Report No. 25662-TP Timor-Leste Poverty Assessment Poverty in a New Nation: Analysis for Action (In Two Volumes) Volume II: Technical Report May 2003 Poverty Reduction and Economic Management Sector Unit East Asia and Pacific Region Joint Report of the Government of the Democratic Republic of Timor-Leste, ADB, JICA, UNDP, UNICEF, UNMISET and the World Bank CURRENCY EQUIVALENTS CurrencyName =US$ FISCAL YEAR July 1-June 30 ABBREVIATIONSAND ACRONYMS ADB Asian Development Bank CFET ConsolidatedFund for East Timor DPT Diphtheria, Pertussis,Tetanus ETTA East Timor Transitional Authority GDP GrossDomestic Product JICA Japan International Cooperation Agency MDG Millennium Development Goal MICS Multiple Indicators Cluster Survey MoPF Ministry of Planning and Finance NDP National Development Plan NGO Non-Government Organization PNG Papua New Guinea PPA Participatory Potential Assessment PPP Purchasing Power Parity PTA Parent Teacher Association SUSENAS Indonesian Socio-economicHousehold Survey TLSS Timor-LesteLiving StandardMeasurement Survey UN UnitedNations UNDP UnitedNations Development Program UNICEF UnitedNations Children'sFund UNMISET United Nations Mission of Support in East Timor UNTAET United Nations Transitional Administration in East Timor Vice President Jemal-ud-din Kassum, EAP Country Director Xian Zhu, EACNF SectorDirector and Chief Economist Homi Kharas, EASPR SectorManager Tamar Manuelyan Atinc, EASPR Task Team Leaders Benu Bidani and Kaspar Richter, EASPR TABLE OF CONTENTS PREFACE ........................................................................................................................... i ACKNOWLEDGEMENTS ........................................................................................... i11 1 . SURVEY DESIGNAND WELFARE MEASUREMENT .................................... 1 Introduction..................................................................................................................... 1 1 Poverty Definition........................................................................................................... Survey Design................................................................................................................. .. 4 Poverty Measurement..................................................................................................... 5 7 Inequality and Social Welfare....................................................................................... Sensitivity Analysis ........................................................................................................ 13 Mobility......................................................................................................................... 15 Summary....................................................................................................................... 17 Appendix: Constructing The Poverty Measure............................................................. 18 2. THE PEOPLE'SPERSPECTIVE ........................................................................ 40 Introduction................................................................................................................... 40 Subjective Well-Being.................................................................................................. 41 43 Winners andLosers....................................................................................................... ChangeinSubjective Well-Being Since The Violence ................................................ 48 PersonalandNational Prlorltles.................................................................................... . . . Policy and ResearchIssues........................................................................................... 50 54 3. WELFARE PROFILE ............................................................................................ 55 Introduction................................................................................................................... 55 Poverty Measures.......................................................................................................... 55 Geography..................................................................................................................... 62 Life Cycle...................................................................................................................... 65 Characteristics ofthe Household Head......................................................................... 67 Assets ............................................................................................................................ 73 Infrastructure ................................................................................................................. 82 Inequality and Social Welfare....................................................................................... 84 Policy and ResearchIssues........................................................................................... 89 4. LABORMARKETS, EMPLOYMENT AND POVERTY ................................. 90 Labor Force Participation and Employment Rates....................................................... 90 92 Unemployment inDili/Baucau..................................................................................... Characteristics of Employment..................................................................................... 95 Household Labor Income Sources................................................................................ 96 Labor Policies andthe Poor .......................................................................................... 97 Child Labor inTimor Leste .......................................................................................... 97 98 Appendix..................................................................................................................... Time use........................................................................................................................ 118 5. EDUCATIONAND POVERTY .......................................................................... 124 The Pre-Existing Conditions....................................................................................... 124 Accomplishments under the Transitional Administration .......................................... 125 136 Priorities...................................................................................................................... Issues inthe Education Sector .................................................................................... 152 DemandSide Issues .................................................................................................... 152 SupplySide Issues that mayAffect Demand.............................................................. 159 Challenges................................................................................................................... 163 Appendix ..................................................................................................................... 165 6 . DISADVANTAGEDGROUPS ............................................................................ 174 Introduction................................................................................................................. 174 175 Female Headship......................................................................................................... Gender......................................................................................................................... 178 Widows ....................................................................................................................... 181 ParentlessChildren ..................................................................................................... 184 Policy andResearch Issues......................................................................................... 187 7. FOODSECURITY ............................................................................................... 188 Introduction................................................................................................................. 188 Prevalence................................................................................................................... 188 Food Security and Poverty.......................................................................................... 193 Coping with Food Shortage........................................................................................ 195 Policy and ResearchIssues ......................................................................................... 200 I 8. DETERMINANTSOF POVERTY ..................................................................... 201 Introduction................................................................................................................. 201 Model.......................................................................................................................... 201 Estimation Results ...................................................................................................... 203 Simulation Methodology ............................................................................................ 206 summary..................................................................................................................... 212 9. REFERENCES ...................................................................................................... 213 TABLES Table 1.1:Number and percentageof householdsby analytical domain........................... 2 Table 1.2: Subjective Well Being: Matrix o f Power Status.............................................. 16 Table 2.1: Subjective Well-being: Adequacy................................................................... 42 Table 2.2: Subjective Well-being: Happiness................................................................... 42 Table 2.3: Change inLiving Standards Since the Violence in 1999................................ 43 Table 2.4: Change inCorruption Since the Violence in 1999.......................................... 44 Table 2.5: Subjective Well-being: Economic andPower Status ...................................... 45 Table 2.6: Matrix Economic Status .................................................................................. 46 Table 2.7: Matrix Power Status ........................................................................................ 46 Table 2.8: Mobility Measures........................................................................................... 47 Table 2.IO: Change inLiving Standards Since the Violence in 1999by Sector..............49 Table 2.9: Winners and Losers: Characteristics............................................................... 51 Table 2.11: PersonalPriorities for Living Standards........................................................ 53 54 Table 3.1:National Poverty Rates .................................................................................... Table 2.12: NationalPriorities for Living Standards........................................................ 55 Table 3.2: Poverty rates at different poverty lines............................................................ 58 Table 3.3: Poverty andHousehold Size............................................................................ 59 Table 3.5 :Poverty by Analytical Domains....................................................................... Table 3.4: Poverty and Geography ................................................................................... 62 63 Table 3.6: Poverty and Demographic Groups................................................................... 65 Table 3.7: Age of HouseholdHeadand Poverty .............................................................. 68 Table 3.8: School Grade Completed of HouseholdHeadand Poverty (YO) ...................... 70 Table 3.9: Employmentof HouseholdHeadand Poverty (%) ......................................... 72 Table 3.10: Landcharacteristics among LandHolders .................................................... 74 Table 3.12: Livestock Holdings........................................................................................ Table 3.1 1:Landaccess and Poverty................................................................................ 75 78 Table 3.13: Livestock Holdings andPoverty.................................................................... 79 Table 3.14: Households Savingsby Type......................................................................... 80 Table 3.15: Household Savings inUrbanandRuralAreas .............................................. 81 Table 3.16: Access to Infrastructure and Poverty............................................................. 83 Table 3.17: Distributionof MonthlyPer Capita Expenditure........................................... 85 Table 3.18: Inequality andGeography.............................................................................. 85 Table 3.19: Inequality andAnalytical Domains ............................................................... 86 Table 3.20: Decomposition o f Inequality ......................................................................... 88 Table 4.2: Labor Force Participation Ratesby Gender, Poverty and Age...................... Table 4.1:Labor Force Participation Rates by Gender, Poverty andRegion...................99 Table 4.3 : Labor Force Participation Ratesby Gender, Age andEducation..................100 100 Table 4.4: ProbitRegressionsof Labor ForceParticipation, Adults 15-64years ..........101 Table 4.5:Employment Rates by Gender, Poverty and Region..................................... 102 Table 4.6: Employment Rates by Gender, Poverty and Age .......................................... 103 Table 4.7: EmploymentRates by Gender, Age and Education...................................... 103 Table 4.8: Probit Regressionsof Employment, Adults 15-64years............................... 104 Table 4.9: Employment Status by Quintile and Gender................................................. 105 Table 4.10: Employment by Gender, Sector, Poverty andRegion................................. 105 Table 4.12: Employment Characteristicsby Gender andPoverty, Adults 15-64 years .106 Table 4.11: Distributionof Workers by Quintile, Gender and Sector............................ Table 4.14: Hourly Wages amongEmployeesby EmployerType (US Dollars) ...........108 Table 4.13 :Probit Regressionsof Wage Employment, Adults 15-64 years ..................106 107 Table 4.15: Monthly Wages for Civil Service by Level (US Dollars) ........................... 108 Table 4.16: Earnings in Self-employment by Sector andUrbadRural(US Dollars).....109 Table 4.17: Comparison of UnemploymentDefinitions................................................. 110 110 Table 4.19: Unemployment Rates................................................................................... Table 4.18: Poverty Rates by Labor Status..................................................................... 111 111 Table 4.21:Probit Regressionsof Unemployment inDiWBaucau................................. Table 4.20: Characteristics of the Unemployed.............................................................. 112 Table 4.22: Household Sources of Labor Income .......................................................... 114 Table4.23: Labor Force Participation Ratesby Gender, Poverty andRegion, Children 10-14 years.............................................................................................................. 115 Table 4.24: Hours worked last week by Gender andPoverty. Children 10-14 years.....115 116 Table 4.26: Time Use last Week by Age, Gender andPoverty ...................................... Table 4.25:ProbitRegressionso f Child Labor (ages 10-14) ......................................... 117 Table 5.1:Characteristics of Schools inTimor-Leste, 2001 .......................................... 126 Table 5.2: Enrollinelitby Single Age ofthe Populationthat remains inTimor.Leste, 1998to 2001 129 Table 5.3: Enrollmentby Age Group o f the Populationthat remains in........................ (%) ........................................................................................................................... 130 Table 5.4: Example of Children who are now enrolled.................................................. 130 Table 5.5: Correlates of Enrollment(ages 5-24) ............................................................ 135 Table 5.6: EstimatedPopulation of Timor-Leste by Age Group.................................... Table 5.7: Highest Grade Completed amongthose who have attended, Ages 19-29.....139 136 Table 5.8: Highest Grade Completed amongthose who have attended, Ages 30 and older 139 Table 5.9: Studentsby Gradeand Age, 2001 ................................................................. ................................................................................................................................. 142 Table 5.10: Studentsby Grade andAge (%), 2001 ........................................................ 143 Table 5.11:Gross andNet Enrollment Ratios ................................................................ 144 Table 5.12: Net Enrollment Ratio Using Age 6 as the starting age (as proposed) .........144 Table 5.13 : Enrollment by Quintile and Grade, 2001..................................................... 145 Table 5.14: Age Distributionby Grade o fthe Poorest Quintile (%) .............................. 147 Table 5.15: Age Distributionby Gradeof the Richest Quintile (%) .............................. 147 Table 5.16: Repetition, Promotion andDropout Ratesby Grade (YO)............................ 149 Table 5.17: Out-of-School Children by Quintile andAge (YO)....................................... 151 Table 5.18: Out-of-School Children by Quintile andAge.............................................. 151 Table 5.19: Numberof EnrolledandRelevant Age Population..................................... Table 5.20: Any Days Absence inPrimary Education withinthe last 3 months............151 156 Table 5.21:Number of Days Absence inPrimary Education within the last 3 months . 156 Table 5.22: Reasons for AbsenteeisminPrimary Education ......................................... 157 157 Table 5.24: Reasons for Absenteeism inJunior Secondary Education .......................... Table 5.23: Numberof Days Absence inJunior SecondaryEducation.......................... 158 Table 5.25: Numberof Days Absence in Senior Secondary Education ......................... 158 158 Table 5.27: Aspects of School Attendance ..................................................................... Table 5.26: Reasons for Absenteeism inSenior Secondary Education.......................... 159 Table 5.28: Schooling Characteristics ............................................................................ 161 Table 5.29: How Obtained Textbooks?, First Source..................................................... 161 Table 5.30: How Obtained Textbooks?, Second Source ................................................ 162 Table 5.3 1: How Obtained Textbooks?, Third Source ................................................... 162 Table 5.32: Has a DesWChair at School? ....................................................................... 162 Table 5.33: Were Teachers inSchool?........................................................................... 162 Table 6.1:Demographic HouseholdComposition.......................................................... 175 Table 6.2: Poverty andGender ....................................................................................... 176 177 Table 6.4: Poverty and Gender ofthe Household Head ................................................. Table 6.3: Welfare and Gender....................................................................................... 179 Table 6.5: Female Headship andWelfare ....................................................................... 180 Table 6.6: Poverty andWidowhood ............................................................................... 182 Table 6.7: Widowhood Status and Welfare .................................................................... 183 Table 6.8: Child Poverty andParentalLiving Status...................................................... 184 Table 6.9: Child Welfare andParental Living Status (%) .............................................. 186 Table 7.1: Food Security: Summaryby Domain............................................................ 189 Table 7.2: Food Security andEx-Ante Coping Strategies.............................................. 196 Table 7.3: Coping StrategiesWhenNot EnoughFood................................................... 197 Table 7.4: Coping Strategies: Intrahousehold Transfers................................................. 198 Table 7.5: Coping StrategiesWhenNot EnoughFood: Who Suffers? .......................... 199 Table 8.1: OLS Regressionson Log Per Capita Consumption....................................... 205 Table 8.2: Actual and PredictedConsumptionandPoverty ........................................... 207 Table 8.3 : Simulations o f % Changes inConsumption andPoverty.............................. 211 FIGURES Figure 1.1:Cumulative Distributionof Per Capita Consumption ...................................... Figure 1.2: Real Per Capita Household ConsumptioninUrbanandRural Areas..............8 9 Figure 1.3:Density Function of Per Capita Consumption............................................... 10 Figure 1.4: Poverty and HouseholdSize: Equivalence Scales ......................................... 12 13 Figure 1.6: Lorenz Curves for Urbanand RuralAreas..................................................... Figure 1.5: Poverty and HouseholdComposition............................................................. 14 Figure 1.7: Generalized Lorenz Curves for UrbanandRuralAreas ................................ 15 Figure 3.1:Cumulative distributionof per capita consumption ....................................... 57 Figure 3.2: Densityfunction o fper capita consumption................................................... 58 Figure 3.3: Poverty andHousehold Composition............................................................. 60 Figure 3.4: Poverty andHousehold Size: Equivalence Scales ......................................... 61 Figure 3.5: FirstOrder Dominance Results...................................................................... 64 Figure 3.6: Poverty andDemographic Groups: Equivalence Scales ................................ 66 Figure 3.7: Poverty and Age of HouseholdHead: Equivalence Scales............................ 68 Figure 3.8: Poverty, Age andGender of HouseholdHead............................................... 69 Figure 3.9: Poverty and Land Size: Ruralversus Urban.................................................. 76 Figure 3.10: Poverty andLandValue: Ruralversus Urban.............................................. 77 Figure 3.11: Poverty and Livestock .................................................................................. 79 Figure 3.12: Poverty and Household Savings................................................................... 82 Figure 3.13: Lorenz Curves andGeography..................................................................... 86 87 Figure 3.15: Generalized Lorenz Curves for Timor-Leste ............................................... Figure 3.14: Lorenz Curves andAnalytical Domains ...................................................... 88 Figure 4.1: Household Sources of Income...................................................................... 113 Figure 4.2: Household Sources of Income inRuralAreas ............................................. 113 Figure 5.1:Illiteracy Rates inTimor-Leste by Place of Birth, 1990.............................. 125 Figure 5.2: School Participation by Age, 1998 and 2001............................................... 127 Figure 5.3 : School Participation by Age, 1999............................................................... 127 Figure 5.5: School Participation of Poorest andRichestUrbanQuintiles, 2001............ 128 Figure 5.4: School Participation by Age, 2001............................................................... 