively flat, showing that many people are in close line of US$1.25 per person per day, converted into local currency vicinity of the basic needs poverty line (sienna solid line). The using purchasing power parity estimates (PPPs) rather than curren- solid green line, which represents the international poverty cy exchange rates. Official World Bank estimates of global poverty line, crosses the consumption distribution close to the 44th use PPPs from 2005 (from the International Comparison Program of 2005). In 2014, a new set of PPPs was released (International Com- parison Program of 2011). At this time, the World Bank has not up- dated its global poverty estimates to be based on the 2011 PPPs; the Table II.1  Poverty Headcount for Alternative World Bank is currently examining the 2011 PPPs in the context of Poverty Lines, 2011/12 global poverty monitoring. In this report we also use the 2005 PPPs to be consistent with official World Bank global poverty estimates. Poverty line (T National poverty Change Sh) Δ (T Sh) headcount (%) 22  The national poverty line is equivalent to about US$1 per capita +0% 36,482 28.2 per day in 2005 PPP. We should highlight that each of these national +5% 38,306 1,824 30.8 and international approaches for the measurement of the pover- +10% 40,130 3,648 34.7 ty line has its strengths and limitations. While international poverty +15% 41,954 5,472 37.9 lines allow comparability between countries and over time, they +20% 43,778 7,296 41.2 remain inevitably arbitrary. The national poverty lines, despite their +25% 45,603 9,121 43.8 limits, are more closely tailored to the actual costs of livings in the Source: HBS 2011/12. country. 20 Tanzania Mainland Poverty Assessment Poverty Estimates in Tanzania Figure II.2   continued to live in the rural zones in 2011/12, relying on and Other Developing Countries subsistence agriculture and low productivity jobs. Around by Percentage 10 million of this population is in poverty and 3.4 million is in extreme poverty, compared to respectively less than 1.9 International poverty headcount (%) million and 750,000 people who live in poverty and extreme South Africa Congo, Rep. poverty in the urban sector. Senegal Chad Ethiopia The specific geographic location also matters for poverty. Uganda Niger As apparent from Tables 2.B-1 and 2.B-2 in Appendix 2.B, Guinea households living in coastal regions have a higher standard Tanzania São Tomé and Principe of living and are less likely to be poor than those located Mali in the south and to a lesser extent the southern highlands. Benin Togo Lesotho Sierra Leone The demographic structure of the household is closely as- Nigeria sociated with poverty. Figure II.4 illustrates some of the key Rwanda Malawi links with family type, number of children and the age of Zambia Madagascar the household head. Households with children, followed by MENA elderly families, have the highest poverty rates. In contrast, East Asia and Paci c South Asia those without children appear to be less poor. Sub-Saharan Africa 0 20 40 60 80 100 Households with a large number of dependents and more Source: PovcalNet estimates for the period 2010–2012. children under the age of 14 are poorer (tables 2.B-1 and Note: Poverty estimates based on the US$1.25 per person per day 2.B-2). Poverty is particularly high among households with international poverty line. Country level poverty estimates are for five or more children. The interaction between family size the period between 2010 and 2012. Region averages are estimat- ed using PovcalNet for the year 2011. These figures are provisional and poverty is bidirectional. On one hand, the large number and subject to be updated. of children and dependents affects the ability of the poor to cover basic food needs and move out of poverty. On the other, poor households tend to have more children to com- income levels. Nevertheless, it remains relatively higher than pensate their inability to invest in the human capital of their in Uganda, Chad, Senegal, Democratic Republic of Congo, kids and as an insurance strategy against infant mortality, and Ethiopia. When compared to other developing regions, trapping them in a vicious circle of poverty. poverty seems much more prevalent in Tanzania. It is around 4 percentage points higher than average poverty levels in Household size dynamics are in part reflected in the ru- South Asia and over 20 percentage points higher than in East ral-urban poverty split. Table II.2 shows that the average Asia, the Middle East, and North Africa region, where average number of children in rural poor households is greater than international poverty rates are estimated, respectively, at 24.5, in families located in Dar es Salaam and to a lesser extent 7.9, and 1.7 percent.23 However, most countries in these re- secondary cities. The number of children is also increasing gions also have much higher average per capita GDP levels. in the rural sectors and among the poor, while it is constant for other urban households and declining in Dar. The Characteristics of the Poor II.  As in many parts of SSA, fertility is very high among poor Geographic location matters—poverty is overwhelmingly families. The average Tanzanian woman is expected to give rural, with more than 80 percent of the poor and extreme birth to five to six children by the end of her lifetime, and poor Tanzanians living in rural areas (Figure II.3). Despite urbanization, over 70 percent of the Tanzanian population 23  Based on PovcalNet estimates for 2011. Poverty Profile 21 Box 2.1 Subjective Indicators of Deprivation While Tanzania’s poverty line appears relatively low by international standards, the level of basic needs poverty (28.2 percent) corresponds reasonably well to subjective indicators of deprivation. For instance, 35–32 percent of the population aged 15+ years in 2010/11 and 2012/13 classified themselves as poor or destitute. Furthermore, around 28–31 percent of the Tanzanian population reported having to rely on lower preference food during the preceding seven days in 2010/11 and 2012/13. Other indicators of food insecurity are somewhat lower. Of course, these subjective indicators measure something different than consumption-based poverty. Nonetheless the indicators suggest that the 2011/12 HBS poverty levels are somewhat in line with common perceptions about material deprivation, at least at the aggregate level. Figure B.2.1  Subjective Indicators of Deprivation Thinking about your current circumstances, In the past 7 days, did you ever ... would you describe yourself as...? 40 35 31 34 33 28 30 30 29 30 25 22 23 22 20 20 17 20 16 15 12 10 10 10 3 3 5 0 0 0 0 Rich Can manage Poor 2010/11 2012/13 2010/11 2012/13 2010/11 2012/13 to get by ... rely on ... limit the ...reduce number less preferred variety of of meals eaten 2010/11 2012/13 foods? foods eaten? in a day? Source: National Panel Survey (NPS) Notes: Food insecurity is population weighted. Subjective well-being is representative of the population 15+ years. Proportion of the Poor by Geographic Domain Figure II.3   2007 2011/12 3.3% 1.4% 11.7% 14.3% Rural Rural Other Urban Other Urban Dar es Salaam Dar es Salaam 85.0% 84.3% Source: HBS 2007 and 2011/12. 22 Tanzania Mainland Poverty Assessment Households’ Demographic Structure Table II.2   Rural Other urban Dar es Salaam Extreme poor Poor households 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 HH size 6.76 7.33 5.96 6.29 5.13 5.47 8.10 8.18 7.40 8.31 Depend. ratio 0.51 0.51 0.44 0.42 0.36 0.33 0.54 0.54 0.53 0.53 No child. <14 yrs 3.30 3.67 2.59 2.60 1.88 1.86 3.72 4.24 3.13 4.28 No adult wom. 3.45 3.64 3.07 3.27 2.61 2.74 4.23 4.09 3.85 4.19 No adult men 3.29 3.67 2.80 2.94 2.41 2.57 3.85 4.08 3.53 4.11 Head women 18.67 18.67 26.20 23.62 22.51 21.16 23.11 21.16 21.16 19.09 Age of head yrs 46.26 47.01 44.20 45.59 43.51 43.99 49.05 48.01 47.94 48.42 Source: HBS 2007 and 2011/12. Note: The dependency ratio is measured by the proportion of children below 14 years old and elderly above 65 years in the household. this number increases to over seven for women in the poor- regression analysis of the determinants of poverty. This may est segments of the population. be explained by the fact that there are two main catego- ries of women-headed households: (i) widows running their Families with many children have been less successful in own household business and mainly located in the rural reducing poverty over time. From 2007 to 2011/12, the pov- areas and (ii) single women working in the private sector erty headcount of families with 0–2 dependent children fell in the urban areas and capital city. The former suffer from by 26 to 33 percent. Families with 3–4 children experienced much higher levels of poverty as compared to the other a reduction in poverty by 19 percent, and families with groups. 5 children or more had the lowest (relative) reduction in the poverty headcount, of just 5 percent. Poverty is associated with lower levels of education of the household head. The head’s level of schooling is closely On the surface, households with younger heads seem to fare related to poverty incidence, suggesting that education is much better than those with older heads. Poverty is lower strongly linked to income-generating opportunities. The and decreasing faster among households with a head 30 incidence of poverty declines considerably among house- years old or younger (Figure II.4). However, this is largely due holds whose head has lower secondary education or above to the fact that young heads are generally better educated (Figure II.7). When one controls for the various sociodemo- and have only just started their family lives and so have few graphic effects in the regression model, education appears children. When one controls for other sociodemographic to be significantly positively associated with consumption, characteristics of the household in a multivariate model, the and the returns to education increase meaningfully with effect of head age on poverty vanishes. This indicates that higher levels of the head’s schooling. the age of the head does not significantly matter of living standards and poverty. (See figures II.5 and II.6 for more.) Education positively affects living standards and poverty re- duction directly, and also indirectly through its impact on Also, there is no significant relationship between the gender health gains, productivity, social integration, and so forth. of household head and economic welfare of the household. The proportion of households headed by women seems to In particular, secondary education appears to be the most be larger among the poor and extreme poor, though this closely associated with higher living standards in both rural appears to be declining over time. While one can think that and urban areas, while primary education seems less import- women-headed households fare worse than male head- ant and is not significant in urban sectors (see tables 2.B-1 ed ones, this effect could not be detected in a multiple and 2.B-2). Although primary education continues to be of Poverty Profile 23 Poverty by Demographic Figure II.4   Poverty Goes in Hand with Large Figure II.5   Structure (%) Family Sizes a. Family type a. Total fertility rate by wealth quintile, 2010 45 8 38.8 7.0 6.8 36.5 40 35.8 6.1 31.7 35 6 5.4 29.7 29.2 4.7 30 25 4 3.2 17.7 20 2 14.4 12.2 15 11.2 10 0 5 Poorest Q2 Q3 Q5 Least Total poor 0 Single Single Couple Couple Elderly parent parent without with kids family no kids with kids kids b. Poverty by number of children (0–14 years), 2011/12 50 45 b. Number of children 40 31 60 30 49.5 50 19 44.8 20 16 13 39.0 40 10 31.4 30 0 23.6 0 1 2 3–4 5+ 16.6 20 10 Poverty headcount Distribution of the poor 0 Source: HBS 2011/12; and DHS 2010. Two kids Between More than or less 3 and 4 kids ve kids c. Age of household head 42.1 45 Poverty Reduction by Number Figure II.6   37.6 37.2 40 of Children (0–14 years), 33.3 35 30.8 2007–2011/12 29.1 29.1 30 25.0 24.1 25 11.19 18.0 20 40% 15 10 20% 5 0 0% Less than 30–39 40–49 50–59 60 years old 30 years and above –9 –20% –19 –26 –28 2007 2011/12 –33 –40% 0 1 2 3–4 5+ Source: HBS 2007 and 2011/12. Note: Single parents with no kids are households composed of Poverty headcount 2007 (%) only adults over 14 years old, where the head is less than 65 years Poverty headcount 2011/12 (%) old and is either never married, divorced, separated, or widowed. Decline in poverty headcount (%) Elderly families are households whose head is aged 65 years old and above. Source: HBS 2007 and 2011/12. 24 Tanzania Mainland Poverty Assessment crucial importance for fighting against poverty, completing Interestingly, households that derive their income from non- primary school seems not enough anymore to move out of agricultural businesses appear to be experiencing a remark- poverty. able decline in poverty. The poverty rate for these house- hold groups dropped by around 9 percentage points (over A surprising result in Figure II.7 is the decline of poverty 30 percent) during the past few years. This suggests that the over time for households with no education. This might development of nonfarm employment can offer a pathway be explained by two facts. On a one hand, there is an in- out of poverty. This effect remains strong and highly signif- crease of ownership of large land plots, as the proportion of icant even after controlling for various other factors related households with no education owning land of more than to household well-being. The regression analysis shows that 5 acres increased from 37 to 47 percent between 2007 and employment in household nonfarm business is positively as- 2011/12. Given that over 70 percent of these households are sociated with greater levels of consumption and therefore engaged in agriculture, this increase helps them to improve negatively linked to poverty. This effect is much higher in the their living standards. On the other hand, the expansion of rural areas than in the urban zones (tables 2.B-1 and 2.B-2). aid and assistance to these household groups, as the pro- portion of families with no education who receive pensions, The regression results also indicate that agricultural employ- remittances, and other transfers, went up from 5 to 20 per- ment is positively correlated with the probability of being cent during the past five years. poor. However, households who own larger land plots and who are able to commercialize their outputs are less likely Wage employment in the private and public sectors is to be poor. These beneficial effects can be large enough to clearly associated with lower poverty for urban households. offset the disadvantages of being engaged in agriculture. Poverty rates are lowest among households headed by gov- Thus, only households engaged in subsistence farming with ernment employees or employees in the private sector and low land holdings suffer from high levels of poverty. NGOs (Figure II.8). The results in figure II.8a are in line with those in figure II.8b, showing that families with cash and There has been a movement out of agriculture between 2007 in-kind revenues from employment, as their main source of and 2011/12, as the proportion of households whose main income, are better off. Less than 20 percent of these families source of income is agricultural activity declined from around live in poverty. 53 percent in 2007 to 39 percent in 2011/12. Even poor households seem less likely to work in agriculture in 2011/12 than they were in 2007, as their proportion declined from 64 percent to 47 percent between the two periods. This seems to Poverty by Education Level of Figure II.7   have contributed to reducing the negative influence of work- the Household Head (%) ing in agriculture on living standards and poverty, probably due the fact that part of those who remained in the sector are 60 more productive and engaged more in cash crop production. 49.1 50 40.9 40.2 40 Households relying on transfers, remittances, and other in- 32.1 30.1 26.8 30 comes as main sources of revenues are experiencing a de- 20 terioration of their living standards (Figure II.8b). The poverty 7.0 7.2 10 rate increased by about 6 percentage points (over 20 per- Upper 3.7 secondary 1.7 0.7 1.0 0 cent) for these households. A rather surprising result is re- No education Less than primary Complete primary Lower secondary University lated to the marked decline of poverty for households with an inactive and unemployed head. This result is probably due to the fact that many of those classified as unemployed 2007 2011/12 work in the informal sector, but further analysis will be need- Source: HBS 2007 and 2011/12. ed to better understand this finding. Poverty Profile 25 Poverty by Sector of Work of Head and Sources of Income of the Household (%) Figure II.8   a. Employment sector of the household head b. Household sources of income 50 46.9 45 42.4 45 40 35.5 34.3 35.4 40 36.6 35 33.3 35 30 30.9 27.1 27.4 30 27.4 25 25 20 19.7 18.8 18.3 20 15.4 15 15 10 8.8 8.5 10 7.0 5 5 0 0 Government Private Self-employed Unemployed/ Cash & Non-agric. Agric. HH transfers/ Other sect & NGO in active in kind HH business business remittances income from empl. 2007 2011/12 Source: HBS 2007 and 2011/12. Recent internal migrants, who moved less than 15 years ago, Poor households tend to have much lower access to infra- are significantly less likely to be poor than nonmigrants. Un- structure than nonpoor ones. Here again data availability is fortunately data on the place of birth and migration status limited to HBS 2011/12, and it reveals that poor households are not available in HBS 2007, so the analysis is limited to tend to have much lower access to private piped water, the 2011/12 survey. It appears from Figure II.9 that poverty electricity and tarmac roads (Figure II.10). The obstacles to is much more prevalent among nonmigrant households. infrastructure and services, particularly electricity and roads, Households whose head migrated less than 15 years but seriously limit the possibilities of the poor to improve their more than five years ago fare the best followed by the very living standards. Table 2.A-2 in appendix 2.A shows that elec- recent migrants. This is in line with most of the literature on tricity access has a very strong income gradient—varying migration, including that on Tanzania, which reveals strong positive linkages between geographic and economic mobili- ty. The next section will explore these linkages more in detail. Figure II.10  Access to Public Infrastructure (%) 50 46.0 45 Figure II.9  Poverty by Migration Status (%) 40 41.3 35 35 30 26.4 32.1 25 30 21.1 20 19.2 20.7 17.0 16.0 25 24.1 15 20 10 6.9 5 2.9 15 14.4 12.8 0 Private Public Public grid Trunkroad Tarmac 10 connection tap (TANES CO) road 5 Access to Access to Access to piped water electricity road infrastrcture 0 (in community) Non Recent Migrant betw. Old migrant migrant migrant 5 and 15 years > 15 years Poor Non-poor < 5 years 2011/12 Source: HBS 2011/12. Note: Connection to piped water is measured for dry season and Source: HBS 2011/12. private connection stands for connection inside and outside house. 26 Tanzania Mainland Poverty Assessment from below 3 percent for the poorest quintile to 42 percent Migration by Gender and Period Table II.3   among the least poor quintile. Total (%) Men (%) Women (%) Connectivity to other soft infrastructure is also found to Lifetime migrant 40.72 43.06 38.94 significantly increase consumption and reduce the risk of Long-run migrant 25.44 22.52 28.12 poverty. Tables 2.B-1 and 2.B-2 show that the presence in Recent migrant 8.44 8.31 8.56 the household’s community of a daily market and mobile — Intraregional 4.45 4.29 4.60 phone signal impact positively on the consumption levels — Interregional 3.99 4.02 3.96 and reduce the probability of poverty. Access to these ser- Source: National Panel Survey, (NPS3 2012/13). vices is still quite limited in rural areas, hampering local op- Note: Recent migrants are individuals that live in a different district in NPS2012 than in NPS2008. Long-run migrants are individuals portunities to reduce poverty. that have migrated to the current district in the past 10 years. Lifetime migrants are individuals that live in a different district than their district of birth. III.  Migration and Poverty The previous section shows that poverty is less prevalent Migration Flows among migrant households, suggesting a potential pos- Lifetime migrants represent about 40 percent of the pop- itive association between internal migration on poverty ulation, and around 25 percent have migrated to their reduction. This evidence is supported by previous studies current district during the past 10 years. Table II.3 reports on internal migration and poverty in Tanzania. However, the the distribution of migrants by gender and according to available literature has mainly focused on the Kagera region the duration of residence in the current location. It shows and might not be conclusive about the benefits of migra- a higher proportion of lifetime male migrants compared tion for poverty reduction at the national level. 24 to lifetime female migrants (see the note below table II.3 for definitions). However, women appear more likely to be This section explores internal migration in depth in all of Tanzania and its economic effects. The analysis relies on 24  See McKenzie et al. (2010), Lokshin et al. (2010), and Gibson et the three available waves of the National Panel Surveys— al. (2011), among others. For the studies on Tanzania, see Beegle NPS1, NPS2, and NPS3—and proceeds in two parts.25 The et al. (2011), who investigated the relation between migration and first investigates the features of migrants and examines the economic mobility. DeWeerdt and Hirvonen (2013) analyzed the determinants of migration. The second explores the impact insurance mechanism in place between migrants and their house- of migration on the living standards of migrants and their holds of origin, and Dimova et al. (2011) explored the relationship families left behind and analyses the relationship between between emigration and child labor. These three studies focused migration, poverty, and remittance flows. on the Kagera region. 25 NPS1, NPS2, and NPS3 stand for National Panel Surveys for 2008/09; 2010/11, and 2012/13, respectively. Appendix 2.C pro- Migrant Features and Determinants A.  vides details on the surveys main characteristics. of Migration 26  See de Weerdt and Hirvonen (2013). Over the past 20 years, Tanzania has experienced great inter- 27  According to the 2002 census, stayers, i.e., individuals who re- nal migration, with half of its population migrating over the side in their region of birth, account for 83 percent of the popu- last two decades.26 According to the latest available migra- lation in Tanzania, with substantial differences between the geo- tion data from the 2002 census, 6.2 percent of the popula- graphic regions. The proportion of individuals born in Zanzibar and still resident at the time of the 2002 census is around 72 percent. tion in Dar es Salaam consists of recent migrants who moved Dar es Salaam is the focal point of attraction for lifetime migrants, between regions in 2001 and 2002 (NBS 2006). The regions as about 49 percent of the population residing in Dar es Salaam of Pwani, Manyara, and Mwanza are the other main desti- in 2002 was born in another region, while about 1.4 percent was nations of recent migrants, while Dodoma is the greatest born in another country (NBS 2006). A study by Muzzini and Linde- sending region, with an out-migration rate of 12.6 percent.27 boom (2008) also shows that migrants’ turnover is very high. Poverty Profile 27 long-run migrants than men. This difference might be re- Table II.4  Characteristics of the Migrants lated to the variation of the reasons of migration between men and women, which will be explored later in the Nonlifetime Lifetime a. migrants migrants Difference chapter. The recent migrants, who moved between the Age 37.94 36.76 ** two waves of the survey, represent only 8 percent of the Female 56.33% 47.32% *** population, suggesting that people migrate over a longer Married/living together 67.11% 60.29% * period of time. They seem more willing to migrate within Literate 72.61% 84.00% *** the same region. Attending school 3.78% 4.76% Dar es Salaam is a popular destination for migrants from Labor active 95.04% 95.03% Pwani, Lindi, Mtwara, and Morogoro. Tables 2.C-2 and 2.C- Work for pay 67.34% 78.16% *** 3 in Appendix 2.C represent the interregional migration Self-employed 15.98% 27.26% *** movements. Dar es Salaam is the most attractive migra- Non-long-run Long-run tion destination, and Mwanza also receives an important b. migrants migrants Difference flow of migrants, particularly from Kagera, Shinyanga, and Age 35.99% 30.46% *** Mara. There is also a high degree of migration within the Female 47.66% 51.99% *** regions in Zanzibar. These patterns are broadly consistent Married/living together 56.04% 58.39% *** with the patterns noted in the 2002 census and indicate Literate 74.41% 82.39% *** migration movements mainly toward the major urban Attending school 8.71% 5.79% *** areas Labor active 89.22% 93.85% *** Work for pay 59.20% 70.17% *** Characteristics of Migrants Self-employed 12.24% 22.92% *** Long-run migrants tend to be significantly younger than Nonrecent Recent non-long-run migrants, they are usually married, and about c. migrants migrants Difference 84 percent are literate. Table II.4 presents the characteristics Age 35.94% 32.44% of migrants by length of the migration experience. Lifetime Female 49.59% 44.94% migrants are slightly younger than nonlifetime migrants Married/living together 57.82% 47.53% and are more likely single. Other differences between life- Literate 75.90% 81.60% time migrants and nonlifetime migrants emerge when look- Attending school 8.43% 5.52% *** ing at the type of activity. Lifetime migrants are more likely Labor active 89.77% 93.75% *** to work for pay and to work as self-employed. Work for pay 62.01% 72.03% *** Self-employed 14.65% 15.94% There is no significant difference between long run and Source: National Panel Survey (NPS3 2012/13). non-long-run migrants in terms of labor activity, however, ***, **, and * denote significance at the 1%, 5%, and 10% level, differences emerge when considering the type of activity. respectively. In line with previous results, a larger percentage of long run migrants works for pay or is self-employed. Lifetime and long-run migrants are mainly motivated by Finally, panel c of Table II.4 presents the results for recent mi- better housing and services conditions as well as marriage grants and shows the difference between recent migrants and family, while recent migrants are motivated by work and nonrecent migrants only in school attendance and la- related reasons. Table II.5 explores the reasons for moving bor activity. The percentage of recent migrants attending for each migration group and shows different motivations school is lower than nonrecent migrants, while the fraction across the migrants’ categories. Stark differences appear be- of recent migrants who are labor active and who work for tween men and women migrants. Men seem more likely to pay is larger than nonrecent migrants. migrate for work reasons and better housing and services 28 Tanzania Mainland Poverty Assessment Table II.5  Reasons for Migrating Relationship to the Head of the Table II.6   Household Total (%) Men (%) Women (%) Lifetime migration Lifetime Long run Recent migrants migrants migrants Work related 10.53 14.83 6.92 Head of household 41.81 32.79 32.44 Studies related 2.49 2.39 2.58 Spouse 26.27 26.3 17.51 Marriage 14.65 0.82 26.27 Son/Daughter 12.9 15.83 34.84 Other family reasons 33.76 34.50 33.14 Step Son/Daughter 0.86 1.28 1.38 Better housing/services 31.76 37.92 26.59 Grandchild 2.52 3.21 4.37 Land/plot 3.80 5.59 2.30 Parent 0.88 0.80 0.95 Other 3.00 3.95 2.21 Other relative 11.2 5.01 6.95 Long-run migration Domestic servant 2.17 2.60 0.42 Work related 8.67 12,82 5.62 Other nonrelative 1.40 2.19 1.14 Studies related 2.68 2.83 2.58 Source: National Panel Survey (NPS3 2012/13). Marriage 18.54 1.11 31.33 Other family reasons 32.21 32.52 31.98 Better housing/services 30.99 40.96 23.68 Determinants of Migration Land/plot 3.47 5.60 1.90 The analysis of the determinants of migration is based on Other 3.44 4.16 2.91 a multinomial logit model. The decision of individuals to Recent migration migrate between NPS1 and NPS3 is categorized into three Work related 14.03 19.65 8.89 types: (i) those stay in the same district in both waves, Studies related 2.15 1.62 2.64 (ii) those who move to a different district within the same Marriage 11.64 0.31 22.01 region, and (iii) those who move to a different region within Other family reasons 28.76 27.33 30.06 the country. The migration decisions are examined against Better housing/services 35.08 40.82 29.82 the individual’s characteristics, the household’s characteris- Land/plot 4.43 5.23 3.71 tics, and the distance between the household’s residence Other 3.90 5.04 2.86 and the district headquarters. Source: National Panel Survey (NPS3 2012/13). Individuals who are less than 30 years old and have high- er education are much more likely to migrate to a different region than others. The head, or the spouse of the house- conditions, while women seem to be more motivated by hold head, appears much less likely to migrate compared marriage and family reasons. Less than 10 percent of wom- to the other household members (Table 2.C-4). The level en seem to migrate for work-related motives. of household assets, measured using Principal Component Analysis of all household assets, is strongly correlated with Lifetime and long-run migrants are mainly heads of house- interregional migration. This suggests that the availability of holds or spouses of the head of household, while recent mi- resources to finance migration strongly affects the migra- grants are predominantly their sons or daughters.28 Table II.6 tion decision. Larger households are associated with a lower shows different patterns in the length of migration experi- probability of migration, possibly because of social networks ence according to the nature of the relationship to the head. Households heads and spouses who are looking for better living conditions are those who migrate over long periods 28 The sons and daughters of all household heads (either migrant of time, while the sons and daughters who are looking for or not migrants) are predominantly the recent migrants, indicat- better job opportunities are those who migrate for shorter ing that recent migration is happening essentially among younger periods of time. cohorts. Poverty Profile 29 Differences between Migrants and Nonmigrants before and after Migration Table II.7   Prior to migration (in NPS1) After migration (in NPS3) Nonmigrants Migrants Difference Nonmigrants Migrants Difference Asset index –0.76 –0.03 0.74*** –0.63 0.30 0.93*** Per capita consumption (log) 3.56 3.38 –0.18 3.89 4.49 0.60*** Per capita income (log) a 3.24 3.14 –0.10 HH size 5.73 6.42 0.70*** 5.41 5.55 0.14 HH members <14 years 2.56 2.72 0.16 2.40 2.30 –0.09*** HH head male 0.76 0.76 0.00 0.75 0.76 0.01 HH head literate 0.76 0.81 0.05*** 0.75 0.82 0.07*** Sources: NPS1 2008/09; and NPS3 2012/13. * p < .10, ** p < .05, *** p < .01. a The income data was taken from the RIGA database, http://www.fao.org/economic/riga. that tie individuals to the local community. The remoteness Asset Differences between Table II.8   of the household location also appears to affect the house- Migrants and Nonmigrants before hold decision to migrate. Individuals who live further away and after Migration from the district headquarters are more likely to migrate to a Asset Intraregional Interregional different region instead of moving to another district in the quintile Nonmigrant migrant migrant Total same region or not moving at all. 1 85.75 7.24 7.00 100.00 2 86.33 5.15 8.52 100.00 The household consumption level seems to increase signifi- 3 87.32 5.25 7.43 100.00 cantly after migration. Table II.7 compares the characteristics 4 79.01 9.13 11.86 100.00 of households before and after they migrate and shows that 5 72.04 11.18 16.78 100.00 migrants tend to have slightly lower consumption and in- Total 83.34 7.17 9.49 100.00 come levels than nonmigrants before migration. However, Sources: NPS1 2008/09; and NPS3 2012/13. migrants appear to have significantly higher consumption levels than nonmigrants after migration, suggesting a posi- tive impact of migration on living standards. The asset own- after migration. It appears that only 40 percent of recent ership seems also to increase significantly after migration migrants work in the agricultural sector, against 62 percent (Table II.8).29 of the nonmigrant population. Migrants were less likely to work in the agricultural sector even before the migration The tabulation of the probability of migration by asset quin- episode. Indeed, about 54 percent of individuals who mi- tiles indicates individuals from households in higher quin- grated between NPS1 and NPS3 report that they worked in tiles are more likely to migrate outside their initial region. agriculture before migrating, while more than 70 percent B.  Economic Impact of Migration There is a movement outside agriculture for migrants. Ta- 29  Prior to migration, migrants seem more likely to live in larger households than nonmigrants. However, the multinomial regres- ble II.9 provides an overview of the occupational choices sion model shows a negative effect of family size on migration de- of recent migrants with respect to nonrecent migrants and cision. This might be explained by the correlation between house- compares their characteristics before and after migration. hold size, the age structure, or assets ownership of the household. The analysis focuses on recent migrants, ages 18 years and When we control for these factors, living in a large households above, and examines their occupational choices before and does not seem anymore to encourage migration. 30 Tanzania Mainland Poverty Assessment Table II.9  Migrant Occupations Prior to migration (in NPS1) After migration (in NPS3) Nonmigrants Migrants Difference Nonmigrants Migrants Difference Agriculture/livestock 71.06% 53.73% *** 62.52% 40.83% *** Family work 4.40% 8.85% *** 5.76% 7.84% ** Private enterprise 3.65% 6.20% ** 6.51% 18.72% *** Self-employed 9.93% 11.67% 11.09% 13.75% Government/parastatal 2.23% 2.52% 2.33% 6.22% *** Student 3.65% 12.52% *** 5.07% 6.87% No job/job seeker 2.16% 2.33% 3.33% 3.18% Other 2.91% 2.19% 3.38% 2.60% Source: NPS3 2012/13. of nonmigrants were involved in the agricultural sector ac- The role of migration in improving living standards in Tanza- cording to the NPS1 data. Working in a private enterprise nia should, not, however be overestimated, as only 8.4 per- appears to be more prevalent among migrants, especially cent of the total population migrated between NPS1 and after migration. Interestingly, migrants are more likely than NPS3. nonmigrants to be classified as students before the migra- tion experience. Finally, migrants are more likely to be work- Remittances ing for the government after migrating. Migrants may improve not only their own welfare but also that of their former households through remittances-in the Migration contributes to a significant increase of the con- form of cash and in-kind transfers. The magnitude of do- sumption level. The (weighted) descriptive statistics in Ta- mestic remittances in Tanzania is neither well known nor ble II.7 reveal that while the average Tanzanian experienced easy to trace, as most remittances flow through semiformal 39.7 percent growth in real consumption between NPS1 and informal channels. Approximately 23 percent of Tan- and NPS3, a typical migrant experienced a 57.3 percent real zanian households report the receipt of remittances in the consumption growth over the same period. However, this 12 months prior to the NPS3.30 Most of these remittances result might be due to the difference of unobservable char- originate from major urban areas in Tanzania such as Dar- acteristics between migrants and nonmigrants, as better es-Salaam, Mwanza, and Arusha (Figure II.11), and only two connections to social networks may help them to find good percent of households report to have received remittances jobs or raise motivations and abilities. The effect of migra- from outside of Tanzania.31 tion on consumption growth is thus further explored using the regression model presented in Appendix 2.C. The results indicate that migrants observe a 21.2 percentage point higher consumption growth in their consumption levels 30 The addition of several questions on remittances in the third than nonmigrants, suggesting that moving to a different wave of NPS greatly contributes to our understanding of the na- district could lead to a significant increase in consumption ture and magnitude of remittances in Tanzania. 31  According to official statistics, Tanzania received $67.3 million in growth even during a relatively short time-period of about four years (see Table 2.C-5). This result holds for different es- international remittances from Tanzanians living abroad. See http:// data.worldbank.org/indicator/BX.TRF.PWKR.CD.DT. International re- timation models. Even after controlling for the endogeneity mittances are mainly received from Rwanda, Uganda, Kenya, UK, resulting from unobservable individual characteristics, there Canada, and the United States. See 2013 information on destina- is significant evidence of the positive impact of migration tions and the migrant stock in http://esa.un.org/unmigration/TIM- on the improvement of consumption. SO2013/migrantstocks2013.htm. Poverty Profile 31 Major Sources of Remittances Figure II.11   Money transfers are increasingly made through mobile Received by Households channels. Approximately 36 percent of all transfers are made through mobile money transfer services such as M-Pesa, 30% 27% Tigo Pesa, EZY Pesa, or Airtel Money (Figure II.12a). That said, Percent of all remittances 25% transfers through more traditional channels such as friends 20% and relatives are just as prevalent as mobile money transfers, 15% while formal channels through the banks, Western Union, or post office services account for a much smaller share of all 10% 7% 7% 5% 5% 5% 4% 4% transfers. 5% 2% 0% The primary use of remittances is for household consump- Dar es Salaam Mwanza Arusha Shinyanga Tanga Mtwara Kilimanjaro Mbeya Foreign country tion, followed by spending on education and health. As apparent from Figure II.12b, only few households use re- mittances to invest in business or agriculture, although the Source: NPS3 2012/13. reason behind this may be the fact that the average amount of remittances received represents a fairly small share of the annual household consumption—they make up only 9 per- Domestic remittances usually come in small amounts. In cent of the annual consumption expenditures for the 23 contrast, households who receive large amounts of remit- percent of all households that received remittances. tances in cash often receive them from a foreign country. Around 50 percent of the households received on average Remittances are strongly correlated with school attendance $67 of domestic remittances during the previous 12 months, for households, suggesting a potentially important channel and 41  percent received less than $50 (see Table  II.10). The through which the benefits of migration accrue to recipient amounts, in cash and in-kind, are evaluated on average to households. Children living in households that received re- around $207, which represent about nine percent of the mittances are 20 to 23 percentage points more likely to at- value of total annual consumption for a typical Tanzanian tend school.32 Moreover, it appears that a 1 percent increase household. in remittances is associated with a greater probability of school attendance by up to 1.7 percentage points (see Table 2.C-6 in appendix 2.C). While remittances have a positive ef- Table II.10  Amount of Domestic Remittances fect on the probability of school attendance, an inverse rela- Received by Households tionship is observed with the impact of migration on school (previous 12 months) attendance. This may be due to the financial costs of migra- Amount Share of households tion or the disruption imposed by migration. These results re- $0–$50 41.3% main consistent after addressing the potential selection bias $51–$100 19.6% using the Propensity Score Matching approach (Table 2.C-7). $101–$500 30.3% $501–$1000 4.7% 32 The impact of remittances on poverty and human capital has $1001+ 4.2% been the subject of a large number of studies, although most of these focused on international remittances. See Adams and Cuec- Total 100% uecha (2013), McKenzie (2005), Yang (2008), de Brauw et al. (2013), Source: NPS3 1012/13. and Lokhshin et al. (2010), among others. Amuedo-Dorantes and Note: The estimates are for the 22.87% of households that Pozo (2010) explored the effects of remittance receipt on children’s reported to have received remittances during the previous 12 school attendance and found a strong positive effect particular- months. The amounts are converted from Tanzanian shillings to U.S. dollars using the official exchange rate for 2013 of 1600.44 ly on the attendance of girls, secondary school-age children, and shillings per dollar. younger siblings. Following these authors, we examine the impact See: http://data.worldbank.org/indicator/PA.NUS.FCRF. of remittances on the school attendance of Tanzanian children. 32 Tanzania Mainland Poverty Assessment Channel and Primary Use of Remittances Figure II.12   a. Channel of remittances b. Primary use of remittances Other Other Mobile money 6% 25% Health transfer 6% 36% Post o ce Education 0% 7% Wetern Union Household 1% consumption 81% Bank account 2% Friends/relatives 36% Source: NPS3 2012/13. Sustainability of Migration The positive effects of migration need, however, to be bal- Migration to larger cities may not necessarily be sustainable anced against the consequences of excessive migration. due to the pressure on receiving households. In a study of in- While it appears from this analysis that migration contrib- ternal migration in Tanzania, Muzzini and Lindeboom (2008) utes to improving the living standards of migrant house- highlight the issue of overcrowding for migrant-receiving holds and their families left behind, these results need to households. Using the 2002 census data, they show that be interpreted in light of the fact that internal migration around 47 percent of migrant receiving households have remains relatively low in Tanzania and is limited to individ- more than two people per room compared to 39 percent in uals from households with higher education and better households without migrants. However, this analysis did not living standards. Even though self-selection effects have capture any overcrowding effect, as the percentage of house- been ruled out in the present analysis, the expansion of holds with more than two people per room is around 37 per- migration to include people from less educated and pros- cent for both migrant and nonmigrant receiving households. perous households as well as the increase of migration flow in urban centers may significantly reduce the beneficial ef- Migrants do not seem to face great obstacles in access to fects found here. Migration can be among the solutions to health care services or higher health care costs. The process address poverty, but excessive migration may worsen the of rapid urbanization might reduce the possibility for mi- problems of city congestion and unemployment, causing a grants to access health care facilities or might impose high- displacement of poverty to the urban zones. Besides migra- er health care costs to migrants compared to nonmigrants. tion, other solutions such as rural diversification and non- However, the analysis does not show significant differences farm development remain needed.33 between the two groups, except the fact that lifetime mi- grants and recent migrants seem more likely to consult a health care provider relative to nonmigrants (see Table 2.C- 8 in Appendix 2.C). Also, lifetime migrants seem to spend, on average, a slightly larger amount on illnesses and injuries than the other groups, which might be due to the fact that 33 See Christiaensen et al. (2013) for a good discussion of these al- they are older. ternative solutions in Tanzania. Poverty Profile 33 34 Tanzania Mainland Poverty Assessment Chapter  3 Economic Growth and Poverty Key Messages ➤➤ During the second half of last the decade, poverty has become more responsive to economic growth ➤➤ The increase of poor households’ living standards is driven mainly by improvements of their endowments in assets and education. Given the returns to these endowments, this has raised the earnings of the poorest.. Poverty is falling and living conditions rising in Tanzania. In December 2014, Tanzania released a new GDP series with However, given Tanzania’s strong economic performance in a base year of 2007, rather than 2001. The revised numbers recent years, the pace of poverty reduction is not as fast as use new and improved data sources to update the nation- might be expected. al accounts series and make it a more accurate reflection of the economy. The new series sees an upward revision of To understand this dynamic, this chapter examines the in- 27.8 percent in the base year 2007. teraction between growth and poverty in Tanzania. It first provides a brief overview of recent economic growth in the The revised GDP figures suggest that Tanzania’s growth country. Second it examines the response of poverty re- has been robust over the past decade. From 2008 to 2013, duction to economic growth and investigates to extent to growth averaged 6.3 percent, but when adjusted by the size which the poor have benefited from growth. of the population this rate drops modestly to 3.5 percent. The new figures show a degree of volatility not seen in the previous series (Figure III.1). The increased volatility for the A Brief Review of Recent I.  most part captures the variations in performance of the ag- Economic Growth ricultural sector, due in part to variable climatic conditions, and possibly selling prices influenced by regional and global At the theoretical level, the relationship between macro- markets. The increased volatility also reflects improved data economic growth and the elimination of poverty is key. A collection across sectors, including industries and services. sound macroeconomic foundation is necessary, but not Extrapolation methods and assumptions were frequently sufficient, to achieve a higher level of per capita income. used in the old series. This section therefore looks at recent economic growth trends. 35 Tanzanian GDP Growth Rate Figure III.1   With the exception of construction, these capital intensive (base year 2007) sectors create limited jobs. In contrast, the rate of growth of the labor-intensive agriculture sector, which employs 9% three-quarters of the workforce, remained far lower than 8% average growth at only 4.2 percent from 2008–13. The ag- 7% riculture sector has continued to underperform compared 6% to the rest of the economy. That said, cash crops, including 5% coffee, tea, cotton, cashews, sisal, and cloves account for a 4% 3% significant proportion of export earnings, and agriculture’s 2% share of current GDP has increased from 27 percent in 1% 2007 to 32 percent in 2013. However, while the volume of 0% major crops has increased in recent years, large amounts of 2008 2009 2010 2011 2012 2013 GDP growth GDP per capita growth Sectoral Real Growth Rates in Figure III.3   Source: National Bureau of Statistics, 2014; World Bank, 2014. Tanzania 14% The level of growth achieved in the past three years is sig- 12% nificantly higher in Tanzania than that achieved by neigh- 10% boring Uganda and Kenya (Figure III.2). 8% 6% The main drivers of Tanzania’s rapid economic growth are 4% 2% a small number of fast growing, relatively capital intensive 0% sectors. Over the period 2008–13, construction, commu- 2008 2009 2010 2011 2012 2013 nications, and financial services all saw a growth rate of Agriculture & shing Industry & construction over 10 percent (see figures III.3 and III.4 and Table III.1). Services Total Source: National Bureau of Statistics, 2014. Comparison of Growth Rates Figure III.2   across Countries Sectoral Composition of Growth Figure III.4   in Tanzania (current market 9% share of GDP) 8% Annual GDP growth rate 7% 100% 6% 90% 5% 80% 70% 4% 60% 3% 50% 2% 40% 30% 1% 20% 0% 10% 2010 2011 2012 2013 0% 2008 2009 2010 2011 2012 2013 Tanzania Kenya Uganda Agriculture and shing Industry and construction Source: World Bank, 2014. Services Other Note: Kenya and Tanzania are rebased figures. Uganda figures are not rebased. Source: National Bureau of Statistics, 2014. 36 Tanzania Mainland Poverty Assessment Real GDP Growth in Tanzania by Sector, 2008–2013 Table III.1   Economic Activity 2008 2009 2010 2011 2012 2013 Agriculture and Fishing 7.5% 5.1% 2.7% 3.5% 3.0% 3.4% Crops 7.8% 5.5% 3.7% 4.8% 3.6% 3.9% Livestock 8.0% 5.3% 1.4% 1.7% 1.9% 2.0% Forestry and Hunting 3.8% 5.0% 3.4% 3.4% 3.5% 4.7% Fishing 7.1% 0.6% 1.0% 3.5% 2.9% 5.3% Industry and construction 6.6% 3.4% 9.0% 12.2% 4.1% 11.2% Mining and quarrying –9.5% 18.6% 7.2% 6.4% 6.7% 3.8% Manufacturing 11.0% 4.7% 8.8% 6.9% 4.1% 6.6% Electricity and water 5.1% 4.4% 7.7% –2.7% 3.1% 8.0% Electricity 8.0% 4.2% 13.2% –4.0% 3.4% 12.9% Water 2.3% 4.5% 2.2% –1.1% 2.8% 2.7% Construction 10.1% –3.5% 10.2% 23.1% 3.4% 18.4% Services 4.2% 5.8% 7.7% 8.4% 7.2% 6.2% Trade, hotels and restaurants 5.6% 2.4% 8.7% 10.4% 3.9% 5.1% Trade and repairs 6.3% 2.7% 9.6% 11.3% 3.6% 5.3% Hotels and restaurants 2.1% 1.0% 3.5% 5.6% 6.0% 3.6% Transport and communication 4.8% 12.6% 15.4% 6.1% 11.0% 8.6% Transport 1.8% 7.1% 10.8% 4.6% 4.4% 5.7% Communications 12.7% 25.4% 24.4% 8.7% 22.0% 12.7% Financial intermediation 17.5% 18.6% 12.7% 14.6% 5.2% 2.9% Real estate and business services 4.1% 3.5% 8.2% 3.4% 6.4% 5.9% Public administration –5.5% 0.1% –4.8% 15.2% 9.2% 7.7% Education 9.4% 9.0% 6.3% 5.6% 7.3% 4.2% Health 5.4% 7.3% 3.3% 5.3% 11.2% 8.7% Other social and personal services 5.1% 4.6% 5.6% 5.9% 6.5% 5.7% FISIM 6.8% 20.1% 8.0% 22.4% 1.3% 0.3% Net taxes 4.5% 12.8% 3.7% 12.1% 1.5% 15.0% Total GDP 5.5% 5.4% 6.3% 8.0% 5.1% 7.3% Source: National Bureau of Statistics, 2014; World Bank National Accounts data, 2014. produce never reach the market. Poor pricing and unreli- enough. Productivity growth and increased diversification able cash flow to farmers continue to inhibit growth in the of the economy will be central to enhancing the expansion agricultural sector. of private firms and their capacity to create productive jobs to improve incomes and reduce poverty. The nature of growth in Tanzania has not created sufficient productive employment for the rapidly growing popula- The new GDP data also highlights some weaknesses in the tion. The capital intensive bias of growth has meant it has Tanzanian economy. Although the revised GDP figures sug- absorbed only a handful of the 700,000 additional workers gest that Tanzania is drawing closer to the $1,045 income who enter the domestic labor market every year. The shift per capita threshold for middle income countries, reaching in labor toward more productive sectors has not been fast $970 in 2013, the level of trade openness at 47.7 percent in Economic Growth and Poverty 37 2013 is well below the regional average of 65.9 percent, and Despite its apparent stability, there are a number of risks Tanzania’s tax revenue to GDP ratio is alarmingly low at 11.5 to Tanzania’s sound macroeconomic position that need to percent in 2013. be considered and carefully managed when developing strategies for poverty reduction. These include the fiscal However, albeit from a low base, Tanzania’s economy has position, including spending pressures from the ongoing become more open, with increasing diversification toward Big Results Now initiative and the upcoming general elec- new products and markets during the past five years. Since tions, the level of government debt, the need for higher 2005, the value of exports of goods and services has nearly tax revenues, continued financial risks in the energy sector doubled. Although the export structure remains largely de- in part due to exogenous factors, the rising level of arrears, pendent on volatile primary commodities such as minerals and fluctuations in commodity prices which will affect the (gold), coffee, tea, cashew, and cotton, the recent surge in trade balance. manufactured exports to the East African Countries (EAC) and the Southern African Development Community has The magnitude and timing of anticipated FDI inflows to the been a notable and welcome development. During the natural gas sectors will also impact the local economy, es- same period, imports tripled, leading to a growing trade pecially in the geographical areas where those investments deficit over time. The current account deficit has been fi- will take place. The new investments are expected to be in nanced by official aid and growing foreign direct invest- the range of US$4 billion to US$5 billion per year. Even if ment (FDI) inflows into the natural resources sector. the majority of these funds are used to purchase imported goods, as is likely, their magnitude will modify the current Inflation has moderated as the Central Bank has followed a equilibrium in the domestic financial markets and possibly tight monetary policy over the past two years. As a result, have an impact on exchange rates. These potential impacts together with falling domestic food prices, inflation reached will have to be carefully managed by the authorities. If man- 6 percent in 2014—a significant achievement, compared aged well, they have the potential to transform Tanzania’s to the 19 percent of December 2011. This is good news for economic future and increase the opportunity for poverty consumers and nonindexed wage workers. The decline has reduction. However, the large scale exploitation of these re- also helped to moderate the appreciation of the real ex- sources is unlikely to begin for at least 7–10 years. It is only change rate, which is positive for exporters, particularly giv- after this point that significant revenues will be generated en the recent fall in commodity prices. However, the mea- from this source. sures implemented to achieve lower inflation have resulted in the cost of credit stabilizing at a higher level, imposing To summarize, Tanzania has made significant econom- increased burdens on borrowers. ic progress, and the macroeconomic position is largely sound. That said, many poor households have not benefit- Fiscal policy provides greater cause for alarm. On the surface, ted from the recent growth, which has largely been driven the government has demonstrated a strong fiscal commit- by non-labor-intensive sectors. The task of broadening the ment, reducing the overall fiscal deficit from 4.2 percent of growth base is key to translate exceptional growth into GDP in 2012/13 to 3.2 percent in 2013/14, below the initial poverty reduction. There are also a number of risks on the target set in the approved budget. This is a significant im- horizon. With this in mind, the government must closely provement after the unexpected slippage in 2012/13. How- monitor its fiscal stance, keep debt service at a reasonable ever, this figure fails to account for a rapidly accelerating level, and determine how best to clear and manage ar- increase in arrears estimated to be about 1 percent of GDP rears. This is essential to ensure continuation of a sound from 2012/13 to 2013/14. Not only is the fiscal deficit argu- macroeconomic base on which to build poverty reduc- ably larger than reported, but fiscal management was not tion. The government should prepare for the influx of gas smooth in 2013/14 and substantial mid-year fiscal adjust- revenues and determine clear structures to manage these ments were required. Deterioration in fiscal management is and maximize their impact for service provision and the a cause for concern. poorest. 38 Tanzania Mainland Poverty Assessment II. The Growth Elasticity of Poverty 1 percent increase in economic growth will reduce poverty headcount by only 1 percent. The previous section reviewed recent macroeconomic trends; this section proceeds to analyze the poverty-growth The difference between the estimates of the growth elas- relationship over the period 2007–2111/12. It starts by esti- ticity of poverty with respect to the measures of economic mating the growth elasticity of poverty reduction. This elas- growth is quite common in developing countries, though ticity measures the percent change in poverty with respect the discrepancy appears larger in Tanzania. A large litera- to a 1 percent change in GDP (or consumption) per capita ture has discussed inconsistencies between national ac- and is a well-known concept for exploring the responsive- counts and household surveys data, for developing coun- ness of poverty to economic growth. tries in general (Ravallion 2001; Adams 2004) and for SSA in particular (Deaton 2005; Christiaensen and Devarajan The response of poverty to economic growth has domi- 2013), and show strengths and weaknesses to both. The nated the recent literature on Tanzania’s economic perfor- discrepancies can be related to variations in the definition mance, with the general conclusion that growth and pover- of consumption in national accounts versus household ty are in many respects “delinked”.34 survey data, inflation adjustment, omission and measure- ment errors, and so forth. While there is no clear consensus There are two broad approaches to measure economic on which of these measures of economic growth is more growth and estimate the growth elasticity of poverty in this accurate, it seems that growth measured from survey data context: one based on national accounts or one based on is more closely related with changes in households’ con- household survey data. The first measures growth as chang- sumption and income and better reflect the spending be- es in GDP per capita in the national accounts. The second havior of the poor. approach is to measure growth directly from the household surveys on which the poverty estimates are based, that is, In Tanzania, the variation of the relationship between pov- as growth in average household consumption per capita. erty and growth for different measures of growth is partly Growth rates estimated from these two sources can dif- related to the price deflators of the two series. Growth in fer significantly, which has implications for the estimated nominal GDP per capita is approximately 20 percent per elasticities. annum for 2007–12, which corresponds reasonably well to the growth of nominal household consumption per capita Economic growth measured by changes in HBS consump- from HBS, estimated at around 19.7 percent per annum. It is tion per capita appears much lower than growth in GDP. only in real terms that the two sources diverge, which mir- Real GDP per capita grew at an average annual rate of 3.6 rors differences in the underlying deflators. In particular, the percent over the period 2007 to 2012. Conversely, household GDP deflator shows inflation of 70 percent over the period consumption per capita in the HBS increased at only an aver- 2007–12, a lower rate than the HBS internal deflator calcu- age annual rate of 0.9 percent between 2007 and 2011/12. lated using unit values, which suggests an increase of prices by approximately 90 percent during the same period. These Economic growth does reduce poverty in Tanzania, but discrepancies in inflation trends between the two series in- the rate of poverty reduction depends on how economic duce a quite significant difference in trends of real house- growth is defined. Measures of poverty reduction appear hold consumption. much more responsive to survey-based household con- sumption growth. When growth is measured by changes in HBS real per capita consumption, the growth elasticity 34 For example, see Atkinson and Lugo (2010), Demombynes and of poverty of –4, that is, a 1 percent increase in the survey Hoogeveen (2007), Hoogeveen and Ruhinduka (2009), Kessy et al. mean will reduce poverty headcount by 4 percent. But (2013), Mashindano et al. (2011), Mkenda et al. (2010), Osberg and when growth is measured by changes in real GDP per capi- Bandara (2012), Pauw and Thurlow (2011), and World Bank (2007, ta the growth elasticity of poverty is –1.02, indicating that a 2013b). Economic Growth and Poverty 39 Survey-based price indices probably better reflect price Growth Incidence Curves, 2001– Figure III.5   variations across regions and over time, but the discrepancy 2007 and 2007–2011/12 between the deflators would need further investigation and is beyond the scope of this report. Growth incidence, Tanzania Mainland, 2001–2007 10 Consumption growth (%) The Distributional Pattern of III.  5 Growth 0 Leaving aside the national accounts data, this study focus- –5 es on the evidence from the 2007 and 2011/12 HBS, using –10 changes in household consumption as the measure of 0 20 40 60 80 100 growth to explore whether the poor have benefitted from Consumption percentiles growth. Growth incidence, Tanzania Mainland, 2007–11/12 10 Consumption growth (%) There are emerging signs of “pro-poor” growth since 2007. 8 The growth incidence curve for 2007–2011/12, which shows 6 the percent change in average consumption for each per- 4 centile of the distribution, are downwardly sloped, indicat- 2 ing higher growth amongst the poorest (Figure III.5). Poor 0 households seem to have benefitted disproportionately –2 from growth, despite the modest increase of real household –4 0 20 40 60 80 100 consumption, which grew by only 0.9 percent per annum. Consumption percentiles The recent pattern of real consumption growth differs from Growth rate by percentile Growth rate in mean the period 2001–07, which shows that growth benefitted mainly the richer groups.35 Source: Hoogeveen and Ruhinduka (2009). HBS 2007 and 2011/12. These results hold even after addressing the data compa- rability problem. The imputed data from the different pre- distribution of consumption across households. This report diction models show downward sloping growth incidence applies the decomposition method proposed by Datt and curves in 2007–2011/12, confirming that consumption Ravallion (1992) to determine the growth and redistribu- growth of poorer households was faster than that of bet- tion components of the decline of poverty. As is apparent ter-off groups.36 from Figure III.6, the reduction in the poverty headcount at the national level was driven by both the increase in mean These positive results are tempered by the limited absolute household consumption (growth effect) and reduction of gains accruing to the poor. People in the 30 percent poor- inequality in the distribution of consumption (redistribu- est groups experienced an increase of their consumption tion effect), with the effect of inequality reduction being of around 20 percent between 2007 and 2011/12. But this increase is from a low base and translates to an addition- al consumption amount of only T Sh 4,300 per adult per 35 See Hoogeveen and Ruhinduka (2009) and Osberg and Bandara month (in 2011/12 prices), which is equivalent to approxi- (2012). mately 10 percent of the cost of basic consumption needs. 36 The National Panel Survey Data show however a different pov- erty trend to that observed in HBS data and do not support the The relationship between growth and poverty involves disproportionate consumption growth for poorer households, see changes both in mean consumption and changes in the more details in Box 3.1 in Appendix 3. 40 Tanzania Mainland Poverty Assessment Growth and Redistribution Figure III.6   decomposed to identify the specific attributes that con- Effects on Poverty Reduction tribute to the changes of consumption. The decomposi- (in percentage points) tion is applied at each decile group of the consumption distribution to understand the patterns of the changes for 0 the different welfare groups.37 –1 –2 The increase of poor households’ consumption is the result –3 –2.5 of improvements in both endowments and returns. One –4 can observe from Figure III.7 an improvement of house- –3.7 –5 holds’ endowments for all the population groups, but the –6 improvements are more marked for the 30 percent poorest –7 –6.2 segments. Change in Growth Redistribution headcount The increase of the endowments is driven by a significant Source: HBS 2007 and 2011/12. expansion of assets ownership, mainly transportation and communication means, and to a lesser extent agricultur- al land. Educational attainment of household’s heads has marginally more important. The growth effect contributes improved as well but less significantly. The access to local by 40 percent (2.5 percentage points) to poverty reduction, infrastructure has deteriorated in general, but access to while the redistribution effect contributes by 60 percent (3.7 local roads seems to have slightly improved for the poor. percentage points). The decomposition indicates also a decline of households’ engagements in business activities, particularly among the The emerging signs of pro-poor growth contrast with the poorest groups. nature of Tanzania’s economic growth. As shown in the first section of this chapter, economic growth in Tanzania was The improvements of households’ endowments were cou- driven mainly by fast growing and relatively capital-inten- pled with an increase of the returns to those endowments, sive sectors with limited job creation capacity. Agriculture, but only for the poorest decile group. Except for the first which represents the main source of livelihood for the vast two deciles, returns appear to have declined over time. But majority of the poor, grew at a much lower rate than the this decline masks divergent trends across the different at- overall economy. With growth centered mainly in national tributes. As observed from the table in figure III.7, the gains accounts sectors where poorer Tanzanians are not so in- from household businesses, essentially nonfarm activity, volved, the observed signs of pro-poor growth are not to increased quite significantly between 2007 and 2011/12 be expected. particularly for the three bottom deciles. Returns to land seem also to have improved over time, though less signifi- In order to explore the basic factors behind the observed cantly for the poor. The returns to community infrastruc- variations in household consumption, this study per- ture also improved, indicating a higher positive influence formed a decomposition of the changes in consumption of access to local markets and roads on needy households’ over time into two components: one component that is living standards. due to improvements in personal characteristics or en- dowments (better education, increased ownership of land and other assets, access to employment opportunities, lo- cal infrastructure, and so forth) and one component attrib- 37 The decomposition approach is based on the Recentered In- utable to changes in the returns to those characteristics fluence Function and unconditional quantile regression method (returns to education, land productivity, returns to busi- proposed by Firpo, Fortin and Lemieux (2009). See Appendix 4 for ness, and so forth). These components are then further more technical details on the approach. Economic Growth and Poverty 41 Returns Effect and Endowments Figure III.7   a continuing constraint on household well-being, although Effects over Time, Tanzania their negative impact appears to have diminished some- 2007–2011 what, as apparent from the positive change in the returns to demographic structure. 0.3 per capita total expenditures 0.2 However, the observed improvements in the returns to Di erence in log real some households’ attributes have been offset by a signifi- 0.1 cant decline of the returns to assets followed by a decline 0 of returns to education, inducing a loss of returns for the moderate poor and better-off households. –0.1 –0.2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Quantities Con dence interval/ Con dence interval/ endowment e ect returns e ect Endowment e ect Returns e ect Extreme Middle poor Poor class Richest Total 0.147*** 0.058*** 0.019* –0.076*** Endowments 0.075** 0.178*** 0.125*** 0.043 Demographic –0.019*** –0.026*** –0.022*** –0.013*** Structure Education –0.001 0.003* 0.003** 0.011*** Wage work 0.001 0.002* 0 0 HH business –0.024*** –0.022*** –0.009*** –0.005 Assets 0.124*** 0.114*** 0.103*** 0.054*** Land 0.006* 0.005* 0.007** 0.011*** Access local markets –0.005** –0.004** –0.002** –0.002 Access local roads 0.037*** 0.052*** 0.028*** 0.005 Returns 0.072** –0.120*** –0.106*** –0.119*** Demographic 0.255*** 0.064 0.025 0.216*** Structure Education –0.186*** –0.017 –0.003 0.066* Wage work –0.003 0.010 0.001 0.012 HH business 0.123** 0.162*** 0.056 0.077 Assets –0.266*** –0.169*** –0.156*** –0.244*** Land 0.016 0.022** 0.019** 0.035** Access local markets 0.055*** 0.049*** 0.021** 0.030* Access local roads 0.011 0.045*** 0.011 –0.022 Source: HBS 2007 and 2011/12. 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. Large household size and numbers of children seem to be 42 Tanzania Mainland Poverty Assessment 43 Chapter  4 Uneven Geographic Decline in Poverty Key Messages ➤➤ Most of the improvements in the poverty indicators occurred in Dar es Salaam, but the absolute number of poor declined more in rural areas. ➤➤ The reduction of poverty outside Dar es Salaam is driven mainly by a decline of inequality. ➤➤ Besides improvements in assets endowments, the growth of poor households’ consumption levels is due to the increase of returns to nonfarm businesses and wage employment in urban zones and improvements of the returns to agricultural businesses and cash crops in rural areas. The previous chapters focused on the analysis of poverty Dar es Salaam and by around 6 percentage points (15 per- trends and the relationship between poverty and growth cent) in the rural sector, while it declined only marginally, by at the national level. The report turns here to the analysis of 1.2 percentage points (5 percent), in the secondary cities and poverty trends and the links between growth and pover- towns. This trend is repeated with extreme poverty, as the ty in the different geographic domains, namely rural areas, highest decline was observed in Dar es Salaam followed by Dar es Salaam, and the other urban zones. The first section rural areas, while there was almost no change in the other examines the evolution of poverty headcounts and pover- urban zones. Clearly, the city and the surrounding adminis- ty depth and severity in the different geographic areas. The trative area have pulled ahead of other parts of the country, second section investigates the distributional patterns of despite the fact that Dar es Salaam already had significantly growth in these areas and explores the factors contributing lower poverty than other areas in 2007. to poverty reduction. Although most of the reduction of the level of poverty oc- curred in Dar es Salaam, the absolute number of poor people Poverty Trends by Geographic I.  declined more in the rural areas (Figure IV.2). The absolute Domains number of poor people declined by around 1.2 million in rural areas against nearly 260,000 in Dar es Salaam, while it slightly Most of the improvements in the poverty indicators occurred increased by approximately 160,000 in the secondary cities. in Dar es Salaam (Figure IV.1). The basic needs poverty head- Likewise, the number of extreme poor, who are not able to count declined by over 10 percentage points (70 percent) in meet their food basic needs, declined by around 400,000 in 45 Basic Needs and Extreme Poverty Headcounts by Geographic Domain (%) Figure IV.1   45 14 13.3 40 39.4 12 11.3 35 33.4 30 10 8.9 8.6 25 22.7 21.5 8 20 14.1 6 15 4 3.2 10 5 4.0 2 1.1 0 0 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 Rural Other urban Dar es Salaam Rural Other urban Dar es Salaam Source: HBS 2007 and 2011/12. the rural zones against nearly 54,000 in Dar es Salaam, while it Dar es Salaam expanded considerably between 2007 and increased by about 81,000 in the other urban zones. 2011/12, increasing by approximately 60 percent, indicat- ing that the benefits of Tanzanian economic growth have But basic needs and extreme poverty remain more perva- largely been reaped by well-off households located in Dar sive in rural areas. HBS 2011/12 indicates that rural areas es Salaam. continue to account for, respectively, 84 and 82 percent of the total poor and extreme poor populations in Tanzania, The depth and severity of poverty declined faster than the against, respectively, 14 and 17 percent in the other urban poverty headcount (Figure IV.4). The decline in the poverty areas and less than 1 percent in Dar es Salaam. gap and poverty severity is more pronounced than the re- duction in the poverty headcount in all the geographic do- The comparison of the rural-urban dynamics over time in mains, indicating a significant improvement in the status of Figure IV.3 shows that over 84 percent of the population those who remain poor in the different regions. Even in the in the bottom quintile (20 percent poorest group) lives in secondary cities where the poverty headcount has declined rural areas and that this proportion remained fairly stable only marginally, one observes a reduction in poverty depth between 2007 and 2011/12. The proportion of population by nearly 2 percentage points, indicating the improvement in the poorest quintile living in Dar es Salaam declined by in consumption by the poor and that the gap between their over 50 percent, while that living in the secondary cities average consumption and the basic needs threshold has nar- increased by over 14 percent. In contrast, the proportion rowed by 2 percentage points (or around 25 percent). In the of the population in the 20 percent richest group living in rural areas, the average consumption level of a poor people attained around 92 percent of the poverty line in 2011/12, while in Dar es Salaam it rose to 99 percent. The quite im- Distribution of Poor Population Figure IV.2   portant decline of poverty severity suggests a significant im- by Geographic Area provement in the consumption level of extremely poor pop- 2007 2011/12 ulation groups, particularly in rural areas where the severity of the poverty index declined by nearly 2.4 percentage points 0.43 7 0.1 1.54 1.7 (or 47 percent). Rural Rural Other Urban Other Urban Dar es Salaam Dar es Salaam The prediction models support the decline of poverty in the 11.2 10.04 different geographic domains but reveal a lower level of pov- erty reduction in Dar es Salaam. The prediction models used Source: HBS 2007 and 2011/12. to adjust for the changes in the HBS methods between 2007 46 Tanzania Mainland Poverty Assessment Population Distribution by Consumption Quintiles and Area of Residence Figure IV.3   a. 2007 b. 2011/12 2.5% 4.5% 6.3% 9.5% 17.7% 1.2% 2.9% 5.5% 12.8% 28.0% 100% 100% 90% 12.6% 11.8% 90% 14.4% 14.7% 15.2% 16.6% 80% 21.1% 80% 21.2% 70% 27.9% 70% 60% 60% 26.9% 50% 50% 40% 40% 84.8% 83.8% 78.6% 69.5% 54.5% 84.4% 82.4% 78.0% 66.0% 45.1% 30% 30% 20% 20% 10% 10% 0% 0% Poorest 2nd 3rd 4th Top Poorest 2nd 3rd 4th Top quintile quintile quintile quintile quintile quintile quintile quintile quintile quintile Rural Other urban Dar es Salaam Source: HBS 2007 and 2011/12. Note: Each quintile represents 20 percent of the population ranked by consumption at the national level. For example, the poorest quin- tile includes the 20 percent of population with the lowest levels of household consumption per adult at the national level, while the top or richest quintile represents the 20 percent of the population at the upper level of the distribution of consumption at the national level. and 2011/12 support the decline of poverty observed in As discussed in “Decline in Poverty and Extreme Poverty Figure IV.1. The decline of poverty at the geographic regions Since 2007” in chapter 1, the prediction models seem to level in Figure IV.5 is very similar to that observed above, ex- attenuate the effects of inflation in food prices inflation cept for the chained method with shows a more important on extreme poverty, implying higher estimates of extreme decline of poverty in the other urban areas (of 5 percentage poverty rates for 2007 and consequently a higher decline in points) and only a slight reduction of rural poverty of around extreme poverty during 2007–2011/12, particularly in rural 2 percentage points. Also, the prediction models confirm areas and other urban zones. that poverty declined faster in Dar es Salaam than in the oth- er regions, but they show a lower level of poverty reduction The different prediction models confirm also the decline of in Dar es Salaam compared to the decline observed above. poverty severity and depth and, as observed, show a more Depth and Severity of Poverty by Geographic Domain Figure IV.4   Depth of Poverty Severity of Poverty 14% 6% 12% 11.8% 5.1% 5% 10% 4% 8% 7.9% 3.3% 7.3% 3% 2.7% 6% 5.5% 2.1% 2% 4% 3.5% 1.3% 2% 1% 0.8% 0.3% 0% 0% 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 Rural Other urban Dar es Salaam Rural Other urban Dar es Salaam Source: HBS 2007 and 2011/12. Uneven Geographic Decline in Poverty 47 Adjusted Poverty Rates for 2007 Figure IV.5   inclusion or exclusion of cell phone ownership. Excluding by Geographic Domain Using cell phones in the prediction of consumption levels and Prediction Methods (%) poverty rates for 2007 seems to introduce a downward bias in the poverty estimates, suggesting low changes in pover- 45 ty between 2007 and 2011/12 (Table IV.1). 39.4 40 38.4 35.4 35 30 Growth and Distributional II.  26.6 23.7 22.6 25 20 Changes by Geographic 16.7 14.9 Domains 14.5 13.7 12.7 15 11.7 11.4 10.7 8.6 10 4.7 4.5 5 The uneven spatial decline of poverty observed is related 1.9 0 to the pattern of economic growth, which was almost en- Rural Other Dar es Rural Other Dar es urban Salaam urban Salaam tirely centered in Dar es Salaam, where are concentrated Povert Extreme poverty most of the expanding and flourishing sectors. When using Semi-parametric (Tarozzi) MI chained HBS changes in real household consumption per capita, Poverty mapping growth is found to average 3.7 per year in Dar es Salaam, Source: HBS 2007 and 2011/12. while there was almost no growth rural areas and second- ary cities, where the annual growth rate was on average of pronounced reduction in these indicators than in the pov- –0.2 percent and -0.1 percent, respectively. The more rapid erty headcount. growth of household consumption in Dar es Salaam reso- nates well with the sectoral composition of real GDP growth Finally, and as discussed in chapter 1, the poverty estimates over the period 2008–12. As shown in “A Brief Review of Re- for 2007 using the prediction models are sensitive to the cent Economic Growth” in chapter 3, Tanzania’s GDP growth Adjusted Poverty Rates for 2007 by Geographic Domain Using Prediction Table IV.1   Methods Semi-parametric MI chained (with cell MI chained (without Poverty mapping (with Poverty mapping (Tarozzi) phone) cell) cell phone) (without cell) Extreme Extreme Extreme Extreme Extreme   poverty Poverty poverty Poverty poverty Poverty poverty Poverty poverty Poverty Headcount Rural 16.7% 39.4% 14.9% 35.4% 13.2% 32.1% 14.5% 38.4% 11.3% 31.6% Other urban 11.4% 23.7% 12.7% 26.6% 10.4% 22.7% 13.7% 22.6% 10.6% 24.8% Dar es Salaam 1.9% 10.7% 4.7% 11.7% 3.7% 10.0% 4.5% 8.6% 3.9% 9.4% Depth of Poverty Rural 4.3% 12.8% 3.2% 9.4% 2.8% 8.3% 3.1% 9.9% 2.2% 7.6% Other urban 3.0% 12.8% 3.3% 7.9% 2.6% 6.5% 2.2% 6.0% 2.2% 7.0% Dar es Salaam 0.4% 2.6% 1.1% 3.0% 0.8% 2.5% 0.8% 2.2% 1.0% 2.5% Severity of Poverty Rural 1.6% 5.9% 1.1% 3.6% 0.9% 3.1% 0.9% 3.6% 0.7% 2.7% Other urban 1.2% 3.9% 1.3% 3.4% 1.0% 2.7% 0.8% 2.4% 0.7% 2.6% Dar es Salaam 0.2% 0.9% 0.4% 1.2% 0.3% 0.9% 0.3% 0.8% 0.4% 1.1% Source: HBS 2007 and 2011/12. 48 Tanzania Mainland Poverty Assessment was essentially driven by construction, communications, the redistribution component contributes by 55 percent, and financial services sectors, which all saw a growth rate of while in the rural and other urban sectors reductions in over 10 percent. With the exception of construction, these poverty are due entirely to improvements in consumption sectors created limited jobs. The agricultural sector, which distribution, with mean consumption changes resulting in employs three-quarters of the workforce and a vast majority slight increases in poverty. of the poor, grew at only 4.2 percent, a much lower rate than average economic growth. Poor households outside Dar es Salaam experienced an in- crease in their consumption levels, albeit from low levels. Despite the limited growth outside Dar es Salaam, poor The Growth Incidence Curves (GIC) in Figure IV.7 show con- households have experienced consumption gains and sumption gains among households in the poorest quintiles poverty has declined quite significantly, particularly in rural in rural and urban areas other than Dar es Salaam. Stagnant areas. This section examines more in detail the relationship average consumption in these areas masks different expe- between growth and poverty in the different geographic riences across the distribution, with poorer households ex- domains and investigates the underlying causes to the de- periencing more rapid increases in consumption (in relative cline of poverty outside the metropolitan city. terms) than the better-off. This is indicated by the downward sloping growth incidence curves. In other urban and rural Poverty reduction outside Dar es Salaam is driven mainly by strata, the better-off experienced declines in consumption, inequality reduction. The decline of poverty in Dar es Salaam whereas the poorest two quintiles in other urban areas and was driven by both an increase in mean consumption and three quintiles in rural areas appear to have experienced an an improvement in consumption distribution (reduction of increase in their consumption levels, albeit from low levels inequality), while poverty reductions in rural and other ur- of consumption initially. The same general pattern (down- ban areas are due entirely to improvements in consumption ward sloping growth incidence curves) applies to Dar es distribution (Figure IV.6). In Dar es Salaam the growth com- Salaam over this period, with all households (across the dis- ponent contributes by 45 percent to poverty reduction and tribution) experiencing increases in real consumption. But poorer households gained more than the better-off—rela- tive to their consumption levels in 2007. Growth and Redistribution Figure IV.6   Components of Changes in The observed signs of pro-poor growth outside Dar es Poverty at the Regional Level Salaam are quite puzzling and need to be investigated in (in percentage points) more detail. Increases in real consumption can be assumed as due either to an improvement in household character- 2 0.9 istics or endowments or increases in the returns to these 0.1 0 endowments. –2 –1.2 –2.1 –4 In order to better understand the factors underlying the –4.5 –6 –5.5 increase of consumption in each geographic region, this –6 –6.2 –8 study decomposes the changes in households’ consump- –10 tion over time into the part explained by improvements –10 –12 in endowments and the part explained by changes in the Change in headcount Growth Redistribution Change in headcount Growth Redistribution Change in headcount Growth Redistribution returns to those characteristics. The decomposition proce- dure is similar to that applied in “The Distributional Pattern of Growth” in chapter 3. Rural Other urban Dar es Salaam The increase of the consumption of rural poor households is Source: HBS 2007 and 2011/12. driven essentially by the improvement of their endowments. Uneven Geographic Decline in Poverty 49 Growth Incidence Curves by Figure IV.7   We observe from Table IV.2 a decline in the access of rural Geographic Domain poor households to business activities, mainly household agricultural businesses, however a more detailed decom- Growth incidence, rural areas, 2007–11/12 position shows an increase in the access to cash crop pro- 10 duction. This supports the findings in the previous chap- 8 ters, suggesting a switch away from agriculture and higher 6 engagement of households who remained in the sector 4 in cash crops and commercial agriculture. These changes 2 apply both to the rural population overall, and to the poor- 0 est three deciles. The access to local infrastructure, mainly –2 roads, also improved but access to local markets remained –4 0 20 40 60 80 100 limited. Growth incidence, other urban areas, 2007–11/12 10 The returns to the endowments of poor rural households 8 increased, but only for the poorest groups. The moder- 6 ate poor as well as nonpoor households experienced a 4 decline of the overall returns to their endowments. How- 2 ever, this decline masks important differences in the di- 0 rection of change of the returns to the specific attributes. –2 For instance, returns to both nonfarm and household –4 agricultural businesses, mainly cash crops, seem to have 0 20 40 60 80 100 expanded for rural poor households. Returns to land also Growth incidence, Dar es Salaam, 2007–11/12 increased slightly. The returns to local markets seem to 10 have improved as well, suggesting that while farming 8 households are not better served by markets than they 6 were in 2007, these currently play a more positive role in 4 their livelihoods. 2 0 The growth of consumption among poor households in the –2 secondary cities is due more to the improvement of their –4 0 20 40 60 80 100 endowments. The consumption level of households in the Growth rate by percentile Growth rate in mean 30 percent poorest groups increased by about 15 percent, essentially due to the improvement of their endowments in Source: HBS 2007 and 2011/12. assets and to a lesser extent in land. One observes important fluctuations of the changes in Rural households in the 30 percent poorest groups experi- returns across the different deciles. The overall returns to enced an increase of their consumption by around 20 per- endowments increased over time for the extreme poor cent between 2007 and 2011/12. This increase was driven segments as well as for better-off household groups, but de- primarily by the improvement of their endowments in as- clined for the moderate poor (see Figure IV.8). However, the sets, mainly, increased ownership of communication and results in Table IV.2 indicate quite significant expansion of transportation means followed by higher possession of ag- the returns to nonfarm activities and wage employment for ricultural land (see Figure IV.8 and Table IV.2). The endow- all households groups, but particularly for the poor. On the ments in education also improved among the moderate other hand, there is evidence from the results of a marked poor segments. deterioration of the returns to education and to assets. 