128 Figure 5.6: School Participation of Poorest andRichest RuralQuintiles, 2001.............129 Figure 5.7: Monthly Per Capita HouseholdExpenditure on Public Primary Education, 1995......................................................................................................................... 132 Figure 5.8: Monthly Per Capita Household Expenditureon Public Primary Education, Figure 5.9: Household Spendingon Primary Education by Quintile, 2001...................132 2001......................................................................................................................... 133 Figure 5.10: Household Spendingon Junior SecondaryEducation by Quintile. 2001.. 133 Figure 5.11:Household Spendingon Senior SecondaryEducationby Quintile. 2001.. 134 Figure 5.12: Ever Attended School by Quintile andAge ............................................... 137 Figure 5.13:Can ReadA Letter by Quintile?. Adults 30 andolder ............................... 138 Figure5.14: CanReadA Letter by Quintile?. Ages 13-15 ............................................ 138 Figure5.15:Enrollmentby Grade and Relevant School-Age Population...................... 141 Figure5.16: Populationby Age andEnrollmentby Grade. 2000/01............................. 141 Figure5.17: Enrollmentby Grade at the RightAge. 2000/01........................................ 141 Figure5.18: Enrollment by Gradeandby Poorest andRichest Quintiles...................... 145 Figure 5.19: Out of School Children and Youth by Age. 2001 ...................................... 150 Figure5.20: Reasons for Never Attended. Ages 5-6. 2001............................................ 154 Figure5.21: Reasonsfor NeverAttended. Ages 7-12. 2001.......................................... 154 Figure5.22: Reasonsfor NeverAttended. Ages 13-15. 2001........................................ 155 Figure 7.1:Household Food Security by Month............................................................ 191 Figure 7.2: Not Enough Foodby Domain...................................................................... 192 Figure 7.3:National Poverty and Interview Date........................................................... 193 Figure 7.4: Regional Poverty andInterviewDate .......................................................... 194 PREFACE This report lays out the challenge of poverty reduction inTimor-Leste. It is based on the first nationally-representative household survey collected during August to December 2001. This work was conducted by the Poverty Assessment Project, a partnership between the Government o f Timor-Leste (with the Ministry o f Planning and Finance providing overall guidance), the World Bank, the Asian Development Bank (ADB), the Japanese International Cooperation Agency (JICA), the United Nations Development Program (UNDP), United Nations Children's Fund (UNICEF) and United Nations Mission o f Support in East Timor (UNMISET). The Poverty Assessment Project was launchedto provide up-to-date information on living conditions after the violence in 1999 as input into the National Development Plan. The Poverty Assessment Project comprised three data collection activities on different aspects o f living standards, which taken together, provide a comprehensive picture o f well-being in Timor-Leste on the eve o f independence: 0 Suco Survey -This is a census of all the 498 sucos inthe country and provides an inventory o f existing social and physical infrastructure, and o f economic characteristics o f each suco, in addition to aldeia level population figures. It was completed between February and April 2001, and the report, written by the ADB, was published inOctober 2001. 0 Participatory Potential Assessment: This qualitative community survey assisted 48 aldeias to take stock o f their assets, skills and strengths, identify the main challenges and priorities and formulate strategies for tackling these within their communities. The field work took place betweenNovember 2001 and January 2002. This activity was managed by UNDP and the report was finalized inMay 2002. 0 Household Survey: The Timor-Leste Living Standards Measurement Survey is a nationally representative survey o f 1800 households from 100 sucos covering one percent o f the population. This comprehensive survey was designed to diagnose the extent, nature and causes o f poverty and analyze policy options for the country. Data collection was undertaken between end- August andNovember 2001. This report, written in two volumes, was a collaborative effort o f the members of the Poverty Assessment Project, with the World Bank taking the lead in the analysis. The objectives o f this report are modest - to set a baseline for the new country on the extent, nature and dimensions o f poverty; to assist the decision making o f the newly elected government and its efforts in formulating, implementing and monitoring its Poverty Reduction Strategy. The objective was not to lay out the elements o f the poverty strategy but rather to present evidence on the basis of which the Timorese can define and refine their own poverty reduction strategy. We hope this isjust the start of a series o f analysis to consider the effects o f government policies on different groups o f people, especially the poor. The preliminary analysis from the household survey was presented at a workshop inDili inFebruary 2002. The early results fed into the National Development Planpresented by the Government at independence. Sector analysis for health, education and agriculture were also presented at the workshop and in more detailed discussions with the relevant Ministries. The full report was discussed with the Government in January 2003. A series o f seminars was organized by the Ministry o f Planning and Finance during January 13- 24,2003. The disseminationtook place before the Ministries embarked on the prioritizing and sequencing o f the National Development Plan for the FY2004 budget. Seminars were held at the Council o f Ministers and several Ministries (Education, Health, Agriculture, Labor and Solidarity and Finance and Planning). A large workshop in Dili and three regional worlcshops in Baucau, Ainaro and Maliana were organized for Government officials from the center and districts, civil society representatives, including the Church, women's, students and youth groups, NGOs, Chefe de Sucos, and development partners. The results froin the UNICEF sponsored Multiple Indicators Survey (MICS) were also presented by their staff and consultants at these workshops, and at the Council o f Ministers and the Ministry o f Health seminars. The report was revised in light of the comments received and the health section was updated using the MICS results. .. 11 ACKNOWLEDGEMENTS This report is a result o f a highly collaborative process betweenthe Government and our donor partners, ADB, JICA, UNDP, UNICEF and UNMISET. The Poverty Assessment Steering Committee chaired by Ms. Emilia Pires, Advisor, Ministry o f Planning and Finance (MoPF), provided overall guidance. We are very grateful to the Steering Committee members for their strategic guidance. The Steering Committee members included Emilia Pires, Robin Boumphrey (ADB Resident Representative), Gwi Yeop Son (Deputy Resident Representative, UNDP), Sarah F. Cliffe (Chief o f Mission, World Bank) and Mr. Takehara Masayoshi (JICA). Following the change o f donor representatives in Dili after Independence, the Steering Committee members were Meeja Hamm (ADB Resident Representative), Shoji Katsuo, (Resident Representative, JICA), Haoliang Xu (Deputy Resident Representative, UNDP), Mr. Yoshi Uramoto (Special Representative, UNICEF) and Elisabeth Huybens (Country Manager, World Bank). We are very grateful to Emilia Pires for her leadership, constant support, and enthusiasm throughout this project. We would also like to express our thanks to Ms. Aicha Bassarewan, Vice Minister, MoPF, for her leadership during the dissemination o f the Poverty Report. We are very thankful to the Statistics team (MoPF) for the great collaboration and partnership. The National Statistics Office team did an outstanding job in implementing the Suco Survey and the household survey under difficult conditions. The core team was led by Manuel Mendonca, Director o f the National Statistics Office, and included Lourenco Soares (Data Manager), Elias dos Santos Ferreira (Field Manager) and Afonso Paixes (Field Manager). It was responsible for implementing the surveys, quality control and supervision, all o f which they managed with great skill. We are also grateful to the survey teams in charge o f fielding the questionnaires. Their names are attached to this acknowledgement. Sonia Alexandrino from the Planning Office provided excellent logistical support inDili, and David Brackfield, Advisor in the National Statistics Office was always ready to lend a competent helping hand. The assistance from Gastao de Sousa and other staff o f the Planning and External Management Assistance Division o f the MoPF is gratefully acknowledged. The World Bank office in Dili consistently provided outstanding support to us. Annette Leith and Diana Isaac always found a way to solve our problems, and the rest of the Dili team helped in innumerable ways, for which we are very thanltful. The Timor-Leste Country team contributed greatly to the entire program o f activities and we gratefully acknowledge contributions from Sofia Bettencourt, Gillian Brown, Lisa Campeau, Alfonso de Guzman, Adrian Fozzard, Dely Gapasin, Francis Ghesquiere, Ronald Isaacson, Natacha Meden, IanMorris, Janet Nassim and KinBingWu. 111 ... The World Bank team comprised Benu Bidani, Kaspar Richter, Martin Cumpa, Juan Mufioz and Rodrigo MuAoz from Sistemas Integrales, Valerie Evans, David Madden, Kathleen Beegle, Paolo Nicolai and Wawan Setiawan. The Asian Development Bank team included Craig Sugden,Zacharias da Costa and Jessie B. Arnucu with Etienne van de Walle from the Manila office. The UNDP team included Antonio Assuncao, Jonathan Gilman, Janne Niemi, Sam Rao, Antonio Serra and Ian White. The JICA team included Charles Greenwald. The UNICEF MICS team included Yoshi Uramoto, Vathinee Jitjaturunt, Stemberg Vasconcelos, Rashed Mustafa, Peter Gardiner and Mayling Oey- Gardiner. This report was written by Benu Bidani and Kaspar Richter with superb overall assistance from Martin Cumpa. Background papers were written by Kin Bing Wu with inputs from Deon Filmer, Kathleen Beegle and Martin Cumpa on Education, Jean Foerster with analysis by Martin Cumpa on Agriculture, Janet Nassiin with analysis by Martin Cumpa on Health, Kathleen Beegle and Martin Cumpa on Labor Markets, and by Kaspar Richter on the Welfare Profile, Disadvantaged Groups, and Food Security. Taranaki Mailei provided assistance with the task and the production o f the report. Walter Meza-Cuadra also helped in formatting the report. The peer reviewers were Pierella Paci and Lant Pritchett. This Report was prepared under the overall guidance of Homi Kharas (Chief Economist and Sector Director, EASPR), Klaus Rohland (Former Country Director), Xian Zhu (Country Director) and Tainar Manuelyan Atinc (Sector Manager, Poverty). The team greatly benefited from advice and guidance from Tamar Manuelyan Atinc. We are also very grateful to Sarah Cliffe (Chief o f Mission) and Elisabeth Huybens (Country Manager) for their consistent guidance and great support inthe field and to Sanjay Dhar (Lead Economist) for his advice in headquarters. We benefited greatly from the extensive comments received from the participants at the dissemination seminars, and the detailed written comments from the Ministry o f Health, Pierella Paci and Lant Pritchett (peer reviewers), Sofia Bettencourt, Elisabeth Huybens, ADB (Meeja Hamm and Craig Sugden), UNDP reviewers, Sam Rao, Caritas and Oxfam. The overall program of activities under the Poverty Assessment Project was funded jointly by the donor partners. The World Bank i s grateful to the Bank-Netherlands Partnership Program and the Norwegian Trust Fund for Environmentally and Socially Sustainable Development for financial support o f this project. Last but not least, our sincere gratitude goes to the people o f Timor-Leste who gave generously o ftheir time to help us collect the information on which this report is based. iv Timor-LesteLivingStandardsSurvey Team Members 1 Altilis Moniz do Rosario Supervisor 2 Antoninho dos Santos Supervisor 3 Manuelda Silva Supervisor 4 Antonio Soares Supervisor 5 Felix Celestino da C. Silva Supervisor 6 BatistaLeos Supervisor 7 D e Francisco Barreto Supervisor 8 Tornas Gusmao Supervisor 9 Juliao da Cruz Enumerator 10 SamuelFatima Enumerator 11 Armando Da Costa Enumerator 12 Gertudesde Amaral Enumerator 13 Henriqueta da CostaBraz Enumerator 14 Armando Martins Enumerator 15 Anibal Cardoso Enumerator 16 Miguel Pereira Enumerator 17 Manuel Ribeiro Enumerator 18 Antonio C. Alves Enumerator 19 Raul Pinto Enumerator 20 Manuel Soares Pereira Enumerator 21 Rodolfo Soares Enumerator 22 Dilvado Rosario de F. da C. Enumerator 23 Januario Ximenes Enumerator 24 Amaro da C Tilman Enumerator 25 EvaFemandes Enumerator 26 DeliaNunes Enumerator 27 Dominggos Moniz Enumerator 28 Filomena M.Guterres Enumerator 29 Julieta F. Silva Enumerator 30 Jaimito do Rego Enumerator 31 Gil Vicente Madeira Enumerator 32 Rogerio Castro Enumerator 33 Sebastiao Dlas Saldanha Enumerator 34 Saozinada Costa Enumerator 35 Maria L.De Jesus Enumerator V Timor-LesteLivingStandards Survey Team Members 1 Akilis Moniz do Rosario Supervisor 2 Antoninho dos Santos Supervisor 3 Manuelda Silva Supervisor 4 Antonio Soares Supervisor 5 Felix Celestino da C. Silva Supervisor 6 Batista Leos Supervisor 7 D e Francisco Barreto Supervisor 8 Tornas Gusmao Supervisor 9 Juliao da Cruz Enumerator 10 SamuelFatima Enumerator 11 Armando DaCosta Enumerator 12 Gertudesde Amaral Enumerator 13 Henriqueta da CostaBraz Enumerator 14 Armando Martins Enumerator 15 Anibal Cardoso Enumerator 16 Miguel Pereira Enumerator 17 Manuel Ribeiro Enumerator 18 Antonio C. Alves Enumerator 19 Raul Pinto Enumerator 20 Manuel Soares Pereira Enumerator 21 Rodolfo Soares Enumerator 22 Dilvado Rosario de F. da C. Enumerator 23 Januario Ximenes Enumerator 24 Amaro da C Tilman Enumerator 25 EvaFemandes Enumerator 26 DeliaNunes Enumerator 27 Dominggos Moniz Enumerator 28 Filomena M.Guterres Enumerator 29 Julieta F. Silva Enumerator 30 Jaimito do Rego Enumerator 31 Gil Vicente Madeira Enumerator 32 Rogerio Castro Enumerator 33 Sebastiao Dias Saldanha Enumerator 34 Saozinada Costa Enumerator 35 Maria L.D e Jesus Enumerator V Timor-LesteLivingStandardsSurveyTeamMembers(contd.) 36 Antonio B.S.Dasilva Enumerator 37 Rogerio Babo Data Entry 37 Inacia Vilena Data Entry 38 Silvina Suares DataEntry 39 Maria Odete DataEntry 40 SuzanaLeong da Costa Data Entry 41 EduardoMartinho Ximenes Data Entry 42 Maria Odete Baros Administration 43 Vicente Leande Jesus FieldAdministration 44 Saul do Carmo Ximcncs FieldAdministration 45 Kintao de Deus FieldAdministration 46 Tomas Pereirra FieldAdministration 47 Rafael C. Lobato FieldAdministration 48 NicolauPereira FieldAdministration 49 Manuel da Costa Silva FieldAdministration 50 Alvaro Maia FieldAdministration vi 1. SURVEYDESIGNANDWELFAREMEASUREMENT INTRODUCTION' 1.1 In developing countries, poverty is often seen as the defining characteristic of underdevelopment, and its elimination as the main purpose o f economic development. Poverty measures are designed to count the poor and to diagnose the extent and distribution o f poverty. The TLSS provides the information required to conduct such analysis. This chapter serves as a reference point for this poverty assessment. It introduces concepts and techniques that are widely used inother chapters. It is primarily written for technical experts in charge o f data analysis as reference on the details o f the analysis conducted to derive the summaries and messages contained in Volume Io f this report. However, it i s also addressed to decision makers who want to define the type o f information they need for monitoring o f poverty reduction and making appropriate policy decisions. The first sections o f this chapter deal with the design and data o f the TLSS. The later parts present the methodology of poverty, inequality, and welfare measurement. SURVEY DESIGN^ 1.2 A survey relies on identifying a subgroup o f a population that is representative both for the underlying population and for specific analytical domains of interest. The main objective o f the TLSS is to derive a poverty profile for the country and salient population groups. The fundamental analytic domains identified are the Major Urban Centers (Dili and Baucau), the Other Urban Centers and the Rural Areas. The survey represents certain important sub-divisions o f the Rural Areas, namely two major agro- ecologic zones (Lowlands and Highlands) and three broad geographic regions (West, Center and East). In addition to these domains, we can separate landlocked sucos (Inland) from those with sea access (Coast), and generate categories merging rural and urban strata along the geographic, altitude, and sea access dimensions. However, the TLSS does not provide detailed indicators for narrow geographic areas, such as postos or even district^.^ The survey has a sample size o f 1,800 households, or about one percent of the total number o f households in Timor-Leste. The experience o f Living Standards Measurement Surveys inmany countries -most o f them substantially larger than Timor- Leste - has shown that samples o f that size are sufficient for the requirements of a poverty assessment. This chapter was written by Kaspar Richter. This section draws on Muiioz (2001). Tiinor-Leste is divided into 13 major units called districts. These are further subdivided into 67 postos (sub-districts), 498 sucos (villages) and 2,336 uldeius (sub-villages). The administrative structure is uniform throughout the country, including rural andurban areas. 1.3 The survey domains were defined as follows. The Urban Area is divided into the Major Urban Centers (the 31 sucos in Dili and the 6 sucos in Baucau) and the Other Urban Centers (the remaining 34 urban sucos outside Dili and Baucau). The rest of the country (427 sucos intotal) comprises the Rural Area. The grouping o f sucos into urban and rural areas is based on the Indonesian classification. In addition, we separated rural sucos both by agro-ecological zones and geographic areas. With the help o f the Geographic Information System developed at the Department o f Agriculture, sucos were subsequently qualified as belonging to the Highlands or the Lowlands depending on the share of their surface above and below the 500 m level curve.4 The three westernmost districts (Oecussi, Bobonaro and Cova Lima) constitute the Western Region, the three easternmost districts (Baucau, Lautem and Viqueque) the Eastern Region, and the remaining seven districts (Aileu, Ainaro, Dili, Ermera, Liquica, Manufahi and Manatuto) belong to the Central Region. Table 1.1: Numberand percentageof householdsby analyticaldomain Agro-ecologic zone Geographicregion Total Highlands Lowlands West Center East Urban 5.446 36,008 5,698 28,317 3,792 41,454 Major Urban Centers 2,236 21,945 - 20,530 3651 24,181 Other Urban Centers 3,210 14,063 5,698 7,787 3,788 17,273 Rural 57,123 81,706 32,749 61,024 45,056 138,829 Total 62,569 117,714 38,447 89,341 52,495 180,283 Altitude Geographicregion Sea access Total Lowlands Midlands Highlands West Center East Inland Coast Urban 3 18 3 4 15 4 16 8 24 Major Urban Centers 2 10 1 0 11 2 I 6 13 Other Urban Centers 1 8 2 4 4 2 8 2 10 Rural 7 37 33 18 49 23 69 21 16 Total 0 55 36 22 54 2s 77 23 100 1.4 Our next step was to ensure that each analytical domain contained a sufficient number o f households. Assuming a uniform sampling fraction o f approximately 1/100, a non-stratified 1,800-household sample would contain around 240 Major Urban households and 170 Other Urban households -too few to sustain representative and significant analyses.' We therefore stratified the sample to separate the two urban areas from the rural areas (see Table 1.1). The rural strata were large enough so that its implicit stratification along agro-ecological and geographical dimensions was sufficient to ensure that these dimensions were represented proportionally to their share o f the population. The final sample design by strata was as follows: 450 households in the Major Urban ~ ~~~ Occasionally, we split Lowlands into Lowlands (of Flatlands), covering sucos below lOOm of altitude, and Midlands, defined as sucos between lOOmand 500m of altitude. The aldeia-level population numbers were collected by the Suco Survey. 2 Centers (378 in Dili and 72 in Baucau), 252 households inthe Other Urban Centers and 1,098 households inthe Rural Areas. 1.5 With the exception of Urban Dili, the sampling of households in each stratum followed a 3-stage procedure: Inthe first stage, a certain number o f sucos were selected withprobabilityproportional to size (PPS). Inthe second stage, 3 aldeias in each suco were selected, again with probability proportional to size (PPS). In the third stage, 6 households were selected ineach aldeia with equalprobability (EP). This implies that the sample is approximately self-weighted within the stratum: all households in the stratum had the same chance o f being visited by the survey. A simpler and more efficient 2-stage process was used for Urban Dili. Inthe first stage, 63 aldeias were selected with PPS and inthe second stage 6 households with equal probability ineach aldeia (for a total sample of 378 households). This procedure reduces sampling errors since the sample will be spread more than with the standard 3-stage process, but it can only be applied to Urban Dili as only there it was possible to sort the selected aldeias into groups o f 3 aldeias located inclose proximity o f each other. 1.6 The final sampling stage requires choosing a certain number o f households at random with equal probability in each o f the aldeias selected by the previous sampling stages. This requires establishing the complete inventory o f all households in these aldeias - a field task known as the household listing operation. Two operational approaches were considered for the household listing. One i s the classical door-to-door (DTD) method that is generally used in most countries for this kind of operations, The second approach -which i s specific o f Timor-Leste -depends on the lists o f families that are kept by most suco and aldeia chiefs in their offices. The prior-list-dependent (PLD) method is much faster, since it can be completed by a single enumerator in each aldeia, working most o f the time in the premises o f the suco or aldeia chief; however, it can be prone to biases depending on the accuracy and timeliness o f the family lists. After extensive empirical testing o f the weaknesses and strengths o f the two alternatives, we decided to use the DTD method in Dili and an improved version o f the P L D method elsewhere. The improvements introduced to the PLD consisted in clarifying the concept o f a household "currently living inthe aldeia", both by intensive training and supervision o f the enumerators and by making its meaning explicit in the forms wording (it means that the household members are regularly eating and sleeping inthe aldeia at the time o f the operation). Inaddition, the enumerators were asked to select a random sample o f 10 households from the list, and visit them physically to verify their presence and ask them a few questions6. The listing operation was completed by a team o f enumerators between May 21 andJune 28,2001. 1.7 The survey was fielded during end August to early December 2001. Eight field teams, each composed o f three interviewers and one supervisor, conducted the household survey. Each interviewer was asked to interview 6 households per week, using a questionnaire that generally required visiting each household several times. Data entry I t is generally a good idea to undertake the listing operation as an independent operation. This reduces incentives on the part of enumerators to not list difficult areas, such as households living on the top o f the mountain, to ensure that they are not selected in the enumeration. 3 took place in the field with the help o f laptops. This not only reduced the time o f data processing, but also allowed for immediate extensive checks on data quality built into the data entry program. Any inconsistency revealed at this stage was to be rectified by revisiting the households while still being in the village. In addition, we implemented a second round o f standard checks on data quality in the project office in Dili upon retrieval o f the data from the field teams. In general, with a few exceptions, the analysis has confirmed the highquality o fthe data entry and validation processes. POVERTY DEFINITION 1.8 The measurement o f poverty i s a contentious science. The poverty literature typically introduces a poverty line, below which people are defined as poor, and above which they are not poor. The simplicity o f a poverty line facilitates the focus on the poor, but it is a crude device. Inparticular, the discontinuity, with poverty on one side and lack o f poverty on the other, i s doubtful and typically not supported by evidence on empirical indicators, be they income, consumption, calories, or assets. 1.9 Poverty involves multiple dimensions o f deprivation, including poor health, low humancapital, andmalnutrition. Inprinciple, eacho fthese deserves separateattention, as the correlations between these categories are far from perfect. Nevertheless, much o f the progress in our understanding o f poverty over the last decades was based on the investigation o f poverty with single summary monetary measures o f poverty. While such measures are clearly limited in scope, they capture a central component o f any assessment o f living standards. The basic choice is between an income and a consumption indicator o f well-being. We follow common practice in most developing countries, and for much of the World Bank's analytical work, and adopt a consumption-based measure. Consumption i s likely to be a more useful and accurate measure o f living standards than income. This judgment already informed the TLSS questionnaire design. While the components o f consumption are included comprehensively, the coverage of income categories is sketchier. 1.10 To clarify the interpretation o f consumption as a welfare indicator, it is useful to review briefly the arguments for and against such a measure. Income, together with assets, measures the potential claims o f a person or household, while consumption captures the level o f living interms o f what living standard individuals actually acquire. From a theoretical point o f view, both concepts can be defended as an approximation to utility. More relevant is the issue o fthe time period over which living standards are to be measured. The main reason for preferring consumption to income as an indicator of living standards is variability (Ravallion 1994). Ina mostly agricultural economy people receive income only infrequently, and the amounts differ across seasons. Households often have consumption smoothing opportunities through savings and community-based risk sharing. This i s confirmed by empirical evidence suggesting that households inlow- income agricultural societies manage to smooth consumption in spite o f highly volatile income receipts (Deaton 1997). Thus, current consumption is likely to be a better indicator o f current well-beingthan current income; and current consumption may also be a better indicator of longer-term welfare, since it reveals information about incomes at other points intime. 4 1.11 To the extent that we collected incomes (e.g. in the agriculture section), the survey design recognized the fact that certain incomes (such as agricultural income) are variable over time and adjusted the reference period accordingly. For example, agricultural data were asked over the past 12 months reflects incomes o f agricultural households more accurately than a shorter period, say a month, even if interviewees recall only imperfectly extended periods. Similarly, consumption is measured using different recall periods reflecting the periodicity o f purchases (e.g. a week for food; but longer periods such a month and a year for more infrequent purchases such as clothing, household supplies, etc). POVERTY MEASUREMENT 1.12 The discussion so far suggests taking consumption as the welfare indicator. We still need to resolve how to convert this indicator into a measure about individual welfare. Following common practice, the TLSS collects expenditures on consumption items at the household rather than individual level. Most purchases occur for the household as a whole (e.g. food), andthe bulk o f food consumption takes placejointly duringmeals. 1.13 Households differ in size and composition. In particular, the needs o f household members differ, especially between adults and children. One option that has been used i s to try use a system o f weights, whereby for example, children count as a fraction o f an adult interms o f needs, and convert all households into the number o f equivalent adults. But there also exist economies o f scale in consumption. Some non-food items (for example, housing to an extent, durable goods) have public goods characters, as their usage by one member o f the household does not reduce their value to other household members. Thus, because people can share goods and services without reducing their welfare, the cost o f attaining a given level o f welfare may be lower in larger households than in smaller households. Simply deflating household consumption by household size ignores these economies o f scale inconsumption. The number o f equivalent adults can be adjusted for economies o f scale to get the number o f "effective" equivalent adults. 1.14 However, attempts to estimate the relative costs, or equivalent scales, faced by different types o f families have failed to establish a generally accepted methodology (Deaton 1997 and Ravallion 1994). Therefore, we follow standard practice and use per capita total household expenditure as the basic welfare indicator, that is, we do not attempt to measure differences in consumption within households and assume families allocate resources equally among their members. For the purposes o f constructing a poverty profile, it i s important to conduct a sensitivity analysis to see to what extent the broad conclusions depend on assumptions regarding equivalent scales. 1.15 Constructing a consumption measure with the TLSS involves going through a series o f steps, guided both by theoretical and practical considerations. Total household consumption i s built up from several components. It includes all reported expenditures on goods and services, and thenadds ina value for consumption that does not go through the market, like home produced consumption items or in-kind receipts from employers and donors. For perishable goods, it i s mostly safe to assume that a household's consumption is closely tied to their purchases. However, for expensive durable goods, a correction has 5 to be made for the difference betweenconsumption and expenditures. Finally, we need to convert all components into real terms usinga price index that accounts for differences in regions and interview dates. The consumption aggregate i s composed o f four main types o f goods and services: food items, non-food items, consumer durables, and housing. For any given household, the shares of these categories depend on a number o f factors, including living standards, demographic composition, location, and tastes. The specific items included in each component, as well as the methodology used to ascribe a consumption value to each o fthese items, are outlined inthe Appendix. 1.16 A poverty line determines the minimum level o f standard of living before a person i s no longer considered to be "poor". Setting poverty lines is often the hardest, and most controversial, step in constructing a poverty profile from household survey data. What method is chosen has implications for the incidence o f poverty, and for policy malting, such as inassessingwhether growth is pro-poor, or indetermining the allocation o f public resources across regions. Following common practice in East Asia, we defined a poverty line that is both "absolute" and "objective". The Appendix provides a brief description o fthe procedure proposed to derive a poverty line for Timor- Leste. 1.17 Followingmost o f the research on poverty measurement, we present three poverty statistics in the analysis. They all belong to the class o f measures proposed by Foster, Greer, and Thorbecke (1984) and are characterized by the following equation: where a is some non-negative parameter, z is the poverty line, y denotes expenditures, i indexes individuals, n equals the total number of individuals in the population, and q is the number o f individuals with expenditures below the poverty line. All measures o f this class are additive inthe sense that aggregate poverty equals the population-weighted sum o f the poverty levels inthe subgroups o f the population. We will use specific members o f this family o fpoverty measures: PI, the head-count index, P2, the poverty gap, andP3, the Foster-Greer-Thorbeclte measure. 1.18 The head-count index gives the share o f the poor inthe total population, in other words, it measures share o f population whose per capita consumption is below the poverty line. This indicator i s by far the most commonly poverty measure. It is appealing because it is simple and easily interpreted. However, it does have limitations. The most important is that the head-count index does not take into account whether the poor have consumption levels just below or far below the poverty line. It i s therefore indifferent to the distribution amongst the poor and insensitive to the degree o f poverty. The poverty gap equals the average expenditure shortfall o f the poor relative to the poverty line. It reflects the distance between the consumption levels o f the poor and the poverty line; the greater the distance the higher will be the poverty gap. The Foster-Greer-Thorbecke measure i s similar in construction to the poverty gap but differs in that it applies an 6 increasing weight to greater distances below the poverty line. This indicator is thus sensitive to the severity o f poverty. SENSITIVITY ANALYSIS 1.19 Although the general procedures for calculating poverty measures are well defined in theory, we have to make compromises between imperfect alternatives in practice. Difficult choices are required inparticular with regard to two issues: the level of the poverty line, and the comparisons across households of different size and composition. We ask how an analysis o f poverty can explore the impact of the assumptions adopted on these two issues. A useful and visually appealing way to investigate both topics i s to use "stochastic dominance'' analysis, which relies on graphical tools to examine the sensitivity o f poverty analysis. The motivation for such an approach is to let the facts speak for themselves as far as possible. Density and cumulative density functions provide a succinct and informative summary o f the distributions o f economic variables that are easily understood. These methods are also useful to indicate hypotheses to explore for explanation. We demonstrate this by comparing distributions for the population as a whole with the estimates for a range of relevant population subgroups. Poverty Line 1.20 H o w does the poverty headcouiit index change as we vary the poverty line? The sensitivity o f this measure to changes in the poverty line can be assessed by plotting the headcount as a function of the poverty line, that is, by drawing the cumulative density function o f per capita consumption relative to the poverty line. Figure 1.1 shows this distribution. The curve indicates on the y-axis what percentage o f the population has a per capita consumption level at or below the level represented on the x-axis. It has also an alternative interpretation. Suppose we increase on the horizontal axis the poverty line from zero to the maximum consumption per capita, and trace on the vertical axis the corresponding headcounts of poverty. The "poverty incidence curve" shows what incidence o f poverty would be associated with a given poverty line on the x-axis (Ravallion 199417. Cumulative density functions are especially useful in order to compare the sensitivity to changes in the poverty line of poverty headcounts of two different distribution, based for example on alternative definitions o f consumption. Similar calculations are possible for the poverty gap and the severity o f poverty. For example, the sensitivity o f the poverty gap measure to the poverty line can be examined by plottingthe areas underthe cumulative density functions, or the `poverty deficit' curve. 7 Figure 1.1: CumulativeDistributionofPer CapitaConsumption 1.00 1 .-*0Q E: `3 0.75 a -1I 0 -1 ru a 0 c 0 G 0.50 rt: 8 - I ~~ , 0 50 100 150 Monthly per capita household expenditure (US Dollars) Source: 2001 TLSS. 1.21 In the analysis of social welfare, we ask to what extent we can say that one distribution of resources i s better than another one.* Cumulative density functions play an important role in social welfare analysis, as it is directly linked to the notion o f first-order stochastic dominance. A cumulative density function 6 first-order stochastically dominates a cumulative density function F2 whenever, for all levels of per capita consumptionx, This condition says that distribution 2 always has more individuals in the lower part o f the distribution, associated with low levels o f per capita consumption. It is naturally linked to the measurement o f poverty with the headcount index, where we want to know whether poverty rankings depend on a specific level o f the poverty line. If the above condition holds, then poverty o f distribution 2 will always be at least as high as poverty of distribution 1, regardless o f the choice of the poverty line. Figure 1.2 shows one example, comparing the urban and rural cumulative density functions o f per capita expenditures. * More formally, the analysis o f social welfare specifies a social welfare function which aggregates the individual welfare levels o f the members o f the population in accordance with certain general principles: social welfare does not decrease ifthe living standards o f an individual rises (Pareto property); does depend on individual welfare levels and not on who has which welfare (anonymity property); and increases with an equalizing transfer from a rich to a poor individuals (transfer property). 8 Figure 1.2: RealPer CapitaHouseholdConsumption inUrban and RuralAreas 1.00 - .*c 0 i- 3 0.75 cd 0 4, a a 4- 0 .-c2 c 0 0.50 .t; l 0 - - I I I 0 25 50 75 100 Monthly per capitahouseholdexpenditure(US Dollars) Source: 2001 TLSS. 1.22 We can illustrate the clustering of households around the poverty line with another connected concept, the density function. It characterizes the distribution by focusing on the concentration o f the population at different points o f the consumption scale. It therefore captures the essential characteristics o f the distributional shape.g Distributional location and clumping can be easily examined using the density function by `bumps' ofconsumptionconcentration at different points alongthe consumption scale. The density function o f per capita expenditures is shown in Figure 1.3. The graph represents a kernel density estimate. Kernel density functions can be thought of as "smoothed" histograms. Histograms are constructed by dividing a range into a fixed number of intervals (`bins`) of equal width, where vertical bars are drawn at each interval with heights proportional to the relative frequencies (`density') o f observations within each bin. Kernel density function gets away from bins by estimating the density at every point rather thanjust for each interval. While ina finite sample there will only be a finite number o f observations, these estimates use mass within an interval or band o f `nearby' points to estimate the density at each point. This technique overcomes the inherent `lumpiness' at the edge o f bins o fhistograms. The area under the density function between two consumption levels is the proportion of the population with consumption within that range. The total area enclosed by the function equals 1 (100 percent of the population). 9 Figure1.3: Density FunctionofPer CapitaConsumption I 0 100 Monthly per capita householdexpenditure (US Dollars) Source: 2001 TLSS. Household Size and Composition 1.23 Inmost countries, an essential characteristic ofthe poor is their large family size. The more people depend on given household resources, the less there is available for each member." Poverty comparisons along the dimension o f household size are complicated by variations in household composition. For example, children will often require less than prime-aged adults to obtain the same level o f living, while old people may need more o f some things, like health services, but less or others, like work-related consumption. These compositional effects could be important enough that members o f large families may be better off than members o f smaller families with the same level o f resources per capita.l1Furthermore, apart from differences in needs, prime-aged adults have a larger earning power than either children or elderly. The inclusion o f a prime-aged adult to a household may increase per capita consumption o f all household members, while the addition o f a child or elderly i s likely to reduce it. loThe linkbetweenhousehold size andpoverty at the micro levelhas a counterpart at the macro level: what is the effect o f demographics on economic performance and poverty alleviation? While no dominant view has yet emergedon this issue, recent evidence shows that countries with higher rates o f population growth have tended to experience less economic growth. In particular, changes in age structures resulting from declining fertility create an one-time 'demographic gift', where the working age population has to support few dependents. This factor may well have contributed to the Asian economic miracle. See Birdsall, Kelley, Sinding (2001). However, it is difficult to establish the correlation empirically, as long as we do not know enough on how household resources are allocated among its members, and therefore how many resources are needed to attain equal living standards for different types of householdmembers. 10 1.24 Unfortunately, there are no generally accepted methods for calculating equivalence scales. Instead, the typically recommended method i s to explore the sensitivity to a range o f some reasonable, but essentially arbitrary, scales. The idea is to transform the number o f persons in a household into the number o f adult equivalents, allowing for relative cost differences and economies o f scale. We adopt this approach and define the number o f adult equivalents (AE) per householdby the formula AE =(A aK)* + where A is the number o f adults in the household, and K i s the number o f children. The parameter a is the cost o f a child relative to that o f an adult, and lies somewhere between 0 and 1. The other parameter, 8, also falls between0 and 1, and controls the extent o f the economies o f scale.'* The base case, used for most o f the analysis inthis report, i s both a and 8 equal to unity: child and adults are assumed to cost the same, and the resources needed to cover the expenses o f households o f different size vary simply in line with the number o f members. 1.25 What are plausible values that a and 8 could take? It i s generally assumed that in developing countries children are relatively cheaper than adults, with costs as low as one third o f an adult per child (Deaton and Muellbauer (1986) and Deaton (1997)). In our sensitivity analysis, we present results for a taking three values, namely 0.33, 0.66, and unity. With regardto economies o f scale, it is often argued that their extend depends on the shared goods within the households, or the household public goods. For example, it all goods are private in consumption, costs should rise in proportion to the number of people in the household. O n the other hand, if all goods are public, then costs are unaffected by the number o f people in the household. In developing countries, the most important good ina household's consumption i s food, which is a private good. The scope for economies o f scale i s therefore small, and 8 i s unlikely to be lower than 0.75. In Figure 1.4, we display the link between different definitions o f household expenditures, based on six combination o f a and 8, and the poverty headcount, relative to household size. l2As the elasticity o f adult equivalents with respect to `effective' size (A+aK) is 8, the measure of economies o f scale becomes (1-e). 11 Figure 1.4: Poverty and Household Size: Equivalence Scales 20 , O I 1 2 5 6 7 8 9 or more Household size a=0.33, 8=0.75 a =0.33,8 =1 ~ a =0.66, 8 =0.75 I a =1, 8 =0.75 a=l,8=1 ~ 1.26 Household size i s not a continuous variable, as by definition it takes only integer values. But how do we display the relationship, say, o f the poverty headcount, to a continuous variable, like the age dependency ratio? Figure 1.5 plots the poverty headcount relative to the age dependency ratio, defined as the number o f dependents (people younger than 15 and older than 64) to the working-age population (those o f ages 15-64). It is estimated using nonparametric regression analysis. The obvious way to calculate such a function would be to derive the average o f all poverty headcounts corresponding to each dependency ratio. However, as we have a finite sample o f households, yet the dependency ratio as a continuous variable (taking all values between 0 and its maximum o f 3.9, we face the same problem as in the estimation o f density functions: how to estimates the poverty rates at points where we have no observations? We adopt the same solution as with the kernel density estimates to average with a specific weighting scheme over the points `near' to a given level o f the dependency ratio (Deaton 1997). 12 Figure 1.5: Povertyand HouseholdComposition 60 ~ , I I 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 Dependencyratio Source: 2001 TLSS. INEQUALITY AND SOCIAL WELFARE 1.27 Inequality i s part o f the general analysis o f welfare. It i s concerned with the dispersion o f the distribution o f economic resources over the whole population, rather than only the individuals or households below a certain poverty line. The middle and top o f the distribution can drive measures o f inequalityjust as much as the bottom tail. Inthis sense, it is a broader concept than poverty. However, it also has a narrower focus than poverty, as it abstracts from the mean of the distribution and instead considers only the dispersion o f the distribution. Poverty and inequality measurement are sometimes combined to capture social welfare, which depends on both the distribution and the means. The main difference between such an approach and poverty measurement is that inthe first case, every individual is considered, even though more weight is assigned to the poor, while in the second case, the non-poor do not get any weight at all. In the following, we will present both summary measures and graphical tools to look at inequality inconsumptioninTimor-Leste. Inequality 1.28 The perhaps simplest way to display inequality i s to compare the resource shares o f individuals at the bottom and top o f the distribution. Just like for poverty measurement, researchers have developed a long array o f summary measures to capture inequality. They include the Gini coefficient, which takes on values between 0 and 1with zero indicating no inequality, and three members o f the Generalized Entropy (GE(a)) 13 class o f inequality measures, which range from 0 to 00, with zero representing no inequality.l3 1.29 H o w should we weigh people at different levels o f per capita expenditures? And to what extend do rankiiigs for specific inequality measures generalize to other inequality measures? The Lorenz curve is the standard graphical tool to address this issue. It captures all information about a distribution, with the exception o f the average level. The Lorenz curve plots the cumulative fraction o f population, starting from the poorest, on the x-axis against the cumulative fraction o f resources on the y-axis. Complete equality is represented by the 45 degree line, with everyone receiving the same. Perfect inequality, implying the richest person having all the resources, generates a Lorenz curve running along the x-axis that jumps to the 45 degree line at the right-outmost point. The further away the Lorenz curve from the 45 degree line, the more inequality there is. The importance o f the Lorenz curve for inequality analysis lies in this property: when two Lorenz curves do not cross, then the upper one represents an unambiguously more egalitarian di~tribution.'