50 Tanzania Mainland Poverty Assessment Sources of Households’ Figure IV.8   The increase of poor households’ consumption levels in Dar Consumption Growth by es Salaam is caused primarily by the improvements of the re- Geographic Domain turns to their characteristics. The consumption level of poor Returns e ects and endowment e ects over time households in Dar es Salaam increased by over 40 percent Rural HBS 2007–2011/12 between 2007 and 2011/12, due mainly to the expansion 0.4 of the returns to employment in public and private sectors per capita total expenditures followed by a slight increase of the returns to nonfarm busi- Di erence in log real 0.2 nesses. Moderate poor households also experienced some gains in their endowment base, essentially endowments 0 in education, but the effect of returns was proportionately stronger. However, one observes a decline of the effect of –0.2 returns at upper deciles, indicating that for better-off house- holds the gains in consumption are explained mainly by the –0.4 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 increase of their endowments. Quantities Returns e ects and endowment e ects over time Other Urban HBS 2007–2011/12 per capita total expenditures 0.4 Di erence in log real 0.2 0 –0.2 –0.4 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Quantities Returns e ects and endowment e ects over time Dar es Salaam HBS 2007–2011/12 per capita total expenditures 0.4 Di erence in log real 0.2 0 –0.2 –0.4 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Quantities Con dence interval/ Con dence interval/ endowment e ect returns e ect Endowment e ect Returns e ect Source: HBS 2007 and 2011/12. Uneven Geographic Decline in Poverty 51 Endowments and Returns Effects of Some Specific Attributes Table IV.2   Rural Other urban Dar es Salaam Extreme Middle Extreme Middle Extreme Middle poor Poor class Richest poor Poor class Richest poor Poor class Richest Total 0.174*** 0.069*** 0.034** –0.097** 0.033* 0.015* –0.020 –0.070* 0.379*** 0.407*** 0.454*** 0.521*** Endow- 0.063** 0.140*** 0.169*** 0.125*** –0.008 0.065* –0.035 –0.485 –0.038 0.127* 0.295*** 0.594*** ments Education 0.000 0.003** 0.004** 0.007** –0.001 –0.004 –0.006 0 0.055 0.061** 0.091*** 0.169** Wage work 0.002 0.001 0.001 –0.001 –0.019 –0.006 –0.001 0 0.004 0 –0.003 0.012 HH –0.002 –0.002 –0.001 –0.001 –0.059** –0.027** –0.012* –0.037** –0.024 –0.003 0.008 –0.052 nonfarm business HH –0.014*** –0.009*** 0.000 0.009 0.002 0.001 0 0 0.002 0.002 0.004 –0.005 agricultural business Assets 0.084*** 0.073*** 0.078*** 0.092*** 0.196*** 0.194*** 0.136*** 0.049 –0.014 –0.001 0.070 0.229 Land 0.004 0.006** 0.007*** 0.014*** 0.016 0.023** 0.018** 0.005 –0.038 –0.004 0.012 0.014 Access local –0.002*** –0.001 0 –0.001 –0.013 –0.004 –0.030** –0.014 –0.006 0 –0.015 0.035 markets Access local 0.037*** 0.046*** 0.039*** 0.017 –0.023 0.007 –0.011 –0.040 0.006 0.009 0.085*** 0.051 roads Returns 0.111*** –0.071*** –0.135*** –0.222** 0.041 –0.050 0.016 0.415* 0.417*** 0.280*** 0.159* –0.073 Education –0.155*** –0.012 –0.004 0.096** –0.605** –0.145* –0.164* –0.092 –0.109 –0.594** –0.481* –0.490 Wage work –0.005 0.006 –0.001 0.008 0.119*** 0.044** 0.036** 0.063*** 0.066 0.145*** 0.122*** 0.251*** HH 0.012 0.022* 0.002 0.009 0.172*** 0.071** 0.058** 0.151 0.170* 0.003 –0.145** 0.344** nonfarm business HH 0.108* 0.089* 0.015 –0.022 0.182*** 0.042 0.077** 0.064 0.020 0.011 –0.087*** 0.096 agricultural business Assets –0.310*** –0.218*** –0.168*** –0.173*** –0.433*** –0.257*** –0.053* –0.194* –1.243* –0.332 –0.184 –0.064 Land 0.003 0.011* 0.003 0.005 –0.044 0.067 0.160* 0.488** –0.104 –0.091* 0.001 –0.164 Access local 0.053*** 0.041*** 0.022* 0.027 –0.097 –0.071* 0.072** 0.090* 0.087 –0.051 0.001 –0.205 markets Access local 0.014 0.018 0.016 –0.060** –0.037 0.107 –0.001 –0.208* –0.328 –0.055 0.486** –0.043 roads Source: HBS 2007 and 2011/12. 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. For rural households the endowments in cash crop agriculture are, respectively, 0.009***, 0.008***, and 0.012*** for the extreme poor, poor, and middle class households. The returns to cash crops are, respectively, 0.003, 0.011**, and 0.015**. 52 Tanzania Mainland Poverty Assessment Chapter  5 Increasing Inequality between Geographic Domains Key Messages ➤➤ Inequality between urban and rural areas as well as between Dar es Salaam and the other regions is increasing. ➤➤ Interregional inequality between poor households is slightly narrowing, but interregional inequality between better-off households is increasing. ➤➤ The interregional differences in returns to household endowments are increasing, but inequality remains due mostly to large differences in households’ characteristics. The previous chapters revealed positive changes in Tanzania, decompose inequality between regions into a component indicating a decline of poverty and emerging signs of “pro- that is due to geographic differences in the distributions of poor” growth. However, the evidence from the results also household characteristics or endowments such as educa- suggests uneven improvements of the living standards and tion, demographic structure, ownership of assets, and so poverty indicators across the geographic regions. Also, the forth, and a component explained by differences in the re- decomposition of inequality by population groups, noted in turns to these characteristics (differences in the returns to “Moderate and Fairly Stable Inequality” in chapter 1, shows education, land productivity, and so forth). More specifical- an increase of urban-rural and interregional inequality over ly, this study decomposes the consumption gap between time. The increasing spatial disparities are worrisome as they geographic regions into (i) a component that is due to can undermine inclusive growth prospects and may jeopar- differences in household characteristics only (endowment dize social cohesion. effects), considering, for example, the gap in consumption that is due to the fact that urban dwellers have higher ed- This chapter examines in detail the extent and sources of ucation levels than rural ones but assuming that people these inequalities. The first section investigates the deter- with same education levels receive the same remunerations minants of urban-rural inequality and the second section across the different locations; and (ii) a component that is explores the sources of inequality between Dar es Salaam due to differences in returns to those characteristics only (re- (metropolitan) and the rest of the regions (nonmetropolitan). turns effect), considering, for example, the gap in consump- tion that is due to the fact that a secondary school graduate This study uses the unconditional quantile regression in the urban areas receives a higher remuneration than a method proposed by Firpo, Fortin, and Lemieux (2009) to secondary school graduate in the rural areas. 53 Decomposing Inequality by Regions Figures V.1   Estimate the unconditional quantiles of log per capita consumption for each region of interest Main steps of the decomposition of spatial inequality Decompose the di erence between regional quantiles into Di erences in households Di erences in the returns to these characteristics (Endowments Gap) characteristics (Returns Gap) Contribution of each speci c households Contribution of each speci c households characteristic to Endowments Gap characteristic to Returns Gap The decomposition proceeds as depicted in Figure V.1 rural rich households than between poor ones was explained (more technical details are in Appendix 4). mainly by larger urban-rural differences in returns at upper quantiles, while in 2011/12 better-off urban households ex- perienced a faster increase of their endowments and returns The Sources of Urban-Rural I.  than their rural counterparts, which induced a widening of Inequality the consumption gap at upper quantiles.38 Urban households are better off than their rural counterparts The urban-rural difference in household endowments was because they have superior endowments such as education, the main source of urban-rural inequality for the poorest family structure, and assets ownership. Inequality between ur- segment of the population in the early and middle of the ban and rural areas is essentially due to the fact urban house- decade, but it seems to be declining in 2011/12. There was holds have higher endowments than their rural counterparts. an important gap in assets ownership and educational at- As shown in Figure V.2, the contribution of the difference in tainment between urban and rural poor households. Start- households’ endowments to the urban-rural gap significant- ing from 2007, education and the possession of assets im- ly dominates the contribution of disparity in returns to those proved for all poor households but improved faster for poor endowments across the entire distribution, indicating that rural households, inducing a shrinking of the urban-rural urban households have higher consumption levels because endowment gap at the lower quantiles. they have characteristics superior to rural ones. The difference between urban and rural areas in market Inequality between better-off urban and rural households is returns to household characteristics does not seem to be larger than inequality between poor urban and rural house- important for poor household groups. This is probably due holds. The difference in real per capita consumption between richest urban and rural households is more than double the 38 Quantiles are values taken at regular intervals from the inverse difference between poorest urban and rural households (see of the cumulative distribution function of per capita real monthly Figure V.2). This is mainly driven, in 2011/12, by larger gaps be- consumption. If there are 5 quantiles then each quantile will cor- tween urban and rural rich households in both endowments respond to a quintile (20 percent of the population) and if there and returns than between urban and rural poor households. are 10 quantiles then each quantile will correspond to a decile (10 In 2001 and 2007, the higher inequality between urban and percent of the population) and so forth. 54 Tanzania Mainland Poverty Assessment to the fact that these households are generally employed structure and in access to basic services between urban and in sectors that pay slightly above the subsistence level. But rural households. The effect of differentials in household this difference in returns is affecting households at upper human capital (measured by the highest number of years quantiles, particularly the wealthiest. As apparent from Fig- of schooling of the household head or his spouse and the ure V.2, the magnitude of the returns effects is increasing experience of the head) increased between 2001 and 2007 proportionately more than the magnitude of endowments and then decreased in 2011/12, particularly for the poorest effects at upper quantiles, showing that even though all quantiles, while differences in the sector of employment of urban households continue to have superior endowments the head kept widening over time. This suggests that de- to those of their rural counterparts, the contribution of dif- spite some improvements in the education level of rural ferences in returns to households’ attributes to inequality is households, the urban dwellers and particularly the richest gaining importance for most well-off households. ones have been more able to access to better job opportu- nities than their rural counterparts. The urban-rural gap between the rich is widening over time while it is slightly narrowing between the poor. Urban-rural Differences in returns to assets and employment are among inequality is increasing over time for the middle-class and the dominant factors accounting for rural-urban gap in re- richest households, driven mainly by widening urban-ru- turns to household characteristics. There is a quite import- ral differentials in households characteristics for both pop- ant difference in the returns to assets between urban and ulation classes and an increasing dispersion of returns to rural areas. This difference contributes more to inequality households attributes for the wealthiest. Rural households between the poor than to inequality between the rich, but at the lower tail of the distribution have observed an im- it is narrowing over time for the poor while it is widening provement in their endowments over time, and there are for the rich. signs of convergence in household endowments between the sectors. This suggests that the development policies The urban-rural gap in returns to human capital showed implemented in Tanzania were appropriate to tackle some a marked increase (particularly at the upper quantiles) in of the rural poor’s problems, such as combating illiteracy 2007 but started declining since then (tables 4-1 to 4-3 in and promoting basic education, facilitating access to assets Appendix 4). Even though urban markets continue to better and land, and so forth. But these policies did not adequate- reward education and experience than rural markets do, the ly address the needs of better-off rural households to help gap seems to have narrowed, particularly for the poorest them catch up with their urban counterparts. Well-off urban and richest segments of the population. households have been better able to improve their endow- ments and to benefit from the opportunities generated by The urban-rural differentials in returns to employment of economic growth than their rural counterparts. the households have widened over time, driven mainly by a more marked increase of returns to wage employment in Differences in the distribution of household demographic the public and private sectors and to a lesser extent to non- characteristics and access to basic services, followed by dif- farm businesses in the urban areas. ferences in the sector of employment of the head, matter the most for inequality between urban and rural households Poor households seem to have benefitted from the policies (Table 4-4 in Appendix 4). In 2001 and 2007, differences in for basic education to catch up with their urban counter- asset ownership, such as land, livestock, cell phones, and parts. However, they continue to suffer from limited access transportation means, are found to significantly contribute to basic services, large family sizes, and large number of de- to the welfare gap between urban and rural households, pendents. Middle-class and well-off rural households have particularly for less well-off segments of the population, slightly reduced their education gap with the urban ones, but the difference in assets possession declined markedly but they have not been able to access better job oppor- over time, mainly for the poor classes. These improvements tunities nor obtain higher returns for their employments have been largely offset by a widening gap in demographic and assets. This points to the possibility that employment Increasing Inequality between Geographic Domains 55  nconditional Quantile Decomposition of Urban-Rural Inequality of Real Figure V.2  U Monthly per Capita Consumption Returns e ects and endowment e ects Returns e ects and endowment e ects by Area for Tanzania 2001 by Area for Tanzania 2007 1.0 1.0 per capita total expenditures per capita total expenditures Di erence in log real Di erence in log real 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 Quantities Quantities Returns e ects and endowment e ects by Area for Tanzania 2012 1.0 per capita total expenditures Di erence in log real 0.5 0 –0.5 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Quantities Con dence interval/ Con dence interval/ Endowment e ect Returns e ect endowment e ect returns e ect 2001 2007 2011/12 Lowest Middle Top Lowest Middle Top Lowest Middle Top percentile percentile percentile percentile percentile percentile percentile percentile percentile Total Gap 0.327 0.390 0.452 0.243 0.385 0.470 0.257 0.427 0.641 (0.016) (0.010) (0.020) (0.026) (0.015) (0.025) (0.017) (0.014) (0.025) Endowments 0.535 0.448 0.379 0.600 0.384 0.393 0.394 0.543 0.545 (0.021) (0.014) (0.025) (0.033) (0.017) (0.030) (0.025) (0.019) (0.037) Returns –0.208 –0.058 0.073 –0.357 0.001 0.077 –0.138 –0.115 0.096 (0.025) (0.016) (0.030) (0.039) (0.020) (0.037) (0.029) (0.021) (0.041) Source: HBS 2001, 2007, and 2011/12. Note: Numbers in parentheses are bootstrap standard deviations based on 100 replications. and profit opportunities are expanding and diversifying higher assets than the rural ones have been more able to more in urban than in rural areas and to the fact that urban take advantage of these opportunities to improve their en- households who were initially better educated and enjoyed dowments and leverage their returns. 56 Tanzania Mainland Poverty Assessment Determinants of Inequality II.  improved for poor households outside Dar es Salaam, they could not be offered returns equivalent to those in the city. between Dar es Salaam and the Other Regions The differences in the distribution of household demo- graphic characteristics and human capital endowments Inequality between Dar es Salaam and the rest of the regions between the geographic locations and the unequal access is increasing because households’ endowments are improv- to private assets and productive employments limited the ing faster in the city. Improvements in households’ endow- ability of the poor to take up the opportunities generated ments in Dar es Salaam outpaced the improvements in the by economic growth and to improve their living standards. rest of country, driven by widening differences in family Households in Dar es Salaam and in urban areas who enjoy structures and access to education and employment oppor- higher endowments have been able to benefit more from tunities between the two sectors. As revealed by Figure V.3, the growth in Tanzania and have seen a larger expansion in the gap in endowments between households living in Dar returns to their attributes. This, combined with the widen- es Salaam and those living in the rest of the country is larger ing differences in characteristics, contributed to increasing and increasing faster than the gap in returns, particularly for interregional inequalities and self-perpetuating poverty in households at upper quantiles. In the early part of the 2000s some regions, mainly rural areas. first decade, metropolitan households were better off than their nonmetropolitan counterparts because markets in Dar Efforts to promote education, family planning, and access es Salaam pay more for their attributes than markets in other to basic services and assets should be further enhanced to regions would. However, in 2011/12 the endowments in hu- improve the endowments of marketable characteristic for man capital (education), employment, and family structure households at the lower end of the income-consumption improved considerably in the metropolitan city compared to distribution. These efforts need to be accompanied, on the the other regions, inducing larger interregional inequalities, one hand, by policies targeting rural and nonmetropolitan particularly among better-off households. households at upper quantiles to help them catch up with their urban and metropolitan counterparts, and on the The education and employment opportunities improved other hand by policies to promote local economies’ devel- for poor households outside Dar es Salaam, but the mar- opment and dynamism and expand productive activities kets in their areas of residence could not offer them the in order to increase the returns to endowments in the less returns they would have obtained in the city. Inequality in favored regions endowments between Dar es Salaam and the rest of the re- gions increased proportionately more for middle class and richest households, while inequality in returns increased more for the poor. This is due to the faster increase over time of returns to the endowments of poor households in Dar (Table  4-5 in Appendix 4).39 It appears, therefore, that 39 The faster increase of returns in Dar can also be observed in Fig- even though education and employment opportunities ure IV.8 and Table IV.2 in the previous chapter. Increasing Inequality between Geographic Domains 57  nconditional Quantile Decomposition of Metropolitan-Nonmetropolitan Figure V.3  U Inequality in Real Monthly per Capita Consumption Returns e ects and endowment e ects Returns e ects and endowment e ects by Region for Tanzania 2001 by Region for Tanzania 2007 2 1.0 per capita total expenditures per capita total expenditures Di erence in log real Di erence in log real 1 –0.5 0 0 –1 –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 Quantities Quantities Returns e ects and endowment e ects by Region for Tanzania 2012 0.8 per capita total expenditures Di erence in log real 0.6 0.4 0.2 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Quantities Con dence interval/ Con dence interval/ Endowment e ect Returns e ect endowment e ect returns e ect 2001 2007 2011/12 Lowest Middle Top Lowest Middle Top Lowest Middle Top percentile percentile percentile percentile percentile percentile percentile percentile percentile Total gap 0.478 0.480 0.529 0.420 0.450 0.533 0.661 0.677 0.767 (0.026) (0.026) (0.040) (0.025) (0.016) (0.030) (0.017) (0.015) (0.028) Endowments 0.302 0.185 –0.531 0.396 0.448 0.232 0.398 0.535 0.473 (0.158) (0.150) (0.232) (0.104) (0.062) (0.168) (0.066) (0.053) (0.104) Returns 0.175 0.295 1.06 0.024 0.002 0.301 0.263 0.143 0.294 (0.160) (0.151) (0.234) (0.105) (0.063) (0.170) (0.067) (0.054) (0.107) Source: HBS 2001, 2007, and 2011/12. Note: Numbers in parentheses are bootstrap standard deviations based on 100 replications. 58 Tanzania Mainland Poverty Assessment Increasing Inequality between Geographic Domains 59 60 Tanzania Mainland Poverty Assessment Chapter  6 Inequality of Opportunity Key Messages ➤➤ Around one-fourth of consumption inequality is explained by family background and circumstances beyond individuals’ control. ➤➤ Inequality of opportunity is higher in urban areas, but increasing in rural zones. ➤➤ Parental education and particularly father’s education contributes the most to the disparity of welfare in Tanzania. Inequality between population groups seems to be increas- The previous chapter revealed that the spatial inequalities ing in Tanzania despite the signs of improving welfare distri- in Tanzania are due mainly to the lack of capacities and en- bution at the national level. These intergroups’ inequalities dowments of households in the rural and disadvantaged manifest themselves in unequal outcomes but also unequal regions. This chapter takes an intergenerational perspective opportunities. To the extent that inequality in opportuni- and explores how family background affects these inequal- ties is high, it will perpetuate the lack of capabilities in the ities. Unlike the previous chapter, where all households’ en- population and the waste of productive potential and will dowments were considered, this chapter focuses only on contribute to poverty and inequality persistence. Hence, de- those inherited and independent of their choices. velopment policies focusing on promoting shared prosper- ity and equity need to address inequality in both outcomes Drawing on data from the National Panel Surveys (NPS) for and opportunity. 2008/09, 2010/11, and 2012/13, the study assesses the ex- tent to which unequal opportunity, resulting from the fam- Inequality of opportunity is defined as the part of inequal- ily and circumstances variables, affects the distribution of ity stemming from circumstances, such as gender, family both consumption and income. background, and place of birth, that are beyond a person’s control and is widely recognized to contribute to the per- All survey waves include a rich information at the household sistence of social and economic inequalities and to con- and individual levels on consumption and income, parental strain inclusive development. It is important to distinguish education, and family circumstances. They include as well inequalities due to unequal opportunities from inequalities a community module that collects detailed information on due to individual choices to better inform policy design and the access to basic services and distance to population cen- institutional arrangements that reduce the unfair influence ters, the presence of local investment projects, infrastruc- of people’s circumstances and favor a more egalitarian dis- ture conditions, and demographic and family characteristics tribution of opportunities. in the communities where the households are located. This 61 information is missing in the Household Budget Surveys characteristics on inequality and compares its impact to that (HBS), which makes them unsuitable of the analysis of in- of family circumstances. Details related to the variables used equality of opportunity. and underlying methodology are in Appendix 5. This study uses the parametric model proposed by Bour- guignon, Ferreira, and Menéndez (2007) and estimate in- Inequality of Opportunity in I.  equality of opportunity as the difference between observed Household Consumption total inequality and the inequality that would prevail if there were no differences in circumstances. Two different welfare Inequality of opportunity levels for household consump- indicators are used for the measurement of total inequality: tion ranges between 0.05 and 0.07 during 2008–12, and (i) real monthly per capita consumption and (ii) real monthly this level is relatively high by international standards. The per capita income. The focus on consumption and income estimated level of inequality of opportunity, reported in is motivated by the desire to investigate the differentiated Figure VI.1, is two times higher than in Egypt and greater impact of the circumstance variables on different household than inequality of opportunity levels in many Latin Amer- welfare dimensions and to get a more comprehensive under- ican countries.40 standing of inequality of opportunity in Tanzania. The circum- stances included are gender, age, mother’s and father’s edu- The degree of inequality of opportunity, estimated using cation, age at which father and/or mother died, and region the mean log deviation (Theil_L) index, should be con- of birth. The chapter explores also the effects of community sidered as a lower-bound estimate of the true level of in- equality of opportunity.41 Despite the relative richness of the circumstance variables in the datasets, many relevant Figure VI.1  C  onsumption Inequality and circumstances, such as parental employment and occupa- Inequality of Opportunity tion status, family wealth, quality of parents’ education, and so forth remain unobserved. Adding more circumstance 0.30 variables would increase the magnitude of inequality of Mean logarithmic deviation opportunity. 0.25 0.20 Around 25 percent of consumption inequality can be at- tributed to unequal opportunities associated with only ob- 0.15 served Tanzanian households’ circumstances. This is a quite 0.10 sizeable share by SSA standards, where inequality of oppor- 0.05 tunity share is estimated at 12 percent in Ghana, 15 percent 2008 2010 2012 in Côte d’Ivoire, and 21 percent in Madagascar.42 It is almost round on par with the levels in Latin American countries.43 Opportunity inequality Overall inequality Con dence/Interval Con dence/Interval 40 See Barros et al. (2009) for inequality of opportunity estimates, Opportunity Opportunity Opportunity inequality inequality inequality inequality inequality inequality based on labor earnings and household consumption and income, for several Latin American countries, and Belhaj Hassine (2011) for Total Total Total inequality of opportunity in labor earnings in Egypt. 0.238*** 0.058*** 0.243*** 0.054*** 0.275*** 0.068*** Theil_L is the only inequality measure with a path-independent 41  (0.007) (0.005) (0.008) (0.005) (0.008) (0.005) decomposition, see Appendix 5 for more details 42  Forthcoming in the poverty flagship report for Africa. Source: NPS 2008, 2010, and 2012. * Significant at the 10 percent level; ** significant at the 5 percent 43 In a study by Ferreira and Gignoux (2011), the opportunity level; *** significant at the 1 percent level. Numbers in parenthe- shares of consumption inequality were found to range between ses are bootstrap standard deviations based on 100 replications. 24 percent in Colombia and 39 percent in Panama. 62 Tanzania Mainland Poverty Assessment Share of Inequality of Opportunity in Tanzania Mainland and by Region Figure VI.2   0.25 Mean logarithmic deviation 0.20 0.15 0.10 2008 2010 2012 round Opportunity inequality share Tanzania Opportunity inequality share rural Opportunity inequality share urban Con dence/Interval Con dence/Interval Con dence/Interval 2008 2010 2012 Tanzania Urban Rural Tanzania Urban Rural Tanzania Urban Rural 0.244*** 0.240*** 0.176*** 0.224*** 0.206*** 0.148*** 0.248*** 0.197*** 0.147*** (0.018) (0.031) (0.017) (0.016) (0.021) (0.019) (0.016) (0.028) (0.013) Source: NPS 2008, 2010, and 2012. * 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. Figure VI.2 shows that the contribution of inequality of op- greater influence on households and individuals with higher portunity to total inequality is increasing over time at the levels of education and engaged in more diversified occu- national level. pations and jobs than is the case in urban sectors. Second, to the extent that some unobserved circumstances (such Unlike HBS data, NPS shows a slight increase in overall con- as family composition, parents’ financial and asset situation, sumption inequality from 0.24 in 2008 to 0.28 in 2012 (and and so forth) shape the opportunity sets for rural Tanzani- from 0.38 to 0.40 using the Gini index). Inequality of oppor- ans, the estimates of inequality of opportunity excluding tunity followed roughly the same pattern but increased these circumstances are significantly biased downward. more steeply, inducing a larger increase of the opportunity shares. In general, the patterns of inequality of opportunity The contribution of unequal opportunities to the consump- are relatively stable due to the little variations in the circum- tion disparity declined over time in the rural and urban sec- stances variable over short periods of time, but the results tors. Overall and opportunity inequalities declined in the here show quite sizeable changes in inequality of opportu- urban areas between 2008 and 2012, and as opportunity in- nity levels over the past four years.44 equality declined more steeply this induced a reduction of the opportunity share. However, in rural areas both overall The incidence of inequality of opportunity is lower in rural areas than in urban sectors. Opportunity shares of inequality 44 Studies by Lefranc et al. (2008), Barros et al. (2009) on several Lat- are almost 1.5 times higher in urban than in rural areas. This in American countries, and Belhaj Hassine (2011) on Egypt show reflects two facts. First, family background variables have quite stable patterns in inequality of opportunity levels over time. Inequality of Opportunity 63 and opportunity inequality increased during 2008–12, but higher than that of community characteristics and is almost overall inequality increased faster, which involved a decline double of this latter in rural areas. of the opportunity share. The factors contributing to the variation of inequality of opportunity in the urban and rural The contribution of family background and community sectors are explored more in detail below. characteristics to inequality of opportunity both increased in 2012 at the national level and in rural areas. Family back- In addition to family circumstances, community character- ground is also increasing slightly in urban areas while the istics also impact on people’s income prospects, and the contribution of community characteristics is declining, sug- disparity of infrastructure facilities and basic services across gesting a possible convergence in infrastructure and service local communities contribute to the disparity of welfare in provision between the urban communities. the country. However, community characteristics cannot be considered as being beyond adult individuals’ control, as- The following material turns to the partial contributions suming that they can migrate, influence public decisions, of individual circumstances, and groups of circumstanc- and so forth, and therefore these cannot be accounted for es, to inequality. Being able to distinguish between these in the opportunity inequality share. sources of inequality of opportunity is important for for- mulating policies that reduce it. The parametric approach Policy actions to address the influence of family back- allows the estimation of the partial effects of individual ground on the distribution of welfare generally differ from circumstances on outcomes, by fixing one or a group of actions to address the influence of community character- circumstances at their mean values while allowing others istics, the first being a longer term mission that is often to vary. more complex. Thus, from a policy perspective it is import- ant to understand how family background and communi- Of all observed circumstance variables, father’s education ty characteristics affect individuals’ income and consump- is associated with the largest shares of consumption in- tion and to compare their effects on the distribution of equality. The analysis of the contribution of individual cir- welfare. cumstances, reported in Figure VI.4, shows that inequality of opportunity related to father’s education increased from This study examines in the following sections the share of 11 percent in 2008 to 15 percent in 2012 at the national consumption inequality arising from family background level. and community characteristics in Tanzania Mainland as well as in urban and rural areas separately.45 Inequality of opportunity resulting from region of birth, which had the largest share in 2008, slightly declined from 12 per- Family background variables explain a greater share of in- cent to 9 percent during 2008–10 and then increased again equality than community characteristics. The share of family to 11 percent in 2012. Mother’s education also plays an im- background exceeds 15 percent at the national level and portant role in determining inequality, accounting for near- is around two times that of community characteristics in ly 10 percent of total inequality for the entire population. It the rural areas. Although the contribution of family back- ground is underestimated due to the absence of informa- 45 Family background group includes father’s and mother’s edu- tion on parental occupation and employment status, their cational attainment, whether one or both parents of the head live financial situation, asset ownership and so forth, it appears with the household, and whether the head lost his father and/or to be associated with the largest shares of overall inequal- his mother before the age of 15. The community characteristics ity (Figure VI.3). Inequality due to family background varies group includes the distance to regional or district headquarters, distance to health centers, distance to primary and to secondary between 15 and 19 percent across the three waves of the schools, distance to main markets; the presence and amounts of survey, while the contribution of community characteris- investment projects for schooling, irrigation water provision, and tics barely exceeded 10 percent. At the urban and rural lev- infrastructure development; the sources of drinking water; and ac- els, the contribution of family background to inequality is cess to electricity. 64 Tanzania Mainland Poverty Assessment Figure VI.3  Contributions of Family Background and Community Characteristics to Inequality 0.20 Mean logarithmic deviation 0.15 0.10 0.05 0 2008 2010 2012 round Family background share Tanzania Family background share urban Family background share rural Community characteristics share tanzania Community characteristics share urban Community characteristics share rural 2008 2010 2012 Tanzania Urban Rural Tanzania Urban Rural Tanzania Urban Rural Family Background 0.157*** 0.160*** 0.0528*** 0.146*** 0.107*** 0.0581*** 0.187*** 0.150*** 0.0876***   (0.016) (0.030) (0.013) (0.015) (0.024) (0.013) (0.016) (0.027) (0.013) Community Characteristics 0.0914*** 0.126*** 0.0246** 0.107*** 0.0937*** 0.0308*** 0.107*** 0.0812*** 0.0358***   (0.006) (0.010) (0.008) (0.006) (0.011) (0.006) (0.006) (0.007) (0.005) Source: NPS 2008, 2010, and 2012. * 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. declined slightly in 2010 but regained importance according natives is shrinking over time. As there seems to be an in- to the latest survey. Gender makes a limited contribution to crease in employment opportunities outside agriculture in inequality but seems to have gained importance during the the rural areas, the influence of parents’ education on efforts last year of the survey, reflecting the appearance of a possible and welfare became more apparent. Although the effect of form of discrimination against women in welfare distribution. parents education remains weak in rural areas compared to its effect in urban regions, as more than 80 percent of Mother’s education is among the most important factors household heads have parents with an education level of shaping opportunity in urban areas. Mother’s education ac- two years or less, it is catching up quickly to the levels in the counts for around 9 percent of urban inequality. However urban areas. The contribution of mother’s and father’s edu- its effect is declining over time while the effect of father’s cation to opportunity inequality in rural areas increased to education is increasing quite importantly. over 6 percent in 2012, getting closer to the levels observed in the urban areas. This effect can be expected to increase as In rural areas, inequality is shaped mainly by the region of the share of rural households with more educated parents birth, but its contribution is declining over time while the is expanding over time. The contribution of gender is also influence of father’s and mother’s education is increasing. increasing over time, indicating that the disadvantage of This indicates that the wide disparity in welfare between being a women is more apparent in the recent years. people who were born in other regions and moved and the Inequality of Opportunity 65 The Contribution of Individual Circumstances to Inequality of Opportunity Figure VI.4   Shares of Individual Circumstance Variables Shares of Individual Circumstance Variables in Consumption Inequality in Consumption Inequality Tanzania Mainland Urban Areas 0.15 0.15 Mean logarithmic deviation Mean logarithmic deviation 0.10 0.10 0.05 0.05 0 0 2008 2010 2012 2008 2010 2012 round round Shares of Individual Circumstance Variables in Consumption Inequality Rural Areas 0.15 Mean logarithmic deviation 0.10 0.05 0 2008 2010 2012 round Contribution of gender Contribution of father education Contribution of mother education Contribution of birth region Tanzania Mainland Urban Rural 2008 2010 2012 2008 2010 2012 2008 2010 2012 Gender 0.014*** 0.008*** 0.014*** 0.017** 0.018* 0.005 0.014*** 0.005* 0.021***   (0.001) (0.002) (0.001) (0.006) (0.009) (0.003) (0.002) (0.002) (0.003) Mother Education 0.115*** 0.087*** 0.110*** 0.114*** 0.060* 0.084** 0.043*** 0.024* 0.060***   (0.016) (0.016) (0.015) (0.023) (0.026) (0.031) (0.010) (0.012) (0.011) Father Education 0.108*** 0.100*** 0.145*** 0.083** 0.047** 0.105*** 0.035** 0.045*** 0.067***   (0.017) (0.014) (0.015) (0.026) (0.025) (0.027) (0.011) (0.011) (0.011) Region of Birth 0.123*** 0.088*** 0.109*** 0.083** 0.065** 0.080*** 0.135*** 0.088*** 0.082***   (0.014) (0.013) (0.011) (0.026) (0.023) (0.020) (0.017) (0.016) (0.012) Source: NPS 2008, 2010, and 2012. * 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. 66 Tanzania Mainland Poverty Assessment Inequality of Opportunity in II.  Rural opportunity inequality shares are much lower than urban ones. As in the consumption-based analysis, the op- Household Income portunity shares are found to be much higher in urban ar- eas than in rural sectors. Inequality of opportunity share in- Overall inequality and inequality of opportunity levels creased slightly in the urban areas between 2008 and 2010 for household incomes are higher than for inequality for and the declined, while it kept declining over time in the household consumptions. Total income inequality is con- rural zones. siderably higher than consumption inequality, supporting the view that consumption expenditures are more accu- Family background is playing a more important role in rately measured and considered to be more reliable than shaping income inequality than are community charac- income data. Moreover, current incomes tend to be more teristics. In a country where private businesses and house- volatile and more sensitive to macroeconomic fluctuations hold enterprises are important sources of livelihoods, one than consumption and expenditures, which are likely to be would expect income disparities to be more affected by closer to permanent income.46 The levels of income-based community characteristics than parental ones, but the inequality of opportunity are higher than the levels of con- results displayed in Figure VI.6 show that the share of in- sumption-based inequality of opportunity, but the gap be- equality associated with family circumstances is higher tween the measures is much lower than the gap between than the share associated with community features. Nev- overall income and consumption inequality measures. (See ertheless, the contribution of community characteristics is figure VI.5 for more.) increasing over time and almost doubled between 2008 and 2012 at the national level and in urban areas. Fam- Income opportunity inequality shares are lower than con- ily background effect increased also over time but less sumption opportunity shares. While the levels of inequality sharply. of opportunity are higher for households’ incomes than for consumptions, the opposite is true for estimates of oppor- Father’s education is once again associated with the largest tunity shares. The share of income opportunity inequality share of opportunity inequality. Figure VI.7 displays a similar varies from the high of 22 percent in 2008 to the low of 13 ranking of the contribution of each circumstance variable to percent in 2012 compared to a share of consumption op- income inequality as that observed for consumption, with portunity inequality of around 24 percent. This is due to the the exceptions that father’s education plays the largest role higher volatility of current incomes and to measurement in shaping opportunities in all areas and that the region of error and idiosyncratic shocks to certain components of birth seems less important. income. Some components of the income aggregates are transitory and cannot be explained by circumstances, and Mother’s education and gender appear also to make a their variance can be misleadingly confounded with the nonnegligible contribution to inequality, and its impact part of income inequality due to effort (Barros et al., 2009; is increasing over time at the national level and in rural Aaberge, Mogstad, and Peragine, 2011). sectors. On can see from Table 5-3 in Appendix 5 that fe- male-headed households have significantly lower incomes The opportunity shares of income inequality are declining than male-headed ones, indicating the engagement of over time. Unlike consumption, total income inequality Tanzanian women in low productivity and low remunera- and opportunity inequality levels are both declining over tion jobs and businesses. This situation does not seem to time, but opportunity inequality is declining more sharply. be improving over time despite some policy measures for This led to a reduction of the share of opportunity inequal- empowering women. ity in producing income inequality. Also, the higher volatil- ity of income compared to consumption induced a higher volatility over time in the estimates of income opportunity shares. 46  See Barros et al. (2009). Inequality of Opportunity 67 Inequality of Opportunity in Income Figure VI.5   0.30 Mean logarithmic deviation 0.25 0.20 0.15 0.10 0.05 2008 2010 2012 round Opportunity inequality share Tanzania Opportunity inequality share rural Opportunity inequality share urban Con dence/Interval Con dence/Interval Con dence/Interval 2008 2010 2012 Tanzania Urban Rural Tanzania Urban Rural Tanzania Urban Rural 0.219*** 0.225*** 0.178*** 0.175*** 0.234*** 0.131*** 0.134*** 0.172*** 0.082*** (0.023) (0.046) (0.026) (0.024) (0.046) (0.020) (0.018) (0.033) (0.017) Source: NPS 2008, 2010, and 2012. * 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. The effect of parental education and family background on affect their chances to move up the economic ladder. With- economic (consumption and income) outcomes indicates out additional policy actions, there are little chances for the significant problems of intergenerational transmission of next generations to spring out of the poverty and inequality inequality and poverty. Father’s and mother’s education to lived by their parents, engendering poverty and inequality a large extent shape opportunities for their children and traps in the country. 68 Tanzania Mainland Poverty Assessment  ontributions of Family Background and Community Characteristics to Income Figure VI.6  C Inequality 0.10 Mean logarithmic deviation 0.08 0.06 0.04 0.02 0 2008 2010 2012 round Family background share Tanzania Family background share urban Family background share rural Community characteristics share tanzania Community characteristics share urban Community characteristics share rural 2008 2010 2012 Tanzania Urban Rural Tanzania Urban Rural Tanzania Urban Rural Family background 0.062*** 0.079* 0.022 0.076*** 0.076** 0.055** 0.084*** 0.072** 0.059***   (0.016) (0.039) (0.018) (0.015) (0.027) (0.020) (0.013) (0.025) (0.010) Community 0.027 0.027 0.042*** 0.038 0.021** 0.005 0.056*** 0.065* 0.051*** characteristics (0.020) (0.023) (0.009) (0.023) (0.007) (0.026) (0.013) (0.029) (0.014) Source: NPS 2008, 2010, and 2012. * 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 of Opportunity 69 Figure VI.7  The Contribution of Individual Circumstances to Income Inequality of Opportunity Shares of Individual Circumstance Variables Shares of Individual Circumstance Variables in Income Inequality in Income Inequality Tanzania Mainland Urban Areas 0.08 0.08 Mean logarithmic deviation Mean logarithmic deviation 0.06 0.06 0.04 0.04 0.02 0.02 0 0 2008 2010 2012 2008 2010 2012 round round Shares of Individual Circumstance Variables in Income Inequality Rural Areas 0.06 Mean logarithmic deviation 0.04 0.02 0 2008 2010 2012 round Contribution of gender Contribution of mother education Contribution of birth region Contribution of father education Tanzania Mainland Urban Rural 2008 2010 2012 2008 2010 2012 2008 2010 2012 Gender 0.001 0.023*** 0.021*** 0.017 0.045*** 0.016 0.004 0.020*** 0.039***   (0.005) (0.004) (0.004) (0.021) (0.012) (0.011) (0.006) (0.005) (0.005) Mother education 0.025 0.047*** 0.037*** 0.051** 0.050* 0.002 0.006 0.028** 0.037***   (0.014) (0.012) (0.010) (0.019) (0.021) (0.018) (0.011) (0.010) (0.007) Father education 0.055*** 0.070*** 0.077*** 0.072** 0.059** 0.060* 0.014 0.055*** 0.050***   (0.013) (0.013) (0.011) (0.023) (0.022) (0.024) (0.014) (0.016) (0.008) Region of birth 0.039* 0.047** 0.038** 0.085 0.057 0.029 0.016 0.041 0.042**   (0.017) (0.015) (0.012) (0.062) (0.031) (0.024) (0.023) (0.025) (0.014) Source: NPS 2008, 2010, and 2012. * 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. 70 Tanzania Mainland Poverty Assessment Inequality of Opportunity 71 Chapter  7 Demographic Pressures Pose a Challenge to Poverty Reduction Key Messages ➤➤ Tanzania is in the very early stages of demographic transition, but demographic pressures will continue to affect economic growth and poverty reduction prospects. ➤➤ Women’s empowerment, through education and employment, can contribute to the control of fertility. The persistence of high population growth in Tanzania Malthus some 200 years back). The “population-alarmist” weighs heavily on the country’s future economic growth view of the 1950s and ‘60s, that rapid population growth and its capacity to reduce poverty. At the national level, de- inhibits economic development, was challenged in the mographic pressures pose challenges for public service pro- 1980s by the revisionists who drew more nuanced conclu- vision, labor markets, land, resources, and so forth and can sions about the impact of population growth and argued put a brake on growth in per capita incomes. At the house- that economies could accommodate demographic change hold level, it affects the ability of families with a large number through compensating technology and institutional of children to reduce poverty. This chapter analyses more in change (Birdsall 2001; Fox 2009).47 During the 1990s, the fo- detail the relationship between demography and poverty in cus of the literature moved away from studying the impact Tanzania. The first section presents the macro perspective of the relationship. The second section analyzes the demo- 47 The change in mindset also reflected, at least in parts, differ- graphic transition in the country. The third section examines ent theoretical models of economic growth, where the emphasis the determinants of fertility and the last section summarizes had shifted from physical capital accumulation to technological the main findings of the chapter and explores directions for change as the key driver of economic progress. Most economic further research on demographic transition and fertility. growth models converge on the view that population growth puts a brake on physical and human capital accumulation, with nega- tive implications on (per capita) income growth, but the nature of I.  Macro Perspective this effect depends on the specification of the production function. Models that assume a fixed capital-output coefficient (such as the traditional Harrod Domar model) or complementarities between At the macro level, the relationship between demography human and physical capital (as some endogenous growth mod- and economic growth has been a heated topic for decades els) tend to find larger effects than the neoclassical Solow model, (or even centuries, considering the early work by Thomas which assumes declining returns to capital (see Klasen 2005). 73 of aggregate population growth to disentangling the dif- growth channel in developing and transitional economies. ferential effects of various components of demographic On the other hand, Iceland (2003) finds that over the period change. Studies by Kelley and Schmidt (1995) and William- 1960–90 poverty became more responsive to demographic son (2001) suggest that increases in population density and shifts in the United States as the elasticity of poverty to in- the share of the working-age population are positively asso- come growth decreases with an increase in income and a rise ciated with growth, while increases in the share of depen- in the number of female-headed households. They find an in- dent children have a negative association. crease in elasticity between poverty and growth in the 1990s, when poverty became more responsive to the upward trends Today a consensus of sorts has emerged that it is crucial of economic growth, as the shifts in family structure slow and for economies to go through a demographic transition in empirically, the effect of family structure disappears. order to benefit from a demographic window of opportu- nity. The passage from the first phase (high fertility, declin- We use a global model to illustrate the potential growth pay- ing mortality) to the second phase (declining fertility) of the offs of the demographic transition for the case of Tanzania. transition goes hand in hand with favorable changes in the This is based on the demographic forecasting model pro- age structure of society, particularly a lower ratio of depen- posed by Lindh and Malmberg (2007). The objective of the dent children and elderly to working-age adults. This affects analysis is not to project actual GDP levels into the future, as growth through three distinct channels: (i) mechanically GDP change would be affected by many factors other than through the higher ratio of the labor force to the total pop- demography, but to isolate the potential impact of demo- ulation, (ii)  through higher savings rates among working graphic change on per capita income growth. The results of adults (who can build up more capital for retirement due this exercise should be viewed as indicative and approximate to the declining number of children), and (iii) through a de- given uncertainties about the underlying parameters, the re- mand-driven investment boom as the working-age popula- duced form nature of the estimation which does not capture tion requires housing, machinery, and infrastructure. Bloom the structural characteristics of the country, and the difficulty and Williamson (1998) argue that as much as half of the East to establish causality in a cross-country framework.48 Asian economic miracle over the period 1965–90 can be ex- plained through the lens of population dynamics. As a first step, a statistical relationship between per capita GDP and demographic variables is estimated using panel The magnitude of this “demographic gift” depends on the data. GDP per capita at 2005 PPPs (from the Penn World Ta- pace of fertility decline and complementary policies. The ble 7.0) is estimated as a function of life expectancy at birth faster the reduction in fertility, the larger the demographic and age structure (from United Nations 2014). The regression gain the country may experience during the transition peri- is estimated on a sample of 108 countries that had at least od. The economic policy framework also plays an important 20 observations for the period 1950–2009. The panel nature role because the economic benefits from a growing labor of the models make it possible to control for unobserved force can only materialize if the economy can absorb the heterogeneity across countries and common time-specific additional workers productively. In this respect East Asia’s effects, such as the world business cycle, through country- export-led and labor-intensive growth model clearly was a and time-specific fixed effects (see Appendix 6 for further factor that contributed to the demographic dividend. details). While this allows for some flexibility, the model still relies on the simplifying assumption that the relationship Research suggests that there are differential effects of de- between per capita GDP and demographic variables is the mographic transition on the elasticity of poverty to growth, same across countries. which may contribute to potential explanations for the pro- poor trends of growth observed in Tanzania. For example, 48 Some caveats are in order. While the model’s forecasting perfor- Lipton and Eastwood (2014) find that the growth effect on mance is adequate on average, it does not predict well Tanzania’s poverty is largest in high-fertility and low-income countries, historical growth trajectory. This reconfirms the notion that coun- when looking at the impact of fertility on poverty via the try-specific factors, particularly policies, play an important role. 74 Tanzania Mainland Poverty Assessment We then use the 2012 United Nations population projec- fertility using data from the 2010 Demographic and Health tions for Tanzania to simulate per capita income trends over Survey; this allows identifying policies that may accelerate the period 2010–50. These population projections, which the demographic transition. are produced by the UN’s Population Division, show demo- graphic trends—in terms of changes in the population’s age The focus on fertility is warranted by the following reasons: composition and life expectancy—under different assump- First, while fertility, mortality, and the age structure of the tions about trends in total fertility. In particular, the high, female population jointly determine population growth at medium and low fertility variants assume that the total fer- the macro level, fertility decline is generally regarded as the tility rate declines from 5.58 in 2005–10 to 3.84, 3.34, or 2.84 primary demographic momentum that triggers the change children per women by 2045–50, while the constant fertility in age structure and induces the second phase of the de- variant assumes that fertility stays at the 2005–10 level. Gen- mographic transition. Second, public interventions that erally, the medium variant, which is based on probabilistic reduce the fertility rate—such as female education, access model of fertility change over time, is considered the most to reproductive health and family planning services, and so likely scenario; the high and low fertility variants are simply forth—are the key means through which governments can projected as 0.5 above and below the medium variant. The lower the rate of population growth. effect of demographic change on economic growth is iso- lated as the difference between simulated GDP per capita growth (over the period 2010–50) under the low/medium/ The Demographic Transition in II.  high fertility variant and simulated GDP per capita growth Tanzania over the same period under the constant fertility variant. In other words, we are interested in the predicted change in With a population growth rate of 2.7 percent per year, Tan- per capita growth induced by a reduction in fertility below zania’s population increases rapidly, albeit at a rate similar to the level that was found in 2005–10. other African counties. The population growth rate reported here, which is taken from the 2012 Population and Housing The results suggest that reductions in fertility significantly Census, matches the average for SSA in the World Develop- accelerate per capita income growth. A reduction in fertility ment Indicator (WDI) database. However, since the WDI are to 3.34 children per women under the medium population typically updated with some lag, population growth is most scenario is predicted to increase per capita income growth likely above the SSA average.49 by 1.3 percentage points per year over the period 2010–50. As expected, the growth pay-off is larger for the low fertil- Population growth will remain high over decades to come. ity variant (+1.9 percentage points per year) and lower for Figure VII.1 shows the 2012 official UN population estimates the high fertility variant (+0.8 percentage points per year). and projections for Tanzania under different scenarios about All this suggests that Tanzania could reap significant eco- fertility trends (low, medium, and high fertility).50 These pro- nomic benefits from a reduction in fertility and accelerated jections suggest that Tanzania’s population will be in the demographic transition, which would accelerate per capita income growth and poverty reduction. 49 The WDI database also still reports a population growth rate of 3 percent for Tanzania. The next section explores where Tanzania stands in terms 50 The figures are based on the 2012 revision, which reports esti- of its demographic transition and analyzes factors that are mates for the period 2005–10 and projections up to 2100. Since associated with fertility. First we draw on UN population we are interested in medium trends we focus on the projections up to 2050. We do not report the constant fertility variant, because projections to examine the demographic trends (in terms this projection assumes that the total fertility rate remains at its of overall population size, population density, and changing level in 2005–10, which is rather unlikely. The projections also make age composition) that Tanzania can expect over the com- assumptions about mortality trends in terms of life expectancy at ing decades (focusing on the period until 2050). Second we birth by sex and international migration. See United Nations 2014 analyze current patterns and intermediate determinants of for details. Demographic Pressures Pose a Challenge to Poverty Reduction 75 Tanzania’s Population Is Figure VII.1    opulation Density Will Be Figure VII.2  P Projected to Reach 100 Million Similar to China’s by 2050 around 2040 200 Persons per square km 150 125 150 100 100 Million 75 50 50 25 0 0 2010 2015 2020 2025 2030 2035 2040 2045 2050 1950 1960 1970 1980 1990 2000 2010 2020 2030 2040 2050 Tanzania-Low fertility China Low fertility Medium fertility High fertility Tanzania-Medium fertility Western Europe Tanzania-High fertility USA Source: United Nations 2014. Source: United Nations 2014. Note: Based on UN medium fertility scenarios for countries other than Tanzania. range of 114 to 145 million by 2050, compared to around 45 million in 2012.51 Even under the low fertility projection, the total population will stabilize only well into the 22nd century.  ortality Has Fallen Rapidly Figure VII.3  M but Fertility Remains High The increase in population size will radically alter Tanzania’s 8 300 Fertility (births per women) economic geography. Population density is expected to in- Under- ve mortality (per 1000 live birth) crease from 48 persons per square kilometer in 2010 to 137 6 200 persons per square kilometer in 2050. The country would 4 then be around 3.3 times more densely populated than the United States and have a similar population density as 100 2 China, though still somewhat lower than Western Europe (Figure VII.2). While this will bring with it certain economic 0 0 1960 2012 advantages—particularly lower unit cost in the provision of Fertility rate (left axis) public infrastructure such as roads, grid electricity or piped Under- ve mortality rate (right axis) water, and greater opportunities for trade—the increase in population density will also exert a significant pressure on Source: World Development Indicators (WDI 2014). agriculture. At present Tanzania is still endowed with large swaths of uncultivated land and past agricultural growth momentum of high population growth would continue for has been largely driven by area expansion. However, there is some time. This is because the fertility rate has surpassed the already evidence that land pressure is emerging in some of replacement rate for many decades, so that an increasing the more productive agricultural regions.52 Tanzania’s high population growth reflects that the coun- 51 The UN projections overestimate the population in 2012 (48 million), compared to 45 million in the 2012 Population and Hous- try is in the early stages of the demographic transition. ing Census. This is because the UN projections were derived before Population growth is naturally high during this phase of the latest census results were released. development. While fertility has not yet come down much, 52 According to the Agricultural Sample Census, the average land child mortality has already fallen rapidly, so that more chil- holding size of rural households in Kilimanjaro region declined by dren survive to adulthood (see Figure VII.3). However, even 22 percent from 2002/03 to 2007/08; Arusha experienced a 17 per- if fertility were to decline immediately, the demographic cent decline. 76 Tanzania Mainland Poverty Assessment Age Structure and Dependency Ratio, 1950–2050 Figure VII.4   Low fertility scenario High fertility scenario 160 1.5 160 1.5 140 140 Dependency ratio Dependency ratio 120 120 100 1.0 100 1.0 Million Million 80 80 60 0.5 60 0.5 40 40 20 20 0 0 0 0 1950 1960 1970 1980 1990 2000 2010 2020 2030 2040 2050 1950 1960 1970 1980 1990 2000 2010 2020 2030 2040 2050 65+ 15–64 0–14 Dependency ratio (right axis) Source: United Nations 2014. number of women will enter the reproductive age group For Tanzania to reap economic gains from a growing labor in each year. force, it needs to accelerate the creation of productive jobs. The total working-age population is projected to increase Tanzania could gain from a demographic dividend starting from the current 23 million in 2012 to between 71 and around 2020–30, but the reduction in the dependency ratio 83  million by 2050—implying that an additional 48  mil- will not match that of East Asia. Figure VII.4 shows that the lion to 59 million people have to be absorbed into the la- dependency ratio is expected to decline by 17 to 34 per- bor force over a 40-year period. As discussed in World Bank cent between 2010 and 2050, depending on the projected (2014), this requires policy actions on several fronts, such as decline in fertility. However, even under the low fertility sce- increasing the growth of nonfarm enterprises, improving nario the rate of decline of the dependency ratio is lower agricultural productivity, and enabling domestic firms to than the rate of reduction that was achieved by Thailand penetrate export markets. and Malaysia over the period 1965–90, suggesting that the economic benefits will also be lower (Figure VII.5). Patterns and Determinants of III.  Fertility in Tanzania The Decline in the Dependency Figure VII.5   From a policy perspective, the key variable needed to re- Ratio Will Be Smaller than in duce population growth and accelerate demographic East Asia change is the fertility rate. While the number of children a family decides to have is not directly amenable to poli- 0.98 1.00 0.92 0.90 cy (with the exception of more coercive policy measures, 0.83 0.77 Dependency ratio 0.80 0.69 0.69 such as China’s one-child policy), there is a large body of 0.60 0.53 evidence that fertility rates respond to the economic and cultural environment. 0.40 0.20 The salience of economic and cultural factors also manifests 0.00 in regional variations in fertility. In 2010, the total fertility rate 2010 2035 (high) 2035 (medium) 2035 (low) 1965 1990 1965 1990 was highest in the Tanzania’s western zone (7.1 children per woman) and lowest in the eastern zone (3.9 children per woman). Families in the eastern part of Tanzania already had Source: United Nations 2014. achieved in 2010 the fertility level projected by the United Demographic Pressures Pose a Challenge to Poverty Reduction 77 Nations for the whole of Tanzania for almost four decades sterility, and postpartum infecundability (affected by into the future (3.84 children per women under the medium postpartum abstinence and duration of breastfeeding). fertility variant in 2045–50). Moreover, while most regions saw a decline in fertility rates between 1996 and 2010, fertili-  Economists traditionally emphasize indirect (or intermedi- ty levels actually increased in the western and central zones. ate) drivers of fertility. Examples are (female) education, (See Figure VII.6). family income, child mortality, culture, the labor force participation of women, and female empowerment. Research has identified the following determinants of fertil- ity at the family level: We follow in the economist tradition and model the relation- ship between fertility and (intermediate) socioeconomic con-  Demographic transition theory emphasizes the causal link ditions. The analysis is based on an econometric approach from high levels of child mortality to high levels of desired developed by the World Bank for Ethiopia (World Bank 2007a) fertility. This link is difficult to pin down from survey and models the total number of children ever born to wom- data, because couples are making their fertility choices en ages 15 to 49 years. The analysis is conducted based on based not on the number of their surviving children but data from the 2010 Demographic and Health Survey. The re- on their perception of the probability of a child’s surviv- gression results are reported in Table 6-3 in Appendix 6. al, which is based on experiences of their community, country, and so forth. However, the link between the Female education is associated with having fewer children two variables has been documented in cross-country in total. Women who have at least completed some prima- analyses (for example, McCord et al. 2010 for SSA and ry education have fewer children than women without any Palloni and Rafalimanana 1999 for Latin America). education. Nonetheless, the effect is relatively small, which may be related to the fact that we are controlling for wheth-  Demographers focus on the direct (proximate) determi- er the woman has ever been married. That is, previous liter- nants of fertility, which are biological and behavioral in ature on other SSA countries (see, for instance, World Bank nature. These include the exposure to the risk of con- 2007a) has found that the main effect of education on fertil- ceiving (percentage of women who are in union), the ity operates is through marital status, such that once marital use of contraceptives (linked in part to the availabili- status was controlled for, the association between educa- ty of services), the rates of abortion and pathological tion and fertility was substantively reduced. Fertility Levels and Trends Differ across Geographic Zones Figure VII.6   8 7.3 7.1 7 6.6 6.5 6.1 6.3 5.9 5.7 5.8 6 5.4 5.4 5.4 5.1 4.9 5 4.6 4.4 4.3 3.9 4 3 2 1 0 Western Central Lake Southern Zanzibar Northern Southern Eastern Total Highlands 1996 2010 Source: DHS Statcompiler 2014. 78 Tanzania Mainland Poverty Assessment Cash employment of women is also linked to lower fertility. Urban location is associated with lower fertility, and other The regressions show that women who receive any cash earn- regional variation also remains important. Rural women ings have fewer children than women who are not employed have had about 0.18 more children than their urban coun- or who receive only in-kind earnings. This effect points to the terparts, everything else being equal. Furthermore, most role of female empowerment, as a women’s cash earnings are regional fixed effects are significant and (apart from Mtwara linked to their bargaining position within the family. and Lindi) positive, indicating that families living in Dar es Salaam (the reference category in the regression model) Early sexual life initiation is associated with higher total fer- tend to have lower fertility levels than families in other parts tility. An increase in the age at first sexual intercourse by one of the country. year is associated with almost 0.2 children less children in total. However, it should be noted that the median age at the first sexual intercourse among women ages 20–49 in the Main Findings and Directions IV.  sample is 17.4, so that large increases in age seem unrealistic. for Further Analysis Poverty is an important correlate of fertility. Women in the The analysis in this chapter indicates that the acceleration richest 20 percent of households have fewer children than of the demographic transition could be beneficial for the their counterparts in the bottom 20 percent. This might be Tanzanian economy. The following are some implications of because better-off people face lower infant and child mor- those findings in terms of possible follow up research and tality rates and, thus, as the demographic transition theory relevant interventions aimed at maximizing the benefits states, have lower desired (and actual) fertility. However, as from the demographic transition. discussed in the introduction to this chapter, the causality might also go in the opposite direction, as high levels of Female education, especially at the secondary level, has fertility (and large numbers of dependent children) make it a strong link with fertility in Tanzania. Furthermore, there more difficult for families to escape from poverty. is a potential positive feedback loop between increased fe- male education and reduction in fertility. That is, when wom- The role of access to family planning services appears in- en receive more education they tend to have fewer children, conclusive. In the regressions, women with an “unmet need which in turn gives them an opportunity to receive more ed- for contraception”—defined as those who do not want to ucation. Interventions to help keep girls in secondary school have any more children (limiters) or want to wait at least may thus have an impact on fertility. These may range from two years before having another child (spacers) but are conditional cash transfer programs to supply-side interven- not using contraception—have more children than other tions such as the expansion of the school system in rural ar- women, which suggests that lack of access to family plan- eas and the expansion of other types of infrastructure that ning might play a role. However, there is some evidence of are known to have large spill-over effects on education (for reverse causality, in the sense that women who have been example, road construction, improvement of sanitary and more fertile in the past are less likely to want any more chil- health infrastructure, and interventions to increase food se- dren at present (and are hence more likely to be “limiters” curity).53 Furthermore, specific programs aimed at empower- or “spacers”). Further analysis also shows that women who ing girls and making them aware of their own worth and hu- are currently using contraception have had more children man rights, as well as gender equality awareness programs than other women. All this casts doubt on whether the lack for all, have been used in similar contexts and added to the of access to family planning methods really predates high fertility. Moreover, only 25 percent of the women in the 53 Unconditional cash transfer programs have also recently been sample use any kind of contraception (including traditional attracting attention from policy actors. In that case, however, no methods), although 85 percent of women know a source of effect has been found on school enrolment. See, for instance, the contraception (private or governmental clinics, NGOs, reli- program administered by the NGO GiveDirectly in Kenya, whose gious associations, and so forth). evaluation is undertaken in Haushofer and Shapiro (2013). Demographic Pressures Pose a Challenge to Poverty Reduction 79 school curriculum in order to respond to parents’ reluctance mortality to high desired fertility. To establish this in the case to send their daughters to school. of Tanzania, one would need to analyze full fertility histo- ries in order to see what the desired number of children for Even controlling for other variables, regional variation each woman was before starting child bearing, and wheth- in fertility remains large in Tanzania. Exploring the pos- er she responded with higher fertility given child mortality sibility of addressing unmet demand for family planning experiences so as to meet her original desired fertility. This may be particularly relevant not only among young wom- is not possible to investigate as women interviewed in a en but also among women in rural areas who do not have given wave of the Demographic and Health Survey are not easy access to health facilities. Policy options in this area can necessarily interviewed in future waves. Redesigning the fu- be considered in ways that do not affect women’s ability to ture DHS surveys would be useful, but meanwhile available make decisions on child bearing and range from the supply cross-country evidence appears sufficient to hypothesize side (for example, ensuring that contraceptives are available that such a links is likely to exist in Tanzania. and affordable throughout the country) to the demand side (for example, carrying out information campaigns on the Experience from East and Southeast Asian countries use of contraceptives and the lack of adverse side effects would help to inform policies to take advantage of the of their use for both men and women). Regarding the latter, expected demographic dividend in Tanzania. In 1960, it may be necessary to design strategies for reaching out to South Korea, Hong Kong, Singapore, and Thailand had to- rural areas that lack access to electricity and where it is not tal fertility rates (TFR) greater than or equal to five children possible to use sophisticated communication means. per woman (and higher than six in Thailand). In 2010, all these countries had TFRs around 2.1 children per woman, While theory and cross-country evidence emphasize and most of them had already reached such low levels in the causal link from high child mortality to high desired the 1990s. During the same time, these countries benefit- fertility, the data at hand are insufficient to establish ed from spectacular economic growth rates. When the first such strong relationship in Tanzania. However, there is window of demographic opportunity became available, considerable scope for further reducing infant and child public authorities in Asia seized the opportunity. They com- mortality and this may arguably have an impact on fertility. plemented demographic changes with energetic policies Indeed, the demographic transition theory and cross-coun- and investments, including but not restricted to the areas of try evidence emphasize the causal link from high child health and education. 80 Tanzania Mainland Poverty Assessment References Adams Jr., R.H. (2004). Economic Growth, Inequality and Beegle, K.G., De Weerdt, J., Friedman, A.J., and Gibson, C.J. Poverty: Estimating the Growth Elasticity of Poverty. (2012). Methods of household consumption measure- World Development 32(12): 1989–2014. ment through surveys: Experimental results from Tan- Adams Jr, Richard H., and Alfredo Cuecuecha. (2013). The zania. Journal of Development Economics, 98: 3–18. impact of remittances on investment and poverty in Belhaj Hassine, N., (2011). Inequality of Opportunity in Egypt. Ghana. World Development 50: 24–40. World Bank Economic Review, 26(2), 265–295. Africa’s Pulse (2013). An analysis of issues shaping Africa’s eco- Birdsall, N. (2001). New Findings in Economics and Demog- nomic future. The World Bank, October 2013 Vol. 8. raphy: Implications for Policies to Reduce Poverty. In Amuedo-Dorantes, C. and Pozo, S. (2010). Accounting for re- National Bureau of Economic Research (NBER), Demo- mittance and migration effects on children’s schooling. graphic and Economic Change in Developed Countries, World Development 38(12): 1747–1759. Columbia University Press. Angrist, J.D. and Pischke, J.S. (2009). Mostly Harmless Econo- Birdsall, N. and Sinding, S.W. (2001). How and Why Popula- metrics: An Empiricist’s Companion. Princeton University tion Matters: New Findings, New Issues. In N. Birdsall, Press. A. C. Kelley and S. W. Sinding (eds.), Population Matters: Atkinson, A. B. and M. A. and Lugo (2010). Growth, Poverty Demographic Change, Economic Growth, and Poverty in and Distribution in Tanzania. Working Paper 10/0831, the Developing World, Oxford University Press. International Growth Centre (IGC), London/Oxford Bloom, D. E. and Williamson, J. G. (1998). Demographic Tran- Barros, Ricardo P., Ferreira, Francisco H.G, Molinas Vega, Jose sition and Economic Miracles in Emerging Asia. World R., Chanduvi, Jaime S., with Mirela de Carvallo, Samuel Bank Economic Review 12(3): 419–455. Franco, Samuel Freije-Rodríguez, and Jérémie Gignoux. Bourguignon, F., Ferreira F.H.G., and Menéndez, M. (2007). (2009). Measuring Inequality of Opportunities in Latin Inequality of Opportunity in Brazil. Review of Income America and the Caribbean. The World Bank, Washing- Wealth 53(4): 585–618. ton D.C. Bradshaw, S. and Quirós Víquez, A. (2008). Women Beneficia- Becker, G. S. (1960). An Economic Analysis of Fertility. In ries or Women Bearing the Cost? A Gendered Analysis National Bureau of Economic Research (NBER), Demo- of the Red de Protección Social in Nicaragua. Develop- graphic and Economic Change in Developed Countries, ment and Change 39 : 823–844. Columbia University Press. Canning, D. (2013). The Demographic Dividend in Africa, Beegle, K.G., De Weerdt, J. and Dercon, S. (2011). Migration Paper presented at the Wilson Center, Washington DC and Economic Mobility in Tanzania: Evidence from a (October 15). Tracking Survey. The Review of Economics and Statis- Christiaensen, L., Lanjouw, P., Luoto, J., and Stifel, D. (2012). tics, 93(3): 1010–1033. Small Area Estimation-based Prediction methods to 81 Track Poverty: Validation and Applications. Journal of Economic Growth, and Poverty in the Developing World, Economic Inequality 10: 267–297. Oxford University Press. Christiaensen, L., and Devarajan, S. (2013). Making the Most Eele, G., Semboja, J., Likwelile, S. and Ackroyd, S. (2000). of Africa’s Growth. Current History, May: 181–87. Meeting International Poverty Targets in Tanzania. De- Christiaensen, L., De Weerdt, J. and Todo, Y. (2013). Urbaniza- velopment Policy Review 18(1) : 63–83. tion and Poverty Reduction: The Role of Rural Diversifi- Elbers, C., Lanjouw, J. and Lanjouw, P. (2003). Micro-Level cation and Secondary Towns. Policy Research WP 6422. Estimations of Poverty and Inequality. Econometrica The World Bank, Washington D.C., USA. 71(1): 355–64. Collier, P., Radwan, S. and Wangwe, S. with Wagner, A. (1986). Ferreira, F.H.G., Gignoux, J. (2011). The measurement of in- Labour and Poverty in Rural Tanzania: Ujamaa and Rural equality of opportunity: theory and an application to Development in the United Republic of Tanzania. Oxford Latin America. Review of Income and Wealth 57(4): University Press, Oxford. 622–657. Cowell, F.A., Jenkins, S.P. (1995). How much inequality can Firpo, S., Fortin, N. M., and Lemieux, T. (2009). Unconditional we explain? A methodology and an application to the Quantile Regressions. Econometrica, 77 (3), 953–973. USA. Economic Journal, 105 : 421–430. Fortin, N. M., Lemieux, T., and Firpo, S. (2010). Decomposition Dabalen, A., Etang, A., Mungai, R., Wambile, A. and Wane, W. Methods in Economics. National Bureau of Economic (2014). Is Poverty Reduction in Africa Underestimated be- Research, WP 16045. cause of Poor Data?. Paper Presented at the CSAE 2014. Foster, J. E., and Shneyerov, A.A. (2000). Path independent Dardanoni, M. and Peracchi, F. (2011). Regression with im- inequality measures. Journal of Economic Theory 91(2): puted covariates: a generalized missing-indicator ap- 199–222. proach. Journal of Econometrics, 162(2) : 362–368. Fox, L. (2009). Poverty and Household Size in Uganda. World Deaton, A. (2005). Measuring Poverty in a Growing World Bank Policy Note, mimeo. (or Measuring Growth in a Poor World). Review of Eco- Gaddis, I. and Hoogeveen, J.G. (2013). Primary Education in nomics and Statistics 87: 1–19. Mainland Tanzania: Poor Kids Left Behind?. World Bank De Brauw, A., Mueller, V. and Woldehanna, T. (2013). Motives Policy Note, mimeo. to Remit: Evidence from Tracked Internal Migrants in Gibson, J., McKenzie, D. and Stillman, S. (2011). The Impacts Ethiopia. World Development, 50: 13–23. of International Migration on Remaining Household Demombynes, G. and Hoogeveen, J.G. (2007). Growth, In- Members: Omnibus Results from a Migration Lottery equality and Simulated Poverty Paths for Tanzania, Program. The Review of Economics and Statistics, 93(4): 1992–2002. Journal of African Economies 16: 596–628.