~The Lorenz curves o f per capita consumption for urban and rural areas are shown inFigure 1.6. Figure1.6: Lorenz Curvesfor Urban and RuralAreas 100-j i / , ru / E Eea 50; /'/ Urban ,," 6% ,/' /' 2 2 5 1 / E ' , Rural , , 0 1 - - I I 0 25 50 75 100 Cumulativeproportionofpopulation Source: 2001 TLSS. 13Any inequality measure that satisfies a set o f five desirable axioms (Transfer, Income Scale Independence, Population, Anonymity, Decomposability) is a member o f the GE class. The parameter ct in the GE class determines the weight given to distances o f expenditures at the tails o f the distribution. A value of 0 gives more weight to the lower tail, a value o f 1 applies equal weights across the distribution, and a value o f 2 gives more weight to gaps inthe upper tail. GE(0) is identical to the mean log deviation, GE(1)to the Theil Index, and GE(2) to ?4the squared coefficient of variation. l4This holds for any inequality measure that satisfies the principle of transfer. Inequality decreases as a result o f an equalizing transfer from a richer to a poorer individual. The Gini coefficient i s related to the Lorenz curve. It equals the area between the Lorenz curve and the 45 degree line relative to the entire area under the 45 degree line. 14 Social Welfare 1.30 Lorenz curves display the degree o f inequality, and are not affected by the mean of the distribution. This can be easily modified by multiplying the y-axis, the cumulative share o f the per capita consumption, by the mean per capita consumption. The generalized Lorenz curve ranks distributions in terms o f social welfare. If a generalized Lorenz curve A lies entirely above another generalized Lorenz curve B, then it implies that all lowest p percent o f the population have more resources intotal in distribution A than indistribution B. Inother words, each lowest quintile of the distribution has more per capita consumption in A than in B, so that each social welfare function that gives more weight to poorer than richer people will rank A over B. Generalized Lorenz dominance is equivalent to second-order stochastic dominance. Since first-order stochastic dominance implies second-order stochastic dominance, a ranking o f two distributions by cumulative density functions implies the same ranlcing by generalized Lorenz curves. These plots o f per capita consumption for urban and rural areas are shown inFigure 1.7 Figure 1.7: GeneralizedLorenzCurves for Urban and RuralAreas Urban Rural I I II 0 25 50 75 100 Cuinulative proportion o fpopulation Source: 2001 TLSS. MOBILITY 1.31 Another dimension o f welfare concern changes over time. For example, in Chapter 4, we will analyze "ladder questions", where persons are asked to rank themselves with regard to economic and power status, both for now and before the violence. The scale ranges from 1 to 9, with 9 indicating "richest" or "most powerful", respectively. H o w does past status relate to current status? For example, those at the bottom o f the economic or power scale today, were they also at the bottom in 1999? In 15 order to investigate this issue, we have to link information on the 1999 status with the current assessment. This can be most conveniently done with the help o f transition matrices. They capture "intra-distributional" dynamics, and reveal differences inupward and downward mobility. More formally, let p" +.."I 11 be a quadratic transition matrix o f order 5 capturing the change ineconomic status. States are ranked from worst to best, and the top four steps are collapsed into one state, reducing the number o f steps from nine to five. For example, element p"12 refers to the probability that an individual at the lowest step in 1999 progresses one step in 2001. Denote by F& and F&,! the 1999 and 2001 density functions o f economic status, represented as column vectors with five elements. Then we can derive the transition matrix P" such that Fioo,= F,;,, P" and each row adding up to unity (P E l=1). An example o f such a matrix, for the change in power status between 1999 and 2001, is shown inTable 1.2 Table 1.2: SubjectiveWell Being: Matrix ofPower Status 200I Lowest 2nd 3rd 4th 5th 26 (2.9) 27 (4.0) 33 (6.5) 59 (10.2) 78 (13.3) Note: Standard errors inparentheses. Source: 2001 TLSS. 1.32 H o w do we interpret these matrices o f 25 transition probabilities? Similar to poverty and inequality measurement, various approaches have been developedto analysis the structure o f mobility as captured by transition matrices (Bartholomew 1982). The simplest indicators are scalar summary measures o f overall mobility, like the immobility measure IM,, which gives the share o f individuals jumping less than k number o f 16 steps," or the jump measure J,s,which equals the absolute value o f the average state in 1999 versus 2001, to be calculated for each state s and the transition matrix as a whole. Such measures treat upward and downward changes symmetrically, even though higher steps are preferred to lower steps. By contrast, the upwardmobility measure Us gives the value o f the average step at 2001, again calculated for each row and the matrix as a whole. 1.33 A transition matrix i s "monotone" (Conlisk 1990) ifa higher step can be obtained from a lower step by moving probability mass to the right. In other words, it stochastically first-order dominates the distribution represented at a lower step. This property is important inthe current context, where states are ranked from worst to best: a monotone matrix preserves the disadvantage o f originating from a low state into the future. Finally, we can compare the two transition matrices in terms o f their upward- mobility assuming that the mobility structure as characterized by the matrix i s constant over time. By multiplying P with itself sufficiently often, we converge to a unique equilibrium probability vector 7c regardless o fthe initial distribution F,,,,16. SUMMARY 1.34 This chapter offered a primer on the design and data o f the TLSS, and basic concepts in poverty, inequality, social welfare, and mobility measurement. In the remainder o f this volume, these definition and techniques will be applied to characterize living standards in Timor-Leste. The following chapters are intended to provide comprehensive information, presented mostly intables rather than figures. Less technical summaries and main messages o fthe analysis inthis volume are provided inVolume I. '*k is no larger than the maximum number of states minus one. Ihf,is linkedto the Shorrocks axiom for transition matrices (Shorrocks 1978): shifting probability mass from the diagonals to off-diagonals increases mobility. More technically, P be the time-invariant mapping of a first-order Markov chain, such that F,; = F,;PE with t2>tI. Under certainregularity conditions,there exists anunique"ergodic" equilibrium distribution rc which is the unique solution to T`= z`P (Quah 1993, Quah 1994, and Kremer, Onatski, and Stock 2001). 17 APPENDIX:CONSTRUCTING THEPOVERTY MEASURE Introduction" 1. This appendix explains how we constructed the consumption measure. The first part lays out the treatment o f the four main types o f goods and services consumed (food, non- food, consumer durables, and housing). The next section presents the procedure for adjusting household consumption to cost o f living differences across time and space. The third part explores the pattern of consumption, in terms o f both regional variation and comparison to other countries. The final section elaborates the procedure to calculate the poverty line. Consumptionmeasure Food items 2. Conceptually, constructing a food consumption aggregate is a straightforward exercise. We need to aggregate the total value o f the food consumed during the recall period. Practical difficulties arise for three reasons. First, households receive food from different sources (purchases, home-production, gifts or remittances, in-kind payments), and all o f them should be included to obtain an aggregate welfare measure, even though they may well be recorded with different recall periods. In the TLSS, households were asked to record the consumption o f a list o f 129 food items and beverages, composed of fourteen food categories/subgroups (cereals, tubers, fish, meat, eggs and milk products, vegetables, legumednuts, fruit, oil and fat, beverages/drinks, spices and honey, miscellaneous foods, alcoholic drinks, tobacco & betel). The common recall period o f all items is the last 7 days. These items were deemed to be purchased relatively frequently so that this short recall period was adequate. The list" and recall period match those from the SUSENAS, the Indonesian household survey, in order to ensure comparability between the TLSS and the SUSENAS. For each item, households were asked separately about the consumption o f purchased, self-produced, and in-kind items to ensure all sources are included. Second, the non-purchased items need to be valued in monetary terms to include them in the welfare measure. This involves typically identifying reference prices at which food quantities can be valued. The TLSS recorded both quantities and Rupiah values for each food itemby source. It was therefore not necessary to refer to price information from other sections or alternative data sources to calculate food expenditures. Third, some less-perishable food items may be stored for a long time, so that food purchases may differ from food consumption. For most items in the TLSS food list, differences between purchases and consumption are likely to be unimportant. We also phrased the questions carefully to emphasize that only quantities and values o f food actually consumed, rather than the total amount and value purchased, should be recorded. 17This appendix was writtenby BenuBidani, MartinCuinpa, andKasparRichter. The list of food items was reviewed to ensure that it reflected the Timor-Leste conditions. A few changes to the list were made to include items that were eaten more commonly in Timor-Leste. Food nameswere also provided inTetunonthe questionnaire. 18 Non-food items 3. The TLSS collected information on consumption o f over 50 non-food categories, belonging to six subgroups (goods and services, including health and education expenditures; clothing, footwear and headgear; durable goods; taxes and insurance; festivities and ceremonies; and other expenses). In line with other household surveys (both the Indonesian SUSENAS and the Living Standards Measurement Surveys), the TLSS asked for information on expenditures only, as most non-food items are too heterogeneous to permit the collection of information on quantities consumed. It recorded expenditures during the past 30 days and during the past 12 months, whether purchased or received in-kind as aid or as payment for work. The computation o f the non-food aggregate involves a simple aggregation over the relevant items. The main difficulties related to which items to include, and which recall period to choose. The items in the non-food list also very closely follow the Indonesian SUSENAS non-food module. 4. Concerning the first issue, the basic principle i s that only those non-food items should be included into the consumption aggregate, which can be considered to add to the consumption o f the household. For example, expenditures on taxes and levies or interest on loans are deductions from income, and therefore not included. In any case, such expenditures are very small and infrequent. Only 14 households inthe data report paying taxes. The average monthly per capita expenditurei s only US$0.0079, which represents only 0.03 percent o f total monthly per capita expenditure. Less than 4 percent o f households pay interest on loans. 5. More complicated i s the issue o f lumpy or infrequent expenditures, such as marriages, dowries, births and deaths. Ideally, we would want to smooth these expenditures linked to rare events over several years but lack the information to do so. Including them would risk to potentially overestimating substantially the longer-term average o f consumption o f those households that happened to incur in such expenditures during the survey period. We therefore followed common practice and excluded such items. 6. By contrast, in line with most poverty assessments, we included expenditure on education and health, even though such items can be viewed as, inthe case o f health, as "regrettable necessities", and, inthe case o f education, as investments, and therefore not directly add to consumption. Yet, excluding them would imply that we make no distinction between two households, both o f whom are sick (or have children in school age), but only one pays for treatment (or sends their children to school). Furthermore, most poverty analysis includes these expenditures. Education and health expenses were recorded notjust inthe consumption section but also inthe education and health sections. Unsurprisingly, the latter sources result inhigher numbers due to more detailed questions. In education, expenditures are asked for each child. However, education and health amount on average to no more than 2 percent o f total expenditures even with the higher numbers. In order to have consistent recall periods, to ensure comparability with the SUSENAS, and to avoid double counting o f related expenditures like transport, we opted 19 to include the expenditure figures from the consumption section." This also ensures that we can construct in future rounds o f the survey a consistent consumption measure, even ifwe do notinclude separatehealthandeducation modules. 7. Another issue for non-food expenditures relates to the choice o f recall period. Non- food expenditures, including health and education, were recorded for both the last 30 days and the last year. We found that nominal per capita consumption expenses for non- food items for the shorter recall period were on average 40 percent higher than for the longer recall period. This evidence is in line with macroeconomic data, which shows that the economy improved substantially during the course o f the year preceding the survey. As we are interested in capturing as well as possible the longer-term well-being o f households at the time of the survey, we decided to stick with the shorter recall period. This brings the recall period for non-food expenditures also in line with the other components o f consumption, food and rent, which are measured (as discussed below for rent) with recall periods of the last 7 days and the last 30 days. ConsumerDurables 8. Finally, durable goods require special treatment as they last typically for several years, so that lumpy and infrequent expenditures on durable goods are not a good indicator o f the utility derived from these goods during the reference period. Instead of including purchases of durable goods, the standard procedure is to estimate the flow o f services accruing to the household from the total stock of durable goods it owns. However, since we only have information on the estimate for the current value o f a durable good, we would need to adopt more or less arbitrary assumptions on the rates o f depreciation and inflation o f a durable good to derive this value. This would add a noisy, and controversial, component to the measure o f longer-term well-being. Furthermore, only very few households report the ownership o f durable goods (see Table A.l.l). Overall, we decided to exclude durable goods from our measure o f consumption inview of their rare occurrence and measurement difficulties. Housing 9. Housing is often the most problematic area to include especially when rental markets are thin, as i s the case in many developing countries. The underlying principle for housing is the same as for other consumer durables. We would like to include in the consumption aggregate a measure o f the flow o f services received by the householdfrom occupying its dwelling. If all households rented their dwelling, and rental markets were well functioning, we could use the value o f rent paid. However, outside Dili, the incidence o f rent payments is very sparse, and even within Dili, only a fraction o f households report rent payments. Many households own the dwelling in which they reside, and others do not pay rent as such. Dili/Baucau the primary urban center reports the highest percentage o f renters (26%). Only 7 percent rent houses in other urban areas, l9We also calculated the consumption and poverty measures using the expenditures from the health and education modules. Total nominal per capita consumption is 1% higher compared to the corresponding measure with health and education from the consumption module. 20 3% in the rural highlands, and 6% in the rural lowlands. Nationally, only 8% o f households rent their houses. 10. While rent payments are reported only for some households, the questionnaire also asked households for estimates o f how much their dwelling could be rented out for. This implicit rental value can in principle be used in the consumption aggregate whenever actual rents are not reported. Implicit rents are a hypothetical concept, and the estimates may not always be credible or usable. We inspected the numbers carefully and identified only a few outliers. In addition, we did a simple cross-check on the validity o f the imputed rent estimates. For those households reporting actual rent payments, we run a typical hedonic housing regression which includes the rental value for households as the dependent variable and characteristics o f the house (such as the construction material o f the house, number o f rooms etc), and used this model to predict rent payments for the other households that did not report rent. We found that predicted rent payments and imputed rent estimates matched each other fairly closely nationally, though there are some differences by different regions. Predicted rents inDili are significantly lower than those reported by households, but imputed values in the rural lowlands and other urban centers are close. For the consumption aggregate, we therefore used actual rents if available, and otherwise imputed rents as estimates for the flow o f services from housing. We plan to do some sensitivity analysis o f our results to different choices related to the housing variable. 21 Table A.l.l: Ownership of Consumer Durables (% population) % owning Stoves 7.9 Refrigerators 2.9 Washing machines 0.3 Sewingiknittingmachines 2.9 Clothes cupboard 32.5 Buffet 16.5 Fans 4.1 Televisions 6.9 Video players 2.9 TapeiCD players 6.8 Cameras, video cameras 1.o PCS 0.1 Radios 32.4 Bicycles 5.2 Motorcyclesiscooters 3.O Cars/trucks 1.8 Motor boat 0.0 Boat without motor 0.5 Generators 0.5 Water dispenser 0.4 Electric rice cookers 2.2 Mosquito nets 45.0 Source: 2001 TLSS. Cost of Living Differences 11. The discussion in the previous section concentrated on the construction o f a consumption aggregate. Before this measure could be used to compare standards o f living o f individuals residing in different parts o f the country, we have to adjust for differences in cost of living. In particular, prices of goods and services vary considerably across different regions and this spatial variation in prices should be taken into account when comparing welfare levels across different parts o f the country. In Timor Leste transportation is difficult and expensive; and local markets are not well connected, giving rise to possibly large variations in the cost o f living. Inthis section we explain how we adjust for differences in the cost o f living due to either temporal or spatial price differences. 12. Adjusting for temporal price differences is in principle straightforward. The survey was implemented over a period o f three and a half months, and we have to account for the changes in the price level over this time span. Households interviewed at the beginning o f the survey period faced a different price vector than households at the end 22 o f the period. This adjustment is especially important in situations o f high inflations or deflations. We only have information on monthly changes in the Consumer Price Index (CPI) for Dili, not for the country as a whole, for which the CPI i s released only quarterly. The price changes were relatively minor: the CPI increased by about 0.5 percent between the beginning o f September to the end o f November. Assuming the time trend in the Dili CPI was representative for other regions o f the country, we deflated consumptionto prices as o f the beginning o f September 2001. 13. In a cross-sectional survey, most price variation i s due to spatial differences. Before we turnto the calculation of the spatial price index, we should clarify our data source for regional price information. The TLSS collected price information in the consumption section and ina separate suco-level price survey. We decided to construct the price index using the implicit price information from the consumption section, obtained by dividing expenditures by quantities. This has a number of advantages over price information from local markets. First, it i s likely to reflect more accurately the prices faced by households. Local consumers may pay different prices than survey enumerators - for example through haggling or because o f their long-term relationship with the vendor. Second, prices quoted at the local inarket within a suco may not be the relevant ones for a household located in this suco, as the household may be closer located to a different market that lies outside its suco. The disadvantage o f using the price information from the consumption section i s that dividing values by quantities gives unit values rather than prices. Better-off households typically purchase higher quality even o f relatively homogenous goods like rice, so that the higher price they face i s at least partially a reflection o f the better quality. We followed the recommended method to deal with this unit value problem by replacing household specific prices with the median of the unit price within each region (Deaton and Zaidi, 1998). 