 1297–1318. De Weerdt, J. and Hirvonen, K. (2013). Risk sharing and in- Green, T. R. (2005). The Political Economy of a Social Transfer ternal migration. Policy Research Working Paper Series Program: Evidence from PROGRESA in Mexico. Paper 6429, The World Bank, Washington D.C., USA. presented at the 63rd Midwest Political Science Asso- DHS Statcompiler (2014). (accessed April 2014) Haushofer, J., and Shapiro, J. (2013). Household Response Dimova, R., Epstein, G. and Gang, I. (2011). Migration, Trans- to Income Changes: Evidence from an Unconditional fers and Child Labor. IZA Discussion Papers 5641, Insti- Cash Transfer Program in Kenya. http://www.iies.su.se/ tute for the Study of Labor (IZA). polopoly_fs/1.162160.1389355591!/menu/standard/ DiNardo, J., Fortin, N.M., and Lemieux, T. (1996). Labor Market file/HaushoferJob%20Market%20Paper.pdf Institutions and the Distribution of Wages, 1973–1992: A Hildebrandt, N. and McKenzie, D. (2005). The effects of migra- Semiparametric Approach. Econometrica, 64: 1001–1044. tion on child health in Mexico. Economia, 6(1): 257–89. Eastwood, R. and Lipton, M. (2001). Demographic Transition Hoogeveen, J.G. and Ruhinduka, R. (2009). Poverty Reduc- and Poverty: Effects Via Economic Growth, Distribution, tion in Tanzania Since 2001: Good Intentions, Few and Conversion. In N. Birdsall, A. C. Kelley and S. W. Sind- Results. Paper Prepared for the Research and Analysis ing (eds.), Population Matters: Demographic Change, Working Group. 82 Tanzania Mainland Poverty Assessment Iceland, J. (2003). Why Poverty Remains High: the Role of Child Mortality and Malaria Ecology. Earth Institute Income Growth, Economic Inequality, and Changes Working Paper, Columbia University. in Family Structure, 1949–1999. Demography, 40(3): McKenzie, D. (2005). Beyond remittances: the effects of mi- 499–519. gration on Mexican households. International migra- Jerven, M. (2011). Growth, Stagnation or Regression? On the tion, remittances and the brain drain, McMillan and Accuracy of Economic Observations, Tanzania, 1961– Palgrave: 123–47. 2001. Journal of African Economies 20(3): 377–394 McKenzie, D., Stillman, S. and Gibson, J. (2010). How Import- Jerven, M. (2013). Poor Numbers: How We are Misled by African ant is Selection? Experimental VS. Non-Experimental Development Statistics and What to Do About It. Studies Measures of the Income Gains from Migration. Journal in Political Economy, Cornell University Press. of the European Economic Association, 8(4): 913–945. Kabeer, N. (2010). Can the MDGs Provide a Pathway to So- McKenzie, D. and Rapoport, H. (2007). Network effects and cial Justice? The Challenges of Intersecting Inequalities. the dynamics of migration and inequality: Theory and New York: UN MDG Achievement Fund: Brighton: IDS. evidence from Mexico. Journal of Development Eco- Kelley, A. C. and Schmidt, R. M. (1995). Aggregate Population nomics, 84(1):1–24. and Economic Growth Correlations: The Role of the McKenzie, D. and Sasin, M. (2007). Migration, remittances, Components of Demographic Change. Demography poverty, and human capital: conceptual and empirical 32(4): 543–555. challenges. Policy Research Working Paper Series 4272, Kessy, F., Mashindano, O., Shepherd, A. and Scott, L. (Eds) The World Bank. (2013). Translating Growth into Poverty Reduction: Be- Merrick, T. (2001). Population Matters: Demographic yond the Numbers. Mkuki na Nyota, Dar es Salaam. Change, Economic Growth, and Poverty in the Devel- Klasen, S. (2005). Population Growth, (Per Capita) Economic oping World. In N. Birdsall, A. C. Kelley and S. W. Sind- Growth, and Poverty Reduction in Uganda: Theory and ing (eds.), Population Matters: Demographic Change, Evidence. mimeo. Economic Growth, and Poverty in the Developing World, Lefranc, A., Pistolesi, N. and Trannoy, A. (2008). Inequality of Oxford University Press. Opportunities vs. Inequality of Outcomes: Are Western Ministry of Finance and Economic Affairs (2009). An Analy- Societies All Alike?. Review of Income and Wealth 54(4): sis of Household Income and Expenditure in Tanzania. 513–46. Poverty and Human Development Report Brief 4: Lindh, T. and Malmberg, B. (2007). Demographically Based Mkenda, A F, E. G. Luvanda, and R. Ruhinduka (2010). Growth Global Income Forecasts Up to the Year 2050. Interna- and Distribution in Tanzania: Recent Experience and tional Journal of Forecasting 23: 553–67. Lessons. Interim Report to REPOA, Dar es Salaam. Lokshin, M., Bontch-Osmolovski, M. and Glinskaya, E. Molyneux, M. (2006). Mothers at the Service of the New Pov- (2010). Work-Related Migration and Poverty Reduc- erty Agenda: Progresa/Oportunidades, Mexico’s Condi- tion in Nepal. Review of Development Economics, tional Transfer Programme. Social Policy and Adminis- 14(2): 323–332. tration 40: 425–449. Machado, J.F., and Mata, J. (2005). Counterfactual Decompo- Muzzini, Elisa and, Wietze Lindeboom. (2008). The urban sition of Changes in Wage Distributions Using Quan- transition in Tanzania : building the empirical base for pol- tile Regression. Journal of Applied Econometrics, 20, icy dialogue. Washington, DC: World Bank. 445–465. National Bureau of Statistics. (2006). Census Analytical Report, Mashindano, O., Kayunze, K., da Corta, L. and Maro, F. (2011). Volume X. Tanzania. Agriculture Growth and Poverty Reduction in Tanzania NBS and ICF Macro (2011). Tanzania Demographic and 2000–2010: Where Has Agriculture Worked for the Poor Health Survey (DHS). National Bureau of Statistics (NBS) and What Can We Learn from This?. Chronic Poverty Re- and ICF Macro, Dar es Salaam and Calverton, MD. search Centre, Working Paper No. 208. Osberg, L. and A. Bandara (2012). Why Poverty Remains High McCord, G. C., Conley, D. and Sachs, J.D. (2010). Improving in Tanzania: and What to do About it. REPOA Special Pa- Empirical Estimation of Demographic Drivers: Fertility, per 12/3, Dar es Salaam. References 83 Palloni, A. and H. Rafalimanana (1999). The Effects of Infant Handbook on the Shadow Economy Edward Elgar Pub- Mortality on Fertility Revisited: New Evidence from Lat- lishing Ltd., Cheltenham UK. in America. Demography 36 (1): 41–58. Seff, I., D. Jolliffe, K. Kafle and J. Kastelic (2014). Dynamics of Pastrana, D. (2005). El Programa Oportunidades Fracasa. Wellbeing in Tanzania 2008–2013: Analysis Based on Mucho dinero, pobres resultados. Masiosare 389, UNAM, Three Waves of the National Panel Survey. Background Mexico, http://www.jornada.unam.mx/2005/06/05/ paper prepared for this report (May 19). mas-daniela.html; Stewart, F., and Langer, A. (2007). Horizontal Inequalities: Ex- Pauw, K. and J. Thurlow (2011). Agricultural Growth, Poverty, plaining Persistence and Change. Centre for Research and Nutrition in Tanzania. Food Policy 36: 795–804. on Inequality, Human Security and Ethnicity. Working Ravallion, M. (1998). Poverty Lines in Theory and Practice. Paper No. 39. Oxford: University of Oxford. Living Standards Measurement Study, Working Paper Tarozzi, A. (2002). Estimating Comparable Poverty Counts No. 133, World Bank, Washington D.C. from Incomparable Surveys: Measuring Poverty in In- Ravallion, M. (2001). Measuring Aggregate Welfare in Devel- dia. Working Paper, Princeton University. oping Countries: How Well do National Accounts and Tarozzi, A. (2007). Calculating Comparable Statistics from In- Surveys Agree?. Policy Research Working Paper 2665, comparable Surveys, with an Application to Poverty in World Bank, Washington D.C. India. Journal of Business and Economic Statistics 25(3): Ravallion, M. (2008). Poverty Lines. In: S. N. Durlauf and L. E. 314–336. Blume (eds), The New Palgrave Dictionary of Economics, UNDP (1990). Human Development Report 1990, Oxford 2nd edition, Palgrave Macmillan. University Press, New York and Oxford. Rocha Menocal, A. (2001). Do Old Habits Die Hard? A Sta- United Nations (2014). World Population Prospects: 2012 Re- tistical Exploration of the Politicisation of PROGRESA, vision. (accessed May 2014) Mexico’s Latest Federal Poverty Alleviation Programme, URT (2009). Household Budget Survey 2007 – Tanzania Main- under the Zedillo Administration. Journal of Latin land, National Bureau of Statistics (NBS). Dar es Salaam. American Studies 33: 513–538. URT (2011). Tanzania National Panel Survey Report – Wave Roemer, J. E. (1998). Equality of opportunity. Cambridge, MA: 2, 2010–11. National Bureau of Statistics (NBS), Dar es Harvard University Press. Salaam. Rosenbaum, P. R. and Rubin, D. B. (1983). The central role of URT (2013a). 2012 Population and Housing Census. Popu- the propensity score in observational studies for casual lation Distribution by Administrative Areas. National effects. Biometrika, 70(1):41–55. Bureau of Statistics and Office of Chief Government Royston, P., White, I.R., (2011). Multiple imputation by chained Statistician, Dar es Salaam and Zanzibar. equations (MICE): implementation in Stata Journal, Sta- URT (2013b). Key Findings 2011/12 Household Budget Sur- ta Software, 45(4), 1–20. vey Tanzania Mainland. National Bureau of Statistics Rubin, D.B.: Multiple Imputation for Nonresponse in Surveys. (NBS), Dar es Salaam. New York (1987) URT (2014). Household Budget Survey Main Report, 2011/12. Sandefur, J. (2013). Africa Rising: Using Micro Surveys to Cor- National Bureau of Statistics (NBS), Dar es Salaam. rect Macro Time Series. CSAE Mimeo (August 9). Uwezo (2013). Are Our Children Learning? Literacy and Nu- Sarris, A., S. Savastano and L. Christiaensen. (2006). The meracy Across East Africa. Uwezo East Africa at Twawe- Role of Agriculture in Reducing Poverty in Tanzania: A za, Nairobi. Household Perspective from Rural Kilimanjaro and Ru- WDI (2014). World Development Indicators. (accessed May 2014). ow economies all over the world: new estimates for WHO (2012). World Malaria Report 2012. World Health Orga- 162 countries from 1999 to 2007. In: F. Schneider (eds.) nization, Geneva. 84 Tanzania Mainland Poverty Assessment Williamson, J. G. (2001). Population Matters: Demographic World Bank (2012a). Tanzania Economic Update: Spreading Change, Economic Growth, and Poverty in the Devel- the Wings: from Growth to Shared Prosperity. World oping World. In N. Birdsall, A. C. Kelley and S. W. Sind- Bank, Washington D.C. ing (eds.), Population Matters: Demographic Change, World Bank (2012b). Zambia Poverty Assessment: Stagnant Economic Growth, and Poverty in the Developing World, Poverty and Inequality in a Natural Resource-Based Oxford University Press. Economy. World Bank, Washington D.C. World Bank (2005). World Development Report 2006. Equity World Bank (2013a). Tanzania Economic Update: Opening and Development. World Bank, Washington D.C. the Gates: How the Port of Dar es Salaam Can Trans- World Bank (2007a). Ethiopia. Capturing the Demographic form Tanzania”, World Bank, Washington D.C. Bonus in Ethiopia: Gender, Development and Demo- World Bank (2013b). Tanzania Economic Update: Raising the graphic Action. World Bank, Washington D.C. Game”, World Bank, World Bank, Washington D.C. World Bank (2007b). Tanzania: Sustaining and Sharing Eco- World Bank (2014). Country Economic Memorandum. Tanza- nomic Growth. CEM and Poverty Assessment, World nia: Productive Jobs Wanted. World Bank, Washington Bank, Washington D.C. DC. World Bank (2008). Tanzania Rapid Poverty Assessment. Yang, D. (2008). International Migration, Remittances and Draft for consultation: November 2008, Poverty Moni- Household Investment: Evidence from Philippine Mi- toring Group, Washington D.C. grants’ Exchange Rate Shocks. The Economic Journal, World Bank (2011). Tanzania: Poverty, Growth, and Public 118(528): 591–630. Transfers Options for a National Productive Safety Net Program. Social Protection Unit, Africa Region, Wash- ington D.C. References 85 86 Tanzania Mainland Poverty Assessment Appendices 87 Appendix 1.A:  Poverty Estimation in the HBS 2007 and 2011/12 This appendix covers technical issues in the design, imple- we draw (mainly) on the HBS data to measure poverty trends mentation and poverty estimation methodology of the two over time, though we make use of the NPS to analyze pover- surveys which affect the analyses and comparability of pov- ty movements and dynamics. erty numbers over time. These issues have been mentioned in the body of the report as well, but are elaborated here. The HBSs are using a diary approach to collect consump- tion, where every individual in a household is asked to re- A. Introduction cord (on a daily basis) all food and non-food consumption Official estimates of poverty in (mainland) Tanzania are transactions that occurred over the course of (approxi- based on the Household Budget Surveys (HBS), which go mately) one month, including consumption of self-pro- back to the early 1990s. The HBSs are a series of repeated duced items.54 Enumerators visited the households regu- cross-sectional surveys conducted by the Tanzania Nation- larly to check and code the individual records. The HBSs al Bureau of Statistics (NBS). As shown in Table 1.A-1 there further included a recall module for non food expendi- have been four HBS rounds so far—1991/92, 2000/01, 2007 tures, particularly (semi-)durables and other irregularly and 2011/12. All HBS collect data on household consump- purchased items. tion, demographics (including education and health), asset ownership, housing, etc. The most recent 2011/12 HBS also The HBS instrument has evolved over time and there contained a detailed labor force and agricultural module. were significant changes between the HBS 2007 and HBS 2011/12. First, while the 2007 HBS recall module for There exists a second survey series suitable for poverty anal- non-food consumption was designed mainly to capture ysis, the National Panel Survey, which has had three round expenditures on semi-durable and durable goods and so far (2008/09, 2010/11 and 2012/13). The NPS is a longi- only probed for a limited number of item categories, the tudinal survey (tracking individuals) conducted every two 2011/12 HBS included a much more detailed and broad- years by the NBS and has a smaller sample size than the HBS. er recall module. Second, the 2007 HBS non-food recall However, the panel nature of the data makes it a particularly attractive survey for studying poverty dynamics and transi- 54 The 2007 HBS used a 28-day diary and staggered the start date tions. Due to differences in the way the HBS and NPS surveys of the diary, while previous HBS fielded the diary over the course capture consumption we follow the NBSs approach in that of one calendar month. Table 1.A–1:  Overview of Consumption Household Surveys in Mainland Tanzania Survey Period Coverage Type No. of households Household Budget Surveys: 1991–92 HBS Dec. 1991 – Nov. 1992 Tanzania Mainland Cross–section ~ 5,000 2000–01 HBS May 2000 – June 2001 Tanzania Mainland Cross–section 22,178 2007 HBS Jan. 2007 – Dec. 2007 Tanzania Mainland Cross–section 10,575 2011–12 HBS Oct. 2011 – Oct. 2012 Tanzania Mainland Cross–section 10,186 National Panel Surveys: 2008–09 NPS Oct. 2008 – Sept. 2009 Tanzania (incl. Zanzibar) Panel 3,265 2010–11 NPS Oct. 2010 – Sept. 2011 Tanzania (incl. Zanzibar) Panel 3,924 2012–13 NPS Oct. 2012 – Sept. 2013 Tanzania (incl. Zanzibar) Panel 5,088 Notes: HBS denotes Household Budget Survey. NPS denotes National Panel Survey. Number of households can differ slightly from official NBS publications. 88 Tanzania Mainland Poverty Assessment module used a uniform recall period of 12 months, while comprehensively in 2007. In the HBS 2011/12 item the 2011/12 HBS used recall periods of 1, 3 and 12 months codes in the diary did not correspond to the recall depending on the type of consumption item. Third, there module and the latter grouped some of the items is some evidence of better supervision in the HBS 2011/12, together that were recorded separately in the diary, which could have affected the capture of food consump- which makes a comparison of expenditures across the tion in the diary. The following paragraphs discuss these two sources more difficult. issues in turn. c. Recall periods: The HBS 2007 uses a uniform 12-month HBS 2011/12 Design and B.  recall period (with the exception of rent), while the HBS Implementation and Comparison to 2011/12 uses recall periods of 1, 3 and 12 months de- 2007 pending on the item (see Table 1.A-2 for an overview). The 2011/12 HBS differs from the preceding 2007 HBS in the A large literature shows that changes in the recall pe- following ways: riod can have effects on measured consumption and poverty (e.g. Beegle et al., 2010; Lanjouw, 2005; Gibson, a. Number of items and aggregation in the recall Huang and Rozelle, 2005). module: The HBS 2011/12 probed for a much larg- er number of items than the HBS 2007. For example, d. Survey supervision: There is evidence of improved the HBS 2011/12 asked explicitly for expenditures on supervision and survey implementation in the HBS 70 different clothing items. Conversely, the HBS 2007 2011/12 compared to the HBS 2007. In particular, the only probed for three broad categories of clothing HBS 2007 diary showed a strong pattern of respon- (of males, females and children), though enumera- dent’s fatigue, as the number of transactions and mea- tors still recorded item-specific expenditures (using sured consumption declined over the course of the codes provided in separate manual). We would expect diary. The HBS 2011/12 does not show such a trend, that the increase in the number of item categories in except for a drop from the first to the second day (Fig- the HBS 2011/12 enhances the household’s recollec- ure 1.A-2), which suggests that efforts to improve the tion of expenditures and hence increases measured quality of data collection have paid off.57 consumption. However, at the same time, the HBS 2011/12 omits certain non-food categories that were included in the HBS 2007, which may counteract the former effect of more non-food consumption being 55  For instance, the HBS 2011/12 does not ask for expenditures captured by the HBS 2011/12.55 On the other hand the on personal care items, though enumerators could record such HBS 2011/12 recall module appears to capture non- expenditures under ‘other personal articles’ or ‘other expenditures’ food consumption more comprehensively than the in an aggregated manner. Conversely, the HBS 2007 included ‘per- HBS 2007 recall module. sonal care items’ as a separate category and enumerators recorded all expenditures item-by-item. b. Diary-recall reconciliation: Both the 2007 and 56 This is despite efforts by the NBS to minimize overlap, see the 2011/12 HBS collect non-food expenditures not only discussion in the Appendix 2.A for further details. through the recall module but also through the con- 57  While the HBS 2007 started each diary on the beginning of the sumption diary.56 In 2007, the diary and recall module month, the HBS 2011/12 staggered the beginning of the diary. This allows disentangling to what effect the pattern in HBS diary con- used the same item codes which allowed compar- sumption is influenced by patterns over the course of the calendar ing reported expenditure for the same item across month (e.g. related to pay days) or to enumerator fatigue. There is the recall and diary (though over a different time no strong pattern of declining consumption over the course of the period). This comparison suggests that none of the calendar month, which suggests that this was not the reason for two sources alone captured non-food expenditures the decline in transactions in the HBS 2007. Appendices 89 Table 1.A-2:  HBS 2007 and 2011/12 Recall Modules HBS 2011/12 HBS 2007 Recall period (months) Recall period (months) Consumption and expenditure categories 1 3 12 1 3 12 Clothing and footwear (COICOP 3) X X Housing and utilities (COICOP 04 + selected other)   Rents X   X Utilities X   X Energy X   X Building maintenance X X Housing equipment (COICOP 05)   Household durables, furniture and furnishings X X Small household appliances X   X Expenditures on domestic workers X   X Health expenditures (COICOP 06) X   X Transportation (COICOP 07)   Vehicle purchases and registration X X Other expenses on vehicles and public transport X   X Communication equipment (COICOP 08) X X Recreation and culture (COICOP 09)   TV/DVD/Hifi equipment and books X X Other leisure (purchases, rentals, entrance fees) X   X Education (COICOP 10)   Education related expenses excl. registration fees X   X Registration fees X X Travel, restaurants and hotels (COICOP 11) X Miscellaneous goods and services (COICOP 12)   Miscalleneaous other X   X   Fees and use charges     X     X Source: Comparison of HBS questionnaires. The HBS 2011/12 Poverty Estimation C.  Calculation of the Consumption D.  Methodology Aggregate The HBS 2011/12 methodology has employed an updat- The Tanzanian poverty estimates are traditionally based on ed methodology to estimate poverty levels in Tanzania. aggregate household consumption as the key welfare in- This section describes the technical features of the HBS dicator. As in many other parts of sub-Saharan Africa, con- 2011/12 poverty estimation methodology. The next section sumption is considered a more reliable indicator of welfare describes how it differs from previous poverty analysis in than income. First, consumption is typically less fluctuating Tanzania (as described in URT 2002 and 2009 for the HBS than income and gives a better and steadier picture of long- 2000/01 and 2007). term welfare. Second, individuals feel more comfortable answering questions related to consumption than to in- come. Third, income measurement in countries with a large 90 Tanzania Mainland Poverty Assessment Transactions by Diary Day – 2007 and 2011/12 Figure 1.A-1:   Average number of transactions by diary day Average number of transactions by diary day 6 6 transactions per household transactions per household Average number of Average number of 4 4 2 2 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 HBS 2007 HBS 2011/12 Source: Household Budget Surveys 2007 and 2011/12. agricultural or informal sector is often highly inaccurate. The categories: (2) Alcoholic beverages and tobacco, (3) Cloth- consumption aggregate captures both food, and non-food ing and footwear, (4) Housing, water, electricity, gas and consumption. other fuels, (5) Furnishings, household equipment, main- tenance of the house, (6) Health, (7) Transport, (8)  Com- a.  Food consumption munication, (9) Recreation and culture, (10) Education, Food consumption is based on the food transactions re- (11) Restaurants and hotels, (12) Miscellaneous goods and corded in the 28-day diary (Form V) of the HBS 2011/12. The services. food consumption aggregate captures food consumed by household members during the day, including consump- The recall periods of these items in Form II are 12 months, tion from purchases and own-production (section B1) 3 months or 1 month, based on the assumed frequency of and food consumed outside the household (section B3). purchase. All spending on non-food goods and services is Households recorded all food consumed either the total converted to monthly expenditure.60 86% of non-food con- amount paid (in the cases of purchases) or an estimate of sumption (unweighted) is based on the non-food recall the monetary value in TZS58 (for own produced food and module (Form II). However, a limited set of diary expendi- gifts received). Total food consumption sums both actual tures were added in the following two cases. expenses and estimated monetary values. Food consump- tion includes the following COICOP categories and con- sists of 175 different items59: (1) Bread and cereals, (2) Meat, (3) Fish, (4) Milk, cheese and eggs, (5) Oils and fats, (6) Fruits, Estimates of the value of own produced goods and gifts were 58  (7) Vegetables, (8) Sugar, jam, honey, chocolate and con- made by the respondents and so have the risk of being over or fectionary, (9) Food products not elsewhere classified, under-estimated. Interviewers were trained to double check esti- (10) Coffee, tea and cocoa, (11) Mineral waters, soft drinks, mates that seemed unrealistic. fruit and vegetable juices. 59  Alcoholic beverages, as usual, were categorised as non-food. 60  28 day diary consumption data were converted to average monthly levels by dividing the consumption amount by 28 to get b.  Non-food consumption the daily amount and then multiplying this amount by 30.416 (365 The non-food consumption aggregate of the HBS 2011/12 days/12 month). Three and twelve month expenditure from Form captures expenditures on the following goods and COICOP II were divided by 3 and 12 respectively. Appendices 91 i.  Non-food consumption from the first ten days of the Adult Equivalence Scale Table 1.A-3:   A2 data file (diary non-food) Diary expenses that were recorded during the first ten days Age (years) Male Female of the diary’s implementation period were added to expen- 0–2 0.40 0.40 ditures already recorded in the recall module.61 3–4 0.48 0.48 5–6 0.56 0.56 Form II was administered on the 10th day of fieldwork and 7–8 0.64 0.64 the first 10 days of the diary implementation period do not 9–10 0.76 0.76 overlap with the recall module. When Form II was admin- 11–12 0.80 0.88 istered enumerators were instructed to be careful not in- 13–14 1.00 1.00 clude large items in Form II that had already been captured 15–18 1.20 1.00 in the diary. For example if in week 1 the household had 19–59 1.00 0.88 happened to purchase a mobile phone and this was in the 60+ 0.88 0.72 diary then it should not be double counted and entered in Source: The scale has been developed by the World Health Orga- Form II as well. An examination of A2 data shows very low nization and is reported in Collier et al (1986). reporting of these larger items. Section B2 in the diary was actually mostly used to record the small non-food items (see URT 2014 for further details). From day 11 onwards all differences in consumption needs between children and non-food is taken from only Form II except for the following adults the following equivalence scale is used:63 items specified below. e. Normalising consumption for differences in Consumption of in-kind water, in-kind firewood and ii.  cost of living tobacco for all 28 diary days Households with the same level of nominal consumption Consumption of the three items mentioned above were (per adult) might have different levels of real consumption added from all days of the diary as these few items were not if they face different costs of living. Nominal consumption specifically captured in Form II. of the household should hence be adjusted for temporal and spatial cost-of-living differences. Temporal price dif- c.  Exclusions from the consumption aggregate ferences are associated with the duration of the fieldwork The consumption aggregate excludes housing related over the course of a full year, while spatial differences are expenditures, neither actual rent or imputed rental val- associated with the location of households interviewed in ues for home owners. The consumption aggregate also the survey. excludes use values for large durable items even though it includes the purchasing values of a fairly large number of smaller, semi-durable goods.62 Finally, household level investments from Section 10 of Form II (purchase of hous- 61 Only for items with a recall period of one month. No adjustment was made for items with recall periods of three months or a year as es, apartments, garages, payments for hiring labour for the 10 day potential overlap would be short in comparison to the own construction, expenditures on ceremonies such as total recall period. weddings, funerals, business expenditures etc.) were also 62 The distinction between durables, semi-durables and non-du- excluded. rable items is based on UNStats.un.org official COICOP classifica- tion in which ND=Non durable, SD=Semi Durable and D=Durable. d. Normalizing consumption for differences in 63 No further allowance is made for possible economies of scale within households. Such economies of scale would assume that household composition consumption requirements of households do not rise linear- To normalise total household consumption for differenc- ly when additional persons are added (because some items in es in household size and composition and to adjust for households, e.g. housing, utilities, durable goods, can be shared). 92 Tanzania Mainland Poverty Assessment The price indices used to adjust nominal consumption are The overall (food and non-food) price deflator is comput- computed entirely from the HBS 2011/12 data—no external ed as the weighted average of food and non-food indices, information (e.g. from the Consumer Price Index database) where the weights are the average budget shares on food/ is used. A price index is a combination of prices and bud- non-food of households in the 2nd to 5th deciles of the distri- get shares in a base and a comparison period. The budget bution of total consumption per adult equivalent.67 shares are the weights that each commodity has in the in- dex and are equivalent to their share in the cost of the bun- iii.  Poverty Lines dle being analysed. The HBS 2011/12 can provide informa- The HBS 2011/12 poverty lines are based on a food basket tion on budget shares and prices (unit values) for all (food concept and correspondingly anchored in nutrition. The and non-food) items captured in the diary. HBS 2011/12 food poverty line (TZSs. 26,085.5 per adult per month) is based on the cost of a food basket that delivers To deflate nominal consumption NBS uses the Fisher ideal 2,200 calories per adult per day. The cost of buying 2200 cal- index. Fisher price indices are more accurate than Laspeyres ories is derived from the food consumption patterns prevail- or Paasche price indices in capturing differences in consump- ing in a reference population—the 2nd to 5th quintile of the tion patterns across domains as a consequence of differences distribution of total consumption per adult equivalent. Con- in relative prices. They also avoid overstating or understating sumed quantities are converted into calories using the NBS’s the true inflation (as it would be the case with Laspeyres and calorie conversion factors and valued at national median Paasche respectively). Separate food and non-food fisher prices (the same as the reference for the Fisher deflators).68 price indices are estimated by geographic stratum (Dar es Salaam, other urban and rural) and quarter (a period of three The non-food component of the basic needs poverty line consecutive months) according to the following formula:64 is based on average non-food consumption of households whose total consumption is close to the food poverty line.69 Fi = Li Pi In the HBS 2011/12 households in this reference group de- voted approximately 71.5% of their total consumption to where i is a combination of stratum and quarter, L refers to a food. Scaling up the food poverty line by this ratio delivers Laspeyres price index and P refers to a Paasche price index. the basic needs poverty line of TZS. 36,482 per adult per The Laspeyres and Paasche price indices are defined as: month (see next section for an assessment of the Tanzanian poverty lines).  −1 n p   n  pik −1  Li = ∑ w 0 k  ik  Pi = ∑ w   64  There are hence 12 price indices in total for each method. K =1  p0 k   K =1 ik  p0k     The diary includes “metre” and “pair” but these measures were 65  never used. where w0k is the average household budget share of item 66 If the household consumed the food item in a unit that does k in the country, wikis the average household budget share not have a metric conversion to the most frequent unit (e.g. piece of item k in stratum and quarter i, p0kis the national median to kg) the respective price is not used for the computation of the deflator. For most items the most frequent unit is kg or liter, but price of item k and pikis the median price of item k in stratum there are some exceptions (e.g. eggs overwhelmingly being con- and quarter i. sumed in units). This intends to make the deflator more tailored to the specific 67  It should be noted that all prices that feed into the deflators consumption patterns of poor households in Tanzania. are computed as unit values (value/quantity) from the HBS 68  As in the context of the Fisher price deflator, only transactions 2011/12 diary. The HBS food diary has six different measure- in the most frequent unit are used for the computation of median ment units for food items65—gram, kilogram (kgr), millilitre prices and to derive the budget shares. (ml), litre(l), piece and unit. Prices are based on the most fre- 69  More precisely, these are households whose total consumption quent unit for each item (with grams being converted to kg lies within the following interval [food poverty line; 1.2*food pov- and ml being converted to l).66 erty line]. Appendices 93 Value of Temporal and Spatial Price Deflators by Survey Quarter and Strata Table 1.A-4   Urban Rural Dar-es-salaam Paasche Laspeyres Fisher Paasche Laspeyres Fisher Paasche Laspeyres Fisher Food Food Food Food Food Food Food Food Food Food I – 10.2011–12.2011 0.980 1.035 1.007 0.918 0.927 0.922 1.021 1.176 1.096 II – 01.2012–03.2012 1.030 1.045 1.037 0.929 0.948 0.939 1.136 1.265 1.199 III – 04.2012–06.2012 1.051 1.084 1.067 0.975 0.989 0.982 1.136 1.279 1.205 IV – 07.2012–10.2012 1.036 1.094 1.065 0.965 0.971 0.968 1.120 1.250 1.183 Urban Rural Dar-es-salaam Paasche Laspeyres Fisher Paasche Laspeyres Fisher Paasche Laspeyres Fisher Non food Non-food Non-food Non-food Non-food Non-food Non-food Non-food Non-food Non-food I – 10.2011–12.2011 0.999 1.042 1.020 0.943 0.936 0.940 1.082 1.365 1.215 II – 01.2012–03.2012 0.993 0.980 0.986 0.941 0.932 0.936 1.065 1.238 1.148 III – 04.2012–06.2012 0.933 0.955 0.944 0.999 0.969 0.984 1.063 1.437 1.236 IV – 07.2012–10.2012 1.015 1.100 1.057 0.979 0.975 0.977 1.041 1.568 1.278 Source: Household Budget Survey (HBS) 2011/12. iv.  Poverty Concepts higher calorie thresholds (e.g. Rwanda with 2,500 or Uganda NBS distinguishes two different poverty concepts — basic with 3,000 calories per adult). needs poverty (often simply referred to as poverty) and food poverty (often also referred to as extreme poverty). A house- The approach used to generate the non-food component hold is considered ‘basic needs poor’ if its consumption per of the poverty line (described in the previous section) is a adult falls below the basic needs poverty line. If consump- variant of the so called ‘lower-bound’ approach (Ravallion, tion per adult also falls below the food poverty line, a house- 1998). In its more conventional application, this approach hold is necessarily consuming less than the minimum food computes average non-food consumption of households requirement and so is considered ‘food poor’ or ‘extreme whose total consumption lies within a small interval around poor’. By definition, a household that is food poor is also ba- the poverty line. Increasing the interval bandwidth itera- sic needs poor. tively and taking the mean of all the averages delivers the non-food component of the poverty line. If we use exactly E.   Evaluation of the 2011/12 Basic this method, the total poverty line amounts to TZS. 35,939 Needs Poverty Line per adult per month, which is just below the official 2011/12 The HBS 2011/12 poverty lines follow the Cost of Basic basic needs poverty line. Needs methodology (Ravallion, 1998; 2008), which is a fre- quently used method to derive poverty lines in Sub-Saha- There also exists an ‘upper-bound’ approach, which looks ran Africa and other developing regions. The food poverty at households whose total food consumption lies within a line (TZS 26,085.5 per adult per month) is based on the cost small interval around the food poverty line (otherwise re- of a food basket that delivers 2,200 calories per adult per peating the steps outlined above for the lower-bound ap- day given consumption patterns prevailing in a reference proach). Intuitively, these households are already consuming population—the 2nd to 5th quintile of the distribution of to- enough food to meet basic nutrition requirements and are tal consumption per adult equivalent. This calorie norm is hence less poor than the reference group under the more within the range of what other countries in the region are austere lower bound approach. If we use this method, the using (e.g. Kenya with 2,250 or Ethiopia with 2,200 calories total poverty line is estimated at TZS 50,967—hence consid- per adult), though there are countries that use significantly erably higher than the 2011/12 basic needs poverty line. 94 Tanzania Mainland Poverty Assessment We can also compare the Tanzanian poverty line to the in- methodology (URT 2014) and the previous poverty analy- ternational 1.25 USD per capita per day poverty line. The TZS sis of the HBS 2000/01 and 2007 data as described in URT 36,482 basic needs poverty line translates into approximately (2002, 2009). It also shows how the new 2011/12 method- 1 USD per capita per day at 2005 purchasing power parities ology was retro-actively applied to the 2007 data to assess (based on 2005–12 CPI inflation in the World Development the poverty trend between 2007 and 2011/12 based on the Indicators), which is lower than the international poverty line. new (2011/12) methodology. It should be noted that the reconstruction of the 2007 consumption aggregate and This shows generally that the HBS 2011/12 basic needs pov- poverty line described here cannot account for differences erty line of TZS 36,482 is at the lower end of the spectrum. in design and implementation This is why we further use As Tanzania continues to increase its per capita income and cross-survey imputation and reweighting methods to trian- move to middle income status, Tanzanian policy makers gulate the change in poverty (see Appendix I.B). might wish to consider revising the poverty line upwards to set itself more ambitious goals in the fight against pov- For areas where the same methods were used in both sur- erty and to meet its vision of a society with a “high quality veys, the description runs across the three columns in the livelihood”. table. For other areas, the details are listed separately for 2011/12 and 2007 in the first and third columns respective- Comparison of the New (HBS F.  ly. The middle column describes how the 2007 dataset was 2011/12) Poverty Estimation re-analyzed to take account of the differences in order to Methodology with the Previous produce a poverty line and headcount based on the same Methodology to Measure Poverty methods as adopted in 2011/12. The following Table gives an overview over similari- ties and differences between the new 2011/12 poverty Table 1.A-5:  Comparison of Poverty Estimation Methodologies Application of 2011/12 methodology Previous methodology HBS 2007 and HBS New methodology in HBS 2011/12 to 2007 data 2000/01 Construction of • Food consumption based on diary (with own produced goods being valued at the estimated monetary values provided by the households) the consump- • Excludes: Rent and housing related expenditures, durable goods (neither expenditures nor use values) and non-consumption expenditures tion aggregate • Includes: Education, health, and communication • Education, health, and communication • Excludes: Education, health, and communica- expenditure expenditure added into the consumption tion expenditure • Consumption standardized to one month aggregate • Consumption standardized to 28 days • Non-food consumption mostly from recall module • Non-food consumption from diary and • Non-food consumption from diary and recall. (except for 10 day diary overlap and few other recall. For each household and item it is For each item a decision is taken whether selected items—see previous section for a checked whether non-food consumption the diary or recall data is deemed a more discussion) is reported (i) only in the recall, (ii) only reliable source of information contingent on in the diary, or (iii) in both sources. In a comparison of reported frequencies and case of (i) and (ii) the reported expen- spending amounts across the diary and recall ditures from either source are included in the 1991/92 and 2000/01 HBS. Non-food in the consumption aggregate, in case consumption for the respective item is than of (iii) a simple average across the two taken only from the source deemed more sources is used—after standardization to reliable for all households in the survey a common reporting period (continues to next page) Appendices 95 Table 1.A-5:  Comparison of Poverty Estimation Methodologies (continued) Application of 2011/12 methodology Previous methodology HBS 2007 and HBS New methodology in HBS 2011/12 to 2007 data 2000/01 Normalizing • Adult equivalence scale based on Collier et al (1986) for household • No allowance for economies of scale at the household level composition • Scale corresponds exactly to Collier et al (1986) • Two incorrect coefficients in 2007 correct- • Scale corresponds to Collier et al (1986) ed to match those in 2011/12 except for variations in two coefficients (uses a coefficient of 0.4 (instead of 0.48) for male children aged 3–4 years and of 0.8 (instead of 0.88) for males aged 60+ years)) Normalizing for • Survey-internal Fisher food and non-food price deflators based on (median) unit values from the consumption diary (only metric units; except for within-survey eggs measured in pieces/numbers) price differences • Non-food Fisher deflator based on a limited number of non-food items • Overall deflator is a weighted average of the food and non-food Fisher deflators • Spatial and temporal price correction (by geo- • Spatial and temporal price correction • Spatial price correction only (by geographic graphic domain and quarter) applied, using food/non-food weights as domain) • The weights of the overall deflator are the share in 2011/12 • The weights of the overall deflator are the of food and non-food spending in the 2 to 5 nd th shares of food and non-food spending deciles of the distribution of nominal consumption amongst the poorest 25% of the popula- per adult equivalent – the same as the reference tion—the same as the reference group for group for the food basket/food poverty line the non-food component of the poverty line) Poverty line • Cost of basic needs (CBN) methodology anchored in nutrition (2,200 kcal per adult per day) • New poverty line computed in 2011/12 • 2007 poverty line is derived by deflating • 2007 poverty line is derived by inflating the • Standardized to one month the 2011/12 poverty line backwards us- 2000/01 basic needs poverty line using a • Food basket based on average expenditure shares ing a survey-internal Fisher deflator, with survey-internal Fisher deflator aggregated across reference population (2nd to 5th food and non-food weighted by the food/ • Standardized to 28 days quintile of the distribution of total consumption non-food ratio of the total distribution • Food basket based on median quantities in per adult equivalent)—i.e. it is not the average the reference population (poorest 50% of the across the proportionate shares of individual population) households • Non-food component based on average bud- • Non-food component based on the average get share spent on non-food items amongst budget share spent on non-food items amongst the poorest 25% of the population households whose total consumption lies within the following interval [food poverty line; 1.2*food poverty line] 96 Tanzania Mainland Poverty Assessment Appendix 1.B: Prediction Methods to Establish Comparability between the 2007 and 2011/12 Data Semi-parametric Approach (Tarozzi, A.  assess changes in poverty over time, in situations where 2002): consumption data are not comparable or where only one The method exploits the existence of consumption and survey collects consumption data (e.g. World Bank 2012a, non-consumption auxiliary variables, which are not affected 2012b, 2012c). Christiaensen et al. (2012) show that the by the changes in the survey design and are related con- small-area estimation technique often performs relatively sistently to total consumption. The consumption distribu- well in tracking poverty over time. tion in 2007 is then recovered based on the distribution of these consumption and other non-consumption auxiliary An issue arises with the cross-survey imputation method variables. In our application here we consider sub-groups of in relation to the choice of predictive variables, particularly food consumption and a range of household characteristics cell phone ownership. It turns out that there is a large dif- as collected comparably across the two surveys. ference in the results when possession of a cell phone (at the household level) is included or excluded in the set of variables used to predict household consumption. This is a The Second Method is a Variant of B.  consequence of the very large increase in the possession the Small Area Estimation (poverty of cell phones over the five years between the two surveys mapping) Methodology Developed and the strong correlation between cell phone ownership by Elbers, Lanjouw and Lanjouw and consumption. Households across a wide range of the (2003): consumption distribution owned cell phones by 2011/12, Unlike the Tarozzi method, this technique does not require compared with 2007 when cell phone ownership tended to that some components of consumption are collected com- be limited to the better off, particularly in urban areas. This parably in the two surveys but relies entirely on (non-mone- sensitivity of the results is not important for the other pre- tary) characteristics of the household. The first step is to iden- dictive variable (i.e. omitting any other individual variable tify a set of household characteristics that were collected in makes very little difference to the final result). the same way in both surveys. It then estimates the relation- ship between these variables and consumption in 2011/12; There are arguments for and against the inclusion of cell that is it calculates the extent to which possession of each phones as predictive variables in the regression model. The of these characteristics by a household predicts their level main argument for the inclusion of cell phones is that they of consumption in 2011/12. This relationship is then used to are an important predictor of consumption and that the impute consumption (per adult) for the 2007 survey house- increase in cell phone ownership captures and proxies for holds by applying these coefficients to the same set of com- a substantive increase in household consumption, which parable household characteristics as observed in 2007. Since may otherwise be overlooked. The main argument against the simulated 2007 consumption distribution is expressed their inclusion is that during a period of rapid cell phone ac- in 2011/12 prices, there is no need to adjust for inflation be- cumulation the relationship between cell phone ownership tween the surveys and the 2011/12 poverty line can be used and consumption is likely to change over time, especially to compute the simulated poverty estimates in 2007. if the increase largely comes from poorer groups (due to relative price changes, etc.). Including cell phones as a pre- This technique has its origin in small area estimation of pov- dictive variable in the model might thus lead to an overesti- erty (‘poverty mapping’), where census and survey data are mation of the decline in poverty. Due to this ambiguity we combined to generate regionally disaggregated poverty show the results for both models, including and excluding maps. However, it has also become a popular method to cell phone ownership. Appendices 97 Appendix 1.C: Welfare Dynamics Table 1.C-1  Trends in Dwelling Material by Area of Residence 2007 2011/12 absolute Δ relative Δ (percent) (percent) (percentage points) (percent) Dwelling material Improved roof material National 55.8 66.2 10.4 18.6 Rural 42.0 54.8 12.7 30.3 Urban 84.6 90.5 5.9 7.0 Dar es Salaam 95.3 99.2 3.9 4.0 Improved floor material National 33.3 38.8 5.6 16.7 Rural 17.0 22.3 5.3 31.4 Urban 62.9 69.2 6.3 10.0 Dar es Salaam 88.1 96.8 8.7 9.9 Improved wall material National 34.1 46.1 12.0 35.3 Rural 21.9 33.1 11.2 51.2 Urban 50.6 67.8 17.2 34.0 Dar es Salaam 85.8 97.1 11.3 13.2 Source: Household Budget Surveys (HBS) 2007 and 2011/12. Table 1.C-2  Trends in Dwelling Material by Quintiles Improved Roof material Improved Floor material Improved Wall material   2007 2011/12 2007 2011/12 2007 2011/12 Poorest Quintile 35.7% 50.1% 12.0% 18.2% 19.0% 36.8% 2nd Quintile 45.4% 59.0% 18.3% 26.4% 22.7% 36.7% 3rd Quintile 55.7% 65.6% 30.5% 35.6% 31.3% 42.1% 4th Quintile 64.1% 76.7% 44.1% 51.7% 41.4% 53.6% Top Quintile 78.0% 86.5% 61.5% 74.1% 56.1% 71.9% Source: Household Budget Surveys (HBS) 2007 and 2011/12. Note: each quintile represents 20 percent of the population. 