14. The literature proposes two main competing methods to calculate price indexes to deflate nominal consumption. They differ inthe choice o f weights. Spatial price indexes compare price vectors at different locations by means o f a set o f quantities or weights. The Paasche Index uses for each household a different set o f weights, namely the purchases o f the household, while the Laspeyres Index uses a fixed set o f reference weights for all households. Inprinciple, the Paascheand Laspeyres indexes give different results in the presence o f either variations in regional price differences or differing expenditure patterns o f households. Nevertheless, in view o f other conceptual and practical problems in the poverty analysis, like accounting for housing in the consumption aggregate or allowing for differences in household composition, the choice o f the deflation techniques is unlikely to be o f paramount importance. We follow standard practice adopted inpoverty analysis in several countries inthe East Asia region and use a Laspeyres Index that uses a fixed consumption bundle.We do however test the sensitivity of our poverty estimates to the choice o f this index and find that the results are remarkably robust.20 *'Dividingnominal consumption by a Paasche Index leads to "money metric utilities", and by a Laspeyres Index gives rise to "welfare ratios". Both concepts have theoretical flaws. Money metric utility violates the transfer principle: an equalizing transfer from a rich to a poor household may widen their gap in money metric utilities, as money metric utility is in general not a concave function o f expenditures. The welfare 23 15. As explained, the Laspeyres Index involves comparing the prices a household living in a particular region faces with a set of reference prices, using a fixed consumption bundle. In terms o f picking regions, we pick regions where prices are relatively homogeneous and people face reflect similar cost-of-living indices; and regions that are disaggregated enough to capture price variations across the country. While a very disaggregated grouping i s desirable, the geographic regions have to be large enough to allow us to get reasonable estimates o f prices. Based on these considerations, we pick five regions: DilUBaucau, other urban areas, and rural areas divided into three groups: the rural central, the rural east and the rural west regions. For the fixed consumptionbundle, we pick the reference basket o f those at the lower end o f the consumption distribution - to capture the tastes o f the poor, not the well-off. Based on these considerations, we pick the group in the 2*ldto 5t11decile based on nominal consumption for Timor Leste as a whole as the reference group. We take the expenditure pattern o f this group and take the average quantities consumed by this group as the fixed consumption bundle. The Laspeyres price index for each region i s computed by comparing the cost o f buying the reference bundle in that region compared to a reference region. The choice o f the reference price vector i s a matter o f convenience. We followed common practice and chose the national median o f the prices observed. The use of medians rather than means limits the sensitivity to outliers. Basing the reference price vector on a national price vector brings our consumption measure closely in line with national income accounting practice, and eliminates results that depend on specific relative price patterns that occur only in some areas. The Laspeyres price index, therefore compares prices in the five regions as discussed above, to the national average. 16. Constructing Laspeyres food price indexes is readily done, as in principle we have price information on each food item for each region. Apart from food, the other major item in the consumption basket i s housing. Since rents, or imputed rents, are highly location specific, it i s important to account for differences in the cost o f living deriving from housing. Inparticular, the same apartment or house i s likely to be more expensive inDili than ina remote rural area. Ignoring such differences would risk overestimating the living standards in urban relative to rural areas. Deriving price indexes is more involved for housing than for food. Inprinciple, we need to identify a reference "housing bundle", and then determine the average price o f this reference bundle for each region (Lanjouw et. a1 1996). However, in practice, defining a reference bundle for housing is more difficult than in the case o f food. In contrast to food items, housing is a heterogeneous bundle o f goods and services comprising different attributes (number and size o f rooms, quality of construction material, accessibility o f services, location, etc.). Inorder to derive a price index for housing usingthe same methodology as for food, we would need to identify housing units across ineach that were exactly alike interms o f all conceivable attributes, and then compare average rental values across regions to derive the housing price index. This would clearly be impossible to implement in practice. Instead, we estimated a hedonic housing regression model using actual rental values for those households in the sample that reported rents and the rents imputed by households that lived in owner-occupied or free housing as the dependent variable. The set o f ratio violates the Pareto principle: it is possible for a policy to make a householdbetter off yet its welfare ratio to decline (Blackorby and Donaldson 1987, Deatonand Zaidi 1998). 24 explanatory variables included a wide range o f housing characteristics, measures o f quality o f housing, regional dummy variables and other factors that helped determine the rental value of dwellings. We then used the parameter estimates o f this model to get a measure o f the "price" o f housing in each region. The model was used to estimate the cost o f renting a typical house, based mostly on mode housing characteristics for the reference group, setting all variables other than the regional dummies to zero.21 The housing price index was then derived by taking the ratio o f the rents in each region to the national mean. 17. The Laspeyres price indexes for food and housing constructed from the TLSS data are presented inTable A.1.2. The TLSS did not collect price data for non-food items, so we could not use the data to construct price indices. As food and housing for the reference group (2nd to decile o f national consumption expenditure) account on average for about 87 percent o f total consumption, we simply ignored the price differences arising from spending on non-food items. To compute the aggregate index, we used fixed weights o f housing and food for the reference group. The fixed weights are 89.8 percent for food and 10.2 for housing. This is like assuming that this expenditure- weighted average o f the Laspeyres food and housing indexes reflects adequately the cost differences for non-food items. 18. Table A.1.2 shows the price indices by region (DiWBaucau, other urban areas, rural east, rural central and rural west.). The food price index shows significant price differences inDiWBaucaurelative to the rest o f the country. Dili/Baucau face prices that are fourteen percent higher than the national average, and the prices other urban areas, the rural east and the rural central regions are slightly lower than the national average, while prices inthe rural west are about 4 percent lower than the national average. 19. Including housing prices alters the picture significantly. The Dili/Baucau housing price index is 70 percent higher than the national average. Prices inthe rural west are 27 percent higher than the national average, while prices in other urban areas are at the national average. The rural east has the lowest housing price index, 40 percent below the national average. 20. Combining both the food and the housing price indices shows that the cost-of-living inDili/Baucau are 20 percent higher than the national average, while prices inthe rest of the country are between 1-5 percent lower than the national average. 21The "reference" house has three rooms, is 36 square meters large, was built in 1997, has bamboo walls, metal sheetslzinc roof, earthlclay floor, no toilet, uses a spring as the main source for bathing and washing, and has a lamp as the main source o f light. 25 Table A.1.2: Regional Laspeyres Price Indices Index Realper capita Food Housing Overall expenditure 1/ (US Dollars per month) Urban 1.071 1.371 1.102 33.78 Dili/Baucau 1.141 1.672 1.196 40.15 Otherurban 0.984 0.994 0.985 25.87 Rural 0.978 0.886 0.969 21.22 Ruralhighland 0.980 0.896 0.972 20.75 Rurallowland 0.976 0.877 0.966 21.58 Ruralcenter 0.988 0.845 0.974 20.54 Ruraleast 0.980 0.629 0.944 24.57 Ruralwest 0.954 1.266 0.986 18.86 Total 1.000 1.ooo 1.ooo 24.17 1/Based on a last month recall period. Note: All Rupiah valuesfrom the survey were converted to USDollars usingan exchange rate ojl0,OOO RupiaWUSDollar. Source: 2001 TLSS. 26 Patternof ConsumptionExpenditure 21. The next two tables display per capita expenditures. Table A.1.3 shows the expenditures deflated by the food price index only and Table A.1.4 shows the expenditures deflated by the food and housing price indices. The results ineach table are presented using the past 30 days expenditure to calculate non-food expenditures. Expenditure calculated based on the past 12 months i s about 4 percent lower than the expenditure based on the last month. Table A.1.3: Monthly Per Capita Expendituresdeflated only by a food price index (US Dollars) Total Urban Rural Dilii Other Rural Rural Nominal Real Baucau Urban Highland Lowland Center East West Rent 5.71 5.46 12.37 3.34 17.46 6.04 2.87 3.71 3.02 3.40 3.98 Utilities 1.31 1.28 1.97 1.06 2.67 1.10 1.03 1.09 0.90 1.50 0.92 Food 15.16 I5 I O 16.06 14.81 17.21 14.64 15.19 14.52 15.10 15.42 13.50 Purchases 8.5 I 8 36 12.25 7 17 15.24 8.53 7.46 6.94 7.06 7.36 7.18 Home production 5.72 5.80 3.19 G G I 1.44 5.35 6.69 6.54 6.83 7.03 5.63 In-kind 0.94 0.94 0.63 1 04 0.53 0.75 1.04 1.03 1.21 1.03 0.69 Non-food 2.06 2 02 3.94 1.43 4.00 3.86 1.20 1.60 0.92 2.87 0.86 Clothing 0.94 0.93 1.18 0.85 1.50 0.80 0.70 0.96 0.44 1.97 0.44 Others I/ 0.92 0 89 2.22 0 48 2.08 2.40 0.40 0.54 0.46 0.64 0.35 Minor durable goods 0.20 0 20 0.53 0.10 0.42 0.67 0.10 0.10 0.02 0.26 0.07 Ediicatioii 0.18 0 17 0.30 0 13 0.45 0.11 0.10 0.16 0.10 0.20 0.14 Health 0.21 0 20 0.22 0.20 0 27 0.15 0.14 0.25 0.21 0.28 0.09 Total 24.63 24.23 34.85 20 98 42.06 25.90 20.53 21.32 20.24 23.67 19.49 Shares (%) Relit 23 23 35 16 42 23 14 17 15 14 20 Utilities 5 5 6 5 6 4 5 5 4 6 5 Food 62 62 46 71 41 57 74 68 75 65 69 Purchases 35 34 35 34 36 33 36 33 35 31 37 Home production 23 24 9 3 1 3 21 33 31 34 30 29 In-kind 4 4 2 5 1 3 5 5 6 4 4 Non-food 8 8 11 7 10 15 6 8 5 12 4 C1othing 4 4 3 4 4 3 3 5 2 8 2 Others 11 4 4 6 7 5 9 2 3 2 3 2 Minor durable goods I 1 2 0 1 3 0 0 0 I 0 Education I I 1 I I 0 0 1 0 1 1 Health I I 1 I 1 1 1 1 1 1 0 Total 100 100 100 100 100 100 100 100 100 100 100 27 Table A.1.4: MonthlyPer Capita Expendituresdeflatedby a food and housing priceindex (US Dollars) Total Uiban Ruial DilU Other Rural Rural Noiniual Real Baucau Urban Higldand Lowland Ceuter East West Rent 5 71 5.38 11.93 3 37 16.67 6.03 2.89 3.74 3.06 3.53 3.85 Utilities 1.31 I 2 7 1.90 1.08 2.55 1.10 1.05 1.10 0.91 1.56 0.89 Food 15.16 15 13 15.62 14.98 16.43 14.62 15.35 14.70 15.32 16.01 13.06 Purchases 8.51 8 32 11.86 7.24 14.54 8.52 7.53 7.02 7.16 7.64 6.94 Home production 5.72 5 86 3.15 6 69 1.38 5.35 6.76 6.63 6.94 7.30 5.45 In-kind 0.94 0 95 0.61 I 0 5 0.50 0.75 1.05 1.05 1.22 1.07 0.67 h'on-food 2.06 2.01 3.84 I 4 6 3.82 3.86 1.22 I.63 0.94 2.98 0.84 Clothing 0.94 0 93 1.15 0.87 1.43 0.80 0.71 0.99 0.44 2.04 0.43 Others I/ 0.92 0.88 2.17 0.49 I.99 2.39 0.41 0.55 0.47 0.67 0.34 Minor durable goods 0.20 0.20 0.52 0.10 0.40 0.67 0.10 0.10 0.02 0.27 0.07 Education 0.18 0.17 0.29 0.14 0.43 0.11 0.10 0.16 0.10 0.21 0.14 Health 0.21 0.21 0.21 0.20 0.26 0.15 0.14 0.25 0.21 0.29 0.09 Total 24.63 24.17 33.78 21.22 40.15 25.87 20.75 21.58 20.54 24.57 18.86 Shares (X,) Rent 23 22 35 I6 42 23 14 17 15 14 20 Utilities 5 5 6 5 6 4 5 5 4 6 5 Food 62 63 46 71 41 57 74 68 75 65 69 Purchases 35 34 35 34 36 33 36 33 35 31 37 Homeproduction 23 24 9 3 2 3 21 33 31 34 30 29 1n-kind 4 4 2 5 1 3 5 5 6 4 4 Non-food 8 8 I 1 7 10 15 6 8 5 12 4 Clothing 4 4 3 4 4 3 3 5 2 8 2 Others I/ 4 4 6 2 5 9 2 3 2 3 2 Minor durable goods I I 2 0 I 3 0 0 0 I 0 Education I l 1 I I 0 0 1 0 1 1 Health 1 1 1 I 1 I 1 1 1 1 0 Total 100 I00 100 100 100 100 100 100 100 100 100 22. Real expenditures are highest inDiWBaucau, followed by other urban areas, the rural east, the rural central and finally the rural west. This pattern i s stable regardless o f the price index used. However the urban-rural differences are higher ifwe deflate only by the food price index. Figure 1 shows the real per capita monthly expenditures by different geographic domains. The largest difference betweenthe two deflators i s in DiWBaucau, with real expenditures being significantly higher when only deflating by the food price index. 23. The expenditure pattern, in terms o f the shares o f expenditure spent on different categories, is also shown in these tables. The results are broadly consistent across the price indices. So for our discussion, we focus on the results using the 30 day expenditure that are deflated by the food and housing index (Table A.1.4, bottom panel). On average, the share o f food intotal expenditure inTimor Leste i s 63 percent, o f which 34 percent is from purchases, 24 percent from home-production and 4 percent fiom gifts/aid or payment in kind. Food shares are highest in the rural center (75 percent) and lowest in DiWBaucau at 41 percent. In contrast to the rest o f the country, in DiWBaucau most of the food consumption comes from purchases. O n average 22 percent o f all expenditures go towards housing rent, whereas another 5 percent i s spent on fuel and other housing utilities. Housing expenditures vary significantly: in Dili/Baucau, rents account for almost half o f all expenditures, whereas inthe rural east they represent only 14 percent. 28 Education and health account for less than 2 percent o f total expenditures. Other non- food goods total the other 8 percent. 24. Table A.1.5 compares the expenditure pattern of Timor Leste to a sample of other developing countries (taken from Deaton and Zaidi 1998). While the aggregates are not necessarily comparable due to differences in survey design, it gives an indication o f the relative importance o f the components. Timor Leste stands out with a high food share, in line with Engel's law, predicting a negative correlation o f food shares and the level of income in the country. Non-food spending share i s the lowest in all countries and spending on consumer durables was not included in the aggregate - a practice also not adopted in South Africa and Brazil. Perhaps surprisingly, the housing share is relatively high. This could be one example o f the transitional rise in the relative price of non- tradable to tradable goods and services as a consequence o f the influx o f some 15,000 expatriates. Table A.1.5: Main components of the aggregateconsumption Tiinor Vietnam Nepal Ghana Kyrgyz Ecuador South Panama Brazil Lese 1992-93 1996 1988-89 1996 1994-95 Africa 1991 1996-97 2001 I993 Food 63 51 64 65 45 50 30 46 28 Purchases a/ 34 34 29 44 33 44 28 40 21 Home production b/ 28 11 35 21 11 5 2 6 7 Non-food items: I O 29 19 28 23 29 4s 46 32 Education 1 3 3 n.a. 2 8 3 8 6 Health 1 6 3 n.a. 1 2 1 5 Other non-foods 8 21 13 n.a. 19 21 40 31 21 Consuiner durables 13 1 2 4 5 5 Housing 27 8 I S 3 30 I 6 25 3 40 Rent 22 6 13 2 18 12 16 2 31 Utilities 5 2 3 1 12 4 9 1 9 Overall IO0 100 IO0 100 100 100 100 100 100 GNP per capita($) c/ 470 170 210 390 550 1,280 2,980 3,080 4,400 u/ Includes n?eo/.staken aiwy from the hoinc. h/ Includes also,fuodreceiiwlfrom other hoii.se/io/dmemher.s,,friend$,ond in the.forin of in-kindpoyment,s. c/GNF' per copitu is token from internulionul .s/u!i,s/ic.s,fbrthe some pur of the surwy, excepr,fhr Ponomu where the lote.s/ uvuiluhle estimoteis ,fir1996. The,figure,fur Timor-Lesler:fer.s IO /hepredicted GDPper ~upilataken,fromthe World Bunk (huntry Economic Ademorondum. Poverty line 25. Following common practice in East Asia, we defined a poverty line that i s both "absolute" and "objective". A poverty line i s absolute ifit fixes a given standard o f living over time and space, or, inthe terminology o f economists, a given level of utility. Such a poverty line guarantees that two individuals with the same standard o f living are always 29 treated in the same way.22 Furthermore, a poverty line i s objective if the standard o f living is anchored inthe attainment of certain basic capabilities, rather than inindividual perceptions o f welfare, as in subjective methods. In particular, we correlate directly the standard o f living with the capability to meet the nutritional requirement for maintaining a certain activity level. The poverty line i s then set so as to meet the cost o f these requirements. 26. The leading method to implement nutrition-based poverty lines is the Cost-of-Basic- Needs (CBN) approach. It sets a consuinption bundle deemed to be adequate for basic consumption needs, and then estimates the costs to obtain such bundle for the relevant population subgroups. A person i s considered to be poor if it cannot meet the cost o f the consumption bundle. Two points are important to bear in mind. First, a person's poverty status i s linked not to whether the actual consumption meets the stipulated needs, but rather to whether the person would have the means to do so. In other words, while nutritional requirements are used to set the reference standard o f living, nutritional status i s not itself the welfare indicator. Second, there are many ways to determine the consumption bundle that provides for the basic needs. Current practice favors to set this bundle with reference to actual consumer behavior. The poverty line i s composed o f two elements, the food- andthe non-food components. Food component 27. First, we need to set the stipulated food-energy requirement. We followed common practice in East Asia and used as basic nutritional requirement 2100 calories per person per day. We defined the food bundle that yields this level o f nutrition by looking at the prevailing consumption patterns. There are a number o f ways to calculate such a bundle. Inparticular, we took the average food bundle consumed by the lowest second to fifth decile o f the population as ranked interms of real consumptionper capita. This reference group i s our first guess for the poverty head-count. Then we used caloric conversion on factors to convert the food bundle into total calories. We identified the caloric content o f the over 100 food items represented inthe food basket o f the reference group, drawing on two sources. Whenever possible, we took caloric conversion factors from Pradhan et a1 (2000), used for the poverty line calculations with Indonesian Susenas data. In case a closely matching food item was missing in Pradhan et a1 (2000), we referred to the detailed nutritional database from the U S Ministry o f Agriculture, which is posted on the web.23 Following standard convention, we excluded alcoholic drinks, tobacco and betel, and residual sub-categories "other". We were left with 102 out o f 129 food items, from which we identified the caloric nutrients o f 93 items. Overall, this covered 99.9 percent o f the food expenditure basket o f our reference group, as shown in Table A.1.6. This table also provides the budget shares o f the main food items and the caloric conversion factors. Finally, we calculate the nutritional content o f the food basket and scaled it proportionately to ensure it provides the required 2100 calories per person. 22 More formally, it guarantees that a Pareto improvement in terms o f welfare, whereby at least one person is better off, and no one else is worse off, cannot increase measured poverty (Ravallion 1998). 23 The website for the nutrient database of the U S Department of Agriculture is located at http://www.nal.usda.govifhicicei-binlnut search.pl. 30 Table A.1.6: Food bundle Code Item U S dollars YO per capita per month 1000 Cereals 2.97 29.6 1010 Tubers 0.98 9.8 1020 Fish 0.23 2.3 1030 Meat 0.6 1 6.1 1040 Eggs and milk product 0.20 1.9 1050 Vegetables 1.51 15.1 1080 Legumeshuts 0.33 3.3 1090 Fruit 0.42 4.2 1110 Oil and fat 0.36 3.6 1120 Beveragesldrinks 0.79 7.9 1130 Ingredients 0.19 1.9 1140 Miscellaneousfood 0.38 3.8 1150 Alcoholic drinks 0.27 2.7 1160 Tobacco and betel 0.78 7.8 Total 10.03 100.0 Note: All Rupiah valuesfioni the survey were converted to USDollars using an exchange rate of 10,000 Rupiah/US Dollar. Source: 2001 TLSS. Non-food component 28. The most controversial part o f setting a poverty line concerns the non-food component. The rationale for allowing a non-food component i s closely tied to the normative judgment involved in choosing the food component. Setting the food-energy needs requires determining an activity level. Yet, maintaining a certain activity level involves participating in society, and therefore, according to prevalent social norms, a minimum level o f spending on clothing, shelter and health care. In order to allow for basic-needs non-food expenditures, common practice is to divide the food component of the poverty line by some estimate o f the budget share devoted to food. H o w do we fix the food share? Standard practice looks at the share o f non-food expenditures o f a person, whose total expenditure i s just enough to reach the food poverty line. This can be interpreted as the minimum necessary allowance for non-food spending, since the person has substituted this spending for basic food needs.24This estimate i s referred to as the 24Under certain assumptions, this method identifies the lower bound of the poverty line. The corresponding upper bound is defined by the food share of households whose actual food spending equals the food poverty line. Once "survival" food needs are satisfied, basic non-food needs will have to be satisfied before basic food needs as total expenditure rises. And food and non-food are "normal" goods, so that their demand increases with total expenditures. They ensure that a person whose food expenditures match the food poverty line has already covered at leastthe basic non-food needs. 31 "lower poverty line". A higher allowance for non-food expenditures looks at those households in which individual food expenditures actually equal the food poverty line. The non-food spendingof these households is added as the allowance for non-foods. The more generous allowance for non-food expenditures gives us the "higher poverty line". 29. We calculated the non-food shares for boththe lower and higher poverty lines with a simple non-parametric technique (triangular kernel density estimation), as suggested in Ravallion (1998)25.First, we considered those households whose overall consumption lie within plus and minus one percent around the food poverty line, and derived their mean non-food expenditure. We then repeated this calculation another nine times, each time increasing the interval on each side by one percent o f the food poverty line. Finally, we took the average o f all the mean non-food share o f expenditures26, which provided us with our estimate for the lion-food components o f the poverty line. Using the food and housing index to get real expenditures, the calculations give us a lower national monthly per capita poverty line o f US$14.45, with a food component o f 75 percent (US$10.89) and a higher national poverty line o f US$15.85. The poverty lines are shown in Table A.1.7. Table A.1.7: Poverty Lines (US Dollars per capita per month) Price index derived from: Foodandhousing Food Upper poverty line Food 10.81 10.73 Non-food 4.63 4.36 Total 15.44 15.10 Lower poverty line Food 10.81 10.73 Non-food 3.60 3.42 Total 14.41 14.15 Source: 2001 TLSS. Poverty Estimates 30. We have finally all components together to calculate the poverty estimates. Tables A.1.8 and A.1.9 present the poverty rates for Timor Leste based on the poverty line developed inthe preceding sections. The poverty measures are presented at the lower and upper poverty lines, using both the food price deflator and the food and housing index 25Alternatively, the food share can be estimated parametrically with an Engel curve. The non-parametric a proach i s both simpler and requires no assumptions onthe functional form of the Engel curve. "This method gives highest weight on the households within the narrowest interval, and lowest weight to households within the widest interval. The weights are declining linearly aroundthe food poverty line. 32 deflator. The ranltings o f the regions are robust to the choice o f the deflator, though the urban-rural differentials in the poverty rates are narrower when we use the food and housing price deflator, which allows for the higher housing prices inurban areas. For the discussion that follows, we only discuss the results using the food and housing price deflator. Table A.1.8: Poverty rates by region, deflated only by a food price index Headcount Poverty Gap Severity % S.E. YO S.E. YO S.E. Upper poverty line Urban 21.7 2.9 5.9 1.1 2.3 0.5 DiWBaucau 9.8 2.4 3.2 0.9 1.3 0.4 Other urban 36.7 5.9 9.3 2.1 3.4 1.o Rural 42.6 3.7 13.0 1.7 5.4 0.9 Highland 43.8 6.2 14.4 3.0 6.3 1.6 Lowland 41.7 4.5 12.0 1.8 4.8 0.9 Center 48.3 5.9 15.6 2.8 6.8 1.5 East 32.5 4.3 9.7 1.7 3.9 0.8 West 42.0 7.5 11.4 3.0 4.4 1.4 Total 37.7 2.9 11.4 1.3 4.7 0.7 Lower poverty line Urban 19.1 2.9 4.9 1.o 1.8 0.4 Dili/Baucau 9.5 2.4 2.8 0.9 1.1 0.4 Other urban 31.2 5.8 7.7 1.9 2.7 0.8 Rural 38.0 3.7 11.2 1.5 4.5 0.8 Highland 40.2 6.4 12.5 2.8 5.3 1.4 Lowland 36.3 4.2 10.2 1.7 3.9 0.7 Center 43.4 5.9 13.5 2.6 5.7 P .3 East 28.6 4.4 8.3 1.6 3.2 0.7 West 37.2 6.8 9.5 2.7 3.5 1.2 Total 33.6 2.9 9.7 1.2 3.9 0.6 Note: Consumption was deflated over time and also spatially using a Laspeyresprice index including onlyfood conzponents. Thestandard errors take into account survey design effects. Source: 2001 TLSS. 33 Table A.1.9: Poverty rates by region, deflated by a food and housing price index Headcount Poverty Gap Severity YO S.E. YO S.E. YO S.E. Upper poverty line Urban 24.8 3.1 6.5 1.1 2.6 0.5 Dilih3aucau 13.9 2.8 3.8 1.o 1.6 0.5 Other urban 38.4 6.1 10.0 2.2 3.7 1.o Rural 44.3 3.6 13.5 I.7 5.7 0.9 Highland 45.4 6.2 14.9 3.0 6.5 1.6 Lowland 43.4 4.2 12.4 1.8 5.0 0.9 Center 49.3 5.7 15.8 2.8 6.9 1.5 East 32.0 4.3 9.4 1.7 3.8 0.8 West 47.5 7.1 13.2 3.2 5.2 1.6 Total 39.7 2.9 11.9 1.3 4.9 0.7 Lower poverty line Urban 19.9 2.9 5.4 1.o 2.1 0.5 Di1i/Baucau 9.8 2.4 3.2 0.9 1.3 0.4 Other urban 32.7 5.7 8.1 2.0 2.9 0.9 Rural 38.7 3.7 11.5 1.6 4.7 0.8 Highland 40.8 6.4 12.8 2.8 5.4 I.4 Lowland 37.1 4.4 10.5 1.7 4.1 0.8 Center 43.6 6.0 13.6 2.6 5.7 1.3 East 27.4 4.3 7.9 1.5 3.0 0.7 West 41.1 7.3 11.0 2.9 4.2 1.4 Total 34.3 2.9 10.1 1.2 4.0 0.6 Note: Consumption was deflated over time and also spatially using a Laspeyresprice index includingfood and housing coinponents. Thestandard errors take into accozint survey design effects. Source: 2001 TLSS. 31. The incidence o f poverty inthe country as a whole i s 40 percent at the higher poverty line when using expenditures deflated by the housing and food index, amounting to 340,000 individuals. Poverty inurban areas i s lower (26 percent) than inrural areas (46 percent). It is lowest in Dili/Baucau (14.4 percent), and highest in rural center and the rural west (51 percent). Since over three quarters o f the population (76.5 percent) resides inrural areas, it is clear that poverty is overwhelmingly a rural phenomenon: 85 percent o f the poor live in rural areas. This conclusion is not overturned when one turns to the other two poverty measures: rural poverty is both deeper and more severe than urban poverty. 