98 Tanzania Mainland Poverty Assessment Table 1.C-3  Trends in Assets Ownership by Location absolute Δ 2007 (percent) 2011/12 (percent) (percentage points) relative Δ (percent) ICT/ Electronics Radio National 65.6 54.6 –11 –16.8 Rural 61.8 51.5 –10.3 –16.7 Urban 72.4 60.5 –11.9 –16.4 Dar es Salaam 78.2 64.7 –13.5 –17.2 TV  National 8.1 13.8 5.7 70.2 Rural 1.8 3.8 1.9 106.3 Urban 15.5 28.4 12.8 82.7 Dar es Salaam 36.9 58 21.1 57.3 Video National 5.2 10.3 5.1 96.5 Rural 1.2 3.4 2.2 181.9 Urban 11.7 20.5 8.8 75.0 Dar es Salaam 20.4 40.0 19.6 96.2 Telephone (landline) National 1.0 0.5 –0.6 –53.3 Rural 0.6 0.1 –0.4 –77.7 Urban 1.8 1.5 –0.4 –20.8 Dar es Salaam 2.7 1.1 –1.7 –61.4 Cell phone National 24.3 55.8 31.6 130.1 Rural 13.8 45.2 31.4 226.6 Urban 42.1 77.5 35.4 84.2 Dar es Salaam 61.4 88.4 27 44.0 Computer National 0.5 1.7 1.3 271.2 Rural 0.1 0.4 0.3 514.5 Urban 0.5 2.6 2.1 449.2 Dar es Salaam 3.2 10.0 6.8 215.9 Transportation Bicycle National 40.1 34.1 –6.0 –15 Rural 45.1 37.9 –7.2 –16 Urban 35.5 33.3 –2.2 –6.3 Dar es Salaam 15.1 7.4 –7.8 –51.3 Car National 1.1 1.2 0.1 13.5 Rural 0.3 0.2 –0.1 –26.0 Urban 2.1 2.5 0.4 18.9 Dar es Salaam 4.3 5.9 1.6 37.3 Motor cycle / National 3.1 3.9 0.8 26.6 moped Rural 2.5 3.8 1.4 54.9 Urban 4.8 5.3 0.4 8.7 Dar es Salaam 4.2 1.9 –2.3 –54.1 (continues to next page) Appendices 99 Table 1.C-3  Trends in Assets Ownership by Location (continued) absolute Δ 2007 (percent) 2011/12 (percent) (percentage points) relative Δ (percent) Household appliances and other items Fridge National 4.8 6.4 1.5 31.6 Rural 1.1 1.3 0.2 17.6 Urban 7.9 11.4 3.4 43.3 Dar es Salaam 24.3 33.7 9.4 38.9 Cooking stove National 41.7 62.2 20.5 49.2 Rural 25.9 51.7 25.8 99.5 Urban 76.3 85.8 9.6 12.6 Dar es Salaam 84.2 90 5.8 6.9 Iron National 26.1 22.6 –3.5 –13.5 Rural 18.3 14.5 –3.8 –20.9 Urban 41.2 35.7 –5.5 –13.4 Dar es Salaam 50.8 55.1 4.3 8.5 Sewing machine National 6.5 6.5 –0.1 –1.1 Rural 4.1 4.6 0.5 11.4 Urban 12.0 11.0 –1.0 –8.5 Dar es Salaam 12.9 10.9 –1.9 –15.1 Water heater National 14.0 4.4 –9.6 –68.8 Rural 14.1 2.7 –11.4 –80.7 Urban 15.4 5.8 –9.7 –62.7 Dar es Salaam 10.7 13.6 2.9 27.5 Mosquito net National 68.9 86.5 17.7 25.6 Rural 61.3 85.6 24.3 39.7 Urban 84.1 88.4 4.3 5.1 Dar es Salaam 92.0 89.1 –3.0 –3.2 Furniture Chair National 79.0 75.3 –3.6 –4.6 Rural 76.6 75.5 –1.0 –1.4 Urban 85.9 77.4 –8.4 –9.8 Dar es Salaam 82.5 69.2 –13.4 –16.2 Sofa National 26.6 12.2 –14.4 –54.1 Rural 14.0 5.3 –8.7 –62.4 Urban 50.0 22.9 –27.1 –54.2 Dar es Salaam 69.1 41.3 –27.8 –40.3 Bed National 90.9 85.9 –5.0 –5.5 Rural 89.5 83.6 –5.9 –6.6 Urban 93.4 90.2 –3.2 –3.4 Dar es Salaam 95.8 93.4 –2.4 –2.5 Table National 70.1 66.0 –4.1 –5.8 Rural 63.6 60.1 –3.5 –5.6 Urban 85.3 79.1 –6.2 –7.2 Dar es Salaam 86.1 82.0 –4.1 –4.8 Source: Household Budget Surveys (HBS) 2007 and 2011/12. 100 Tanzania Mainland Poverty Assessment Table 1.C-4  Trends in Some Assets Ownership by Quintiles Cell Phone TV Radio Mopped Bicycle Mosquito Net   2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 2007 2011/12 Lowest Quin. 5.4% 35.7% 0.8% 1.9% 48.2% 44.0% 0.0% 0.9% 34.6% 34.3% 58.9% 83.0% 2nd Quintile 11.2% 48.1% 1.8% 4.6% 60.3% 49.5% 0.0% 1.1% 41.6% 36.9% 64.3% 85.5% 3rd Quintile 22.1% 55.7% 4.7% 9.7% 66.9% 55.4% 0.0% 1.6% 43.0% 36.9% 67.6% 86.7% 4th Quintile 33.1% 68.1% 9.9% 20.4% 73.4% 61.2% 0.1% 1.8% 43.1% 34.3% 74.7% 89.1% Top Quintile 49.5% 78.3% 23.4% 41.4% 79.0% 64.8% 0.1% 2.8% 38.2% 23.0% 78.8% 88.9% Source: Household Budget Surveys (HBS) 2007 and 2011/12. Note: each quintile represents 20 percent of the population. Appendices 101 Appendix 1.D: Static Decomposition of Inequality The static decomposition carried out, in the first section, and (iii) inactive, disabled or retired. The employment sec- to examine the importance of certain individual and fami- tor comprises six categories: (i) Government; (ii) Private sec- ly characteristics in determining the level of consumption tor, NGOs and international companies; (iii) self-employed inequality is based on eight household attributes: the gen- with others; (iv) self-employed alone; (v) household duties; der, age, educational attainment, activity status and sector and (vi) unemployed & inactive. The regional locations are of employment of the head, the regional location, the ur- the 21  regions in the HBS surveys.70 Households are also ban/rural status, and the demographic composition of the grouped into five categories by the demographic types: household. (i) “single parent with no kids”, (ii) “single parent with kids”, (iii) “couple with no kids”; (iv) “couple with kids”, and (v) “fam- The gender of the household head is simply male or female. ilies of elderly whose head is aged 65 years old or above”. His age is split into five categories: (i) under 30, (ii) 30–39, (iii) 40–49, (iv) 50–59, and (v) 60+ years. The head education- al attainment is classified into six categories: (i) no educa- 70 The high number of regions (low number of observations in tion & illiterate; (ii) less than completed primary; (iii) com- each group) may induce biases in the between groups inequality pleted primary; (iv) lower secondary; (v) upper secondary or estimates. However, even when the regions are grouped into five equivalent; and (vi) university. Three groups are considered main zones, a similar trend of sharply increasing interregional in- for the head activity status: (i) employed; (ii) unemployed; equalities over the last ten years is observed. 102 Tanzania Mainland Poverty Assessment Appendix 2.A: Characteristics of the Poor and Poverty Correlates Table 2.A-1  Household Characteristics by Poverty Status, Quintile and Region, 2011/12 Poverty Status Quintile Region Non- Richest Dar-es-   Poor poor Poorest Q2 Q3 Q4 quintile Rural Urban Salaam All Share of the population (%) 28.2 71.8 20.0 20.0 20.0 20.0 20.0 71.2 18.7 10.1 100.0 Age of the household head         Mean 48.4 45.7 48.8 47.5 46.9 45.7 43.4 47.0 45.6 44.0 46.4 Median 46 43 47 45 45 43 41 45 43 41 44 Household size       Mean 8.3 6.4 8.5 7.7 6.9 6.2 5.4 7.3 6.3 5.5 6.9 Median 7 6 8 7 6 6 5 6 6 5 6 Number of children (below 15 years)         Mean 4.3 2.9 4.4 3.8 3.3 2.8 2.1 3.7 2.6 1.9 3.3 Median 4 2 4 3 3 2 2 3 2 2 3 Education of head (years)       Mean 4.3 6.2 4.4 4.7 5.3 6.1 7.9 4.9 7.2 8.7 5.7 Median 7 7 7 7 7 7 7 7 7 7 7 Number of migrants       Mean 0.8 1.3 0.7 1.0 1.0 1.2 1.6 0.8 1.4 2.5 1.1 Median 0 0 0 0 0 0 1 0 0 2 0 Employment sector of household head Government employee 1.0 5.4 0.8 1.7 3.4 4.7 10.3 2.4 8.8 8.5 4.2 Private/NGO/other employee 1.7 7.1 1.5 2.7 3.3 6.6 13.7 1.8 8.9 25.6 5.6 Self-employed (with employees) 5.7 7.4 4.7 6.5 6.0 5.9 11.6 5.4 11.3 9.9 6.9 Self-employed (own-account) 77.9 65.9 79.3 75.6 73.2 67.3 50.9 78.3 54.2 33.5 69.2 Unpaid family worker, household duties 2.1 3.5 2.1 1.9 2.4 3.7 5.4 1.6 4.8 10.9 3.1 Unemployed/Inactive 11.6 10.7 11.7 11.6 11.8 11.7 8.1 10.6 12.0 11.6 11.0 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Source: Household Budget Survey (HBS) 2011/12. Note: Population quintiles and population weighted. Appendices 103 Table 2.A-2  Access to Public Infrastructure by Poverty Status, Quintile and Region, 2011/12 Poverty Status Quintile Region Non- Richest Dar-es-   Poor poor Poorest Q2 Q3 Q4 quintile Rural Urban Salaam All Share of the population (%) 28.2 71.8 20.0 20.0 20.0 20.0 20.0 71.2 18.7 10.1 100.0 Access to piped water (dry season, %)       Private connection (inside/outside house) 6.9 17.0 6.8 8.0 10.3 17.3 28.2 5.3 37.2 33.2 14.1 Public tap 19.2 20.7 20.6 17.4 20.3 21.4 21.8 19.2 23.4 22.3 20.3 Access to electricity (%)       Public grid (TANESCO) connection 2.9 21.1 2.8 5.3 10.5 19.1 42.2 3.5 34.8 69.4 16.0 Access to road infrastructure (%)       Trunk road (in community) 41.3 46.0 41.2 39.4 43.8 45.0 54.1 40.5 51.4 61.8 44.7 Tarmac road (in community) 16.0 26.4 15.0 18.0 21.7 25.5 37.3 15.8 34.0 57.6 23.4 Access to local markets (%) Daily markets (in community) 26.0 35.2 25.7 27.9 32.1 34.6 42.9 28.6 44.3 39.4 32.6 Weekly markets (in community) 28.5 27.9 28.3 29.7 28.9 27.1 26.3 31.6 25.8 7.7 28.1 Source: Household Budget Survey (HBS) 2011/12. Note: Population quintiles and population weighted. Access to road infrastructure is missing for six enumeration areas (1.5 percent of the population). 104 Tanzania Mainland Poverty Assessment Private Productive Assets and Durable Goods by Poverty Status, Quintile and Table 2.A-3   Region, 2011/12 Poverty Status Quintile Region Non- Richest Dar-es-   Poor poor Poorest Q2 Q3 Q4 quintile Rural Urban Salaam All Share of the population (%) 28.2 71.8 20.0 20.0 20.0 20.0 20.0 71.2 18.7 10.1 100.0 ICT and electronics (ownership, %)         Cell phone 42.5 69.5 39.7 55.5 58.5 71.1 84.6 52.2 82.8 90.8 61.9 Radio 46.6 61.5 45.4 52.6 56.1 62.7 69.7 54.2 63.4 67.5 57.3 TV 2.0 19.4 1.8 3.7 8.9 17.6 40.4 3.8 29.9 61.2 14.5 Transportation assets (ownership, %)         Car 0.1 2.3 0.0 0.0 0.1 0.4 7.7 0.3 3.3 8.1 1.6 Motor cycle/ moped 2.7 6.7 2.2 4.6 2.6 6.5 11.9 5.6 6.9 2.7 5.6 Bicycle 40.4 40.3 39.7 45.4 40.6 41.3 34.8 45.0 39.2 9.6 40.4 Other household items (ownership, %)         Cooking stove (electric, gas or traditional) 48.1 69.2 46.8 56.2 61.4 70.1 81.6 52.0 88.4 95.6 63.2 Mosquito net 84.3 88.8 83.5 86.8 85.8 90.3 91.1 86.8 88.9 89.9 87.5 Bed 83.8 87.8 83.3 85.4 85.1 88.0 91.6 84.6 90.8 93.9 86.7 Table 57.8 72.8 55.9 64.6 67.6 72.5 82.2 62.8 82.0 84.3 68.5 Dwelling characteristics (ownership, %)         Improved roof 52.5 73.1 51.3 57.9 67.0 74.7 85.6 56.7 90.4 99.1 67.3 Improved wall 36.0 51.2 36.7 35.4 41.0 51.6 69.7 34.4 67.3 97.2 46.9 Improved floor 17.5 45.0 15.3 23.0 32.0 46.3 69.7 21.0 67.2 96.6 37.3 Land and livestock         Any owned land (%) 86.7 67.9 87.8 82.5 78.9 68.0 48.7 89.1 47.2 8.9 73.2 Any rented land (%) 7.8 8.8 7.2 11.6 8.4 8.4 7.0 10.1 6.7 0.5 8.5 Any livestock (%) 69.9 57.9 69.3 68.9 67.1 57.4 43.7 74.2 38.3 12.6 61.3 Owned land (mean acres)* 7.6 5.4 8.0 7.2 5.3 5.3 4.5 7.4 4.0 0.6 6.0 Rented land (mean acres)* 0.2 0.3 0.2 0.4 0.2 0.3 0.4 0.3 0.3 0.0 0.3 Source: Household Budget Survey (HBS) 2011/1. Note: Population quintiles and population weighted. * Mean includes households with zero land. Appendices 105 Appendix 2.B: Multivariate Regression We perform a regression analysis to examine the main fac- It is worth mentioning that the direction of causality is tors affecting households’ consumption and poverty. This sometimes difficult to establish in these kinds of analysis. allows us to identify the main correlates of poverty. The results below allow the identification of variables close- ly related with poverty, but the direction of causation will We use two regression models. The first examines the im- necessitate more sophisticated analysis.71 pact of the household socioeconomic characteristics on the logarithm of real per adult equivalent household consump- tion, and the second investigates the determinants of the probability of being poor. The first model is estimated using 71 Identifying how important each explanatory variable is in a re- gression of this sort has to consider two main factors: first, the im- the Ordinary Least Square (OLS) method and the second us- pact on the dependent variable, given by the size of the estimated ing the probit approach. The estimation results are reported coefficient; second, the statistical significance of the coefficient— respectively in Tables 2.B-1 and 2.B-2. typically whether it is significantly different from zero. Table 2.B-1  Correlates of Consumption, 2011/12 (1) (2) (3) National Rural Urban Household characteristics       Household size –0.0272*** –0.0225*** –0.0516*** (–6.794) (–5.175) (–9.833) Share of members aged 0–14 years –0.379*** –0.318*** –0.384*** (–9.330) (–6.181) (–8.815) Share of members aged 65+ years –0.0124 0.0615 –0.124 (–0.209) (0.903) (–1.281) Education of the head (Omitted: no education) Less than completed primary 0.0539* 0.0647** 0.0117 (1.876) (2.153) (0.176) Completed primary 0.120*** 0.102*** 0.141** (4.826) (4.055) (2.241) Lower secondary 0.385*** 0.362*** 0.390*** (11.55) (7.654) (6.265) Upper secondary 0.681*** 0.709*** 0.647*** (12.27) (6.266) (8.983) Migrant household 0.134*** 0.0661 0.167*** (4.378) (1.427) (6.190) Economic activity and assets Household activity (Omitted: no reported working hours) Mainly engaged in agriculture –0.104** –0.0927* –0.134** (–2.478) (–1.828) (–2.157) Mainly engaged in non-farm enterprise 0.157*** 0.135** 0.157*** (3.553) (2.341) (3.053) (continues to next page) 106 Tanzania Mainland Poverty Assessment Table 2.B-1  Correlates of Consumption, 2011/12 (continued) (1) (2) (3) National Rural Urban Mainly engaged in wage work 0.154*** 0.0632 0.175*** (3.330) (0.940) (3.440) Uses irrigation 0.0928 0.0959 0.0883 (1.645) (1.369) (1.519) Sells agricultural output 0.0790*** 0.0731** 0.136** (2.654) (2.368) (2.588) Size of landholdings (square root) 0.0418*** 0.0524*** 0.0338*** (4.293) (3.954) (3.087) Has any livestock –0.0128 –0.0139 0.0554* (–0.495) (–0.497) (1.719) Community characteristics Daily market 0.0703* 0.0624 0.0768* (1.926) (1.303) (1.948) All season passable road 0.0731 0.0836 –0.0418 (1.539) (1.587) (–0.637) Mobile phone signal 0.0812 0.0678 0.0890* (1.530) (1.089) (1.717) Geographic zone (Omitted: Coastal) Northern Highlands –0.0956 –0.0224 –0.0967 (–1.403) (–0.228) (–1.412) Lake –0.132** –0.0691 –0.169*** (–2.470) (–0.807) (–2.880) Central –0.180*** –0.125 –0.109 (–3.231) (–1.446) (–1.122) Southern Highlands –0.200*** –0.174* –0.119* (–3.021) (–1.700) (–1.844) South –0.392*** –0.259** –0.553*** (–5.100) (–2.589) (–7.256) Constant 10.92*** 10.79*** 11.17*** (139.5) (105.4) (141.3) Observations 9,930 4,064 5,866 R-squared 0.314 0.163 0.416 Source: HBS 2011/12. Notes: t-statistics in parentheses. Standard errors corrected for clustering and stratification. OLS. Dependent variable is log consumption per adult. *** p<0.01, ** p<0.05, * p<0.1 Appendices 107 Table 2.B-2  Correlates of Poverty, 2011/12 (1) (2) (3)   National Rural Urban Household characteristics       Household size 0.0600*** 0.0513*** 0.114*** (5.628) (4.406) (6.208) Share of members aged 0–14 years 0.748*** 0.817*** 0.249 (5.312) (5.073) (1.382) Share of members aged 65+ years 0.0961 0.0438 0.275 (0.520) (0.210) (0.897) Education of the head (Omitted: no education) Some primary –0.132* –0.139* –0.0754 (–1.700) (–1.705) (–0.422) Completed primary –0.289*** –0.264*** –0.318* (–4.338) (–3.830) (–1.891) Lower secondary –0.883*** –0.971*** –0.903*** (–6.332) (–4.562) (–4.045) Upper secondary –1.529*** –2.168*** –1.105*** (–7.608) (–5.171) (–4.406) Migrant household –0.227** –0.101 –0.417*** (–2.323) (–0.803) (–4.678) Economic activity and Assets Household activity (Omitted: no reported working hours) Mainly engaged in agriculture 0.168 0.172 0.133 (1.592) (1.335) (0.850) Mainly engaged in non-farm enterprise –0.470*** –0.439*** –0.484*** (–3.992) (–2.925) (–3.725) Mainly engaged in wage work –0.353*** –0.0978 –0.582*** (–2.718) (–0.531) (–4.900) Uses irrigation –0.282* –0.304* –0.133 (–1.956) (–1.862) (–0.694) Sells agricultural output –0.237*** –0.221** –0.475*** (–2.771) (–2.458) (–2.865) Size of landholdings (square root) –0.0749** –0.0939** –0.0351 (–2.417) (–2.364) (–0.830) Has any livestock –0.0182 0.00730 –0.201** (–0.234) (0.0865) (–2.201) Community characteristics Daily market –0.148 –0.148 –0.136 (–1.385) (–1.139) (–1.210) All season passable road –0.235* –0.266* 0.276 (–1.839) (–1.943) (1.422) Mobile phone signal –0.110 –0.0677 –0.252 (–0.713) (–0.386) (–1.379) (continues to next page) 108 Tanzania Mainland Poverty Assessment Table 2.B-2  Correlates of Poverty, 2011/12 (continued) (1) (2) (3)   National Rural Urban Geographic zone (Omitted: Coastal) Northern Highlands 0.0957 –0.0759 0.368* (0.473) (–0.292) (1.848) Lake 0.269* 0.145 0.552*** (1.747) (0.711) (4.085) Central 0.387** 0.265 0.594*** (2.278) (1.204) (3.011) Southern Highlands 0.587*** 0.568** 0.358* (3.356) (2.463) (1.765) South 0.914*** 0.671*** 1.354*** (4.719) (2.830) (6.970) Constant –0.770*** –0.652** –1.285*** (–3.526) (–2.424) (–5.636) Observations 9,930 4,064 5,866 t-statistics in parentheses. Standard errors corrected for clustering and stratification. Probit. Dependent variables equals to unity if house- hold is below the poverty line. Table reports coefficients (not marginal effects). *** p<0.01, ** p<0.05, * p<0.1 Appendices 109 Appendix 2.C: Migration A.  Data Description and Migration Flows between NPS1 and NPS3. We see that short term migration, defined as migration over a four-year period here, is relative- Data description ly small. As one would expect, people to be more likely to The analysis in this section is based on the three waves of the move over a longer period of time. Table 2.C-3 is a matrix National Panel Survey (NPS) described in the previous appen- that compares the current region of residence of individuals dices. For the simplicity of notation, the three waves will be with their previous region if they migrated in the previous labeled respectively NPS1, NPS2 and NPS3. Table 2.C-1 sum- 10 years. This gives a picture of migration over a longer pe- marizes the sample size and recontact rates of NPS. At the riod of time. Here we see that individuals are much more end of NPS3, 90% of the original respondents that were still likely to have migrated over a longer period of time. Approx- living were re-interviewed; 95% of the original households imately 40% of individuals lived in a different region 10 years were also recontacted in NPS3. This survey is ideal for analyz- ago than now. We see similar magnitudes of migration if we ing migration patterns in Tanzania as it tracks individuals over compare the region of individuals’ birth with their current time rather than households. However, this survey does not location (not presented here). allow us to examine international migration since individuals are recontacted only if they are present in Tanzania. Migration Decision Based on how far individuals have moved from their usu- The migration decision is examined using a multinomial al location of residence, each of the above definition has logit model. The outcome variable is the decision to migrate three categories: non-migrant, intra-regional migrant, and between NPS1 and NPS3, and includes three categories: inter-regional migrant. If an individual has lived in the same 1)  Not to migrate; 2) Migrate to a different but within the district at two points in time, then this person is classified same region; 3) Migrate to a different region. The explan- as a non-migrant. If an individual has moved to a different atory variables include individuals and households charac- district in the same region, then this person is an intra-re- teristics, and the distance of household’s residence from the gional migrant. Finally, if an individual has moved to another district headquarters that affect the migration decision in region, then this person is an inter-regional migrant. NPS1. The results are presented in Table 2.C-4. A migration matrix representing location in two different Impact of Migration on Individuals II.  points in time could be informative of the patterns of mi- and Households Welfare gration. Table 2.C-2 is a matrix of inter-regional migration The Effects of Migration on Consumption Growth The analysis of the impact of the migration decision on con- sumption growth relies on the methodology proposed by Sample Size and Recontact Table 2.C-1   Rates of Tanzania National Beegle et al. (2011). We consider the growth rate of per cap- Panel Survey ita consumption between NPS1 and NPS3. The estimated equation is the following:   NPS1 NPS2 NPS3 Households 3,265 3,924 5,011 ΔInCit+1,t = a + bMi + gXit + δh + eih(1) Individuals 16,709 20,559 25,413 Re-interviewed from previous wave — 15,597 18,968 Where, ΔInCit+1,t is the difference in logarithm of consump- Not interviewed (deceased or untraced) — 1,112 1,591 tion between t and t+1 (NPS1 and NPS3 in our dataset). Mi is New individuals — 4,962 6,445 a dummy variable indicating whether an individual moved Recontact rate, individuals*   94% 90% out of the district between NPS1 and NPS3. Xit includes in- Recontact rate, households*   97% 95% dividual characteristics at the baseline that could also affect * Eligible individuals (not deceased) or households from NPS1 consumption growth. δh is the initial household fixed effect 110 Tanzania Mainland Poverty Assessment Table 2.C-2  Inter-regional Migration between NPS1 and NPS3 Region in NPS3 (columns) Dodoma Arusha Kilimanjaro Tanga Morogoro Pwani Dar es salaam Lindi Mtwara Ruvuma Iringa Mbeya Singida Tabora Rukwa Kigoma Shinyanga Kagera Mwanza Mara Manyara KASKAZINI UNGUJA KUSINI UNGUJA MJINI/MAGHARIBI UNGUJA KASKAZINI PEMBA KUSINI PEMBA Total Region in NPS1 (rows) Dodoma 94 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 5 0 0 0 0 0 100 Arusha 0 93 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 100 Kilimanjaro 0 0 98 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 100 Tanga 0 1 0 96 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 100 Morogoro 1 1 0 0 96 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 Pwani 1 0 0 0 0 94 4 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 100 Dar es salaam 0 0 1 1 0 1 92 1 0 0 1 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 100 Lindi 0 0 0 0 0 0 0 98 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 Mtwara 0 0 0 1 0 0 0 1 98 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 Ruvuma 0 0 0 0 0 0 0 0 0 99 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 Iringa 0 0 0 0 0 0 0 0 0 0 99 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 100 Mbeya 0 0 0 0 0 0 1 0 0 0 1 97 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 Singida 0 0 0 0 1 0 0 0 0 0 0 0 94 1 0 0 4 0 0 0 0 0 0 0 0 0 100 Tabora 0 0 0 0 0 0 0 0 0 0 0 0 1 94 0 0 2 0 1 0 0 0 0 0 0 0 100 Rukwa 0 0 0 0 0 0 0 0 0 0 0 0 0 0 97 0 0 0 2 0 0 0 0 0 0 0 100 Kigoma 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 0 0 100 Shinyanga 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 97 0 0 1 0 0 0 0 0 0 100 Kagera 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 94 5 0 0 0 0 0 0 0 100 Mwanza 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 2 1 95 0 0 0 0 0 0 0 100 Mara 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 98 0 0 0 0 0 0 100 Manyara 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 98 0 0 0 0 0 100 KASKAZINI UNGUJA 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 99 0 0 0 0 100 KUSINI UNGUJA 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 86 14 0 0 100 MJINI/MAGHARIBI UNGUJA 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 3 0 0 0 0 1 0 94 0 1 100 KASKAZINI PEMBA 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 99 1 100 Appendices KUSINI PEMBA 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 100 111 112 Table 2.C-3  Did you move to your current region in the last 10 years? Region in NPS3 (columns); Dodoma Arusha Kilimanjaro Tanga Morogoro Pwani Dar es salaam Lindi Mtwara Ruvuma Iringa Mbeya Singida Tabora Rukwa Kigoma Shinyanga Kagera Mwanza Mara Manyara KASKAZINI UNGUJA KUSINI UNGUJA MJINI/MAGHARIBI UNGUJA KASKAZINI PEMBA KUSINI PEMBA Total Region in NPS1 (rows) Dodoma 57 3 2 0 10 0 3 0 0 0 2 0 1 1 0 1 0 2 0 0 17 0 0 0 0 0 100 Tanzania Mainland Poverty Assessment Arusha 0 53 21 2 1 0 3 0 0 0 0 1 7 0 0 1 1 0 0 1 8 0 0 0 0 0 100 Kilimanjaro 0 23 38 8 8 1 9 0 0 0 2 0 4 0 1 2 1 0 2 0 2 0 0 0 0 0 100 Tanga 0 1 1 68 0 7 5 0 5 0 2 0 3 0 0 0 0 1 0 0 6 0 0 0 1 0 100 Morogoro 6 1 4 3 45 1 6 0 4 1 2 2 3 4 0 2 14 2 0 0 0 0 0 0 0 0 100 Pwani 9 0 3 1 2 47 31 2 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 1 0 1 100 Dar es salaam 3 1 4 4 5 4 63 2 1 0 2 1 1 1 0 1 1 1 3 1 0 0 0 0 0 0 100 Lindi 0 0 0 0 3 4 5 66 18 1 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 100 Mtwara 3 0 0 2 0 0 6 9 72 1 0 4 0 1 0 0 0 0 0 0 0 0 0 0 0 0 100 Ruvuma 0 1 0 0 0 0 4 2 5 69 1 0 1 15 0 0 0 0 2 0 0 0 0 0 0 0 100 Iringa 4 0 2 0 1 0 2 0 0 3 70 10 0 0 2 0 0 0 5 2 0 0 0 0 0 0 100 Mbeya 1 0 0 0 0 0 5 0 0 3 8 71 1 2 6 0 0 0 3 0 0 0 0 0 0 0 100 Singida 0 0 3 0 1 0 1 0 0 0 0 2 66 13 0 0 12 0 0 0 1 0 0 0 0 0 100 Tabora 0 0 0 0 0 0 0 0 0 1 1 4 4 51 1 4 28 1 4 1 0 0 0 0 0 0 100 Rukwa 0 0 1 0 0 0 0 0 0 0 0 12 0 1 58 2 10 0 14 1 0 0 0 0 0 0 100 Kigoma 0 0 3 0 0 0 1 0 0 0 0 0 0 1 1 89 0 1 3 2 0 0 0 0 0 0 100 Shinyanga 1 0 0 2 0 0 2 0 0 1 0 0 3 3 0 2 65 3 14 3 0 0 0 0 0 0 100 Kagera 0 0 1 0 0 0 3 0 0 0 0 0 0 0 0 0 4 72 18 2 0 0 0 0 0 0 100 Mwanza 0 0 1 1 2 1 1 0 0 0 0 0 0 1 0 1 9 8 68 7 0 0 0 0 0 0 100 Mara 0 0 0 1 1 0 6 0 0 0 0 0 1 0 0 3 8 0 9 70 0 0 0 0 0 0 100 Manyara 1 25 11 1 0 0 3 0 0 0 0 0 4 0 0 0 2 0 0 0 52 0 0 0 0 0 100 KASKAZINI UNGUJA 0 0 0 2 0 0 5 0 0 0 0 0 0 23 0 0 2 0 1 0 0 44 4 11 1 6 100 KUSINI UNGUJA 0 0 0 0 4 0 9 0 0 0 0 0 0 5 0 0 0 0 26 0 0 0 27 27 0 2 100 MJINI/MAGHARIBI UNGUJA 0 0 0 1 0 1 6 1 0 0 0 0 0 0 0 0 0 0 0 0 0 8 3 65 3 9 100 KASKAZINI PEMBA 0 0 0 1 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4 0 7 68 17 100 KUSINI PEMBA 4 0 0 3 0 0 3 1 0 0 0 0 0 0 0 0 2 0 0 0 0 1 0 10 20 56 100 Table 2.C-4  Multinomial Logit Model of Migration Decisions Non-migrant Intra-regional Migrant Inter-regional Migrant β s.e. β s.e. β s.e. Asset index –0.003** (0.001) 0.001 (0.001) 0.002** (0.001) Years of education –0.002 (0.001) –0.000 (0.001) 0.002** (0.001) Household size 0.002** (0.001) –0.002** (0.001) 0.000 (0.000) Age group (Base: 0–14 years) Age group: 15–29 years –0.052*** (0.011) 0.028*** (0.008) 0.025*** (0.008) Age group: 30–44 years –0.034** (0.014) 0.021** (0.010) 0.014 (0.011) Age group: 45–59 years –0.000 (0.016) 0.012 (0.011) –0.011 (0.013) Age group: 60+ years 0.011 (0.020) 0.019* (0.011) –0.030* (0.018) Female –0.001 (0.009) 0.003 (0.006) –0.002 (0.007) Unmarried male 0.001 (0.014) –0.007 (0.010) 0.006 (0.010) Unmarried –0.008 (0.012) –0.000 (0.008) 0.009 (0.009) Household head or spouse 0.049*** (0.013) –0.025*** (0.009) –0.025** (0.010) Child of household head 0.016 (0.011) –0.008 (0.007) –0.008 (0.008) Male child of household head 0.022 (0.014) –0.006 (0.010) –0.016 (0.010) Distance to district HQ, km 0.000 (0.000) –0.000** (0.000) 0.000*** (0.000) Observations 10,681 10,681 10,681 Sources: National Panel Survey, NPS1 and NPS3. Note: A key assumption of the multinomial logit model is the Independence of Irrelevant Alternatives (IIA), which suggests that the choice between two alternatives is unaffected by the addition or subtraction of alternative choices. The Small-Hsiao test of IIA suggests that this assumption, that the odds between two choices are independent of other alternatives, cannot be rejected. * , **, and *** indicate statistical significance at the levels of 10, 5 and 1 percent respectively. The figures represent marginal effects. and εih is a random error term. β is our variable of interest The results of this IHHFE estimation are presented in Table and represents the impact of migration on consumption 2.C-5. Column (1) is a parsimonious model without any co- growth between t and t+1. variates. In this specification, we see that migrants had a 15.7 percentage point higher growth of consumption than Although estimating equation (1) by differences-in-differ- non-migrants. This is an economically significant impact of ences would wipe out the effects of time-invariant individ- migration, given the fact that the sample average consump- ual-specific characteristics that could affect consumption tion growth between NPS1 and NPS3 is 41.8%. growth, this is not sufficient to tease out the impact of mi- gration on consumption growth. Initial household charac- Column (2) contains a set of individual characteristics at teristics such as assets and social networks could influence the baseline as conditioning variables, but the coefficient consumption growth. Since we know the original house- on migration is virtually identical to the parsimonious hold that all respondents belonged to in NPS1, we use this model. Although column (2) controls for household-lev- feature to identify the effect of migration on consumption. el heterogeneity and individual-level heterogeneity, we Following Beegle et al. (2011), we identify the impact of mi- could still worry that it does not control for unobservable gration on consumption growth by estimating equation (1) individual-specific characteristics such as motivation and using initial household fixed effects (IHHFE), effectively us- social network that could also affect consumption growth. ing variation within the initial household to control for initial We address this endogeneity concern by implement- growth paths of households. ing the instrumental variable estimation. The estimation Appendices 113 procedure proceeds in two stages: the estimation of the log of remittances received by the household. To uncover migration decision in the first stage and the estimation of the effect of remittances on children’s schooling, the anal- the consumption growth in the second stage. The key to ysis controls for various individual and household-specific identification is a set of instruments that affect the migra- characteristics that may affect school attendance. Given tion decision but not the consumption growth. Similar to that receiving remittances may be associated with the Beegle et al. (2011), we use the following push and pull migration status of the household, the analysis controls factors as instruments of migration: head or spouse, child also for this variable considering as migrant household a of head, distance of household from district headquarters household in which at least one member moved to a dif- interacted with age, and an index of self-reported shock ferent district in NPS3. index (constructed through Principal Component Anal- ysis) interacted with age. Column (4) presents the esti- The estimation results in Table 2.C-6 may be biased by dif- mates from the first stage of 2SLS estimation. These results ferences between households who receive remittances and suggest that the set of instruments chosen pass the un- those who do not. Controlling for households characteristics der-identification and weak-identification tests. does not help to fully address this selection bias. This prob- lem is difficult to handle in the absence of a randomized Column (3) presents results from the 2SLS estimation with experiment, but we tried to address it through the quasi-ex- IHHFE. These results suggest that even after controlling for perimental method of Propensity Score Matching (PSM). the endogeneity resulting from unobservable individual The approach consists in matching the treatment group to characteristics, migration has a significant impact on con- a comparison one within the sample of non-participants us- sumption growth. Migrants have 21.2 percentage point ing the propensity score (the predicted probability of partic- higher consumption growth in consumption in this estima- ipation given observed characteristics). Although the PSM tion, which is slightly higher than the IHHFE estimates. Taken approach does not completely solve the problem of selec- together, Table 2.C-5 suggests that migrating to a different tion bias, it helps to attenuate it and in our case it provides district could lead to a significantly increase in consumption a consistent estimation of the impact of remittances on the growth than staying in the same district, even during a rela- probability of attending school.72 tively short time-period of about 4 years. The PSM results are presented in Table 2.C-7 and show that The Effect of Remittance Receipt on Children receiving remittances is associated with a higher probability School Attendance of school attendance. Children living in households that re- The impact of remittance receipt on the school atten- ceived remittances in the previous 12 months were 3.3–5.8 dance of children in Tanzania is examined using a probit percentage points more likely to attend school compared regression model, where the outcome variable is a bina- with the counterfactual. This impact is much smaller than ry indicator of whether or not a school-age child is cur- the impact reported in Table 2.C-6, suggesting that selec- rently attending school. The primary explanatory variable tion issues may be important. Similar to the probit model, is whether the child’s household receives remittances, the coefficient on migration is the opposite of the coeffi- which captures the effect of receiving remittances on cient on remittances, suggesting that the beneficial impact relaxing the financial constraints of the household and of remittances on school attendance may be mitigated by allowing children spend less time in income-generation the adverse effects of migration. activities. The analysis considers both, whether the house- hold receives remittances and the amounts received. The 72  Rosenbaum and Rubin (1983) propose that under the assump- estimation results are presented in Table 2.C-6, where col- tions of selection on observables and the overlap condition, the umns 1–3 include a dummy variable representing wheth- difference between the mean outcomes for treatment and control er the child’s household received remittances in the pre- groups at each level of the propensity score provides an unbiased vious 12 months and columns 4–6 consider the natural estimate of the average treatment effect on the treated (ATT). 114 Tanzania Mainland Poverty Assessment Table 2.C-5  The Impact of Migration on Consumption Growth (1) (2) (3) (4) IHHFE IHHFE 2SLS with IHHFE 2SLS with IHHFE First Stage Moved outside of district 0.157*** 0.160*** 0.212** (0.038) (0.009) (0.105) Characteristics at baseline: Years of education 0.005*** 0.005*** 0.006*** (0.001) (0.001) (0.001) Age, years 0.001 0.000 0.003** (0.001) (0.001) (0.001) Age, squared –0.000 –0.000 –0.000** (0.000) (0.000) (0.000) Unmarried male –0.007 –0.004 –0.028* (0.012) (0.013) (0.015) Unmarried 0.040*** 0.039*** 0.074*** (0.010) (0.015) (0.013) Male 0.009 0.007 –0.022*** (0.006) (0.007) (0.008) Instruments for migration: Head or spouse –0.116*** (0.017) Child of head –0.058*** (0.011) Distance from HQ X 15–30yrs 0.001*** (0.000) Shock index X 15–30 years 0.010* (0.005) Constant 0.241*** 0.199*** (0.042) (0.013) Observations 14,473 12,127 10,380 10,380 Cragg-Donald statistic 16.56 Anderson LM statistic 65.80 Sargan statistic 17.77 Source: National Panel Survey, NPS1 and NPS3. Standard errors in parentheses. * p < .10, ** p < .05, *** p < .01 Appendices 115 Table 2.C-6  Determinants of Whether a Child (6–18 years) is Currently Attending School (1) (2) (3) (4) (5) (6) Remittance-recipient household 0.230*** 0.198*** 0.205*** (0.056) (0.056) (0.060) Log of remittances received 0.017*** 0.015** 0.015** (0.006) (0.006) (0.007) Migrant household –0.189** –0.193** (0.077) (0.076) Age (years) –0.114*** –0.111*** –0.114*** –0.111*** (0.008) (0.008) (0.008) (0.008) Male –0.534*** –0.451*** –0.553*** –0.472*** (0.134) (0.143) (0.135) (0.144) Male X Age 0.045*** 0.038*** 0.046*** 0.039*** (0.010) (0.011) (0.010) (0.011) Household size –0.012* –0.010 –0.012* –0.011 (0.007) (0.007) (0.007) (0.007) Household dependency ratio 0.276** 0.242* 0.277** 0.246** (0.122) (0.125) (0.121) (0.123) Household head is literate 0.251*** 0.254*** 0.251*** 0.253*** (0.048) (0.049) (0.048) (0.049) Household asset index 0.115*** 0.116*** 0.116*** 0.118*** (0.014) (0.015) (0.014) (0.015) Urban 0.139** 0.151** 0.142** 0.156** (0.068) (0.070) (0.069) (0.070) Constant 0.404*** 0.929*** 0.916*** 0.435*** 0.952*** 0.938*** (0.030) (0.192) (0.200) (0.030) (0.192) (0.199) Observations 8009 7970 7245 8009 7970 7245 Source: National Panel Survey, NPS3. Note: Standard errors in parentheses. * p < .10, ** p < .05, *** p < .01. All specifications use survey weights. Columns 2, 3, 5, and 6 also control for the region of individual’s residence. Table 2.C-7  Determinants of School Attendance, Age 6–18 years (Propensity Score Matching) (1) (2) (3) Treatment variable=> Household received remittances Household received remittances Migrant household ATE in population2 0.033** 0.058*** –0.058*** (0.018) (0.017) (0.019) ATE on the treated 0.052*** 0.063*** –0.087*** (0.018) (0.019) (0.022) Control variables: Migrant household No Yes No Other control variables 1 Yes Yes Yes Standard errors in parentheses. * p < .10, ** p < .05, *** p < .01. 1 The following variables were used to generate the propensity score: age of child in years, male, male X age, household size, household dependency ratio, household head is literate, and household asset index. 2 ATE refers to the Average Treatment Effect 116 Tanzania Mainland Poverty Assessment Sustainability of Migration Table 2.C-8  Access to Health Care Non-lifetime migrants Lifetime migrants Difference Consulted health care provider in the last 4 weeks 18.30% 19.71% Hospitalized in the last 12 months 6.29% 6.01% Medical exemption 4.45% 3.15% ** Mean amount spent on illnesses and injuries (TZSS) 2,136 6,017 ** Non-long run migrants Long run migrants Difference Consulted health care provider in the last 4 weeks 14.13% 18.99% *** Hospitalized in the last 12 months 4.85% 5.59% Medical exemption 2.86% 3.33 Mean amount spent on illnesses and injuries (TZSS) 2,668 2,648 Non-recent migrants Recent migrants Difference Consulted health care provider in the last 4 weeks 14.84% 18.59% ** Hospitalized in the last 12 months 4.41% 4.26% Medical exemption 3.33% 3.55 Mean amount spent on illnesses and injuries (TZSS) 2,426 2,883 Appendices 117 Appendix 3: Comparison of Poverty Trends Using NPS and HBS Data A Divergent Perspective of Poverty Trends: National Panel Survey vs Household Box 3.1  Budget Survey Tanzania has a second survey series that collects consumption data and is hence suitable for poverty analysis, the National Panel Sur- vey (NPS), with three rounds so far (2008/09, 2010/11, and 2012/13). The NPS is a longitudinal survey (tracking individuals) conducted every two years by the National Bureau of Statistics. The NPS is representative for the whole of Tanzania (including Zanzibar), and has a smaller sample size than the HBS. The panel nature of the data make it a particularly attractive survey for studying poverty dynamics and transitions, and the survey series is used in this poverty assessment for this and other purposes. The NPS shows a different trend in poverty than the HBS series. In particular, the poverty headcount for mainland Tanzania in the NPS increased from 11.4 percent in 2008/09 to 21.1 percent in 2012/13 and growth incidence curves show a decline in consump- tion that is most pronounced for the poorest groups. 