34 32. Since the poverty rates are based on sampled data, it i s important to take into account the standard errors o f the estimates. In calculating the standard errors we took into account the sampling structure o f the TLSS survey. As this survey involved both stratification as well as clustering these two features o f the survey were incorporated into the standard error formulae (Howes and Lanjouw 1995). The standard errors indicate that the poverty rankings between urban and rural areas can be inferred with great confidence, and that DiliiBaucau is the richest region. The geographical rankings between the other urban areas, rural highland and the rural lowlands cannot be inferred with great confidence. 33. At the lower poverty line, the poverty rate for Timor Leste as a whole is 34 percent o f the population or around 280,000 people. Poverty in urban areas is 20 percent and it is 38.5 percent in rural areas. The rankings o f the different regions remain broadly unchanged at the lower poverty line. 35 Table A.l.lO: Food bundle Code Item Unit Calories Per capitaexpenditure Per US Dollars Y O unit per month TOTAL FOODEXPENDITURE 10.03 100.00 1002 Unhuskedrice kg 3,614 0.04 0.41 1003 Imported rice kg 3,614 118 11.71 1004 Corn kg 3,200 1.03 10.25 1005 Wheat flour kg 3.330 0.00 0.03 1006 Corn flour kg 3,200 0.01 0.06 1012 Sweet potatoes kg 1,252 0.27 2.71 1013 Sago (ambon sago) kg n.a. 0.02 0.20 1014 Taro kg 1,120 0.23 2.25 1015 Potatoes kg 270 0.03 0.33 1022 V. small sea fish (sardines,teri, etc) kg 824 0.07 0.72 1023 Other fresih fish kg 824 0.07 0.14 1024 Salted fish kg 824 0.01 0.15 1025 Canned fish 100 gms 82 0.02 0.18 1026 Squid kg 920 0.01 0.11 1027 Freshshrimp kg 1,060 0.02 0.22 1028 Dried shrimp 100gins 106 0.00 1031 Beef kg 2,070 0.25 2.52 1032 Buffalo ineat kg 990 0.04 0.42 1033 Goat kg 1,090 0.02 0.16 1034 Pork kg 4,165 0.13 1.32 1035 Chicken kg 3,020 0 11 1.14 1036 Cannedineat kg 2,070 0.00 1037 Meat scraps and bones kg n.a 0.00 0.03 1041 Chicken eggs each 66 0.09 0.91 1042 Other eggs each 66 0.00 0.01 1043 Fresh inilk litre 630 0.01 0.12 1044 Canned sweet milk 390 gms 1,334 0.05 0.50 1045 Powdered milk kg 5,090 0.00 0.00 1046 Baby inilk 400 gins 1,984 0.04 0.39 1047 Other eggs/milk and dairy 100 gms 0.00 36 TableA.l.lO: Food bundle Code Item Unit Calories Per capitaexpenditure Per US Dollars Y O unit per month 1050 Vegetables 1.51 15.06 1051 Spinach 114 0.02 0.24 1052 Kangkung 220 0.08 0.76 1053 Cabbage 250 0.05 0.49 1054 Light mustardgreen 260 0.14 1.43 1055 Dark mustardgreen 260 0.08 0.82 1056 String bean 276 0.01 0.09 1057 Tomato 671 0.01 0.15 1058 Carrot 430 0.00 0.03 1059 Cucumber 125 0.00 0.00 1061 Cassavaleaves 635 0.25 2.50 1062 Eggplant 260 0.01 0.10 1063 Squash 285 0.03 0.31 1064 Papaya, young 345 0.17 1.66 1065 Papayaflowers 345 0.18 1.78 1066 Lettuce 130 0.01 0.06 1067 Pumpkin 260 0.02 0.21 1068 Pumpkin leaves 190 0.02 0.20 1069 Kabura n.a. 0.02 0.17 1071 A Timor veg 635 0.05 0.51 1072 Tips of banana plants 644 0.05 0.47 1073 Green bitter melon 320 0.00 0.04 1074 Onion(big) 1,236 0.17 1.66 1075 Garlic 1,490 0.10 0.99 1076 Redpepperichili 659 0.00 0.04 1077 Sukun n.a. 0.02 0.16 1078 Other vegetables 0.02 0.16 1081 Soya bean kg 4,160 0.03 0.26 1082 Mung bean kg 300 0.06 0.64 1083 Cashews 100gins 587 0.00 0.01 1084 Peanuts kg 5,670 0.04 0.42 1085 Kidney bean kg 3,330 0.16 1.58 1086 Tofu & tempe kg 1,350 0.00 0.00 1087 Other legumeshuts kg 0.04 0.40 37 Table A.l.lO: Foodbundle Code Item Unit Calories Per capita expenditure Per USDollars YO unit per month 1090 Fruit 0.42 4.23 1091 Orangeitangerines 455 0.00 0.02 1092 Mango 365 0.07 0.67 1093 Apples 590 0.00 1094 Avocado 1,610 0.02 0.17 1095 Pineapple 490 0.01 0.14 1096 Banana 920 0.17 1.74 1097 Papaya 345 0.08 0.76 1098 Jambu air n.a. 0.00 0.02 1099 Goiabas n.a. 0.00 0.02 1101 Watermelon 320 0.01 0.06 1102 Soursop 660 0.00 1103 Jackfruit 940 0.01 0.13 1104 Markka n.a. 0.00 0.04 1105 Cannedfruit n.a. 0.00 0.01 1106 Coconuts 3,363 0.05 0.47 litre 6,960 0.08 0.80 1112 Pork oil litre 6,960 0.01 0.07 1113 Other cooking oil litre 6,960 0.27 2.70 1114 Dry coconut kg 6,960 0.00 0.05 1115 Butter andmargarine 100 gms 717 0.00 1116 Other oil and fat litre 0.00 0.01 1120 Beveragesldrinks 0.79 7.89 1121 sugar 100 gms 375 0.34 3.37 1122 Palm sugar 100gms 375 0.00 0.01 1I23 Tea 100 gms 466 0.02 0.21 1124 Coffee 100 gins 1,243 0.43 4.27 1125 Cocodchocolate powder 100gms 288 0.00 0.01 1126 Soda drinks (Sprite. Coke) litre 403 0.00 0.03 1127 Other beverages 0.00 u litre 1130 Ingredients 0.19 1.88 1131 Salt 100 gms 0 0.08 0.84 1132 Honey kg 3,040 0.00 0.01 1133 Candle nut 100 gms 2,245 0.00 0.01 1134 Paprika 100gins 289 0.04 0.35 1135 Soy sauce sweetkour 140mi 77 0.00 0.01 1136 MSG gram 0 0.07 0.66 1137 Other ingredients/spices kg 0.00 0.00 38 Table A.l.lO: Food bundle Code Item Unit Calories Per capita expenditure per US Dollars Y O unit per month 1141 Instant noodles 80 gms - 356 0.26 2.57 1142 Macronie 100gms 360 0.01 0.11 1143 White bread small piece 53 0.01 0.14 1144 Sweet bread each 162 0.05 0.53 1145 Biscuits 100gms 325 0.01 0.14 1146 Sweetslcakes each 37 0.03 0.33 1147 Snacks portion n.a. 0.00 0.00 1148 Other food 0.00 1149 Prepared food and drink 0.00 0.00 1151 Beer 620 ml 0.00 0.02 1152 Wine 620 ml 0.00 0.02 1153 Tua mutin litre 0.10 1.04 1154 Tua sabu litre 0.16 1.55 each 0 0.16 1.60 1162 Clove cigarette, lionfilter each 0 0.01 0.08 1163 Tobacco cigarette, filter each 0 0.00 0.02 1164 Tobacco cigarette, noli filter each 0 0.01 0.11 1165 Tobacco 100 gins 0.19 1.89 1166 Betel fruit stick 0.03 0.29 1167 Betel nuts 100 gins 0.07 0.65 1168 Betel leaves grams 0.14 1.36 1169 Areca nut stick 0.18 1.79 Note: All Rupiah valuesjronz the strrvey were converted to USDollars using an exchange rate of 10,000 Rupiah/US Dollar. Source: 2001 TLSS. 39 2. THE PEOPLE'S PERSPECTIVE INTRODUCTION2' 2. I Happiness, satisfaction, and well-being with life are broad notions that go beyond purely material endowments. Welfare indicators like income, or expenditure, fail to capture this multi-dimensionality o f happiness. Furthermore, we cannot make interpersonal comparisons o f welfare by looking solely at the "revealed" preferences o f people as evident from their demand behavior. This has severe consequences: as i s well know from Arrow's Impossibility Theorem (1950), it i s impossible to construct a social welfare function inthe absence o f interpersonal comparisons o f individual welfare. 2.2 The shortfalls o f standard welfare measures are well recognized, but the conventional methods have remainedpopular, as it proved difficult to propose convincing measurable concepts that comprise the wider concepts o f happiness. However, over the last three decades or so, a substantivevolume o f research has emerged that uses people's own assessments to get at notions o f individual happiness and satisfaction. The underlyingidea is to rely on individuals and households themselves to define their level o f well-being. Even though precise definitions o f these concepts still remain elusive, psychologists and economist have used self-assessments as proxy measures o f welfare and well-being. The approach has become so successful that subjective questions are now routinely included in household surveys, along with objective measures. TLSS followed this praxis and collected subjective information on life satisfaction both in general and with respect to various domains of life, such asjobs, food security, health, education, and empowerment. 2.3 The distinguishing feature o f subjective measures i s that they are based on a person's self-assessment. This raises the questionwhether they are consistent, and change systematically with objective measures. For example, a person's self-assessed economic situation may stay unchanged even though her consumption increases, either because her expectations have increased, or because her position relative to her reference group has remained the same. To avoid such pitfalls inthe assessment o f well-being across a group o f individuals on the basis o f subjective indicators, we need to assume that individuals: are able to understand and answer consistently questions about own situation, and e provide responses that are comparable. 2.4 While some evidence suggests that these conditions typically hold within a common cultural context, it i s generally difficult to verify these assumptions. Therefore, 27 This chapterwas written by KasparRichter. 40 an analysis o f subjective measures i s best conducted jointly with an investigation o f objective indicators. In this chapter, we have drawn on subjective measures to cross- check evidence from objective indicators. We investigate four specific areas: 0 Subjective well-being post-violence; 0 Change insubjective well-being since the violence; 0 Characteristics o f "winners" and "losers"; and 0 Personal and national priorities. SUBJECTIVE WELL-BEING 2.5 Happiness is much more than income alone. Participatory poverty assessments in Timor-Leste and around the world have shown that the good life or well-being is multidimensional with both material and psychological dimensions.28Well-being is peace o f mind; it i s good health; it i s belonging to a community; it i s safety; it is freedom of choice and action; it i s a dependable livelihood and a steady source o f income; and it is food. Absence o f poverty i s the capability to cover one's essential needs. To be poor "...is to be hungry, to lack shelter and clothing, to be sick and not cared for, to be illiterate and not schooled" (World Development Report 2000-2001). Ill-being is notjust lack o f material things - o f food, but also work, money, shelter and clothing. It i s also living and working in often unhealthy, polluted and risky environments, and with bad experiences and bad feelings about the self. 2.6 A standard tool to assess subjective well-being are "adequacy" questions covering the different categories o f family needs. InTable 2.1, we display the answers o f heads o f households to questions regarding their family requirements. It shows the percentage shares o f each o f the three (less than adequate; just adequate; and more than adequate) possible answers along the dimensions o f food, shelter, clothing, health care, education, and income. '*One example is the World Bank's participatory research initiative, Voices of the Poor, which collected the voices of more than G0,OOO poor women and men froin GO countries (World Bank 2000). 41 Table 2.1: Subjective Well-being: Adequacy Food Housing Clothing Health Education Income (%) (%) ("/) care (%> ("/) Less than adequate 59.2 48.9 64.0 35.9 55.9 75.4 (2.3) (2.6) (2.1) (2.9) (2.4) (1.8) Just adequate 40.5 50.8 35.9 62.8 43.4 24.3 (2.3) (2.6) (2.1) (2.9) (2.3) (1.8) More than adequate 0.3 0.3 0.1 I.3 0.7 0.3 (0.1) (0.1) (0.1) (0.4) (0.3) (0.11 hlote: Standarderrors inparentheses Source: 2001 TLSS. 2.7 The striking feature i s one o f widespread inadequacy and severe hardship o f everyday life. Whatever specific aspect o f living standards we consider, 99 in 100people inTimor-Leste feel at bestjust adequately endowed, and betweenover one thirdto three quarters believe to be less than adequately covered. The concern is largest for clothing, followed by food, children's education, and housing, and least for the provision of health care. Inaddition, more than three infour persons live inhouseholds where total income i s deemed inadeq~ate.~' Table 2.2: Subjective Well-being: Happiness Very satisfied Rather satisfied Neithermor Somewhat unsatisfied Very unsatisfied Note: Standarderrors inparentheses. Source: 2001 TLSS. 2.8 The household head was the respondent to these adequacy questions, assessing the economic situation o f the entire family. Yet, perceptions o f happiness are personal, vary from one household member to the next, and go beyond just purely economic 29 It is not clear whether respondents viewed total income as a summary measure capturing other dimensions, or a separate dimension o f living standards itself. 42 notions o f well-being. In Table 2.2, we show the assessment o f all individuals aged 15 years or older in terms o f happiness. In spite o f widespread deprivation, the population displays a surprisingly degree o f satisfaction with life in general. More than one third o f the individuals are very or rather satisfied, compared to just over one fifth who are somewhat or very unsatisfied. CHANGE IN SUBJECTIVE WELL-BEINGSINCE THEVIOLENCE 2.9 H o w has life changed since 1999 inthe people's own assessment? Inthe survey, the people o f Timor-Leste were asked to assess the changes since before the violence in 1999 along different dimensions: living standards, corruption, economic status, and power status. This section looks at the evidence from these questions. Our focus is on exploring what could account for the high degree o f general satisfaction with life in the face o f economic hardship. We will show that part o f the explanation lies with the improvement in non-economic dimensions o f life that have improved substantially over the last two years. 2.10 Living standards are closely linked to economic conditions, and they have remained difficult since the violence. In Table 2.3, we show the responses o f all individuals aged 15 years or older when asked about the change in living standards. About three in ten persons believe living standards have deteriorated, compared to only one in ten persons saying they have improved. This underlines the message o f the adequacy questions, pointing to substantial material hardship. Table 2.3: Change in Living Standards Since the Violence in 1999 Improvement 10.5 (1.0) Same 60.4 (2.4) Deterioration 29.1 (2.4) Note: Standarderrors in parentheses Source: 2001 TLSS. 2.1 1 Corruption i s a core poverty issue. For example, the World Bank's Voices o f the Poor recorded reports by poor people o f hundreds o f incidents of corruption as they attempt to seek health care, educate their children, claim social assistance, get paid, attempt to access justice or police protection, and seek to enter the marketplace. Intheir dealings with officials, poor men and women are subject to insults, rudeness, harassment, and sometimes assault by officials. Harassment o f vendors in urban areas is widespread. Politicians, state officials, and public servants are rarely viewed as effective, trustworthy, or participatory. Corruption also matters for the broader performance o f a country. Iti s an 43 obstacle to economic and social development. It distorts the rule o f law and weakens the institutional foundation on which economic growth depends. These harmful effects are especially severe on the poor, who suffer most from economic decline, are most reliant on the provision o f public services, and are least capable o f paying the extra costs associated with bribery, fraud, andthe misappropriation o f economic privileges. Table 2.4: Change in Corruption Since the Violence in 1999 Age-gender groups Geography --- Total 15 24 - 25 49 - 50 plus National R U MUC OUC RW RC RE RL RM RH RI RS F M F M F M F M More 18 17 23 30 14 18 18 14 8 17 19 17 16 16 20 18 21 18 22 11 15 (1.4) (1.8) (1.8) (2.4) (2.6) (4.5) (2.6) (2.4) (4.7) (3.0) (2.1) (1.9) (4.8) (1.4) (1.6) (2.1) (2.3) (1.5) (1.8) (1.7) (2.3) Same 42 44 36 39 33 42 47 41 40 40 49 45 38 43 41 39 43 43 39 49 43 (1.9) (2.4) (3.0) (2.6) (6.0) (5 2 ) (3.7) (3.3) (3.5) (3.3) (3.8) (2.7) (3.9) (2.1) (1.9) (2.7) (2 4) (2.2) (2.2) (3.5) (2.9) Less 40 39 40 31 53 40 35 45 51 43 32 38 46 40 39 42 36 40 39 40 42 (2.5) (3.1) (3.5) (2.8) (7.1) (6 5) (4.5) (5.2) (7.7) (4.5) (4.3) (3.5) (5.7) (2.7) (2.4) (3.3) (2.9) (2.7) (2.6) (3.5) (3.3) 2.12 People's perception on the change in corruption since 1999 are shown in Table 2.4. Overall, people feel corruption i s less o f an issue now than in 1999. Only one fifth of the population aged 15 years or older believes corruption has worsened since violence, compared two fifth who feel corruption has declined. Across the board o f geographic and age-gender categories, more people believe corruption is less prevalent now. However, there are important differences. Most strikingly, in major urban centers, three in ten people feel corruption has become worse. Inrural areas, the issue appears to be larger in rural west and center than in rural east, and in rural mid- and highlands than in rural lowlands. With regard to gender, men are more pessimistic than women about the progress made in corruption prevention, as are persons younger than 50 years of age compared to those older than 50 years o f age. One possible explanation o f this pattern could be involvement in commercial and administrative tasks. Inhabitants o f Dili and Baucau, and prime-age men are likely to be more exposed to such activities. Interestingly, the more optimistic view on change incorruption inthe rural east and rural lowland coincides with lower poverty than inthe other rural domains. 2.13 Living standards and corruption are important for both economic status and empowerment. Table 2.5 displays the responses to "ladder questions", where persons are asked to rank themselves with regard to economic and power status, both for now and before the violence. Let us consider the economic dimension first. Looking back to before the violence in 1999, the vast majority view themselves as poor: one third o f the respondents believe they were on the lowest step, another third on the second lowest step, and another 30 percent between the third to fifth lowest steps. Less than two percent ranked themselves on the top four steps. By comparison, today's situation has improved, especially for the lowest third. The share at the lowest step has significantly decreased, boosting the shares o f the second and third lowest steps, with the rest remaining unchanged. Overall, the economic situation o f the lowest two thirds has improved or 44 remained unchanged, while the one for the highest third has remained ~nchanged.~'This more detailed assessment o f the changes o f the economic status leads to a more positive evaluation o f the alteration since the violence than the single question about changes in living standards discussed inthe previousparagraph. Table 2.5: Subjective Well-being: Economic and Power Status Economic Power 2001 1999 2001 1999 Lowest 22 32 5 62 (1.7) (1.9) (0.7) (2.3) 2nd 38 34 20 22 (1.8) (1.5) (1.8) (1.5) 3rd 25 19 21 9 (1.6) (1.2) (1.3) (0.9) 4th 11 9 25 4 (1.1) (1.0) (1.7) (0.6) 5th 3 4 15 2 (0.5) (0.5) (1.0) (0.5) 6th 1 1 8 1 (0.2) (0.3) (0.9) (0.4) 7th 0 1 3 0 (0.0) (0.2) (0.7) (0.1) 8th 0 0 1 0 (0.0) (0.1) (0.3) (0.0) Highest 0 0 1 0 (0.0) (0.0) (0.3) (0.1) Note: Standard errors inparentheses. Source: 2001 TLSS. 2.14 The questions regarding power status reveal a clear picture. In 1999, today's population viewed themselves as powerless, with six in ten placing themselves on the lowest step, and another two inten on the second lowest step. Essentially nobody ranked herself on the top four steps. The situation in 2001 i s substantially different. Only one in twenty people believe they are completely powerless, and close to three in ten believe they rank on the top five steps.31 These numbers suggest that, while the economic situation has improved primarily at the bottom tail, the advances in power status have affected almost the entire population. The transition matrices for economic and power status are shown in Table 2.6 and Table 2.7. For example, p i equals 39 percent, 30Assumingthat the means of the two distributions are unchanged, we can apply the concept of first-order stochastic dominance. The 2001 cumulative density function is not higher than the one for 1999 up to the 95 percentile, and therefore first-order dominatesthe 1999 distribution up to this percentile. 31The 2001 cumulative density function i s not higher than the 1999 cumulative density function over the entire range, and therefore first-order dominatesthe 1999 distribution over the entire range. 45 indicating that about 4 in ten person who belonged to the lowest step in 1999 moved to the second lowest step in2001. The corresponding number for power status i s 22 percent. Table 2.6: Matrix Economic Status 200I Lowest 2nd 3rd 4th 5th 1999 Lowest 43 39 14 3 1 (3.5) (3 .o> (1.8) (1.1) (0.5) 2nd 16 50 26 7 1 (2.0) (3.1) (2.8) (1.1) (0.6) 3rd 8 30 40 19 2 (1.9) (3.4) (3.7) (2.9) (0.8) 4th 7 23 29 31 9 (2.6) (4.1) (4.5) (5.6) (4.2) 5th 3 14 20 17 45 (2.4) (4.6) (5.4) (4.5) (6.0) Note: Standard errors in parentheses. Source: 2001 TLSS. Table 2.7: Matrix Power Status 2001 Lowest 2nd 3rd 4th 5th 26 (2.9) 27 (4.0) 33 (6.5) 59 (10.2) 78 (13.3) Note: Standard errors inparentheses. Source: 2001 TLSS. 2.15 The values for mobility measures are shown in Table 2.8. Summing over the entries on the main diagonal, we find substantial mobility. Only four in ten persons 46 remain on the step for economic status, and only less than three in ten for power status. The other immobility indicators also suggest lower mobility with regard to economic status than power status. The same holds overall for the jump measures, even though those for the top highest steps are lower for power status than for economic status. For the entire matrix, economic status changes on average by half a step, compared to 0.7 steps for power status. Table 2.8: Mobility Measures I J U e Economic status 1 42.0 0.8 1.8 16.5 2 79.2 0.3 2.3 37.5 3 94.0 0.2 2.8 28.2 4 99.1 0.9 3.1 13.4 5 1.1 3.9 4.5 Total 0.5 2.8 2.5 Power status 1 27.9 2.5 3.5 2.4 2 60.1 1.5 3.5 12.3 3 81.7 0.7 3.7 11.3 4 94.5 0.2 4.2 10.2 5 0.5 4.5 63.8 Total 0.7 3.9 4.2 Note: The mobility measures are defined in the mobility section in Chapter 1 of this volume. Source: 2001 TLSS. 2.16 From the perspective o f social welfare, more important than overall mobility i s upward mobility. For economic status, the average state rank in2001 is 2.8, compared to the average state rank o f 3, suggesting an overall downward movement. However, in contrast to the top three states, the two lowest states show upward mobility.32By contrast, for power the average state rank in 2001 i s 3.9, implyingupward mobility. Inparticular, the improvement inthe lowest rank is dramatic, with an average step rank o f 3.5. We also findthat bothtransition matrices are "monotone" (Conlisk 1990) so that the disadvantage o f originating from a low state i s preserved into the future. 2.17 The equilibrium vectors, deriving from a first-order Markov chain, are shown in Table 2.8. The distribution on economic status shows the bulk o f the population on the second and third step, and no more than 5 percent on the top step. By contrast, close to two thirds o f the population end up on the top ladder for power status. Furthermore, the 32 Note that by construction o f a transition matrix, the lowest state cannot display downward mobility, while the highest state cannot display upwardmobility. 47 power status equilibrium distribution first-order dominates the one for economic status. This confirms the previous conclusions: on the one hand, the economic situation has improved primarily for the poorest families, while little amelioration is evident for other households; on the other hand, the population feels broadly empowered as a result o f the changes since 1999. WINNERS AND LOSERS 2.18 Timor-Leste has made significant strides since the violence surrounding the referendum o f September 1999. This transformation has changed the direction o f life courses for literally every citizen, and affected their material and emotional well-being, including own perceptions o f self-worth. However, as the previous section suggests, not everybody has benefited inthe same way, and some even feel to have lost out. Although the fundamental achievement o f gaining independence is overwhelmingly positive, the vast structural changes were accompanied by conflict, destruction, and migration with negative impacts on parts o f the population. 2.19 Inparticular, as the experience inthe former Soviet Unionhas demonstrated, the economic transformation can trigger conflict through creating `winners' and `losers', challenging traditional values or authority structures, or raising the stakes o f economic competition. A full appreciating o f the scale of these changes would require a thorough investigation of political and economic transformation, social reconstruction and empowerment, and the institutional capacity to manage or resolve violent conflict and to promote tolerance, and buildpeace and human security. 48 Table 2.9: Winnersand Losers: Characteristics Economic Status Power Status Upward Downward Upward Downward Poverty headcount ("h) 0.319 0.276 0.341 0.279 (0.033) (0.03 1) (0.027) (0.055) Poverty gap ("h) 0.086 0.070 0.100 0.080 (0.013) (0,010) (0.012) (0.020) Poverty severity (%) 0.034 0.025 0.041 0.03 1 (0.006) (0.004) (0.006) (0.009) Age (years) 34.5 35.7 35.6 34.8 (0.458) (0.470) (0.282) (0.965) Male ("h) 0.507 0.494 0.502 0.507 (0.008) (0.012) (0.006) (0.021) Attended school (%) 0.504 0.503 0.480 0.439 (0.024) (0.025) (0.016) (0.045) Average grade ifschooling (1 - 22) 9.05 8.84 8.70 8.93 (0.246) (0.234) (0.151) (0.509) Farmer 0.378 0.408 0.399 0.5 10 (0.016) (0.019) (0.01 1) (0.038) Labor force 0.553 0.576 0.549 0.610 (0.014) (0.022) (0.012) (0.031) Household size 5.9 5.7 5.7 5.0 (0.157) (0.183) (0.112) (0.269) Dependencyratio 0.952 0.934 0.940 0.814 (0.039) (0.038) (0.027) (0.086) Urban 0.284 0.258 0.256 0.176 (0.026) (0.026) (0.009) (0.036) Major urban centers 0.180 0.138 0.145 0.087 (0.021) (0.016) (0.007) (0.026) Other urban centers 0.103 0.120 0.111 0.089 (0.014) (0.019) (0.006) (0.024) Rural west 0.206 0.268 0.166 0.305 (0.056) (0.066) (0.042) (0.087) Rural center 0.330 0.359 0.348 0.393 (0.055) (0.061) (0.050) (0.084) Rural east 0.180 0.116 0.230 0.126 (0.049) (0.034) (0.046) (0.051) Rural lowland 0.330 0.472 0.428 0.581 (0.054) (0.062) (0.049) (0.079) Ruralmidland 0.295 0.380 0.356 0.508 (0.053) (0.064) (0.049) (0.084) Rural highland 0.386 0.270 0.317 0.243 (0.060) (0.057) (0.050) (0.073) Rural seaside 0.099 0.154 0.147 0.222 (0.033) (0.048) (0.039) (0.077) Population(%) 35 23 84 6 Note: Standard errors in parentheses. Source: 2001 TLSS. 49 2.20 While such an agenda goes beyond the scope o f this analysis, we can shed some light on this issue with regardto self-perceived changes in economic and power status as captured by the ladder questions. What distinguishes the winners from the losers o f this two-year transition? InTable 2.9, we classify individuals according to whether they have climbed, or dropped back, on the ladders o f economic status and power status, and display summary statistics o f basic personal, household, and geographical characteristics. 2.21 Let us consider economic position first. Over one third o f the citizens aged 15 or older believe their situation has improved, compared to just below one quarter who experienced downward mobility. Interms o f mean group characteristics, the two groups are comparable in terms of age, gender, and household composition. Furthermore, the differences in terms o f education and labor force are too small to be significant. By contrast, the geographical incidence shows important variations. The upwardly mobile are more urban, and correspondingly less likely to be farmers than the downwardly mobile. Furthermore, in urban areas, the winners are concentrated in the major urban centers, and in rural areas in the east, the highland, and inland. In other words, the winners come from parts of the country with low poverty (DiWBaucau and rural east) and highpoverty (rural highland andrural inland). However, the poverty statistics reveal that the upwardly mobile are overall poorer than the downwardly mobile. This suggests that the economic winners o f the transition come over-proportionately from the poorer segment o f the population in 1999. Their gains may have narrowed the material gap to the rest o fthe population, butnot eliminated it. 2.22 H o w does this picture differ for power status? Again, we find little evidence for differences in terms of age and gender, but other features are important. Overall, the urban-rural gap i s substantially larger than for economic status. Correlated with this division, those feeling more empowered a more likely to have attended school, less likely to be farmers, and have larger household size.33In terms o f geographical breakdown in urban and rural areas, the pattern i s in line with the picture on economic status, with winners originating from major urban centers, rural highland and rural inland. Equally, we find that the poor were especially empowered, again indicating that the least advantaged in 1999 feel included inthe gains inpower status. PERSONALAND NATIONAL PRIORITIES 2.23 Changes in self-perceived economic and power status can be triggered by various factors. Economic well-being i s tied to issues like employment, housing, and business climate, while empowerment relates to aspects like participation in the community and absence o f fear o f violence. Exploring these factors allows us to gain a deeper understanding on the perceived successes and failures o f the transformation since the referendum o f 1999.Contrasting the evaluation o f the past performance with the personal and national priorities looltiiig forward gives us an indication about to what extend the agenda o fthe past should be modified for the future. 33Note that in Timor-Leste, in contrast to most developing countries, households have more members in urbanthan inruralareas. 50 2.24 This section draws on ranltings o f individuals aged 15 years or older o f ten categories with regards to changes inthe past, and personal and national priorities for the future. In four questions, interviewees were asked to give the two main areas o f improvement and deterioration over the last two years, and o f individuals and national priorities for improvement o f living standards as o f today. In Table 2.10, we display the backward-looking results. We summarize the responses by calculating the difference in the percentage rates o f improvement and deterioration for each sector, nationwide and separated by salient geographic and age-gender groupings. A positive number suggests more feel the specific area has improved than deteriorated, while a negative number indicates that those perceiving a worsening outnumber those seeing an improvement. Table 2.10: Change in Living Standards Since the Violence in 1999 by Sector Geogiapliy Age-gender groups -- 15-24 25 - 49 50 plus National MUC OUC RW RC RE RL RM RH F M F M F M Safety 39 45 31 67 23 40 46 39 37 33 41 39 36 43 44 (4.7) (5.1) (14.5) (7.4) (9.5) (7.1) (11.8) (7.5) (10.3) (5.4) (5.5) (5.0) (5.1) (5.2) (6.0) Political participation 26 32 15 23 30 26 39 26 25 28 26 25 26 28 30 (3.0) (4.2) (8.6) (7.1) (6.6) (4.4) (5.9) (4.9) (6.7) (2.9) (4.2) (3.2) (3.9) (3.2) (3.6) Education 19 19 29 27 19 8 6 16 22 20 18 20 19 16 19 (2.3) (3.0) (7.4) (5.4) (4.5) (4.3) (5.4) (4.3) (4.3) (3.8) (3.5) (2.6) (2.7) (3.4) (4.0) Status in community 9 6 7 1 6 8 9 6 8 1 4 7 9 10 10 9 8 (1.5) (1.8) (2.8) (3.8) (2.5) (4.1) (7.4) (2.6) (2.8) (2.1) (2.0) (1.6) (1.6) (2.4) (2.0) Health care 2 8 7 -3 5 -7 -11 2 0 4 - 1 2 3 - 1 1 (2.7) (2.3) (8.1) (5.0) (5.6) (5.6) (5.6) (5.3) (5.1) (3.0) (3.4) (3.3) (2.8) (4.5) (3.5) Access to land -5 1 -4 -1 1 -3 -5 -5 -7 -5 -4 -4 -5 -4 -5 -5 (1.2) (1.2) (3.4) (3.4) (1.7) (2.9) (3.7) (2.3) (2.1) (1.5) (1.6) (1.5) (1.3) (2.7) (2.4) Infrastructure -12 -2 -12 -15 -12 -16 -9 -14 -14 -13 - 1 I -13 -10 -12 -14 (1.6) (2.1) (4.4) (4.4) (2.8) (3.5) (7.3) (2.4) (3.4) (2.2) (2.4) (1.7) (2.0) (2.5) (3.1) Employment -19 -40 -15 -29 -11 -11 -17 -16 -15 -20 -18 -19 -21 -17 -17 (2.0) (3.9) (6.1) (4.2) (3.4) (3.7) (4.5) (3.3) (4.3) (3.3) (3.3) (2.5) (2.7) (3.3) (3.1) Demandfor products -26 -17 -19 -28 -30 -24 -17 -27 -32 -22 -28 -26 -24 -30 -26 (2.4) (2.7) (5.1) (6.3) (4.2) (5.9) (7.9) (4.7) (4.4) (3.4) (3.3) (3.0) (2.8) (3.2) (3.2) Housing -34 -48 -38 -45 -29 -22 -39 -28 -33 -33 -32 -33 -35 -33 -40 (2.1) (3.0) (6.1) (5.2) (3.7) (3.7) (10.2) (3.7) (3.7) (3.2) (2.6) (2.6) (2.5) (3.6) (3.2) Noie: MlJ~.',siun~s,fi,rMujorUrhun Cenler.v,O ~ K ' / uOIher ilrhun Cenieia. RWfor Rrirul We~lRCfiir Rarul ('enler, REfiir Rjir~IEmf, r , m,fbrRwulI,owhmd,M / o r Ri,ruiA4idlmd, I?H/or Rimi Highlund, 1,'frirFenrulem7dMfiJrMule. Slundurderrurx inlmrenrhrses. Som-cc. 2Olil TLSS. 2.25 According to this ranking, safety comes out on top, while housing ranks bottom. Inparticular, three fifths believethat safety has improved since the violence, comparedto one fifth who feel safety has deteriorated. Therefore, the excess o f those feeling an improvement relative to those perceiving a deterioration i s two fifths, the highest share across all categories. Other areas, where more interviewees saw an improvement rather than a deterioration, are, ranked inorder o f importance, political participation, education, status in community, and health care. For the remaining five areas, more people experienced a deterioration than an improvement: access to land, infrastructure, employment, demand for products, and housing, which ranked bottom scoring-34. 2.26 This pattern highlights both the achievements, and disappointments, during the transformation since 1999. On the one hand, the areas o f improvement like safety, political participation, and status incommunity are directly associated with overcoming a 51 history o f violence and suppression, and the move towards an independent and democratic Timor-Leste. Furthermore, the positive scores for education and health also reflect the appreciation o f the population for the substantial social investments made duringthe last years. This emphasis is likely to have contributed to the overall positive assessment o f the transitional period by the p ~ p u l a t i o n . ~ ~ 2.27 O n the other hand, the negative scores for housing and infrastructure reflect the destruction occurring inthe immediate aftermath o f the 1999 referendum. It confirms that inview ofthe large-scale devastation the considerable reconstructionefforts havenot yet been able to fully repair the damage. Finally, the low ranking o f land access, employment, and demand for products point to the disruptive impact o f the transition period on economic activities. 2.28 Table 2.10 also shows a breakdown o f the ranking by geographic and age-gender groups. Overall, the differences between regions, and especially age-gender groups, are relatively minor, with a number o f noteworthy exceptions. Due to its proximity to Indonesia, the rural west was especially affected by violence and destruction in 1999. This shows up with high scores on safety, status in community, and education, and low scores on access to land, infrastructure, employment, and housing. A similar pattern is discernible for Dili and Baucau, which, as major urban agglomerations, were also focal points o f disruption and conflict. 2.29 We now turn to the priorities of the population looking forward. The results are shown in the Table 2.11 and Table 2.12, displaying the main personal concerns and concerns for Timor-Leste, respectively. The numbers state the percentage o f individuals aged 15 or older indicating an area as first or second priority. Top o f the list o f personal concerns are economic and social factors. Number one i s employment, quoted by three fifths o fthe interviewees. This i s followed by improvements insocial services (education, health care, and housing), and demand for products. In contrast, the main achievements o f the past years (safety, political participation, and status in community) rank lowest in terms o f importance for individual living standards for the future. Separating regions, employment matters substantially more for the rural west and rural highland compared to rural east and rural lowland, while the order o f priority i s reverse for demand for products. Interms o f age-gender groupings, employment is more important for men than women, the young are concerned particularly about education, while the old worry more about health care and housing. Perhaps surprisingly, demand for products turns out to be more a concern for women than men. 34 A specific emphasis on social policies, relative to sectoral and macroeconomic policies, is recommended as policy priority for post-conflict countries, based on a series of empirical research conducted by the Conflict Prevention and ReconstructionUnit inthe World Bank. 52 Table 2.11: PersonalPrioritiesfor LivingStandards Geography Age-gender groups -- 15 -24 25 -49 50 plus National MUC OUC RW RC RE R L RM RH F M F M F M Employment 62 63 65 73 67 43 49 57 69 59 63 61 12 47 59 (1.8) (2.8) (4.2) (5.0) (2.0) (3.0) (6.8) (3.1) (3.2) (2.8) (2.8) (2.1) (2.0) (3.0) (2.7) Healthcare 36 25 40 28 40 40 33 38 38 32 33 39 33 43 38 (1.6) (2.1) (5.1) (5.4) (2.5) (2.7) (6.3) (2.5) (3.5) (2.2) (2.5) (2.0) (1.8) (2.7) (2.8) Education 30 32 31 37 24 33 27 31 29 40 44 26 27 22 25 (1.5) (2.6) (3.4) (4.5) (2.7) (2.9) (6.1) (2.7) (3.1) (2.2) (2.5) (1.9) (1.8) (2.3) (2.4) Housing 23 29 23 29 17 24 31 23 18 23 19 22 22 29 27 (1.5) (2.4) (4.5) (4.8) (2.0) (2.9) (4.6) (2.9) (2.5) (2.3) (2.0) (1.7) (1.5) (3.1) (2.8) Deinand for products 21 21 i 7 15 20 30 32 23 19 20 16 25 16 31 24 (1.3) (2.1) (2.9) (3.9) (1.9) (1.8) (5.0) (2.5) (1.7) (2.1)(1.9) (1.6)(1.4) (2.5) (2.5) Safety I O 8 12 3 16 5 2 9 12 9 8 10 10 10 9 (1.3) (1.3) (3.4) (1.2) (2.6) (1.4) (0.9) (2.0) (2.9) (1.8) (1.6) (1.4) (1.4) (2.2) (2.0) Infrastructure 9 8 8 10 9 12 11 10 9 6 7 8 11 10 12 (0.8) (1.4) (1.9) (2.2) (1.3) (2.0) (2.1) (1.3) (1.8) (1.2) (1.3) (1.0) (1.2) (1.8) (1.9) Access to land 3 2 3 5 2 6 8 4 2 4 2 4 3 4 3 (0.5) (0.7) (1.2) (1.6) (0.6) (1.3) (2.1) (1.0) (0.7) (0.8) (0.6) (0.7) (0.6) (1.2) (0.8) Political participation 1 3 0 0 1 2 1 1 0 2 2 1 1 1 1 (0.3) (0.8) (0.3) (0.1) (0.3) (0.9) (0.7) (0.6) (0.1) (0.5) (0.7) (0.3) (0.2) (0.4) (0.6) Status in coininunity 1 4 0 1 1 1 0 1 1 2 2 1 1 1 0 (0.3) (0.9) (0.2) (0.5) (0.6) (0.4) (0.0) (0.4) (0.6) (0.7) (0.5) (0.3) (0.4) (0.7) (0.3) Note:MU(:stunds,ji,rMujor Urhun Cen2er.v. Oil('/or Other lirhun (.'en/cv\, RW,Ji,rRirrul We.st,RC,fr,r Rvrul (.'enter, RE fix Rard La.st. RLfor Rrrul Loii~lund,RMfor Rerul Mldluiid, RH for Riirul Higlilund, I,'/or Fentole mid M,fi,r Mule. Stundurd errws hipureiitl?c~cs. Sourcc. 2001 TLSS. 2.30 The priorities for Timor-Leste's living standards are broadly in line with individual preferences. The bottom three categories are exactly the same, and the same three categories appear in the top three, even if their internal ranking i s reversed. The most striking difference i s the emphasis on education as key to national prosperity, listed by seven in ten individuals, compared to only three in ten for personal preferences. Employment, housing, and demand for products are listed by fewer people as national priorities than individual priorities. Overall, this suggests that the immediate individual economic concerns are viewed as less important for the national agenda. Inboth personal and national rankings, economic and social concerns dominate aspects linked to empowerment, perhaps a reflection o f the achievement in this area over the past few years. 53 Table 2.12: National Priorities for Living Standards Geography -- Age-gender groups 15-24 25 49 - 50 plos National MUC OUC RW RC RE RL RM RH F M F M F M Education 70 66 72 75 70 70 75 67 74 71 70 70 73 63 71 (1.5) (2.4) (3.1) (4.7) (2.7) (2.8) (3.1) (2.7) (3.0) (3.0) (2.4) (1.8) (1.7) (2.6) (2.4) Employment 46 49 47 53 48 36 36 42 51 47 45 45 48 46 44 (1.9) (2.9) (6.0) (5.2) (3.0) (3.4) (6.4) (2.6) (4.0) (2.8) (2.9) (2.2) (2.3) (2.9) (3.2) Health care 39 23 38 42 42 45 39 44 42 41 39 44 33 43 37 (1.6) (3.1) (2.7) (5.4) (2.4) (3.4) (6.1) (2.4) (3.5) (2.4) (2.7) (2.1) (2.0) (2.9) (2.7) Safety 20 15 22 13 23 22 20 24 16 16 20 17 21 20 26 (1.5) (1.5) (4.7) (3.7) (2.9) (2.9) (5.8) (2.6) (2.9) (2.1) (1.9) (1.4) (2.1) (2.6) (2.8) Housing 7 I 1 6 5 6 8 3 6 7 8 8 7 7 6 5 (0.6) (1.2) (1.0) (1.8) (0.9) (1.4) (1.7) (1.1) (1.1) (1.3) (1.2) (0.9) (0.9) (1.3) (1.0) Infrastroctme 5 I O 6 3 4 7 11 5 3 6 6 4 5 7 6 (0.7) (1.8) (2.0) (1.7) (0.9) (1.6) (3.3) (1.1) (0.8) (1.1) (1.2) (0.7) (0.8) (1.4) (1.2) Demandfor products 5 14 4 2 4 4 6 3 3 6 7 5 4 7 4 (0.6) (2.1) (1.3) (1.3) (0.9) (0.7) (1.8) (0.8) (0.9) (1.0) (1.4) (0.7) (0.7) (1.3) (0.8) Access to land 2 3 2 2 1 6 7 3 1 1 2 3 3 4 3 (0.4) (0.7) (1.3) (0.9) (0.3) (1.3) (1.9) (0.9) (0.2) (0.5) (0.5) (0.6) (0.6) (1.1) (0.8) Politicalparticipation 2 4 2 4 2 1 2 3 1 3 2 2 3 2 2 (0.5) (0.7) (1.2) (2.6) (0.5) (0.5) (1.4) (1.3) (0.4) (0.6) (0.6) (0.5) (0.6) (0.7) (1.3) Statusin coininunity 0 2 1 0 0 0 1 0 0 I 1 0 1 0 0 (0.1) (0.6) (0.4) (0.1) (0.1) (0.1) (0.5) (0.1) (0.1) (0.3) (0.4) (0.1) (0.2) (0.1) (0.2) NCJWA4[l~:,slunrls,li,rMujur ilrhuii (:enIei!s,Oil(' /or O//ierilrhun enlo lo^. RWjiJrRum1 Wesf,RCfor Rtirul (:enler,I1E.for Rtird l?u.sl, RL,for Rarul Lowland,RMfor Riirul Midland. I Male Female Urban Rural age spline ages 5-9 0.120 0.062 0.102 0.088 (10.28) ** (5.25) ** (8.06) ** (7.96) ** age spline ages 10-14 0.110 0.068 0.092 0.087 (14.78) ** (9.32) ** (11.77) ** (12.63) ** age spline ages 15-19 0.057 0.024 0.049 0.038 (11.25) ** (4.70) ** (9.31) ** (7.86) ** age spline ages 20-24 0.022 -0.004 0.019 0.004 (5.28) ** (1.02) (4.46) ** (0.91) urban =1, else 0 0.160 0.114 (5.53) ** (3.74) ** male =1, else 0 0.048 -0.025 (1.98) * (1.07) log hhpc expend (nominal) 0.034 0.022 0.061 0.008 (1.88) (1.13) (3.99) ** (0.39) Sample 2,095 1,915 1,810 2,200 Note: Probit models with marginal effects shown. Absolute value of z-statistics inparentheses. * Signifcant ut 5% level ** Significant at 1% level Source: 2001 TLSS. TableA.11: CorrelatesofEnrollmentby quintile,2001(ages 5-24) Poorest Quintile 2 Quintile 3 Quintile 4 Richest age splineages 5-9 0.076 0.085 0.083 0.123 0.104 (4.44) ** (4.35) ** (4.29) ** (6.00)** (5.58) ** age splineages 10-14 0.076 0.085 0.086 0.116 0.095 (6.97) ** (7.10) ** (7.15) ** (8.71)** (8.21) ** age splineages 15-19 0.03 0.037 0.037 0.06 0.047 (3.90) ** (4.40) ** (4.46) ** (6.73)** (6.09)** age splineages 20-24 -0.003 0.001 0.004 0.025 0.015 (0.43) (0.17) (0.62) (3.51)** (2.52)* urban=1, else 0 0.005 0.105 0.16 0.181 0.172 (0.09) (1.75) (3.22) ** (4.32)** (4.84)** male =1, else 0 -0.05 1 0.027 -0.074 0.01 0.078 (1.31) (0.63) (1.86) (0.27) (2.29)* log hhpc expend(nominal) 0.019 0.214 -0.166 -0.167 0.011 (0.21) (1.OO) (0.79) (1.12) (0.36) Sample 770 676 754 801 1,009 Note: Probit models with marginal effectsshown. Absolute value of r-statistics inparentheses * Significant 5% level at +*Signifcant at 1%level Source: 2001 TLSS. 171 FigureA.3: EstimatedPopulationby Age, 2001 35 ' 30 'w 43 -g 25 ' s0c 20 /+- i - v 3 ~ -.o c + 15 c3 0 ?L 10 5 0 1 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63 66 69 72 75 78 81 85 90 Source: 2001 TLSS. Table A.12: Reasonsfor never attendingschool, Ages 5-6 Poorest Quintile 2 Quintile 3 Quintile 4 Richest All Below school age 68.7 76.4 65.9 72.9 72.8 71.3 Completed studies 0.0 0.0 0.0 0.0 2.2 0.3 Too expensive 0.5 3.5 0.0 1.6 2.8 1.7 No interest 14.7 9.0 5.9 12.5 11.0 10.6 Work at home 0.0 0.0 1.7 0.0 0.0 0.3 Other work 0.0 0.0 1.2 0.0 0.0 0.3 School too far 11.1 8.9 19.6 9.4 6.8 11.4 Noteacher 0.0 2.2 0.0 0.0 0.0 0.6 No supplies 3.5 0.0 1.4 0.0 0.0 1.1 School not functional 0.0 0.0 1.4 0.0 0.0 0.3 Illness 0.0 0.0 0.0 0.6 0.0 0.1 Family illnessideath 1.2 0.0 0.0 0.0 0.0 0.3 Displaced 0.4 0.0 0.0 1.1 0.0 0.3 Other 0.0 0.0 2.9 2.0 4.4 1.5 Source: 2001 TLSS. 172 Table A.13: Reasonsfor neverattendingschool, Ages 7-12 Poorest Quintile 2 Quintile 3 Quintile 4 Richest All Below school age 21.9 33.9 31.5 23.6 18.5 27.2 Too old 1.5 0.0 0.0 0.0 0.0 0.4 Too expensive 0.9 11.3 0.9 1.7 0.9 3.6 No interest 32.4 22.5 33.6 33.1 25.7 29.9 Agricultural work 0.0 0.0 0.0 3.4 0.0 0.7 Work at home 0.0 4.0 0.0 2.7 18.9 2.8 School too far 16.4 13.8 24.6 18.8 12.1 17.8 No teacher 2.6 7.2 0.0 0.0 0.0 2.4 No supplies 5.8 0.0 2.5 0.0 0.0 2.0 School not functional 0.0 1.6 0.0 0.0 0.0 0.4 Illness 0.6 2.9 2.6 4.1 3.0 2.5 Family illness/death 4.3 0.0 0.0 0.0 0.0 1.1 Displaced 6.1 2.8 2.8 0.0 0.0 2.8 Safety 0.0 0.0 0.0 1.8 0.0 0.4 Harassment 3.1 0.0 0.0 2.3 1.9 1.4 Other 4.4 0.0 1.5 8.7 19.0 4.5 Source: 2001 TLSS. Table A.14: Reasonsfor never attendingschool, Ages 13-15 Poorest Quintile 2 Quintile 3 Quintile 4 Richest All Too expensive 6.3 0.0 0.0 1.2 32.6 6.5 No interest 38.8 36.1 35.2 0.0 42.0 31.4 Agricultural work 6.8 0.0 0.0 4.0 8.4 3.9 Work at home 6.4 12.4 42.6 11.1 0.0 14.6 School too far 20.9 0.0 0.0 24.2 0.0 10.6 No teacher 8.2 10.0 0.0 0.0 0.0 4.5 School not functional 0.0 12.4 0.0 0.0 0.0 2.4 Illness 5.8 2.1 5.9 8.3 0.0 4.7 Displaced 0.0 11.8 10.2 0.0 0.0 4.3 Harassment 6.8 6.4 0.0 12.0 0.0 5.4 Other 0.0 8.7 6.2 39.3 17.0 11.8 Source: 2001 TLSS, 173 6. DISADVANTAGED GROUPS 6.1 In many Asian countries, some groups are excluded from the benefits of economic developments. Parentless children, elderly, widows, and women are often found to be vulnerable, as economic, social, cultural, and institutional barriers combine to result in low living standards. These groups depend particularly on cooperation from others. Identifying disadvantaged groups is a first step towards developing support strategies that prevent poverty, marginalization, and social disintegration. 6.2 Welfare i s a characteristic of individuals, not o f households. Deprivation can affect entire households or certain members within a household. Families draw both on joint household resources, like housing and land, and on individual receipts, such as wages. Common funds may be distributed unevenly within the family, and salary recipients may not redistribute these earnings to other members. If women receive systematically less than men, or children and old people are worse off than prime-age adults, we will be overstating distribution-sensitive welfare by assuming equal allocations. 6.3 In this chapter, we take a closer look at the TLSS evidence on social and economic inequities experienced by specific groups. We use demographic and family characteristics to categorize the p ~ p u l a t i o n , ~and investigate whether particular ~ household groups, or segments within a household, are especially disadvantaged. Information on household composition i s shown in Table 6.1. In most households, different generations live together.73 About nineteen in twenty individuals live in such families. Furthermore, over nine in ten persons reside in households with both prime- aged adults and children. The typical household structure i s a two-generational family with prime-aged adults and children. In addition, almost three in ten individuals live in three-generational households. Effectively all children stay with prime-aged adults, and two thirds live with elderly. The highincidence of multi-generational households implies that the fates o f different generations are closely intertwined. 