0 –2 –4 –6 –8 0 20 40 60 80 100 Growth rate by percentile Growth rate in mean Source: National Panel Survey (NPS), 2008/09 and 2012/13. These discrepancies to the HBS are striking and not easy to explain. Part of the difference in poverty trends is likely to be related to various technicalities of the underlying survey data and poverty estimation methodology. For instance, the NPS uses a food price deflator to update the poverty line, while the HBS uses a combined food and nonfood price deflator. Since food price inflation outpaced nonfood price inflation in recent years, the NPS poverty lines escalate more rapidly than the HBS poverty lines, which contributes to the increase in poverty in the NPS. Differences in the way the surveys collect data on consumption are also likely to play a role but require further investigation and triangulation. (See URT 2011 for a discussion). However, more substantive explanations also need to be considered. In particular, the NPS base year (2008/09) does not perfect- ly coincide with the HBS base year (2007). This difference can matter, given the sensitivity of the poverty estimates to price changes and the high variability of incomes in agriculture and the informal household enterprise sector. In fact, the NPS data show a signif- icant amount of churning, i.e. individual movements into and out of poverty. This confirms once again that many Tanzanians are vulnerable to poverty, and that what is observed at any given time is just a snapshot of what is in reality a rapidly evolving scenario. It is important to note, however, that while the NPS shows an increase in consumption poverty 2008/09–2012/13, the data still show improvements in housing conditions, assets, and access to basic services. In sum, we believe that the decline in poverty in the HBS is more plausible and more consistent with other indicators of well-being than the increase in poverty in the NPS, but further analytical work is need to cross-triangulate HBS and NPS poverty estimates and explain the discrepancies between the two data sets. 118 Tanzania Mainland Poverty Assessment Appendix 4: The Unconditional Quantile Regression Model & Analysis of Spatial Inequality The static decomposition of inequality by population The Recentered Influence Function (RIF) regression approach groups is a useful descriptive analysis and can be informa- recently proposed by Firpo, Fortin and Lemieux (2009) ad- tive regarding the role played by certain household char- dresses these shortcomings and provides a simple regres- acteristics in inequality. However, it has several limitations. sion-based procedure for performing a detailed decompo- First, handling an important number of population groups sition of different distributional statistics such as quantiles, with different categories for each population partition is variance and Gini coefficient. The RIF-regression model is often unwieldy and limits the reliability of the estimates. called unconditional quantile regression when applied to the Second, it does not allow to infer causality in the relation- quantiles. The technique consists of decomposing the wel- ship between inequality and the different household attri- fare gaps at various quantiles of the unconditional distribu- butes. Some of the variables used to explain inequality may tion into differences in households endowment characteris- themselves be determined by the welfare patterns and the tics such as education, age, employment etc., and differences direction of causation cannot be determined from the de- in the returns to these characteristics. These components are scriptive analysis. Third, and most importantly, the decom- then further decomposed to identify the specific attributes position gives little information regarding the importance which contribute to the widening welfare gap. of the welfare gaps across the various quantiles of the distri- bution and about the sources of these gaps. We apply the RIF unconditional quantile regression to exam- ine the rural-urban as well as the metropolitan-nonmetropol- We attempt to address this drawback via the unconditional itan welfare differentials at various points of the consumption quantile regression model. The model analyzes the sources of distribution. The procedure is carried out in two stages. The inequality between rural and urban areas, and between met- first stage consists of estimating unconditional quantile re- ropolitan and non-metropolitan locations. The procedure al- gressions on log real per capita monthly household consump- lows to understand how the difference in the distributions tion for rural and urban (metropolitan and non-metropolitan) of observed household characteristics between the locations households, then constructing a counterfactual distribution contribute to the welfare gap and how the marginal effects that would prevail if rural (non-metropolitan) households of these characteristics vary across the entire distribution. have received the returns that pertained to urban (metropoli- tan) area. The comparison of the counterfactual and empirical Popular approaches used in the decomposition of distribu- distributions allows to estimate the part of the welfare gap tional statistics and the analysis of the sources of inequal- attributable to households characteristics differentials, the ity include the standard Oaxaca–Blinder decomposition endowment effect, and the part explained by differences in method, the reweighting procedure of DiNardo, Fortin, and returns to characteristics, the return effect. The second stage Lemieux (1996) and the quantile-based decomposition ap- involves dividing the endowment and return components into proach of Machado and Mata (2005). The main drawback the contribution of each specific characteristic variable. of the Oaxaca–Blinder technique is that it applies the de- composition to only the mean welfare differences between The method can be easily implemented as a standard linear two population sub-groups and yields an incomplete repre- regression, and an ordinary least squares (OLS) regression of sentation of the inequality sources. The other conventional the following form can be estimated: methods extend the decomposition beyond the mean and permit the analysis of the entire distribution, nevertheless RIF(y,Qθ)=Xb+e(1) they all share the same shortcoming in that they involve a number of assumptions and computational difficulties (For- where y is log real per capita monthly household consump- tin et al., 2010). tion, and RIF(y,Qθ) is the RIF of the θth quantile of y estimated Appendices 119 by computing the sample quantile Qθ and estimating the household surveys includes observations on migration status density of y at that point by kernel method: and access to drinking water that are absent in the previous surveys, and that the categories of sector of employment and RIF ( y ,Qθ ) = Qθ + ( θ − I { y ≤ Qθ }) source of income variables differ somewhat in 2011/12 data. fY (Qθ ) , is the marginal den- However, the results remain consistent to different specifica- sity function of y and I is an indicator function. RIF can be tions, whether considering the same variables and categories estimated by replacing Qθ by θth sample quantile and esti- across the three waves or using the more detailed informa- mating fY by kernel density.73 tion available in the latest survey. X is the regressors matrix including the intercept, β is the We estimate model (1) for the 10th to 90th quantiles and use regression coefficient vector and ε is the error term. The re- the unconditional quantile regression estimates to decom- gressors include eight groups of variables: (1) the household pose the rural-urban inequality, as well as the metropoli- demographic and general characteristics variables includ- tan-nonmetropolitan, inequality into a component attribut- ing household size, the proportion of household members able to differences in the distribution of characteristics and a aged below fourteen years and the proportion of those aged component due to differences in the distribution of returns over 65 years, and the gender of the household head; (2) the as follows: household human capital measured by the number of years of schooling of the more highly educated of the head or his ˆ i −Q Qθ ˆ i' = Q θ { θ }{ ˆ i −Q ˆ* + Q ˆ * −Q θ θ } ( ˆ i + X i' β ˆ i' = X i − X i' β θ ) ˆ i −β θ (θ ) ˆ i' (2) θ spouse, and the head’s years of experience.74 The choice of the years of schooling variable is motivated by capturing where Qˆ is the θth unconditional quantile of log real per θ the influence that family members with more education capita monthly household consumption, X represents the may have in household decision making; (3) the household vector of covariate averages and β ˆ the estimate of the θ head employment sector and other attributes, which include unconditional quantile partial effect. Superscripts i, i’ and * a dummy variable indicating whether the head is over 65 designate respectively the urban (or metropolitan), rural (or years old, his marital status, and his sector of employment nonmetropolitan) and counterfactual values. recoded as a categorical variable: : (i) Government; (ii) Pri- vate sector, NGOs and international companies; (iii) self-em- ˆ i is the counterfactual quantile of the uncondi- ˆ * = X i' β Q θ ployed with others; (iv) self-employed alone; (v)  household tional counterfactual distribution which represents the dis- duties; (vi) farming and fishing; and (vii) unemployed & inac- tribution of welfare that would have prevailed for group i’ tive; (5)  asset ownership including the area of land owned, (rural/non-metropolitan households) if they have received rented and provided for free; dummy variables indicating group i (urban/metropolitan households) returns to their respectively whether the household owns livestock, bicycle, characteristics.75 cell phone, telephone, computer; and dummies capturing the housing conditions; (6) the sources of income, captured The first term on the right-hand side of equation (2) rep- by categorical variables indicating the main source of in- resents the contribution of the differences in distributions come of the household and including: (i) cash and in-kind income from employment; (ii) income from non-agricultural household business; (iii) income from agricultural household 73  For more details see Firpo, Fortin and Lemieux (2009). business; (iv) transfers/assistance/remittances; and (v) other The squared schooling years and the squared experience were 74  not significant in any equation, thus we excluded them. sources; (7) access to basic services measured by categorical 75  The decomposition results may vary with the choice of the variables indicating the sources of lighting and of drinking counterfactual distribution. For example, if the counterfactual used water; (8)  external factors to the household capturing the is the distribution that would have prevailed for group i if they have community characteristics such as access to transportation, received group i’ returns we would obtain different results. The schooling and hospital facilities as well as geographic loca- choice of the counterfactual in this analysis is motivated by the aim tion fixed effects. It is worth mentioning that the 2011/12 of emphasising household groups living in disadvantaged areas. 120 Tanzania Mainland Poverty Assessment of household characteristics to inequality at the θth uncondi- tional quantile, denoted endowment effect. The second term ˆ i −Q Qθ θ k ( ˆ * = ∑ X i − X i' β k k ) ˆ * −Q ˆ i and Q θ ,k θ ˆ i' = ∑ X i' β θ k k ( ˆ i− β θ ,k ) ˆ i' k : 1K θ,k ˆ* =∑ ˆ i −Q of the right-hand side of the equation represents Q θ θ k theX (i inequal- k k ˆi − X i' β ) θ ,k ˆ * −Q and Q θ ˆ i' θ = ∑ X (β k i' k ˆ i ˆ i' ) k : 1K (3) θ ,k − βθ,k ity due to differences (or discrimination) in returns to the household characteristics at the θ unconditional quantile, th where k designates the individual specific household characteristics. The endowment and return effects can be further decom- posed into the contribution of individual specific households Tables 4-1 to 4-3 below, present the estimation results of characteristics (or group of some characteristics) as follows: equation (1) for the survey years 2001, 2007 and 2011/12, respectively. The decomposition results of equations (2) and (3) are summarized in tables 4-4 and 4-5. Appendices 121 122 Table 4-1  Estimated Coefficients for Unconditional Quantile Regression Model, 2001 2001 Urban Rural Metropolitan Nonmetropolitan 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile HH size –0.056*** –0.065*** –0.062*** –0.062*** –0.044*** –0.057*** –0.068** –0.08*** –0.084*** –0.057*** –0.048*** –0.05*** –(4.580) –(12.200) –(8.570) –(4.840) –(7.510) –(7.060) –(3.310) –(6.800) –(5.160) –(4.750) –(8.640) –(7.620) pch14 –0.402*** –0.576*** –1.09*** –0.437** –0.671*** –1.313*** –0.099 –0.346* –0.932** –0.448*** –0.653*** –1.183*** –(4.120) –(7.670) –(9.510) –(3.210) –(9.220) –(8.270) –(0.550) –(1.850) –(3.370) –(3.830) –(10.000) –(9.590) Tanzania Mainland Poverty Assessment peld65 0.038 –0.498** –0.888*** –0.36 –0.35** –0.283 0.659 –0.832** –1.674*** –0.246 –0.222 –0.262 (0.140) –(3.470) –(4.740) –(1.190) –(2.420) –(0.900) (1.230) –(2.350) –(4.060) –(0.900) –(1.600) –(1.020) Male head –0.005 –0.096** 0.17** –0.044 –0.003 0.263** 0.068 –0.121 0.249 –0.052 0.003 0.24** –(0.060) –(2.210) (2.490) –(0.400) –(0.060) (2.130) (0.590) –(1.080) (1.410) –(0.570) (0.060) (2.790) Head over 65 yrs –0.192 0.02 0.144** 0.078 0.096 –0.025 –0.527 –0.032 0.481** 0.028 0.053 –0.025 –(1.390) (0.350) (2.330) (0.610) (1.430) –(0.250) –(1.510) –(0.220) (2.740) (0.240) (0.850) –(0.320) Marital status (Omitted=married) Never Married 0.091 0.026 0.601*** –0.057 0.124** 0.698** 0.036 0.106 0.636** –0.031 0.104** 0.733*** (1.210) (0.570) (6.020) –(0.680) (2.250) (3.300) (0.250) (0.890) (2.850) –(0.460) (2.230) (4.830) Divorced 0.087 0.114** 0.311*** –0.055 0.094* 0.463*** 0.419*** 0.32** 0.359** –0.021 0.062 0.408*** (1.050) (2.410) (4.250) –(0.450) (1.780) (3.660) (3.570) (2.990) (2.070) –(0.210) (1.310) (4.440) Widowed 0.049 –0.021 0.182** –0.022 –0.003 0.295* 0.349* –0.043 0.279 0.009 –0.051 0.248* (0.420) –(0.330) (2.240) –(0.180) –(0.040) (1.720) (1.850) –(0.290) (1.240) (0.090) –(0.760) (1.930) Max. Education 0.022** 0.019*** 0.023** 0.003 0.012** 0.032** 0.021* 0.03** 0.025 0.007 0.012** 0.027*** (3.310) (4.020) (3.220) (0.350) (2.100) (3.320) (1.810) (2.970) (1.530) (0.830) (2.400) (3.930) Sector of Employment (Omitted= Farming & Fishing) Government 0.073 0.238** 0.31 0.072 0.225** 0.711** 0.302 0.247 –0.145 0.079 0.202** 0.345* (0.950) (3.480) (1.620) (0.520) (2.360) (2.330) (1.200) (1.210) –(0.580) (0.770) (2.800) (1.750) Private & NGO 0.128* 0.208*** 0.165** –0.119 0.108 0.212 0.361 0.357** –0.024 –0.068 0.106** 0.205 (1.790) (3.590) (2.040) –(0.770) (1.500) (1.010) (1.420) (2.430) –(0.130) –(0.660) (1.980) (1.550) Self-emp.(others) 0.072 0.281*** 0.151 0.212 0.152 0.068 0.111 0.434** 0.075 0.104 0.136 0.16 (0.680) (3.760) (1.440) (1.430) (1.050) (0.350) (0.410) (2.380) (0.430) (1.090) (1.390) (0.960) (continues to next page) Table 4-1  Estimated Coefficients for Unconditional Quantile Regression Model, 2001 (continued) 2001 Urban Rural Metropolitan Nonmetropolitan 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile Self-emp. (alone) 0.075 0.211*** 0.119** –0.031 0.055 0.212 0.299 0.29** 0.039 –0.028 0.06 0.089 (1.050) (4.380) (2.180) –(0.340) (1.000) (1.640) (1.220) (2.060) (0.260) –(0.470) (1.410) (1.110) HH duties –0.012 0.045 –0.063 –0.083 –0.133* –0.084 0.029 –0.256 –0.269 –0.101 –0.096 –0.122** –(0.100) (0.650) –(0.870) –(0.490) –(1.740) –(1.160) (0.090) –(1.280) –(1.260) –(0.750) –(1.440) –(2.240) Unemployed 0.09 0.192** –0.012 –0.061 –0.036 0.071 0.34 0.371** –0.055 –0.049 –0.008 0.04 (0.830) (2.830) –(0.180) –(0.550) –(0.520) (0.440) (1.330) (2.310) –(0.300) –(0.500) –(0.130) (0.310) Main source of income (Omitted= Income from agricultural HH business) No Income –0.518* –0.185 –0.114 –1.164** –0.333** 0.243 –0.601 –0.186 –0.309 –0.916** –0.333** 0.169 –(1.730) –(1.420) –(0.480) –(2.400) –(2.970) (0.410) –(1.270) –(0.590) –(1.320) –(2.410) –(3.300) (0.380) Cash & in kind from 0.062 0.006 –0.014 –0.036 –0.068 0.006 0.165 –0.175 0.031 –0.041 –0.027 0.062 employment (0.720) (0.100) –(0.210) –(0.220) –(1.170) (0.040) (0.620) –(0.940) (0.150) –(0.340) –(0.550) (0.590) Non-agricultural HH business 0.106 0.019 0.093* 0.03 0.089** 0.156* 0.249 –0.161 0.102 0.03 0.07* 0.135** (1.210) (0.410) (1.950) (0.410) (2.260) (1.910) (0.930) –(0.890) (0.530) (0.440) (1.880) (2.110) Agr. & nonagric. Cooperatives –0.17 –0.008 –0.017 0.316* –0.042 –0.12 0.566* –0.543** 0.139 0.305* –0.02 –0.085 –(0.540) –(0.070) –(0.110) (1.790) –(0.220) –(0.800) (1.810) –(2.680) (0.570) (1.860) –(0.110) –(0.720) Transfer & assistance 0.06 0.058 0.149 0.155** –0.027 –0.17** 0.289 0.102 0.44 0.165** –0.02 –0.138** (0.490) (0.820) (1.390) (2.300) –(0.420) –(2.320) (1.020) (0.440) (1.480) (2.690) –(0.350) –(2.440) Other 0.135 –0.119 –0.134 0.168 –0.121 0.228 0.272 –0.222 –0.085 0.124 –0.114 –0.007 (1.230) –(1.110) –(1.180) (1.040) –(0.950) (0.780) (0.910) –(0.800) –(0.350) (1.170) –(1.320) –(0.040) Source of lighting (Omitted= Kerosene) Electricity 0.105** 0.188*** 0.314*** –0.115 0.126 0.267 –0.052 0.079 0.07 –0.021 0.149** 0.455*** (2.190) (4.730) (4.320) –(0.690) (1.170) (1.320) –(0.590) (0.890) (0.880) –(0.380) (3.400) (4.340) Firewood –0.566* –0.124 0.123 0.079 –0.07 0.045 –0.648 –0.2 –0.006 0.088 –0.107** 0.073 –(1.870) –(1.280) (0.750) (0.840) –(1.250) (0.500) –(1.540) –(0.960) –(0.020) (0.970) –(2.140) (0.920) No of sleeping rooms 0.059 –0.014 –0.013 0.157** 0.076* 0.033 0.142 –0.044 0.076 0.13** 0.037 0.016 (1.040) –(0.400) –(0.180) (2.420) (1.780) (0.390) (1.520) –(0.530) (0.620) (2.290) (0.980) (0.240) Dwelling roof (metal & 0.269** 0.163*** 0.021 0.259*** 0.228*** 0.301*** –0.187* –0.299** –1.095** 0.218*** 0.264*** 0.225*** concrete) (3.290) (4.830) (0.290) (5.230) (6.230) (3.950) –(1.830) –(2.100) –(2.150) (4.070) (7.860) (3.830) Appendices (continues to next page) 123 124 Table 4-1  Estimated Coefficients for Unconditional Quantile Regression Model, 2001 (continued) 2001 Urban Rural Metropolitan Nonmetropolitan 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile Area of land 0.202* 0.103 0.276 0.088 0.076* 0.185* 0.285 0.119 0.700 0.097 0.091** 0.190* (1.77) (1.02) (1.08) (1.18) (1.85) (1.78) (0.39) (1.16) (0.350) (1.40) (2.11) (1.89) Well 0.193* –0.116 –0.042 0.398*** 0.054 0.164 0.142 0.861*** –0.545** 0.394*** 0.082 0.003 (1.880) –(0.710) –(0.360) (4.470) (0.820) (1.250) (1.340) (5.760) –(2.210) (5.040) (1.280) (0.030) Tanzania Mainland Poverty Assessment Livestock 0.074 0.085* 0.124 –0.043 0.008 –0.038 0.171 0.03 0.789* –0.064 0.013 0.006 (1.030) (1.940) (1.360) –(0.690) (0.250) –(0.650) (1.640) (0.240) (1.740) –(1.090) (0.420) (0.140) Radio 0.379*** 0.233*** 0.03 0.155** 0.101** 0.154** 0.166 0.17* 0.129 0.178** 0.112** 0.125** (5.670) (6.430) (0.500) (2.240) (2.930) (2.560) (1.510) (1.970) (1.230) (2.820) (3.460) (2.600) Computer 0.118 0.279 0.331 0.292*** –0.149 –0.275* 0.045 0.159 0.509 0.254*** –0.056 –0.236* (1.290) (1.620) (1.530) (5.390) –(1.160) –(1.660) (0.460) (0.840) (1.090) (5.050) –(0.440) –(1.910) Bicycle 0.075 0.074** 0.217** 0.21** 0.076** 0.083 0.063 –0.02 0.065 0.19** 0.074** 0.086* (1.640) (2.410) (2.830) (2.750) (2.170) (1.330) (0.720) –(0.210) (0.490) (2.820) (2.330) (1.750) Tel (land line) 0.02 0.276*** 1.056*** –0.081 –0.157 1.33** 0.175** 0.468*** 0.888*** –0.1 0.065 1.137*** (0.360) (4.610) (5.630) –(0.510) –(1.160) (2.900) (2.300) (4.540) (3.710) –(1.290) (1.290) (7.080) Geographic Zone (Omitted=Lake) Coastal 0.19** –0.043 –0.119** 0.232** 0.002 –0.189** 0.237*** –0.024 –0.166** (3.010) –(1.180) –(2.350) (3.380) (0.040) –(2.090) (3.890) –(0.580) –(2.440) North. Highland 0.149** –0.021 –0.156* 0.167* 0.004 –0.314** 0.188** –0.007 –0.257*** (2.290) –(0.370) –(1.680) (1.700) (0.070) –(3.440) (2.110) –(0.140) –(3.670) Central 0.146* –0.012 –0.09 –0.023 –0.257*** –0.382*** –0.005 –0.245*** –0.262*** (1.770) –(0.260) –(1.040) –(0.220) –(6.060) –(5.290) –(0.050) –(5.920) –(4.380) South. Highland 0.253*** 0.077 –0.051 0.147** –0.042 –0.057 0.164** –0.019 –0.05 (3.820) (1.330) –(0.360) (2.190) –(0.930) –(0.610) (2.690) –(0.450) –(0.670) South 0.083 0.091** –0.146** –0.007 –0.127** 0.025 –0.014 –0.09** –0.005 (1.000) (2.490) –(2.350) –(0.080) –(2.580) (0.250) –(0.170) –(2.060) –(0.070) (continues to next page) Table 4-1  Estimated Coefficients for Unconditional Quantile Regression Model, 2001 (continued) 2001 Urban Rural Metropolitan Nonmetropolitan 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile Constant 8.062*** 9.413*** 10.39*** 8.591*** 9.472*** 10.277*** 8.501*** 9.927*** 11.507*** 8.55*** 9.462*** 10.225*** (47.700) (115.970) (94.960) (57.220) (125.810) (57.610) (22.720) (32.690) (18.760) (66.110) (132.080) (78.810) No of Obs. 14,550 14,550 14,550 7,622 7,622 7,622 11,067 11,067 11,067 21,005 21,005 21,005 Adjust_R2 0.135 0.274 0.17 0.104 0.189 0.133 0.148 0.271 0.233 0.103 0.216 0.148 Source: Household Budget Surveys (HBS) for 2001. * Significant at the 10 percent level; ** significant at the 5 percent level; *** significant at the 1 percent level. Numbers in parentheses are Student-t. Appendices 125 126 Table 4-2  Estimated Coefficients for Unconditional Quantile Regression Model, 2007 2007 Urban Rural Metropolitan Nonmetropolitan 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile HH size –0.069*** –0.043*** –0.07*** –0.06*** –0.049*** –0.055*** –0.113*** –0.07*** –0.09*** –0.046*** –0.039*** –0.055*** –(4.740) –(6.180) –(9.560) –(4.210) –(8.710) –(7.110) –(5.090) –(7.740) –(6.440) –(3.870) –(5.550) –(8.040) pch14 –0.716*** –0.772*** –1.158*** –0.648*** –0.629*** –0.915*** –0.503** –0.731*** –1.119*** –0.675*** –0.655*** –1.131*** –(5.700) –(13.370) –(12.280) –(5.070) –(9.090) –(8.280) –(3.160) –(8.140) –(6.390) –(6.200) –(10.010) –(11.550) Tanzania Mainland Poverty Assessment peld65 0.555 –0.169 –0.843*** –0.36 –0.149 0.451** 0.58 –0.375 –0.981** –0.259 –0.055 0.324 (1.460) –(1.190) –(4.370) –(1.630) –(1.330) (2.040) (1.310) –(1.390) –(3.410) –(1.300) –(0.510) (1.560) Male head 0.051 –0.011 0.081 –0.12 –0.035 0.043 0.155 0.072 –0.06 –0.106 –0.023 0.067 (0.570) –(0.340) (1.590) –(1.460) –(0.840) (0.610) (1.300) (1.500) –(0.680) –(1.560) –(0.670) (1.120) Head over 65 –0.191 –0.041 0.173** 0.154 0.066 –0.095 –0.285 0.028 0.058 0.179 0.048 –0.035 –(0.930) –(0.600) (2.260) (1.100) (0.950) –(1.100) –(1.160) (0.270) (0.470) (1.410) (0.700) –(0.430) Marital status (Omitted=married) Never Married –0.163 0.116** 0.519*** 0.065 0.015 0.368** –0.052 0.171** 0.315** –0.075 0.052 0.529*** –(1.240) (2.510) (6.160) (0.520) (0.220) (2.600) –(0.480) (3.270) (2.870) –(0.740) (1.030) (4.800) Divorced –0.006 0.058 0.281*** 0.05 0.017 0.08 –0.485** 0.149* 0.345** 0.038 –0.002 0.19** –(0.040) (1.080) (3.600) (0.430) (0.250) (0.830) –(2.070) (1.950) (2.490) (0.400) –(0.040) (2.260) Widowed 0.017 0.076* 0.165** –0.009 –0.014 0.091 0.111 0.243*** 0.132 0.016 0.004 0.137* (0.150) (1.690) (2.640) –(0.080) –(0.260) (1.140) (0.690) (3.490) (1.050) (0.180) (0.090) (1.940) Max Education 0.033** 0.02*** 0.05*** 0.028** 0.014** 0.015* 0.045*** 0.03*** 0.091*** 0.026** 0.014** 0.017** (3.050) (4.760) (6.740) (2.620) (3.070) (1.860) (3.940) (4.420) (6.980) (2.760) (3.420) (2.460) Experience –0.002 0 0.002 –0.001 –0.001 –0.003 0.006 0.002 0.009** –0.002 –0.001 –0.004** –(0.750) –(0.390) (0.800) –(0.350) –(0.500) –(1.520) (1.320) (1.060) (2.630) –(0.880) –(0.770) –(2.480) Sector of Employment (Omitted= Farming & Fishing) Government 0.249* 0.244*** 0.096 –0.112 0.146* 0.586** –0.185 0.255** 0.075 –0.069 0.193** 0.338** (1.940) (4.060) (0.990) –(1.090) (1.870) (2.460) –(0.780) (2.200) (0.400) –(0.850) (3.430) (2.370) Private & NGO 0.235** 0.109** 0.117* 0.05 0.114 –0.037 –0.112 0.181* 0.015 0.031 0.115** 0.108 (1.970) (2.370) (1.820) (0.560) (1.530) –(0.390) –(0.520) (1.750) (0.120) (0.460) (2.370) (1.340) (continues to next page) Table 4-2  Estimated Coefficients for Unconditional Quantile Regression Model, 2007 (continued) 2007 Urban Rural Metropolitan Nonmetropolitan 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile Self-emp.(Others) 0.17 0.046 0.268** 0.095 0.18* 0.287 0.011 0.286** 0.132 0.075 0.139** 0.321** (1.450) (0.830) (2.480) (1.600) (1.820) (1.370) (0.060) (2.370) (0.800) (1.310) (2.130) (2.170) Self-emp. (Alone) 0.252** 0.045 0.037 –0.018 0.045 0.175** 0.014 0.138 –0.007 –0.03 0.069* 0.086 (2.770) (1.100) (0.800) –(0.190) (0.960) (2.340) (0.070) (1.350) –(0.060) –(0.390) (1.890) (1.470) HH duties –0.251 –0.119 –0.001 0.075 –0.135 –0.029 –0.22 0.01 –0.137 –0.024 –0.065 0.124 –(0.810) –(1.420) –(0.010) (0.210) –(0.840) –(0.190) –(0.670) (0.090) –(1.020) –(0.080) –(0.480) (0.820) Unemployed –0.158 –0.039 –0.016 –0.233* –0.084 –0.024 –0.216 0.051 –0.017 –0.188* –0.016 –0.026 –(1.000) –(0.700) –(0.250) –(1.800) –(1.510) –(0.360) –(0.950) (0.480) –(0.120) –(1.660) –(0.280) –(0.440) Main source of income (Omitted= Income from nonagricultural HH business) No Income 0.021 –0.202** –0.401*** –0.726* –0.239** –0.259** –0.009 –0.436*** –0.508*** –0.431 –0.158* –0.267** (0.070) –(2.290) –(3.640) –(1.660) –(2.020) –(2.130) –(0.040) –(3.990) –(4.080) –(1.420) –(1.720) –(2.500) Cash & in kind –0.208** –0.123** –0.155** 0.027 –0.074 –0.216** 0.094 –0.079 –0.01 –0.082 –0.099** –0.184** employment –(2.520) –(3.320) –(2.960) (0.300) –(1.400) –(2.380) (0.820) –(1.560) –(0.130) –(1.180) –(2.430) –(2.560) Agricultural HH business –0.175* –0.079** 0.093 –0.128** –0.133*** –0.118** 0.014 –0.013 –0.028 –0.113** –0.093** –0.079* –(1.810) –(2.000) (1.510) –(2.140) –(4.060) –(2.250) (0.080) –(0.110) –(0.150) –(2.160) –(3.110) –(1.780) Agr. & nonagric. Coop. 0.281** –0.035 0.18 –0.124 –0.347** –0.442*** 0.387 –0.043 0.828 –0.091 –0.328* –0.374*** (2.060) –(0.130) (0.580) –(0.390) –(1.980) –(4.540) (1.340) –(0.170) (1.270) –(0.300) –(1.840) –(4.040) Transfer & assistance –0.081 0.031 0.216** 0.123 –0.008 –0.048 –0.253 –0.155* 0.18 0.007 0.005 0.152 –(0.660) (0.630) (2.740) (1.130) –(0.120) –(0.400) –(1.170) –(1.770) (1.250) (0.090) (0.100) (1.520) Other –0.006 –0.114 –0.156 –0.533 –0.297** –0.11 –0.194 –0.197** –0.223** –0.272 –0.205** –0.011 –(0.040) –(1.590) –(1.390) –(1.500) –(2.060) –(0.490) –(1.000) –(2.520) –(2.340) –(1.170) –(2.240) –(0.060) Source of lighting (Omitted= Kerosene) Electricity 0.07 0.172*** 0.306*** –0.115 0.163** 0.796** 0.256*** 0.202*** 0.101 –0.044 0.109** 0.61*** (1.520) (5.700) (6.710) –(1.490) (2.580) (2.990) (3.840) (5.230) (1.370) –(1.220) (3.320) (5.670) Firewood –0.262 0.209** 0.588*** 0.081 –0.038 0.238** –0.093 –0.029 –0.155 0.069 –0.013 0.344*** –(1.500) (3.260) (3.860) (0.760) –(0.750) (2.670) –(0.380) –(0.310) –(1.610) (0.670) –(0.230) (4.310) (continues to next page) Appendices 127 128 Table 4-2  Estimated Coefficients for Unconditional Quantile Regression Model, 2007 (continued) 2007 Urban Rural Metropolitan Nonmetropolitan 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile Dwelling Roof 0.424*** 0.066* –0.025 0.17*** 0.073** 0.095** 0.025 0.06 0.176 0.157*** 0.089** 0.061* (3.700) (1.750) –(0.670) (3.510) (2.570) (2.370) (0.120) (0.480) (1.240) (3.540) (3.390) (1.780) livestock 0.271*** 0.096** 0.097 0.181** 0.125*** 0.014 0.027 0.078 0.359 0.209*** 0.126*** 0.063* (3.530) (2.490) (1.570) (3.440) (4.450) (0.330) (0.200) (0.680) (0.980) (4.470) (4.810) (1.740) Tanzania Mainland Poverty Assessment Area of land 0.001 0.004*** 0.004** 0.076* 0.078** 0.207*** 0.447 0.387 0.437** 0.004 0.005 0.006 (0.137) (9.450) (2.530) (1.79) (2.53) (12.51) (0.400) (0.580) (2.780) (1.480) (1.590) (0.060) well 0.242** 0.134* 0.291** 0.031 0.143** 0.389** 0.052 –0.072 0.41 0.063 0.138** 0.439** (1.970) (1.730) (2.060) (0.200) (2.000) (2.680) (0.310) –(0.510) (0.920) (0.520) (2.120) (3.370) cell 0.264*** 0.292*** 0.232*** 0.14** 0.297*** 0.397*** 0.205** 0.173*** 0.165** 0.147** 0.318*** 0.439*** (4.290) (10.260) (6.690) (2.640) (7.410) (5.070) (2.210) (4.100) (3.470) (3.460) (9.770) (7.410) bicycle 0.239*** 0.118*** 0.038 0.103* 0.106*** 0.104** 0.228** 0.14** –0.021 0.144** 0.116*** 0.108** (4.220) (4.080) (0.900) (1.850) (3.780) (2.290) (2.260) (2.440) –(0.250) (3.090) (4.810) (2.910) radio 0.255*** 0.063** 0.024 0.273*** 0.128*** 0.092** 0.294** 0.022 –0.055 0.248*** 0.138*** 0.098** (3.530) (2.320) (0.710) (4.270) (4.420) (2.320) (3.190) (0.560) –(0.910) (4.480) (5.360) (3.040) computer –0.143** 0.121** 1.422*** –0.378** –0.262* 1.447** –0.06 0.183** 1.493*** –0.093 0.022 1.843*** –(2.160) (2.350) (5.890) –(2.410) –(1.920) (2.800) –(0.860) (2.900) (5.760) –(1.170) (0.360) (4.640) Geographic Zone (Omitted=Coastal) North. Highland –0.141 –0.061 0.006 –0.201** –0.15** 0.135 –0.221** –0.132** 0.109 –(1.380) –(1.510) (0.090) –(2.240) –(2.700) (1.150) –(3.040) –(2.870) (1.170) Lake –0.265*** 0.023 0.124** –0.176** –0.201*** –0.045 –0.195** –0.183*** 0.052 –(3.670) (0.680) (2.460) –(2.640) –(5.090) –(0.720) –(3.320) –(4.780) (1.030) Central –0.192* –0.153** –0.059 –0.184** –0.287*** –0.074 –0.193** –0.253*** –0.047 –(1.950) –(2.720) –(0.820) –(2.090) –(6.040) –(1.030) –(2.600) –(6.060) –(0.800) South. Highland 0.123* –0.023 0.041 –0.029 –0.072* 0.196** –0.06 –0.07* 0.208** (1.740) –(0.570) (0.670) –(0.470) –(1.670) (2.570) –(1.080) –(1.810) (3.430) South –0.644*** –0.198*** 0.001 –0.273** –0.282*** –0.121 –0.284*** –0.283*** –0.085 –(4.990) –(5.000) (0.010) –(3.100) –(5.980) –(1.510) –(3.740) –(6.770) –(1.300) (continues to next page) Table 4-2  Estimated Coefficients for Unconditional Quantile Regression Model, 2007 (continued) 2007 Urban Rural Metropolitan Nonmetropolitan 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile Constant 9.003*** 10.235*** 11.016*** 9.486*** 10.319*** 11.121*** 9.109*** 10.081*** 10.671*** 9.497*** 10.238*** 11.156*** (42.880) (125.080) (83.240) (57.970) (121.230) (79.790) (27.210) (55.630) (52.580) (66.780) (133.310) (91.430) No of Obs. 7,119 7,119 7,119 3, 345 3, 346 3, 347 3,456 3,456 3, 456 7,008 7,008 7,008 Adjust_R2 0.177 0.423 0.292 0.207 0.116 (0.236) 0.184 0.287 0.226 0.109 0.248 0.196 Source: Household Budget Surveys (HBS) for 2007. * Significant at the 10 percent level; ** significant at the 5 percent level; *** significant at the 1 percent level. Numbers in parentheses are Student-t. Appendices 129 Table 4-3  Estimated Coefficients for Unconditional Quantile Regression Model, 2011/12 2011/12 Urban Rural Metropolitan Nonmetropolitan 10th 50th 90th 10th 50th 90th 10th 50th 90th 10th 50th 90th pctile pctile pctile pctile pctile pctile pctile pctile pctile pctile pctile pctile HH size –0.087*** –0.057*** –0.061*** –0.022** –0.026*** –0.025** –0.086*** –0.064*** –0.059*** –0.032*** –0.029*** –0.022** –(5.150) –(7.650) –(6.170) –(3.470) –(5.030) –(2.440) –(4.980) –(5.510) –(4.550) –(4.250) –(5.590) –(2.020) pch14 –0.172 –0.651*** –1.01*** –0.458*** –0.721*** –0.935*** –0.279** –0.732*** –0.804*** –0.434*** –0.674*** –1.082*** –(1.430) –(9.600) –(9.900) –(6.140) –(12.250) –(9.190) –(2.560) –(9.200) –(5.620) –(5.790) –(13.030) –(12.160) peld65 –0.387 –0.131 –0.557** –0.085 –0.004 0.42** 0.049 –0.269 –0.618** –0.072 0.047 0.053 –(0.970) –(0.830) –(2.520) –(0.770) –(0.040) (2.210) (0.130) –(1.100) –(2.090) –(0.590) (0.520) (0.310) Male 0.086 0.059 0.185** –0.114** –0.012 –0.021 0.136** 0.091** 0.175** –0.073 0.037 –0.02 (1.090) (1.290) (3.190) –(2.330) –(0.280) –(0.390) (2.000) (2.080) (2.400) –(1.600) (1.060) –(0.420) old 0.188 0.057 0.13 0.031 0.044 –0.046 –0.173 0.133 0.262** 0.086 0.046 0.079 (0.990) (0.800) (1.370) (0.460) (0.760) –(0.540) –(0.950) (1.440) (2.040) (1.100) (0.870) (1.030) Marital status (Omitted=married) Never Married –0.05 0.114** 0.72*** –0.096 –0.015 0.534*** –0.008 0.129** 0.657*** –0.069 –0.009 0.545*** –(0.410) (1.980) (7.030) –(1.000) –(0.270) (3.890) –(0.150) (2.340) (5.830) –(0.980) –(0.170) (4.800) Divorced –0.095 0.112* 0.276** –0.127 –0.007 0.2** 0.062 0.068 0.195* –0.187* 0.017 0.185** –(0.760) (1.770) (2.900) –(1.370) –(0.120) (2.090) (0.670) (1.050) (1.830) –(1.920) (0.330) (2.100) Widowed 0.226** –0.021 0.118* –0.015 0.073 0.101 –0.159* 0.057 0.17* –0.014 0.074 0.069 (2.180) –(0.410) (1.730) –(0.260) (1.340) (1.630) –(1.690) (0.910) (1.910) –(0.250) (1.600) (1.240) maxedu 0 0.015** 0.029*** 0.001 0.016*** 0.03*** 0.022** 0.026*** 0.039*** –0.001 0.013*** 0.025*** (0.030) (3.110) (3.750) (0.240) (3.940) (4.290) (3.060) (4.830) (3.840) –(0.140) (3.790) (3.800) hdexp –0.003 0 0.001 0 –0.001 –0.002 0.003 –0.002 0 –0.002 –0.001 –0.002 –(1.150) –(0.250) (0.380) (0.200) –(0.830) –(0.940) (1.180) –(1.000) –(0.110) –(1.070) –(1.420) –(1.310) Migration Status (Omitted= non migrant) Recent Migrant –0.151* 0.089* 0.244** –0.07 –0.047 0.291** –0.146* 0.061 0.207 –0.089 –0.01 0.235** (below 5 yrs) –(1.960) (1.700) (2.600) –(0.800) –(0.640) (2.280) –(1.910) (1.030) (1.590) –(1.280) –(0.190) (2.270) btw 5 & 15 yrs –0.065 0.102** 0.156** 0.038 0.017 0.212 –0.082 0.031 0.117 –0.023 0.036 0.312** –(1.050) (2.180) (2.230) (0.560) (0.270) (1.620) –(1.250) (0.620) (1.220) –(0.380) (0.810) (2.760) Above 15 yrs 0.078 0.057 0.086 0.076 –0.05 0.066 –0.083 0.026 0.044 0.085** –0.015 0.118* (1.410) (1.400) (1.470) (1.580) –(1.160) (1.010) –(1.430) (0.600) (0.590) (2.040) –(0.440) (1.940) Total migrants 0.056** 0.015 0.003 0.024* 0.017 –0.009 0.014 0.003 0.025 0.033** 0.02** 0.001 HH members (3.180) (1.270) (0.140) (1.960)q (1.560) –(0.650) (0.630) (0.220) (0.860) (2.750) (2.250) (0.060) Sector of Employment (Omitted=Self Employed alone) Government 0.159* 0.11** 0.31** 0.12** 0.206** 0.483** –0.025 0.074 0.299 0.171*** 0.146** 0.38** (1.840) (2.140) (2.340) (2.340) (3.320) (2.980) –(0.390) (1.010) (1.410) (3.750) (2.910) (3.140) Private & NGO 0.092* 0.077* 0.322*** 0.037 0.013 0.214 –0.026 0.065 0.165 0.093* 0.063 0.212* (1.690) (1.840) (4.020) (0.440) (0.190) (1.410) –(0.450) (1.360) (1.570) (1.780) (1.440) (1.920) Self-emp. –0.09 0.076* 0.21** 0.043 0.001 0.273** –0.082 0.115** 0.114 0.022 –0.009 0.215** (Others) –(1.180) (1.820) (3.010) (0.770) (0.010) (2.130) –(1.010) (2.180) (1.120) (0.450) –(0.220) (2.330) HH duties 0.024 0.027 0.079 0.162* 0.031 0.092 –0.054 0.049 0.099 0.156** 0.049 0.158 (0.300) (0.540) (1.060) (1.730) (0.400) (0.710) –(0.690) (0.880) (1.070) (2.410) (0.890) (1.430) (continues to next page) 130 Tanzania Mainland Poverty Assessment Table 4-3  Estimated Coefficients for Unconditional Quantile Regression Model ... (continued) 2011/12 Urban Rural Metropolitan Nonmetropolitan 10th 50th 90th 10th 50th 90th 10th 50th 90th 10th 50th 90th pctile pctile pctile pctile pctile pctile pctile pctile pctile pctile pctile pctile Unemployed 0.049 –0.057 –0.029 0.07* –0.032 –0.063 0.05 0.003 –0.04 0.048 –0.012 –0.114** (0.580) –(1.400) –(0.520) (1.680) –(0.880) –(1.310) (0.550) (0.050) –(0.450) (1.090) –(0.400) –(2.910) Main source of income (Omitted=Cash & inkind from employment) Nonagr. HH 0.08 0.002 0.165* 0.125* 0.179** 0.315** –0.087 –0.041 0.095 0.186** 0.196** 0.117 businesses (0.910) (0.030) (1.710) (1.810) (2.750) (2.690) –(0.750) –(0.520) (0.910) (2.700) (3.260) (1.390) (manuf.) Nonagr. HH 0.13** 0.086** 0.305*** 0.131** 0.168** 0.293** –0.02 0.048 0.2* 0.138** 0.153*** 0.318*** businesses (2.230) (2.140) (3.810) (2.310) (3.090) (2.860) –(0.310) (1.000) (1.830) (2.680) (3.680) (3.740) (sales) Nonagr. HH 0.133* –0.072 0.256** –0.011 0.021 0.046 –0.061 –0.072 0.209* 0.04 0.052 0.121 business (1.950) –(1.440) (2.260) –(0.120) (0.320) (0.470) –(0.620) –(1.240) (1.950) (0.510) (0.990) (1.380) (services) Agric. HH –0.06 –0.047 0.102 0.075 0.05 0.093 –0.175 –0.234** –0.261 0.09* 0.037 0.047 business –(0.490) –(0.790) (1.210) (1.540) (1.280) (1.490) –(0.890) –(2.480) –(1.540) (1.860) (1.090) (0.840) Transfer & 0.09 –0.05 0.087 –0.026 0.001 0.134** –0.152** –0.052 0.095 0.023 0.008 0.091 assistance (1.140) –(1.170) (1.320) –(0.460) (0.030) (2.050) –(2.170) –(1.010) (0.830) (0.460) (0.250) (1.510) Other –0.208* –0.085* 0.17** –0.148 0.034 0.24* –0.096 –0.051 0.235* –0.188 –0.037 0.095 –(1.810) –(1.940) (2.010) –(1.120) (0.500) (1.790) –(1.080) –(0.860) (1.790) –(1.610) –(0.720) (1.160) Source of lighting (Omitted= Kerosene) Electricity 0.219*** 0.224*** 0.22*** 0.003 0.076 0.463** 0.248*** 0.198*** 0.103** 0.014 0.129*** 0.491*** (5.180) (7.250) (4.980) (0.060) (1.580) (3.390) (4.390) (5.280) (1.970) (0.390) (4.250) (6.510) Firewood 0.045 –0.023 0.155** –0.055 –0.035 0.052 0.063 0.051 0.044 –0.047 –0.018 0.065 (0.410) –(0.570) (2.300) –(1.550) –(1.310) (1.160) (0.650) (0.940) (0.460) –(1.240) –(0.720) (1.550) Main source of drinking water (Omitted= Public sources & well) Piped water 0.09* 0.076** 0.206** –0.083 0.009 0.24* 0.005 0.064 0.242** –0.053 0.046 0.188** inside dwelling (1.810) (2.420) (3.200) –(0.790) (0.150) (1.850) (0.120) (1.540) (2.390) –(0.840) (1.320) (2.510) Piped water 0.148** 0.025 0.018 0.041 0.059 0.054 0.043 0.018 –0.007 0.056 0.007 –0.035 outside dwel. (2.630) (0.690) (0.350) (0.640) (0.680) (0.320) (1.000) (0.450) –(0.110) (1.250) (0.140) –(0.420) Dwelling roof –0.02 0.113** 0 0.082** 0.107*** 0.025 0.051 0.005 0.203 0.063* 0.11*** 0.016 –(0.140) (2.350) (0.000) (2.580) (4.350) (0.680) (0.250) (0.060) (1.410) (1.780) (4.780) (0.450) Area of Land 0 0.268** 0.170*** 0.122 0.177** 0.475*** 0.534 0.670 0.262** 0.179* 0.187 0.633** (0.030) (2.11) (4.14) (1.29) (2.28) (4.03) (2.14) (1.34) (2.14) (1.70) (1.62) (2.08) livestock 0.161** 0.048 0.077 –0.049 –0.006 –0.052 0.107* 0.081 0.221 0.026 0.021 –0.023 (2.790) (1.370) (1.400) –(1.410) –(0.240) –(1.360) (1.670) (1.520) (1.630) (0.750) (1.050) –(0.670) Radio 0.155** 0.072** 0.004 0.042 0.029 0.029 0.112** 0.025 0.024 0.055* 0.047** –0.007 (2.960) (2.810) (0.100) (1.360) (1.200) (0.740) (2.460) (0.810) (0.500) (1.760) (2.170) –(0.190) computer –0.145** 0.243*** 1.543*** 0.046 –0.097 0.388 0.069 0.36*** 0.871*** –0.036 –0.069 0.907*** –(3.040) (6.770) (9.690) (0.680) –(0.590) (1.040) (1.510) (6.890) (5.340) –(0.800) –(1.030) (4.260) (continues to next page) Appendices 131 Table 4-3  Estimated Coefficients for Unconditional Quantile Regression Model ... (continued) 2011/12 Urban Rural Metropolitan Nonmetropolitan 10th 50th 90th 10th 50th 90th 10th 50th 90th 10th 50th 90th pctile pctile pctile pctile pctile pctile pctile pctile pctile pctile pctile pctile Bicycle 0.061 0.057* 0.077 0.084** 0.063** –0.003 0.173** 0.113** 0.413** 0.074** 0.064** 0.031 (1.000) (1.720) (1.400) (2.710) (2.450) –(0.070) (2.780) (2.040) (2.930) (2.290) (2.830) (0.780) Cell 0.24** 0.217*** 0.124** 0.265*** 0.189*** 0.234*** 0.293** 0.169*** 0.085 0.255*** 0.193*** 0.167*** (3.170) (7.430) (3.100) (7.580) (7.490) (6.360) (3.420) (4.350) (1.600) (7.140) (8.420) (4.840) Geographic Zone (Omitted=Coastal) North. –0.08 –0.078 0.002 –0.029 0.043 –0.033 0*** 0*** 0*** –0.019 0.014 –0.003 Highland –(1.510) –(1.520) (0.030) –(0.540) (0.930) –(0.410) (0.000) (0.000) (0.000) –(0.420) (0.350) –(0.040) Lake –0.029 –0.214*** –0.098* –0.039 –0.053 –0.046 0*** 0*** 0*** –0.035 –0.101** –0.074 –(0.610) –(5.600) –(1.730) –(0.900) –(1.360) –(0.570) (0.000) (0.000) (0.000) –(0.930) –(3.040) –(1.000) Central –0.33*** –0.226*** –0.007 –0.004 –0.032 –0.127 0*** 0*** 0*** –0.017 –0.072* –0.117 –(4.050) –(4.890) –(0.080) –(0.070) –(0.750) –(1.550) (0.000) (0.000) (0.000) –(0.370) –(1.950) –(1.620) South. –0.064 –0.227*** –0.082 –0.085 –0.112** –0.007 0*** 0*** 0*** –0.09** –0.125*** 0.008 Highland –(1.060) –(5.180) –(1.270) –(1.620) –(2.600) –(0.080) (0.000) (0.000) (0.000) –(2.010) –(3.560) (0.100) South –0.92*** –0.456*** –0.219*** –0.099* –0.148** –0.137 0*** 0*** 0*** –0.225*** –0.225*** –0.243** –(6.430) –(9.770) –(3.560) –(1.690) –(3.180) –(1.610) (0.000) (0.000) (0.000) –(3.620) –(5.810) –(3.430) Constant 10.227*** 10.85*** 11.356*** 10.086*** 10.711*** 11.297*** 10.226*** 11.038*** 11.138*** 10.108*** 10.698*** 11.493*** (47.590) (106.730) (76.610) (100.850) (126.300) (84.020) (40.930) (83.670) (48.850) (97.390) (145.000) (94.200) No of Obs. 6,039 6,039 6,039 4,120 4,120 4,120 3,011 3,011 3,011 7, 148 7, 149 7, 150 Adjust_R2 0.268 0.354 0.258 0.105 0.188 0.141 0.195 0.322 0.228 0.107 0.223 0.185 Source: Household Budget Surveys (HBS) for 2011/12. * Significant at the 10 percent level; ** significant at the 5 percent level; *** signifi- cant at the 1 percent level. Numbers in parentheses are Student-t. 132 Tanzania Mainland Poverty Assessment Table 4-4  Quantile Decomposition of Urban-Rural Real Monthly per Capita Consumption 2001 2007 2011/12 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile Observed Gap 0.327 0.390 0.452 0.243 0.385 0.470 0.257 0.427 0.641 (0.016) (0.010) (0.020) (0.026) (0.015) (0.025) (0.017) (0.014) (0.025) Endowment effects attributable to HH characteristics 0.074 0.104 0.135 0.103 0.105 0.164 0.191 0.158 0.193 (0.006) (0.005) (0.008) (0.012) (0.007) (0.012) (0.015) (0.011) (0.021) Head educ. & exp. 0.055 0.047 0.058 0.089 0.054 0.115 0.017 0.041 0.073 (0.007) (0.005) (0.009) (0.014) (0.007) (0.013) (0.011) (0.008) (0.016) Head other characteristics 0.063 0.125 0.119 0.118 0.062 0.068 0.002 0.051 0.151 (0.015) (0.010) (0.018) (0.025) (0.013) (0.023) (0.016) (0.012) (0.024) Asset Ownership 0.189 0.104 –0.031 0.162 0.094 0.065 0.000 0.081 0.003 (0.021) (0.013) (0.025) (0.033) (0.017) (0.030) (0.024) (0.018) (0.036) Source of Income 0.041 0.002 0.012 0.032 0.006 –0.083 0.027 0.020 0.003 (0.015) (0.009) (0.018) (0.025) (0.013) (0.023) (0.016) (0.012) (0.023) Access to basic services 0.072 0.075 0.107 0.032 0.050 0.079 0.124 0.116 0.098 (0.009) (0.005) (0.010) (0.013) (0.007) (0.012) (0.014) (0.010) (0.020) Geographic region 0.042 –0.010 –0.021 0.064 0.015 –0.014 0.034 0.075 0.025 (0.006) (0.004) (0.007) (0.010) (0.005) (0.008) (0.011) (0.008) (0.014) Total endowment 0.535 0.448 0.379 0.600 0.384 0.393 0.394 0.543 0.545 (0.021) (0.014) (0.025) (0.033) (0.017) (0.030) (0.025) (0.019) (0.037) Returns effects attributable to HH characteristics 0.088 –0.184 –0.065 0.101 –0.033 –0.247 –0.187 –0.142 –0.166 (0.071) (0.044) (0.089) (0.110) (0.061) (0.107) (0.071) (0.054) (0.102) Head educ. & exp. 0.095 0.035 –0.034 0.081 0.038 0.340 –0.135 0.013 0.090 (0.027) (0.016) (0.034) (0.129) (0.070) (0.120) (0.083) (0.064) (0.119) Head other characteristics 0.010 0.015 –0.021 –0.009 0.004 0.034 0.033 0.019 0.031 (0.016) (0.010) (0.020) (0.028) (0.015) (0.027) (0.019) (0.015) (0.028) Asset Ownership 0.131 0.090 0.027 0.153 –0.101 –0.158 0.131 0.083 0.094 (0.030) (0.019) (0.037) (0.056) (0.030) (0.054) (0.049) (0.037) (0.071) Source of Income 0.015 –0.003 0.001 –0.045 0.033 0.153 –0.045 –0.082 –0.005 (0.014) (0.009) (0.018) (0.051) (0.028) (0.050) (0.046) (0.036) (0.066) Access to basic services –0.049 –0.003 0.007 –0.019 0.017 0.014 0.044 0.010 0.020 (0.007) (0.004) (0.008) (0.009) (0.005) (0.008) (0.015) (0.011) (0.021) Geographic region 0.036 0.055 0.038 –0.045 0.121 0.047 –0.127 –0.150 –0.026 (0.020) (0.013) (0.026) (0.051) (0.028) (0.050) (0.036) (0.028) (0.050) Constant –0.534 –0.064 0.121 –0.472 –0.078 –0.106 0.148 0.136 0.058 (0.093) (0.058) (0.117) (0.182) (0.100) (0.176) (0.131) (0.100) (0.187) Total returns –0.208 –0.058 0.073 –0.357 0.001 0.077 –0.138 –0.115 0.096   (0.025) (0.016) (0.030) (0.039) (0.020) (0.037) (0.029) (0.021) (0.041) Source: Household Budget Surveys (HBS) for 2011/12. Numbers in parentheses are Standard deviations. Appendices 133 Quantile Decomposition of Metropolitan-Nonmetropolitan Real Monthly p.c Table 4-5   Consumption 2001 2007 2011/12 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile 10th pctile 50th pctile 90th pctile Observed Gap 0.478 0.480 0.529 0.420 0.450 0.533 0.661 0.677 0.767 (0.026) (0.026) (0.040) (0.025) (0.016) (0.030) (0.017) (0.015) (0.028) Endowment effects attributable to HH characteristics 0.058 0.118 0.198 0.211 0.197 0.286 0.199 0.211 0.248 (0.017) (0.018) (0.026) (0.019) (0.012) (0.022) (0.027) (0.022) (0.041) Head educ. & exp. 0.061 0.094 0.077 0.093 0.070 0.197 0.051 0.092 0.122 (0.024) (0.023) (0.036) (0.019) (0.011) (0.023) (0.015) (0.012) (0.024) Head other 0.228 0.230 0.014 –0.040 0.118 0.030 –0.052 0.044 0.127 characteristics (0.085) (0.080) (0.125) (0.048) (0.029) (0.058) (0.028) (0.022) (0.044) Asset Ownership –0.138 –0.204 –0.877 0.001 0.002 –0.329 –0.020 –0.027 –0.226 (0.129) (0.122) (0.190) (0.099) (0.059) (0.164) (0.058) (0.047) (0.092) Source of Income 0.095 –0.104 0.020 0.013 –0.031 –0.002 0.078 0.100 0.129 (0.092) (0.087) (0.135) (0.057) (0.034) (0.069) (0.043) (0.035) (0.069) Access to basic services –0.002 0.049 0.038 0.118 0.092 0.050 0.143 0.116 0.073 (0.029) (0.027) (0.042) (0.021) (0.012) (0.025) (0.020) (0.016) (0.031) Geographic region Total endowment 0.302 0.185 –0.531 0.396 0.448 0.232 0.398 0.535 0.473 (0.158) (0.150) (0.232) (0.104) (0.062) (0.168) (0.066) (0.053) (0.104) Returns effects attributable to HH characteristics 0.199 –0.219 –0.188 –0.125 –0.190 –0.392 –0.175 –0.253 0.038 (0.130) (0.121) (0.190) (0.106) (0.063) (0.125) (0.077) (0.061) (0.120) Head educ. & exp. 0.072 0.101 –0.007 0.377 0.192 0.873 0.297 0.057 0.143 (0.045) (0.041) (0.064) (0.121) (0.072) (0.143) (0.091) (0.072) (0.140) Head other 0.085 0.064 0.008 –0.079 0.057 –0.024 –0.095 0.031 0.010 characteristics (0.042) (0.039) (0.061) (0.036) (0.021) (0.042) (0.023) (0.018) (0.035) Asset Ownership –0.108 –0.140 –0.133 –0.015 –0.100 0.377 0.180 0.002 0.540 (0.115) (0.109) (0.169) (0.118) (0.070) (0.181) (0.097) (0.078) (0.153) Source of Income 0.085 –0.081 0.021 0.088 0.041 0.052 –0.173 –0.157 –0.136 (0.070) (0.066) (0.103) (0.068) (0.041) (0.082) (0.058) (0.046) (0.090) Access to basic services –0.054 –0.011 –0.027 0.009 0.005 –0.061 0.054 0.025 –0.040 (0.011) (0.011) (0.017) (0.010) (0.006) (0.011) (0.018) (0.014) (0.028) Geographic region –0.075 0.041 0.085 0.152 0.150 –0.044 0.055 0.092 0.071 (0.009) (0.006) (0.012) (0.029) (0.017) (0.032) (0.021) (0.016) (0.031) Constant –0.031 0.537 1.290 –0.383 –0.153 –0.481 0.119 0.345 –0.334 (0.292) (0.273) (0.426) (0.195) (0.116) (0.232) (0.176) (0.141) (0.276) Total returns 0.175 0.295 1.060 0.024 0.002 0.301 0.263 0.143 0.294 (0.160) (0.151) (0.234) (0.105) (0.063) (0.170) (0.067) (0.054) (0.107) Source: Household Budget Surveys (HBS) for 2011/12. Numbers in parentheses are Standard deviations. 134 Tanzania Mainland Poverty Assessment Appendix 5: Inequality of Opportunity: The Parametric Decomposition Method of Decomposition ln(yi) = Cia + Eib + vi(3) The approach to estimate the degree of opportunity in- equality associated with the distribution of both consump- Ei = ACi + ei tion and income is based on the framework of Bourguignon et al. (2007). The method is based on the separation of the where α and β are two vectors of coefficients, А is a matrix determinants of household’s outcome (consumption or in- of coefficients specifying the effects of the circumstance come), yi, into a set of circumstances variables, denoted by variables on effort and εi is an error term. Model (3) can be the vector Ci ; efforts variables, denoted by the vector Ei and expressed in reduced from as: unobserved factors, represented by vi. The outcomes func- tion can be specified as: ln(yi) = Ciδ + ηi  (4) y i = f ( C i , E i ,v i ) i : 1.....N (1) where δ = a + b + A and ηi = ei b. The circumstances variables are economically exogenous Inequality of opportunity can be measured using equation since they are outside the individual’s control but effort (2) where the counterfactual distribution is obtained by re- factors may be endogenous to circumstances as an indi- placing yi with its estimated value, from equation (4), and ∼ vidual’s actions may be influenced by its gender, parental which can be expressed as: yi = exp Cδ ∼ ( ˆ +η ) ˆ i . In this decom- background etc. position, the variation in y can i= exp be(ˆ +η δ ˆ Cinterpretedi ) as the influ- ence of effort because circumstances are set to be equal for Equality of opportunity occurs, in the Roemer’s (1998) sense, all households, and inequality of opportunity is measured when outcomes are independently distributed from cir- as a residual. cumstances. This independence implies that circumstanc- es have no direct causal effect on outcomes and no causal Inequality of opportunity can also be measured directly by impact on efforts. The degree of opportunity inequality can eliminating the contribution of effort to outcomes, using therefore be determined by the extent to which the con- the smoothed distribution, obtained from the predicted val- ditional distribution of outcomes on circumstances, F(y|C), ues of outcomes based on circumstances in equation (4) differs from F(y). while ignoring the remaining variation in the residuals: Inequality of opportunity can be estimated as the difference ∼ = exp( C δ z ˆ (5) i i ) between the observed total inequality in the distribution of consumption or income and inequality that would prevail if The share of inequality of opportunity can thus be mea- ∼ there were no differences in circumstances. I F ∼ r Let( ( ))be the y sured by: ΘP = 1− I (F ( y )) counterfactual distribution of outcomes when circumstanc- ∼∼ es are identical for all individuals. The opportunity share of d ΘP = ( I F( z ) )(6) inequality can be defined as: I (F ( y )) ∼∼ ∼ The subscripts d and r, in ΘP d ,= ( I F( z ) ) denote respectively that in- r (I F (∼ y) ) I (F ( y )) ΘP = 1− I (F ( y )) (2) equality of opportunity is estimated directly or residually by eliminating the contribution of effort or circumstances to The first step for computing ΘP consists on estimating a outcomes. The direct and residual methods can yield differ- specific model of (1), which can be expressed in the follow- ent figures of opportunity inequality and the only inequality ing log-linear form: measure for which the two methods give the same results Appendices 135 is the mean log deviation (Theil_L), which has a path-inde- and in-kind wages, income from agricultural and nonagri- pendent decomposition when the arithmetic mean is used cultural household businesses, crop sales, rental of proper- as the reference income or consumption (Foster and Shney- ties, remittances, transfers and pensions. erov, 2000). By using the mean log deviation inequality in- dex the residual and direct methods give the same oppor- The circumstance variables used in the analysis include fa- tunity inequality measures. ther’s and mother’s education and their residence and vi- tal status, the gender, age and region of birth of the head. The parametric approach allows the estimation of the par- Parental education is coded into six categories (none, did tial effects of one or some circumstance variables on out- not finish primary school, completed primary, did not fin- comes, while controlling for the others, by simulating distri- ish secondary, completed secondary, above secondary). butions such as: Parents’ residence and vital status are captured through dummies indicating whether the father and/or mother live ∼ ( ˆ h≠ j + η ˆ j + C h≠ j δ yij = exp C j δ ˆi ) with the household and dummies indicating whether the father and/or mother died before the household head at- ∼ ∼j where F y( ) is the counterfactual outcomes distribution tains the age of 15 years old. The region of birth includes the 26 regions of the survey. It would have been interesting to obtained by keeping circumstance Cj constant. limit the place of birth to urban and rural sectors, but this information is not available in the survey. In order to check The inequality share specific to circumstance j can be com- the possibility for biased results due to the large categories puted residually by: in the place of birth, we estimated opportunity inequality we estimated opportunity inequality grouping these vari- ΘPj = 1− ( ( )) ∼ j I F ∼ y ables into five main zones and obtained quite close results I (F ( y )) to those displayed here. Data We explore also the effects of community characteristics The analysis uses data from the National Panel Surveys on inequality and compare its impact to that of family (NPS) of 2008, 2010 and 2012. The surveys were conduct- circumstances. The community characteristics include a ed on nationally representative samples of households, set of variables capturing the access to basic services in and methodology and data were selected to ensure com- the community of residence of the household. It would parability. They include information on household charac- have been more consistent to use this type of informa- teristics; household consumption and income, individuals tion for the community of birth, but this information is education, and employment status; and parents’ educa- unfortunately too difficult to obtain. The community char- tion and vital status. In addition, all survey waves include acteristics include distance to: head regional or district a community module that collects detailed information headquarters, government and private primary schools, on the access to basic services, the presence of local in- government and private secondary schools, health cen- vestment projects, infrastructure conditions and family ters, and markets, all recoded into four categories (within characteristics in the commune where the households are the village, outside at less than 5 km, outside between 5 located. and 10 km, and outside more than 10 km). The communi- ty characteristics include also variables on the presence of Inequality of opportunity is derived from two outcomes: investment projects for construction and maintenance of consumption and income. Household’s consumption is schools, water irrigation provision and infrastructure de- measured as real monthly per capita consumption of food velopment (including roads, health centers, markets etc.) and non-durables and excludes expenses on housing and all recoded into four categories (no projects, projects of durable goods. Household income is measured as real less than 1M TZS, projects between 1M and 10 M TZS, and monthly per capita income from all sources including cash projects over 10 M TZS). They also include the number of 136 Tanzania Mainland Poverty Assessment household that permanently migrated out of the village Computing the opportunity share of earnings inequality for during the last 12 months to capture some of the family the entire country is important to the design of equal-op- issues inside the village, the main sources of drinking wa- portunity policies, but it fails to capture the differential ter in the village, the main source of lighting as well as the intensity of opportunity inequality across areas and pop- type of toilet facilities.76 ulation groups. Because heterogeneity in population com- position across the urban and rural areas may distort the As with most samples, NPS surveys include missing obser- aggregate picture of inequality of opportunity, opportunity vations that need to be treated with caution. The variables inequality indices are also computed for urban and rural reporting on family background include quite a few miss- subgroups. ing values by individuals who sometimes cannot recall their parent’s education correctly. While the percentage of miss- Table 5-1 presents descriptive statistics for selected cir- ing observations barely exceeds 11 percent in each wave, cumstance variables used in the analysis. Consumption dropping all households with missing data on these vari- and income are higher in urban areas and are expanding ables would disregard information available on the other over time, except a slight drop in income observed in the variables, and would likely introduce bias because missing last survey. Father’s and mother’s education are signifi- values are not completely random. cantly higher in urban areas. While the number of house- holds with parents having completed primary education Dealing with missing values generated by nonresponse is a is expanding more in the rural areas, those with secondary well-known problem in survey-based research (Dardanoni school (or higher) graduates parents are expanding more and Peracchi, 2011), more so in the biomedical literature in urban sectors. than in economics. We follow the procedure suggested by Royston and White (2011) known as Multiple Imputations The community characteristics variables indicate success- Chained Equations (implemented in STATA with the ICE ful strategy for the promotion of primary education and command), in which multiple imputations of missing data to a lesser extent secondary education apparent through are generated as new data sets, stacked, and then used in the expansion of access to government and even private estimation. schools particularly in the rural zones. However, the efforts to facilitate access to health, water and electricity seem This method is built on the so-called “missing at random” to be still slow. There seems also to be important initia- assumption, which means that “any systematic difference tives to improve the infrastructure and facilitate access to between the missing values and the observed values can schools, health centers and markets particularly in rural be explained by differences in observed data”. This is a less sectors. While the big investment projects (of over 10M) stringent assumption than complete randomness, which for building the infrastructure are expanding over time, is unlikely to fit the NPS data. For example, missing values those for schools construction and maintenance seem to of parental education are more likely to occur for less well- be declining. off and less educated households, which is non-random and explainable by observed values. Clearly, why an ob- The estimation results, by ordinary least squares (OLS), of servation has missing values matters for how it is “filled in”, equation (4) are presented in tables 5-2 and 5-3. Because and the bias from a particular method may be worse than of space limitations, we limit the presentation to the most using the complete case data. The literature does not of- significant variables in the regression results. fer clear guidance on how to judge the size of this bias. Fortunately, the size of the estimates of inequality of op- portunity change little when imputed values are added (compared to complete case estimates) and the main con- 76 We also estimate the model including information on the edu- clusions are consistent to the different methods of treating cation and occupation of the head of the village but the variables missing values. were not significant. Appendices 137 Descriptive Statistics for Selected Variables Table 5-1   2008 2010 2012 Rural Urban Total Rural Urban Total Rural Urban Total Mean Monthly per capita 32441.32 67535.60 40234.73 37093.01 73478.12 46534.07 49009.12 105132.30 63848.41 Consumption (TZS) (21692.13) (56158.32) (35760.44) (27126.53) (59192.25) (41327.21) (38044.27) (86516.96) (60462.32) Mean Monthly per capita 35242.65 93205.14 48114.42 52930.64 112552.00 68407.42 45461.34 92789.33 58184.85 Income (TZS) (80406.97) (155219.10) (104674.40) (125887.90) (159763.80) (137978.50) (89862.78) (161241.40) (114870.90) Father Education (%) Did not go to School 66.79 39.47 61.07 61.71 35.59 55.01 61.18 32.16 53.72 Did not finish Primary School 18.49 21.06 19.03 20.71 25.74 22.00 18.67 20.59 19.10 Finished Primary School 13.15 27.83 16.23 15.07 29.60 18.79 18.14 34.42 22.29 Did not finish Secondary School 0.70 2.72 1.12 0.66 1.99 1.00 0.29 1.97 0.71 Finished Secondary School 0.70 5.26 1.65 1.44 4.99 2.35 1.20 6.31 2.64 Higher than Secondary School 0.17 3.67 0.90 0.42 2.09 0.85 0.52 4.55 1.54 Mother Education (%) Did not go to School 82.72 58.42 77.45 79.46 52.50 72.68 77.78 51.81 70.82 Did not finish Primary School 8.09 13.47 9.26 9.18 16.29 10.97 9.14 12.10 9.89 Finished Primary School 8.94 23.39 12.07 10.65 26.29 14.58 12.27 31.04 17.27 Did not finish Secondary School 0.13 1.49 0.42 0.17 1.37 0.47 0.24 0.88 0.40 Finished Secondary School 0.12 2.28 0.59 0.48 3.18 1.16 0.44 3.10 1.24 Higher than Secondary School 0.00 0.94 0.20 0.05 0.37 0.13 0.13 1.08 0.38 Main Drinking water source (%) Piped water 1.04 15.86 4.33 1.39 7.93 3.09 2.20 15.76 5.81 Standpipe/tap & vendor 21.00 55.94 28.76 22.76 63.13 33.24 24.14 58.48 33.16 Well water 35.45 19.77 31.97 39.06 20.02 34.11 37.76 19.52 33.00 River & rainwater & other 42.51 8.43 34.94 36.79 8.92 29.56 35.90 6.23 28.03 Electricity (%) No access to Electricity 97.13 56.39 88.08 94.23 56.02 84.31 91.23 48.08 79.71 Public, Solar & other 2.87 43.61 11.92 5.77 43.98 15.69 8.77 51.92 20.29 Government primary schools (%) No facility 0.83 5.03 1.74 4.37 5.61 4.67 0.00 1.25 0.31 Within the village 94.78 66.47 88.65 91.90 65.00 85.41 96.97 74.12 91.26 Outside village less 5km 2.94 27.39 8.24 3.61 28.20 9.53 2.60 24.63 8.11 Out. village btw 5 & 10km 0.32 1.12 0.50 0.03 1.02 0.27 0.00 0.00 0.00 Out. village more 10km 1.12 0.00 0.88 0.09 0.17 0.11 0.44 0.00 0.32 Private primary schools (%) No facility 28.86 18.44 26.58 15.15 16.25 15.41 2.83 7.39 3.95 Within the village 62.44 62.60 62.48 81.31 64.94 77.36 91.69 77.70 88.21 Outside village less 5km 3.44 18.96 6.83 2.56 17.62 6.19 1.72 14.41 4.90 Out. village btw 5 & 10km 0.42 0.00 0.32 0.85 1.19 0.93 1.04 0.00 0.78 Out. village more 10km 4.84 0.00 3.78 0.14 0.00 0.10 2.72 0.50 2.16 (continues to next page) 138 Tanzania Mainland Poverty Assessment Descriptive Statistics for Selected Variables (continued) Table 5-1   2008 2010 2012 Rural Urban Total Rural Urban Total Rural Urban Total Government secondary schools (%) No facility 4.10 20.37 7.62 4.78 9.99 6.03 0.20 4.44 1.26 Within the village 39.24 25.33 36.23 64.68 51.21 61.43 66.15 51.97 62.64 Outside village less 5km 18.35 43.72 23.84 10.57 35.81 16.65 9.76 39.48 17.15 Out. village btw 5 & 10km 23.81 8.95 20.59 11.55 2.92 9.47 13.73 3.43 11.20 Out. village more 10km 14.49 1.64 11.71 8.42 0.07 6.41 10.15 0.67 7.76 Private secondary schools (%) No facility 55.43 40.87 7.62 51.27 23.54 44.59 21.75 30.13 23.93 Within the village 4.32 17.40 36.23 8.01 19.25 10.72 6.35 21.79 10.18 Outside village less 5km 3.66 26.80 23.84 5.48 30.88 11.60 7.39 28.58 12.66 Out. village btw 5 & 10km 11.23 9.22 20.59 8.67 8.74 8.69 8.92 10.30 9.22 Out. village more 10km 25.35 5.71 11.71 26.57 17.60 24.41 55.60 9.20 44.00 Health Centers (%) No facility 2.09 7.34 3.23 4.32 6.21 4.78 1.18 0.63 1.04 Within the village 53.60 54.55 53.81 59.69 45.58 56.29 60.73 61.38 61.03 Outside village less 5km 16.03 34.67 20.06 14.17 44.99 21.60 15.94 31.62 19.81 Out. village btw 5 & 10km 19.45 2.06 15.69 17.52 3.15 14.06 15.01 4.94 12.44 Out. village more 10km 8.83 1.37 7.22 4.30 0.07 3.28 7.14 1.42 5.68 Investment for schools construction & renovation (%) No inv project 14.42 41.49 20.42 24.45 46.03 29.65 36.53 69.50 44.69 inv project less than 1M 7.00 14.55 8.67 9.94 6.21 9.04 8.00 6.75 7.74 inv project btw 1 & 10M 34.00 16.28 30.07 29.73 17.15 26.70 25.31 12.51 22.13 inv project more 10M 44.58 27.68 40.83 35.88 30.61 34.61 30.17 11.23 25.43 Investment for infrastructure building (%) No inv project 43.55 70.72 49.58 49.74 72.70 55.27 48.81 75.96 55.58 inv project less than 1M 12.63 13.02 12.71 9.56 10.89 9.88 9.11 6.39 8.53 inv project btw 1 & 10M 19.93 5.12 16.64 9.25 5.52 8.35 11.48 9.62 10.97 inv project more 10M 23.89 11.15 21.06 31.45 10.89 26.50 30.59 8.04 24.93 Source: NPS surveys for 2008, 2010 and 2012. Numbers in parentheses are standard deviations. Results are weighted by appropriate sam- pling weights to reflect the characteristics of the Tanzanian population. Appendices 139 Regression of Consumption on Circumstances Table 5–2   ·2008 2010 2012 Rural Urban Total Rural Urban Total Rural Urban Total Female Head –0.034 –0.128** –0.048* –0.086** –0.134** –0.098*** –0.059* 0.022 –0.036 –(1.080) –(2.730) –(1.780) –(2.760) –(3.020) –(3.770) –(1.900) (0.580) –(1.420) Age head 0.003** –0.002 0.003** 0 –0.003 0 0.002* –0.001 0.001 (2.730) –(1.180) (2.340) (0.260) –(1.370) –(0.320) (1.820) –(0.870) (1.550) Father Education (Omitted: Did not go to School) Under Primary School 0.071* –0.039 0.07* 0.118** 0.002 0.105** 0.043 0.073 0.065** (1.680) –(0.530) (1.910) (3.170) (0.030) (3.130) (1.160) (1.340) (2.070) Finished Primary 0.069 0.133** 0.115** 0.149** 0.042 0.139** 0.105** 0.045 0.092** School (1.220) (2.040) (2.490) (3.200) (0.580) (3.280) (2.370) (0.810) (2.600) Under Secondary –0.222 0.349** 0.175 0.355** 0.244 0.351** 0.339** 0.178* 0.319** School –(1.230) (2.940) (1.350) (2.910) (1.610) (3.530) (2.290) (1.680) (3.420) Finished Secondary 0.16 0.198* 0.23** 0.298** 0.236** 0.315*** 0.302** 0.26** 0.296*** School (0.960) (1.800) (2.690) (2.720) (2.460) (4.360) (3.340) (2.870) (4.210) Higher than Second. 0.13 0.136 0.192* 0.382** 0.339** 0.448*** 0.52*** 0.26** 0.381*** School (0.950) (1.320) (1.920) (1.970) (2.260) (3.820) (4.390) (2.570) (4.810) Mother Education (Omitted: Did not go to School) Under Primary School 0.043 0.008 0.041 0.041 0.125* 0.066 0.171*** 0.012 0.137** (0.780) (0.100) (0.900) (0.770) (1.910) (1.630) (3.550) (0.190) (3.520) Finished Primary 0.181** 0.078 0.165** 0.079 0.144** 0.11** 0.147** 0.131** 0.164*** School (2.830) (1.160) (3.280) (1.450) (2.130) (2.420) (2.710) (2.360) (4.210) Under Secondary 1.224** –0.027 0.287 –0.039 0.376** 0.155 0.231 0.191 0.235* School (2.330) –(0.200) (1.490) –(0.160) (2.050) (1.010) (1.080) (1.210) (1.820) Finished Secondary 0.377 0.517*** 0.571*** 0.295 0.238* 0.28** 0.071 0.141 0.154* School (1.050) (4.060) (3.940) (1.530) (1.850) (2.520) (0.410) (1.400) (1.730) Higher than Second. 0.011 0.585** 0.662** 0.727** 0.686** 0.616** 0.448* 0.459** 0.437** School (0.050) (3.550) (3.070) (2.690) (3.270) (2.880) (1.780) (2.210) (2.750) Place of Birth (Omitted=Dar es Salaam) Dodoma –0.389** –0.223* –0.374*** –0.343 –0.124 –0.239** –0.772*** –0.186** –0.421*** –(2.650) –(1.870) –(4.390) –(1.270) –(1.100) –(2.680) –(4.540) –(1.980) –(4.930) Arusha 0.041 0.099 0.021 –0.035 0.243** 0.075 –0.36** 0.007 –0.082 (0.240) (0.690) (0.200) –(0.130) (2.220) (0.860) –(2.120) (0.060) –(0.940) Kilimanjaro 0.02 0.18* –0.05 –0.075 0.256** 0.072 –0.458** –0.02 –0.164** (0.120) (1.770) –(0.550) –(0.280) (2.800) (0.900) –(2.800) –(0.260) –(2.350) Tanga –0.059 0.06 –0.045 –0.201 –0.073 –0.133* –0.296* –0.189** –0.095 –(0.390) (0.530) –(0.550) –(0.750) –(0.770) –(1.680) –(1.890) –(2.010) –(1.390) Mororgoro –0.268* –0.183* –0.251** –0.12 –0.064 –0.056 –0.36** –0.07 –0.115* –(1.770) –(1.850) –(3.150) –(0.450) –(0.710) –(0.690) –(2.270) –(0.950) –(1.760) Pwani 0.143 0.116 0.076 0.145 0.127 0.155* –0.281 –0.091 –0.062 (0.870) (1.090) (0.840) (0.530) (1.180) (1.680) –(1.650) –(1.170) –(0.880) Lindi –0.359** –0.008 –0.284** –0.244 0.079 –0.101 –0.586*** –0.217** –0.31*** –(2.350) –(0.070) –(3.170) –(0.920) (0.790) –(1.250) –(3.770) –(1.990) –(4.540) (continues to next page) 140 Tanzania Mainland Poverty Assessment Regression of Consumption on Circumstances (continued) Table 5–2   ·2008 2010 2012 Rural Urban Total Rural Urban Total Rural Urban Total Mtwara –0.164 –0.137 –0.213** –0.364 0.103 –0.195** –0.568*** –0.105 –0.289*** –(1.160) –(1.320) –(2.820) –(1.390) (1.110) –(2.490) –(3.630) –(1.150) –(4.510) Ruvuma –0.366** 0.117 –0.342*** –0.44* 0.115 –0.271** –0.769*** –0.218** –0.427*** –(2.420) (1.060) –(4.130) –(1.660) (0.950) –(3.150) –(4.850) –(2.300) –(6.150) Iriniga –0.111 –0.018 –0.119 –0.123 –0.055 –0.093 –0.45** –0.113 –0.166** –(0.740) –(0.160) –(1.400) –(0.470) –(0.570) –(1.140) –(2.820) –(1.300) –(2.410) Mbeya –0.071 –0.096 –0.104 –0.147 0.136 –0.052 –0.368** 0.067 –0.041 –(0.510) –(0.810) –(1.330) –(0.560) (1.010) –(0.670) –(2.380) (0.590) –(0.600) Singida –0.252 –0.334** –0.253** –0.266 –0.047 –0.167* –0.512** –0.101 –0.227** –(1.580) –(2.250) –(2.590) –(1.000) –(0.370) –(1.880) –(3.170) –(0.710) –(2.790) Tabora –0.379** –0.193 –0.329*** –0.166 –0.211 –0.172* –0.58*** –0.257** –0.32*** –(2.550) –(1.280) –(3.710) –(0.620) –(1.300) –(1.830) –(3.820) –(2.890) –(4.890) Rukwa –0.425** –0.431** –0.52*** –0.568** –0.222* –0.447*** –0.807*** –0.371** –0.481*** –(2.750) –(3.060) –(5.740) –(2.130) –(1.670) –(4.980) –(4.990) –(2.180) –(6.090) Kigoma –0.571*** –0.186 –0.534*** –0.516* –0.36** –0.44*** –0.716*** –0.261** –0.409*** –(3.710) –(1.200) –(6.220) –(1.940) –(2.660) –(5.150) –(4.620) –(3.000) –(5.950) Shinyanga –0.203 0.004 –0.222** –0.209 –0.012 –0.134* –0.39** –0.079 –0.155** –(1.370) (0.030) –(2.640) –(0.800) –(0.100) –(1.670) –(2.580) –(0.850) –(2.470) Kagera –0.017 –0.046 –0.048 –0.072 –0.089 –0.049 –0.441** –0.058 –0.162** –(0.110) –(0.400) –(0.600) –(0.270) –(0.740) –(0.590) –(2.780) –(0.530) –(2.310) Mwanza –0.294* 0.029 –0.27** –0.315 –0.164 –0.236** –0.508** –0.306** –0.312*** –(1.900) (0.260) –(3.040) –(1.200) –(1.020) –(2.800) –(3.320) –(3.400) –(5.190) Mara –0.381** 0.199 –0.265** –0.46* 0.081 –0.236** –0.765*** –0.204** –0.447*** –(2.400) (1.510) –(2.810) –(1.680) (0.590) –(2.340) –(4.530) –(2.000) –(5.360) Manyara –0.209 0.357 –0.176* –0.283 0.31 –0.141 –0.505** 0.051 –0.192* –(1.290) (1.330) –(1.710) –(1.040) (1.380) –(1.370) –(2.840) (0.350) –(1.940) Kaskazini Unguja –0.359** –0.577*** –0.512*** –0.247 0.142 –0.221* –0.697*** –0.632*** –0.525*** –(2.430) –(4.540) –(5.970) –(0.920) (0.370) –(1.940) –(4.320) –(4.210) –(6.460) Kusini Unguja –0.331* –0.235 –0.44*** –0.104 –0.019 –0.203* –0.685*** –0.571*** –0.504*** –(1.960) –(1.170) –(3.830) –(0.380) –(0.070) –(1.930) –(3.810) –(3.630) –(5.440) Mjini/Magharibi –0.339* –0.523*** –0.55*** –0.306 –0.362** –0.402*** –0.758*** –0.312** –0.455*** –(1.810) –(4.630) –(5.840) –(1.100) –(1.970) –(3.830) –(4.380) –(2.720) –(4.680) Kaskazini Pemba –0.707*** –0.458** –0.721*** –0.155 –0.148 –0.225** –0.69*** –0.318** –0.41*** –(4.720) –(2.950) –(7.170) –(0.580) –(0.740) –(2.430) –(4.330) –(2.290) –(5.430) Kusini Pemba –0.511** –0.659*** –0.653*** –0.353 –0.527*** –0.444*** –0.881*** –0.527*** –0.609*** –(3.310) –(4.970) –(7.080) –(1.310) –(3.990) –(4.860) –(5.280) –(4.180) –(7.360) Father does not live 0.047 –0.021 0.029 0.032 –0.013 0.019 –0.022 0.035 –0.007 with HH (1.350) –(0.420) (0.990) (1.020) –(0.270) (0.720) –(0.620) (0.770) –(0.250) Mother does not live 0.006 0.03 0.02 –0.08** 0.023 –0.06** 0.023 –0.026 0.009 with HH (0.170) (0.620) (0.680) –(2.570) (0.450) –(2.260) (0.790) –(0.580) (0.380) (continues to next page) Appendices 141 Regression of Consumption on Circumstances (continued) Table 5–2   ·2008 2010 2012 Rural Urban Total Rural Urban Total Rural Urban Total Mother died before 0.005 0.005 0.01 –0.044 –0.061 –0.058 –0.031 0.014 –0.026 age 15 yrs of head (0.090) (0.060) (0.210) –(0.930) –(0.860) –(1.440) –(0.560) (0.220) –(0.560) Father died before age 0.015 –0.074 0.011 0.018 0.13** 0.05 –0.005 0.068 0.017 15 yrs of head (0.380) –(1.360) (0.320) (0.480) (2.400) (1.530) –(0.120) (1.480) (0.540) No of min 2063 1202 3265 2583 1263 3846 3159 1731 4886 observations Number of imputations 10 Source: NPS 2008, 2010, 2012. Note: The dependent variable is the logarithm of real monthly per capita consumption. Numbers in parentheses are bootstrapped stu- dent-t based on 100 replications. * Significant at the 10 percent level; ** significant at the 5 percent level; *** significant at the 1 percent level. 142 Tanzania Mainland Poverty Assessment Regression of Income on Circumstances Table 5–3   2008 2010 2012   Rural Urban Total Rural Urban Total Rural Urban Total Female Head –0.271** –0.228* –0.463** –0.276** –0.245** –0.318* –0.218** –0.198** –0.207**   –(2.680) –(1.850) –(3.120) –(3.230) –(2.410) –(1.930) –(3.160) –(2.240) –(2.130) Age head –0.013** –0.015** –0.003 –0.009** –0.01** –0.005 –0.006** –0.009** 0.002   –(3.440) –(3.280) –(0.410) –(2.630) –(2.770) –(0.420) –(2.380) –(3.060) (0.410) Father Education (Omitted: Did not go to School)             Under Primary School 0.336** 0.33** 0.167 0.226** 0.213* 0.237 0.218** 0.224** 0.115   (3.090) (2.520) (0.790) (2.090) (1.740) (0.900) (2.400) (2.040) (0.830) Finished Primary School 0.518*** 0.418** 0.58** 0.44** 0.502*** 0.276 0.118 0.074 0.174   (3.890) (2.470) (2.820) (3.420) (3.900) (0.830) (1.200) (0.640) (1.240) Under Secondary School 0.237 –0.55 0.374 0.763* 0.686 0.752 0.466 0.592 0.438   (0.780) –(1.150) (1.190) (1.960) (1.480) (1.170) (1.200) (0.840) (0.960) Finished Secondary School 0.399 0.596 0.414 0.535** 0.463 0.743* 0.451** 0.463 0.66**   (1.500) (0.970) (1.340) (2.370) (1.310) (1.730) (2.710) (1.490) (3.210) Above Secondary School 0.707** 0.747* 0.708* 0.818** 1.156** 0.529 0.537** 0.478 0.631**   (2.270) (1.930) (1.940) (2.630) (2.140) (1.220) (2.680) (1.040) (2.980) Mother Education (Omitted: Did not go to School)           Under Primary School 0.202 0.384** –0.203 0.136 0.12 0.1 0.105 0.145 0.014   (1.490) (2.210) –(1.060) (1.120) (0.760) (0.540) (0.940) (1.070) (0.090) Finished Primary School 0.131 0.234 –0.015 0.061 0.058 0.065 0.106 0.134 0.067   (0.800) (1.040) –(0.070) (0.480) (0.400) (0.280) (0.980) (0.960) (0.480) Under Secondary School –0.334 2.24** –0.911 0.804 0.424 1.118* 0.749** 1.339** 0.075   –(0.510) (2.480) –(1.420) (1.660) (0.540) (1.780) (2.070) (3.270) (0.170) Finished Secondary School 0.449 –0.331 0.267 0.108 –0.13 0.036 –0.118 –0.246 0.012   (1.600) –(0.360) (0.980) (0.420) –(0.290) (0.120) –(0.490) –(0.430) (0.050) Above Secondary School –0.16 –0.819 –0.186 0.546 0.332 0.902* 0.095 –0.68 0.34   –(0.160) –(0.450) –(0.280) (1.500) (0.500) (1.680) (0.320) –(1.310) (1.080) Place of Birth (Omitted=Dar es Salaam)               Dodoma –0.468 –1.059** –0.28 –0.285 –0.74* –0.646** –0.391** –0.68* –0.193   –(1.410) –(2.670) –(0.670) –(1.130) –(1.940) –(2.310) –(2.140) –(1.910) –(0.850) Arusha –0.507 –1.255** 0.323 –0.208 –0.866** –0.017 –0.144 –0.591 0.449*   –(1.460) –(2.790) (0.810) –(0.840) –(2.370) –(0.050) –(0.710) –(1.630) (1.870) Kilimanjaro –0.273 –0.668 0.212 –0.294 –0.911** –0.003 –0.662*** –1.222** –0.182   –(0.900) –(1.630) (0.650) –(1.410) –(2.370) –(0.020) –(3.770) –(3.360) –(0.970) Tanga –0.272 –0.931** 0.003 –0.405* –1.049** –0.24 –0.235 –0.239 –0.727***   –(0.850) –(2.170) (0.010) –(1.810) –(2.940) –(0.840) –(1.480) –(0.700) –(3.560) Mororgoro 0.022 –0.459 –0.123 –0.2 –0.775** –0.03 –0.147 –0.353 –0.212   (0.070) –(1.190) –(0.390) –(0.960) –(2.120) –(0.140) –(0.880) –(1.010) –(1.190) Pwani 0.185 –0.541 0.385 –0.066 –0.589 –0.054 0.187 0.35 –0.313   (0.540) –(1.180) (1.110) –(0.280) –(1.600) –(0.220) (1.050) (0.890) –(1.630) Lindi –0.092 –1.088** 0.382 0.213 –0.369 0.624** –0.227 –0.484 –0.294   –(0.310) –(2.850) (1.090) (1.030) –(1.090) (2.150) –(1.330) –(1.400) –(1.440) (continues to next page) Appendices 143 Regression of Income on Circumstances (continued) Table 5–3   2008 2010 2012   Rural Urban Total Rural Urban Total Rural Urban Total Mtwara –0.263 –0.811** –0.025 –0.238 –0.845** 0.215 –0.329* –0.673** –0.325 –(0.920) –(2.390) –(0.060) –(1.180) –(2.670) (0.930) –(1.960) –(1.990) –(1.270) Ruvuma –0.717** –1.295** 0.102 –0.461** –1.152** 0.387 –0.174 –0.431 –0.119 –(2.030) –(3.380) (0.170) –(2.030) –(3.320) (1.520) –(1.000) –(1.240) –(0.540) Iriniga –0.032 –0.505 –0.032 –0.315 –0.853** –0.172 0.128 –0.084 –0.078 –(0.110) –(1.450) –(0.080) –(1.430) –(2.500) –(0.590) (0.770) –(0.240) –(0.450) Mbeya –0.478 –1.131** 0.049 –0.151 –0.651** 0.284 –0.044 –0.142 –0.245   –(1.570) –(3.200) (0.100) –(0.730) –(2.010) (0.990) –(0.270) –(0.430) –(1.070) Singida –0.971** –1.686*** –0.471 –0.108 –0.472 –0.529 –0.265 –0.572 0.125   –(2.560) –(3.560) –(1.010) –(0.450) –(1.330) –(1.070) –(1.030) –(1.370) (0.430) Tabora –0.547 –1.169** –0.341 0.003 –0.65** 0.579* –0.271 –0.321 –0.731**   –(1.640) –(3.010) –(0.770) (0.010) –(1.970) (1.860) –(1.410) –(0.980) –(2.110) Rukwa –0.966** –1.33** –0.75* –0.494* –1.127** 0.092 –0.372 –0.61 –0.358   –(2.790) –(3.100) –(1.660) –(1.870) –(3.080) (0.190) –(1.600) –(1.600) –(0.980) Kigoma –0.55* –1.03** –0.081 –0.164 –0.456 –0.473 –0.33** –0.637* –0.164   –(1.840) –(2.620) –(0.210) –(0.720) –(1.380) –(1.460) –(2.000) –(1.860) –(0.780) Shinyanga –0.151 –0.698* 0.544 0.129 –0.276 –0.12 –0.12 –0.345 0.009   –(0.500) –(1.930) (1.470) (0.650) –(0.910) –(0.370) –(0.790) –(1.070) (0.040) Kagera –0.862** –1.513*** 0.04 –0.431 –0.637** –1.38 –0.425** –0.695** –0.441*   –(2.780) –(3.810) (0.100) –(1.230) –(1.970) –(1.230) –(2.750) –(2.130) –(1.670) Mwanza –0.507 –1.166** 0.549 0.251 –0.141 –0.219 –0.082 –0.323 –0.358*   –(1.550) –(2.910) (1.380) (1.210) –(0.450) –(0.680) –(0.580) –(0.990) –(1.740) Mara –0.727** –1.431** 0.291 –0.718** –1.638*** –0.068 –0.451** –0.782** –0.149   –(2.070) –(3.080) (0.750) –(2.450) –(3.600) –(0.230) –(2.200) –(2.050) –(0.670) Manyara –1.118** –1.845*** 1.249 –0.724** –1.218** –0.273 –0.307 –0.636* –0.297   –(3.140) –(4.350) (1.400) –(2.330) –(3.030) –(0.450) –(1.510) –(1.750) –(1.020) Kaskazini Unguja –0.805** –1.226** –0.406 0.403 –0.157 0.993 –0.716** –1.295** –0.29   –(2.060) –(2.570) –(1.010) (1.070) –(0.390) (1.190) –(3.200) –(3.350) –(0.940) Kusini Unguja 0.071 0.138 –0.687 0.329 –0.069 1.079 –0.64** –1.121** –0.742**   (0.200) (0.290) –(1.270) (1.030) –(0.170) (1.340) –(2.170) –(2.530) –(1.970) Mjini/Magharibi –0.447 –0.574 –0.477 –0.03 –0.704 0.146 –0.357* –1.084** –0.242   –(1.420) –(1.200) –(1.390) –(0.080) –(1.450) (0.250) –(1.820) –(2.490) –(1.120) Kaskazini Pemba –0.629* –0.801 –0.702* 0.146 –0.27 –0.166 –0.602** –1.178** –0.303   –(1.740) –(1.620) –(1.820) (0.450) –(0.650) –(0.360) –(2.810) –(2.900) –(1.110) Kusini Pemba –0.741** –1.104** –0.407 –0.762** –1.204** –0.937 –1.361*** –2.175*** –0.493**   –(2.070) –(2.310) –(0.910) –(2.290) –(3.070) –(1.420) –(5.530) –(4.840) –(2.400) Father does not live with 0.159 0.224* 0.001 0.09 0.147 0.099 –0.002 –0.013 –0.078 HH (1.600) (1.870) (0.010) (1.100) (1.470) (0.660) –(0.030) –(0.150) –(0.760) Mother does not live 0.135 0.086 0.287** –0.051 –0.135 0.202 0.013 0.04 –0.093 with HH (1.340) (0.720) (1.980) –(0.580) –(1.470) (0.960) (0.180) (0.470) –(0.960) (continues to next page) 144 Tanzania Mainland Poverty Assessment Regression of Income on Circumstances (continued) Table 5–3   2008 2010 2012   Rural Urban Total Rural Urban Total Rural Urban Total Mother died before age 15 0.079 0.005 0.371 –0.059 –0.109 0.064 –0.004 0.15 –0.253* yrs of head (0.460) (0.030) (1.340) –(0.440) –(0.690) (0.260) –(0.030) (0.870) –(1.720) Father died before age 15 –0.243* –0.393** –0.463** –0.05 0.005 –0.19 –0.142 –0.21* 0.072 yrs of head –(1.850) –(2.450) –(2.160) –(0.440) (0.040) –(0.910) –(1.450) –(1.680) (0.620) Constant 9.889*** 10.329*** 10.392*** 10.887*** 11.409*** 10.484*** 10.32*** 10.793*** 10.146*** (26.760) (22.760) (19.880) (40.520) (30.570) (11.310) (48.360) (27.090) (36.440) No of min observations 1872 1109 2981 2385 1203 3588 2964 1658 4707 Number of imputations 10                   Source: NPS 2008, 2010, 2012. Note: The dependent variable is the logarithm of real monthly per capita income. Numbers in parentheses are bootstrapped student-t based on 100 replications. * Significant at the 10 percent level; ** significant at the 5 percent level; *** significant at the 1 percent level. Appendices 145 Appendix 6: Demography The Demographically Based Forecasting spurious results less likely. However, the crucial argument is Model for per Capita Income that the forecasting performance of the model out-of-sam- This Appendix builds on the work of Lindh and Malmberg ple is quite good on average and yields very reasonable (2007), who have developed a demographically based long-term predictions for growth rates. Spurious regression forecasting model for GDP. The model includes a number parameters would not perform that well. Furthermore, the of demographic variables and allows for some systematic impact of demographic variables depends on several fac- country heterogeneity as well as for time-specific effects. tors, such as policies that are conducive (or not) to the in- Denoting y the level of real GDP per capita, e0 life expectan- crease of employment and labor force participation as the cy at birth, and a each age group’s share in the population, supply of potential workers increases, and some favorable the regression equation has been specified as: or less favorable circumstances, which might be related, for example, to geography or the prevalence of diseases. 65+ yit = α ln e0it + ∑ ( βk + γk ln e0it )ln akit + ηi + νt + εit (1) To the extent that such circumstances are inherent and con- k =0−14 stant disadvantages, this will be picked up by country-spe- The interaction terms allow for changing age-share coeffi- cific intercepts in the regressions, but when these factors are cients contingent on how far the demographic transition episodic and changing over time we would expect them has progressed. The ηi and νt account for country- and to turn up in the form of systematic underperformance or time-specific effects. The subdivision into age groups is as over-performance relative to the model we estimate. Hav- follows: children 0–14 years old, young adults 15–29 years ing estimated equation (1) using the whole sample of 108 old, mature adults 30–49 years old, middle-aged adults 50– countries we next subdivided it into two: over-performers 64 years old, and old dependents 65 years and older. (countries with a higher average growth rate than the one predicted by the model) and underperformers. Table 6-1 re- Following Kelley and Schmidt (1995), life expectancy is in- ports estimation results for the full sample (column (1)) as cluded to capture human capital effects. Increases in life well as for the two subsamples (columns (2) and (3)). The expectancy and years of schooling are mutually reinforcing results of the full sample were used to produce the forecasts (longer life span encourages greater investment in educa- for Tanzania presented in this report. tion, and the other way around), and in many countries the relationship between them is nearly linear.77 Controlling for country-specific effects allows for some country heterogene- 77 Technological change and other trends are also accounted for ity, especially for that which could be accounted for by omit- by this variable, at least to some extent. ted variables remaining constant over the estimation period. 78 However, there will always be more complex heterogeneity, Controlling for time-specificity allows for influences in time such as differences in technology and preferences that vary over which are common to all countries, such as the world busi- time and across countries. The estimation result must therefore be ness cycle, world market price fluctuations, etc.78 Equation (1) interpreted as valid for an average country conditional on the con- was first estimated as a panel on a sample of 108 countries trols. In the sample individual countries will be distributed around with sufficiently long time series (minimum 20 years) for an- the average model with deviations that may be more or less im- nual purchasing power parity GDP, the dependent variable. portant. To take an obvious example, the genocide in Rwanda causes large deviations from the average model. To the extent that this has affected life expectancy and age structure, it is accounted The fact that the variables are trended raises questions of for in the model, but the disturbance to production of that kind spurious regression. Lindh and Malmberg (2007) show that of event is much larger than the demographic repercussions can the age variables can probably be treated as if co-integrated account for. Events like the tsunami in the Indian Ocean will also with GDP. Even if this were not true, the panel context makes cause deviations from the average model. 146 Tanzania Mainland Poverty Assessment For the full sample of 108 countries, column (1) in Table 6-1 Table 6–1  Demographically Based shows that most coefficients are different from zero at con- Forecasting Model for Real GDP ventional significance levels. Furthermore, the coefficient per Capita pattern indicates that with increasing life expectancy, the (1) (2) (3) positive correlations of the younger active age groups will Full sample Over-performers Under-performers tend to become smaller or even negative. Life expectancy –1.480 3.203 –7.224*** (le) (1.779) (2.304) (2.306) The difference in actual and predicted growth rate be- Population shares [males 65+] tween 1987 and 2009 was then calculated and two subsa- 0–14 5.302** 1.572 13.86*** mples were created. Over-performing (under-performing) (2.220) (2.966) (2.830) countries were defined as having a higher (lower) average 15–29 11.53*** 8.650*** 17.74*** growth rate than the one predicted by the model. This left (1.634) (2.146) (2.040) 54 countries in each subsample. Table 6-2 shows the coun- 30–49 8.928*** 3.881** 16.41*** tries that belong to each group. (1.478) (1.964) (1.838) 50–64 –7.719*** –7.247*** –6.183*** Equation (1) was then estimated for each sub-sample. The (0.862) (1.136) (1.100) results are shown in the last two columns of Table 6-1 (col- females 65+ –0.742 –2.393*** –2.159*** umns (2) and (3)). Overall, the pattern and magnitude of the (0.511) (0.614) (0.835) coefficients is similar in both subsamples and do not greatly Interactions (le * pop shares) differ from the full sample regression. Moreover, comparing le * 0–14 –1.360*** –0.603 –3.309*** (0.507) (0.676) (0.650) the predicted GDP paths for Tanzania the resulting models le * 15–29 –2.675*** –1.955*** –4.088*** are not that different. (0.376) (0.493) (0.470) le * 30–49 –1.900*** –0.712 –3.612*** (0.343) (0.460) (0.427) le * 50–64 1.948*** 1.792*** 1.643*** (0.205) (0.270) (0.261) le * females 0.213* 0.549*** 0.551*** 65+ (0.123) (0.147) (0.201) Observations 6,027 3,009 3,018 R-squared 0.730 0.785 0.792 Number of 108 54 54 countries Source: Own estimation using data from the Penn World Tables 7.0 (GDP per capita) and the World Population Prospects 2012 Revision (life expectancy at birth and population shares). All vari- ables (life expectancy, population shares and GDP are in natural logarithms). Estimations include time and individual fixed effects. The omitted category is shown in brackets. Standard errors in pa- rentheses. Asterisks denote the significance level (double sided): 10%, **: 5%, ***: 1% Appendices 147 List of Countries in the Sample Table 6–2   List of Countries in the Sample Table 6–2   Over-performing Under-performing Over-performing Under-performing Angola Algeria Malawi Italy Argentina Austria Malaysia Ivory Coast Australia Barbados Mali Jamaica Bangladesh Benin Mauritania Japan Belgium Brazil Mauritius Jordan Bolivia Burundi Mozambique Kenya Botswana Cameroon Namibia Korea, Republic of Burkina Faso Canada Nepal Madagascar Cape Verde Central African Republic Niger Mexico Chad Colombia Nigeria Morocco Chile Comoros Norway Netherlands China Congo, Dem. Rep. Pakistan New Zealand Cyprus Congo, Republic of Philippines Nicaragua Denmark Costa Rica Rwanda Panama Dominican Republic Ecuador Senegal Paraguay Egypt El Salvador Singapore Peru Ethiopia Fiji South Africa Portugal Gambia, The Finland Sri Lanka Romania Ghana France Sweden Sierra Leone Guatemala Gabon Tanzania Spain Guinea-Bissau Greece Thailand Switzerland India Guinea Trinidad & Tobago Syria Indonesia Haiti Uganda Togo Ireland Honduras United Kingdom Tunisia Israel Hong Kong United States Turkey Lesotho Iceland Uruguay Venezuela Luxembourg Iran Zambia Zimbabwe 148 Tanzania Mainland Poverty Assessment The Correlates of Fertility Correlates of Fertility, Women Aged 15–49 Table 6–3   TableYears,C2010 6–3   orrelates of Fertility, Women Aged 15–49 Year Total children born Total children born Current age [40–49] Tanga 0.481*** 15–19 –5.608*** (0.151) (0.0982) Morogoro 0.240** 20–24 –4.621*** (0.107) (0.0828) Pwani 0.235** 25–29 –3.386*** (0.113) (0.0756) Lindi –0.267 30–34 –2.261*** (0.169) (0.0770) Mtawara –0.690*** (0.155) 35–39 –1.056*** Ruvuma 0.125 (0.0866) (0.116) Education [none] Iringa 0.266** Some primary –0.212*** (0.116) (0.0702) Mbeya 0.625*** Primary or more –0.230*** (0.115) (0.0576) Singida 0.573*** Age 1st intercourse –0.192*** (0.136) (0.00656) Tabora 0.500*** Marital status [never married] (0.129) Ever married 0.680*** Rukwa 1.082*** (0.0417) (0.129) Unmet need for contraception 0.803*** Kigoma 0.915*** (0.0503) (0.104) Earns cash –0.177*** Shinyanga 0.807*** (0.0391) (0.106) Kagera 0.725*** Wealth quintile [poorest] (0.127) Poor –0.0410 Mwanza 0.562*** (0.0584) (0.0986) Middle –0.0245 Mara 0.912*** (0.0601) (0.156) Richer –0.198*** Manyara 0.655*** (0.0643) (0.123) Richest –0.469*** Zanzibar North 0.646*** (0.0745) (0.114) Rural residence 0.181*** Zanzibar South 0.427*** (0.0530) (0.127) Region [Dar es Salaam] Town West 0.881*** Dodoma 0.439*** (0.103) (0.153) Pemba North 1.185*** Arusha 0.325*** (0.113) (0.112) Pemba South 1.308*** (0.124) Kilimajaro 0.278*** (0.104) (continues to next page) Appendices 149 Correlates of Fertility, Women Aged 15–49 Years, 2010 (continued) Table 6–3   Total children born Constant 8.397*** (0.186) Mean births / prob. 2.989041 (.0292352) Observations 9672 R–squared 0.701 Source: Demographic and Health Survey (DHS) 2010. Note: Dependent variable is total children ever born to a woman aged 15–49 years. 150 Tanzania Mainland Poverty Assessment