71This chapter was written by KasparRichter. l2Poor health, migration, and ethnicity are other features that can define disadvantage groups. 73Children include all individuals less than 15 years of age, prime-age adults cover individuals betweenthe ages of 15 and49, and elderly are 50 years and older. 174 Table 6.1: DemographicHouseholdComposition Householdtype Percentage Three-generational 29 Two-generationalPrime & Children 56 Two-generationalPrime & Elderly 7 Two-generationalElderly & Children 2 One-generationalPrime 4 One-generationalElderly 2 Children with no parent deceased 89 Children with at least one parentdeceased 11 Source: 2001 TLSS, GENDER 6.4 Gender i s an important aspect in the debate on development. Policy researchers and development practitioners have begun building a body o f evidence and experience that links attention to gender in policies and projects to equitable, efficient, and sustainable outcomes in development. There i s growing evidence that societies that discriminate on the basis of gender tend to experience more poverty, slower economic growth, and a lower quality o f life than societies in which gender inequality is less pronounced. In all countries, but particularly in the poorest, giving women and men the same rights - allowing them equal access to education, jobs, property and credit, and fostering their participation in public life - produces positive outcomes, such as decreased child mortality, improved public health, and a strengthening o f overall economic growth. 6.5 Attempts to estimate the number of women living in poverty has generated a considerable amount of debate around the world. The main stumblingblock is the lack of an acceptable indicator for gender comparisons. The basic poverty measures inthis report are based on household resources, and incorporate the essentially arbitrary assumption o f equal distribution within the household. They do not capture any female poverty deriving from intra-household inequality. With this caveat inmind, it is nevertheless useful to ask whether, under this "conservative" assumption, there i s evidence for gender bias in poverty. 175 Table 6.2: Povertyand Gender National 0 to 6 7 to 14 15 to 49 50 or older Female Male Female Male Female Male Female Male Female Male Headcount 39.7 39.7 42.6 44.7 49.1 45.7 36.0 35.3 31.1 33.1 (3.0) (2.8) (3.5) (3.5) (3.8) (3.5) (2.8) (2.7) (3.6) (3.4) Poveity Gap 12.0 11.7 13.0 13.5 15.1 14.1 10.8 10.2 9.3 8.8 (1.4) (1.3) (1.7) (1.6) (1.8) (1.6) (1.3) (1.1) (1.5) (1.2) Severity 5.1 4.8 5.5 5.6 6.4 6.0 4.5 4.1 3.8 3.3 (0.7) (06) (0.9) (0.8) (0.9) (0.9) (0.7) (0.5) (0.7) (0.5) Memorandum items: Household size 6.0 6.1 6.3 6.4 6.7 6.8 5.9 5.8 4.5 4.9 Dependency ratio (%) 125 125 158 166 161 164 103 92 73 81 Populationshare 49 51 11 12 10 I 1 22 21 6 6 Sotme: 2001 TLSS 6.6 We display poverty statistics disaggregated by age and gender inTable 6.2. There are no significant differences across poverty rates. Household demographics differ little within each age category, implying that this result is robust to changes in equivalence scales.74This finding reflects that, even once age groups are distinguished, females do not live systematically indifferent households than males. The exception i s females 50 years or older, who live in smaller households with lower child share than males o f the same age-group. Upon closer inspection, it emerges that average female poverty increases relative to male poverty as we move from the crude headcount measure to more distribution sensitive measures. For the severity o f poverty, female statistics are consistently higher than male statistics, and the difference widens as we go from children to the elderly. For example, girls younger six years old or younger face 1 percent less severe poverty than boys, while elderly women experience 15 percent more severe poverty than elderly men. Nevertheless, due to high standard errors the difference remains statistically insignificant, and allowing for economies o f scale would narrow the gap. Overall, assuming equality inthe distribution o f household resources across gender, we find at best weak evidence that women face more severe poverty than men. 74By the same token, household demographics vary across age groups, so the poverty ranking across age groups i s affected bythe choice of equivalence scale, as discussed inChapter 1, Volume 11. 176 Table 6.3: Welfare and Gender 0 to 6 7 to 14 15 to 49 50 or older Female Male Female Male Female Male Female Male Immunization BCG 52.2 55.8 (4.1) (3.8) Polio 57.9 61.1 (4.3) (3.9) DPT 53.3 57.0 (4.9) (4.2) DPT3 8.3 9.1 (1.1) (1.4) Measles 51.7 49.0 (4.8) (4.5) Vitamin A 6.5 7.6 (1.2) (1.4) Health No health complaints last month 73.2 72.7 86.8 87.4 79.0 83.0 61.3 58.2 (1.7) (2.0) (1.8) (1.5) (1.5) (1.5) (2.5) (2.7) Sihjective health status (1 to 5) 3.97 3.91 3.85 3.91 3.56 3.60 (0.0) (0.0) (0.0) (0.0) (0.0) (0.0) Education Net Primary Enrollinent Rate 63.4 60.5 (1.9) (2.3) Net Primary Class Enrollment Rate 19.2 17.0 (1.8) (1.4) Schooling 82.1 77.1 47.9 66.2 2.9 12.8 (1.7) (2.0) (1.8) (2.0) (0.7) (1.5) Grade completed (1 to 6) 1.9 2.3 1.0 1.2 (0.0) (0.1) (0.0) (0.0) Literacy (%) 49.8 67.3 6.1 14.3 (1.8) (2.0) (1.0) (1.7) Subjective Welfare Happiness (1 to 5) 3.2 3.2 3.1 3.1 (0.0) (0.0) (0.1) (0.0) Change in living standard since violence (1 to 3) 1.8 1.8 1.9 1.8 (0.0) (0.0) (0.0) (0.0) Economic stahis (1 to 9) 2.4 2.4 2.2 2.4 (0.1) (0.1) (0.1) (0.1) Change in economic statns since violence (-8 to 8) 0.1 0.2 0.1 0.1 (0.1) (0.1) (0.1) (0.1) Power status (1 to 9) 3.8 3.9 3.5 3.1 (0.1) (0.1) (0.1) (0.1) Change inpower stahis since violence (-8 to 8) 2.1 2.2 1.9 2.0 (0.1) (0.1) (0.1) (0.1) Note: SIandard errors in parenihcsm Source: 2001 TLXS. 6.7 More evidence on gender bias is provided in Table 6.3. It focuses on non- consumption indicators, including immunization, health, education, and subjective well- being. Looking at these indicators notjust shades light on these dimensions of poverty. It also has conceptual advantages, as education and health measures are gathered for individuals rather than households (Case and Deaton 2002). We observe the actual value per individual, rather than havingto rely on anarbitrary assumption of equal allocation of household level resources across individuals, as we have to for household consumption. The precise set of indicators is aligned to the specificity of eachage group. 177 6.8 Before turning to gender differences, we consider the variation across age groups to see whether it accords to our expectations. Health indicators deteriorate as we move from children, to prime-age, and the elderly. Similarly, younger age-groups outscore older ineducation, inline with the broad improvement inschooling over the last decades. The same ranking is evident in the indicators o f subjective welfare. Overall, these regularities give us some comfort inthe quality o f the data. 6.9 For children under the age o f 7,boys tendto show higher immunization rates than girls, but the differences are not statistically significant. No consistent pattern in health emerges for the age group 7 to 14, while education indicators are better for girls than boys, but again the gap i s too small to be significant. For prime-age adults, men are better off than women both in health and education. Finally, the differences inhealth reveal no clear ranking for the old-age by gender, but elderly men are better educated and score higher in terms o f today's subjective well-being, while the evidence on the perceived changes over the last two years shows no consistent pattern. Overall, female adults are less well educated and perceive to have lower economic and power status, especially at old age, than their male counterparts. 6.10 This section documents that evidence on gender bias in Timor-Leste is mixed. First, women do not live in poorer households than men. This finding comes however with a strong caveat. TLSS provides no information on the gender allocation o f consumption within the household. More research into intra-household distribution is required to conclude that this household-level finding translates into an absence o f gender bias at the individual level. In addition, we also find little systematic differences across gender-age groups. Immunization rates are higher for boys, and education indicators better for girls, but the gaps are statistically insignificant. For adults, male educational standards are generally higher, which says more about gender inequalities inthe past than today. Finally, subjective indicators tend to rank men higher than women, especially for those 50 years or older, but the differences are small, and the evidence on changes since the violence inconclusive. FEMALE HEADSHIP 6.11 The analysis so far focuses on characteristics o f gender-age groups cutting across households. It does not capture deprivations linked to particular households features. One salient household characteristic i s the gender o f the household head. In this section, we focus on differences inwelfare between male and female-headed households. We want to explore, whether, as a result o f economic and perhaps cultural constraints, female-headed households experience lower welfare than male-headed households. 6.12 In Timor-Leste, cultural values in general, and traditions of family life specifically, are primarily based on catholic beliefs. In this context, female headship arises for two main reasons. First, some families have lost their male breadwinner as a result o f the years o f violence during the Indonesian period and the time o f the referendum. Second, women have a higher life expectancy than men. Overall, over 19 in 20 female heads are widows. 178 6.13 Both factors suggest that female-headed households have fewer household members than male-headed households, while the second aspect implies that female heads are on average older than male heads, and in turn are likely to have lower child shares. Overall, more than one in seven household heads are women. Female headed households are indeed smaller than male headed households (4.1 members relative to 6.3 members), so in terms o f population, about one in ten individuals live in households whose head i s a woman. For male headed households, seven in ten individuals have a head who is younger than 50. The corresponding number for female headed households is only 5 in 10. The child share inmale headed households is on average 20 percent higher than infemale headed households. Table 6.4: Povertyand Gender of the HouseholdHead National 0 to 6 7 to 14 15 to 49 50 or older Female Male Female Male Female Male Female Male Female Male liead head head head head liead head head head head Headcount 29.7 40.8 31.9 44.5 43.8 47.7 26.2 36.6 19.4 34.9 (3.9) (3.0) (5.8) (3.4) (6.0) (3.4) (3.9) (2.8) (3.4) (3.5) Poveity Gap 8.3 12.2 8.9 13.6 13.7 14.6 6.4 10.9 5.6 9.8 (1.5) (1.4) (2.3) (1.6) (2.6) (1.6) (1.3) (1.2) (1.1) (1.4) Severity 3.4 5.1 3.8 5.7 5.8 6.2 2.4 4.5 2.2 3.9 (0.8) (0.7) (1.1) (0.9) (1.4) (0.9) (0.6) (0.6) (0.5) (0.6) Memorandum items: Household size 4.1 6.3 4.8 6.5 4.7 7.0 4.1 6.1 2.9 5.1 Dependency ratio ("A) 126 125 213 159 197 158 82 99 68 79 Children (% household size) 38 46 61 57 59 56 28 40 18 27 Population share 10 90 2 22 2 19 4 39 2 9 Nnfe: Slandardermr.sinparentheses. S~nirce:2001 TLSS. 6.14 When taking on the role as household head, women can face difficulties if they have limited education and job opportunities. However, the figures shown in Table 6.4 appear to suggest that female headship is associated with lower poverty. Poverty is between one third to one half higher for male headed households, and the standard errors imply that the differences are significant. However, as the previous paragraph indicates, male and female headed households differ in size and composition. Especially, allowing for economies o f scale can reverse the ranking as male headed households are one third larger than female headed households. In addition, as male headed households have a higher child share, factoring in a needs-discount for children would further reduce the gap. For example, assuming economies o f scale o f 25 percent, and a cost ratio of children to adults o f one third, and the poverty headcount for both headship categories becomes equal. We conclude that the poverty ranltings o f male and female headed households are not robust to changes inequivalence scales across a plausible range. 179 Table 6.5: FemaleHeadshipand Welfare 0 to 6 7 to 14 15 to 49 50 or older Female Male Female Male Female Male Female Male head head head head liead head head head Immunization B C G 39.2 54.9 (10.0) (3.6) Polio 40.1 60.6 (10.0) (3.8) DPT 40.7 56.1 (10.2) (4.2) DPT3 5.6 8.9 (2.2) (1.0) Measles 37.5 51.1 (10.4) (4.2) Vitamin A 2.4 7.3 (1.4) (1.0) Health N o Iiealth complaints last month 70.7 73.1 83.7 87.5 77.2 81.4 55.3 60.7 (4.4) (1.6) (3.9) (1.4) (2.6) (1.4) (4.4) (2.3) Subjective health status (1 to 5) 3.9 3.9 3.8 3.9 3.3 3.6 (0.0) (0.0) (0.0) (0.0) (0.1) (0.0) Education Net Primary Enrollment Rate 57.1 62.4 (3.9) (1.7) Net Primary Class Enrollment Rate 12.8 18.6 (2.7) (1.2) Schooling 76.2 80.2 53.5 57.3 2.3 9.2 (3.7) (1.5) (3.4) (1.8) (1.1) (1.1) Grade completed (1 to 6) 2.1 2.1 1.0 1.1 (0.1) (0.0) (0.0) (0.0) Literacy ( O h ) 54.1 58.9 5.5 11.3 (3.5) (1.8) (1.7) (1.3) Subjective Welfare Happiness (1 to 5) 3.0 3.2 3.0 3.1 (0.1) (0.0) (0.1) (0.0) Change in living standard since violence (1 to 3) 1.8 1.8 1.9 1.8 (0.0) (0.0) (0.0) (0.0) Economic status (1 to 9) 2.2 2.4 1.9 2.3 (0.1) (0.1) (0.1) (0.1) Change in economic status since violence (-8 to 8) 0.2 0.1 0.0 0.1 (0.1) (0.1) (0.1) (0.1) Power status (1to 9) 3.7 3.9 3.3 3.6 (0.1) (0.1) (0.1) (0.1) Change inpower status since violence (-8 to 8) 2.1 2.2 1.9 2.0 (0.1) (0.1) (0.1) (0.1) NOIL': Standard e r n m inparenthcses. ,%u,urce: 2001 TLSS. 6.15 We turn towards broader notions o f well-being, covering education, health, and subjective well-being. Again, we note that indicators deteriorate as we move from younger to older age groups, in line with our expectations. The findings shown in Table 6.5 are surprisingly clear-cut: male-head households are better off than female headed households across all dimensions. Children under 6 in male-headed households have significantly higher immunization rates for all six indicators. Children of school age 180 report more health complaints and have worse educational indicators compared to those living in female headed households. The same holds for both prime age adults and the elderly. Finally, the subjective welfare indicators suggest that adults in male-headed households feel to have a higher economic and power status. Encouragingly, with regard to the changes in living standards since the violence, female-headed households score slightly better, eventhoughthe differences are not generally significant. 6.16 To summarize, classifying households by gender o f the householdhead brings out a clear pattern. Male-head households are consistently better off than female-headed households, with the exception o f consumption poverty - which again is subject to the caveat o f lack o f information on intra-household distribution. However, better welfare in male headed households may not be linked to gender bias, but in fact simply reflect that female-headed households are deprived o f one important breadwinner. The subsequent sections will shed more light on this issue, when we look at the welfare o f widows and children who lost their parents. WIDOWS 6.17 As discussed in the previous section, the large incidence o f female headed households is owed to Timor-Leste's violent recent past. Almost all female heads are widows, but about one third o f all widows are not head o f households. Widowhood not just alters the family structure, butoften changes the economic andsocial roles o fwomen inhouseholds and communities. Itcan affect the physical safety, identity andmobility of women and children, their access to basic goods and services necessary for survival, and their rights to inheritance, land and property. Widows may become responsible for her late husband's dependants, but she may also be taken in by his family. The death o f the main breadwinner can cause a breakdown in the familiar division o f labor because women take over roles traditionally carried out only by men. 6.18 Before we analyze welfare indicators, it i s instructive to review basic demographic information. Among married women (`wives') up to age 50, the average age at marriage i s 21 years. Most women get married betweenthe ages o f 15 to 25. Inthe following, we restrict attention to women aged 15 and older, and contrast welfare of widows and wives. Overall, three infive of women aged 15 or older are married and one insix widowed. Inorder to limitdifferences inthe average age across groups, we split the sample at the age o f 50. 181 Table 6.6: Poverty and Widowhood Aged 15 to 49 Aged 50 or older Married Widowed Married Widowed Headcount PovertyGap Severity Memorandumitems: Householdsize 5.8 4.2 4.6 4.4 Dependencyratio (%) 118 132 63 79 Children(YOhouseholdsize) 46 41 21 24 Age 33 37 56 61 Populationshare 50 5 9 11 Note: Population refers to women aged 15 or older. Standarderrors inparentheses. Source: 2001 TLSS. 6.19 For women younger than age 50, wives are poorer than widows, however, with the exception o f the headcount, the differences are not statistically significant (see Table 6.6). Furthermore, wives live in substantially larger households with slightly lower dependency ratio, even though the child share i s higher. Alternative assumptions regarding equivalence scales affect the ranking, in particular once we allow for economies o f scale. For example, assuming that the economies to scale are 25 percent instead o f zero, the poverty headcount for widows becomes 4 percent higher than for wives. Widows aged 50 or older display a slightly higher poverty gap and severity measure than wives o f the same age group. Again, the poverty ranking is not clear-cut, depending on the exact choices of poverty indicator andequivalence scale. 182 Table 6.7: Widowhood Status and Welfare Aged 15 to 49 Aged 50 or older Married Widowed Married Widowed Education Schooling (%) 1.2 (0.5) Grade completed(1 to 6) 1.o (0.01) Literacy (%) 4.8 (1.3) Health No healthcomplaints last month (%) 60.1 (3.3) Subjective healthstatus (1 to 5) 3.5 (0.1) Subjective Welfare Happiness (1 to 5) 3.0 (0.1) Change inlivingstandardsince violence (1 to 3) 1.9 (0.0) Economic status(1 to 9) 2.0 (0.1) Change ineconomic statussince violence(-8 to 8) 0.1 (0.1) Power status(1 to 9) 3.3 (0.1) Changeinpower status since violence (-8 to 8) 1.8 (0.1) Note: Standarderrors inparentheses Source:2001 TLSS. 6.20 Non-income dimensions o f welfare include education, health, and subjective well- being. A stark anduniform picture emerges. Wives are better and have experienced a larger improvement than widows (Table 6.7).76With regard to schooling, wives have more schooling, higher degrees, and are more literate than widows. Since these indicators are unlikely to change as a result o f widowhood, they suggest that widowhood affects disproportionately less educated women. The education indicators for the prime age group are higher than for the elderly, reflecting the large increase in school enrollment during Indonesian time. The absolute differences across widowhood status are larger for the prime age group, but the differences remain statistically significant even at old age. Wives report fewer health complaints and a (marginally) better subjective health status. Finally, they are happier, and enjoy a higher subjective economic and power status than widows. They also report a greater improvement since the violence in terms of general livingstandards andpower status. 75We cannot establish whether widows are worse off than wives as a result of the loss of their spouse or whether widows lived already indisadvantagedfamilies whentheir husbandwas still alive. 76We also find that elderly women, almost uniformly, are worse offthan younger women. 183 6.21 Women in Timor-Leste have shown commendable courage, resourcefulness and resilience in carrying on despite the trauma o f their loss, the isolation imposed by being widows and the difficult tasks o f earning a living and protecting themselves and their dependent family members. While many o f the complex implications o f widowhood cannot be adequately analyzed with the survey, the data shows that widows live in adverse circumstances resulting inlower welfare thanwives. PARENTLESSCHILDREN 6.22 The counterpart o f widows, from the point o f view o f the children generation, is boys and girls without living fathers. In any country, one o f the most disadvantaged groups is children without parents. In Timor-Leste, as a legacy o f a long history o f violent conflict, over one in ten children have only one or none living parent. The largest group i s the children without fathers, accounting for four infive o fthe children without at least on parent.77This part discusses the welfare o f parentless children. Table 6.8: Child Poverty and Parental Living Status Father and Father dead, Father alive, mother alive mother alive mother dead Headcount 51.2 (6.1) Poverty Gap 15.7 (2.9) Severity 6.8 (1.4) Memorandum items: Household size 6.7 5.2 5.8 Dependency ratio (YO) 159 202 180 Population share 84 6 3 Note: Children are all individuals less than 15years of age. Population refers to children. Standard errors in parentheses. Source: 2001 TLSS. 6.23 A simple way to identify the impact o f having lost a parent is to compare the welfare o f children with and without fathers and mothers. InTable 6.8, we separate three groups: those with both parents alive, those whose father has died and whose mother is still alive, and those whose mother has died and whose father is still alive.78 The categories represent 89 percent, 6.5 percent, and 3.5 percent o f all children under the age ~ ~~~ l7 Out of the children with both natural parents alive, more than nine in ten of these children live together with both ofthem, and almost all o f them with at least one o fthem. 78 Among the children below the age o f 15, 19 in 20 children have a living mother. 184 o f 15, re~pectively.~~ us first consider the two largest groups, children with both Let parents alive versus those with a living mother and a deceased father. Fatherless children live inhouseholds without the typical main breadwinner, so we expect highpoverty. This i s indeed the case. Child poverty rates are 13 to 17 percent higher for those without a living father than for those where the father has deceased. This ranking is robust to chances inthe equivalence scale. Children without fathers live insmaller households with a higher dependency ratio and child share than the other children. As a result, allowing for economics o f size or differences in needs by age groups leaves the ranking unchanged.80 6.24 Let us turn to children with a living father and a deceased mother. Poverty is slightly lower than for children with both parents. However, the differences are not statistically significant. Furthermore, changes in the equivalence scales reverse the ranking, since motherless children live in smaller households with higher dependency ratio." While there i s no clear pattern in the comparison to children with both parents alive, children without mothers are poorer than children without fathers. This result holds regardless o f the choice o f the equivalence scale. Overall, we find that fatherless children with living mothers are worse of than children with living fathers. For children with living fathers, no clear patternemerges relative to the living status o fthe mother. 6.25 Does this pattern carry over to other notion o f well-being, like education and health? Let us consider education first. Table 6.9 shows three educational enrollment indicators. They consistently show that children without either father or mother are worse off than children with both parents alive: they are less like to have received any schooling; have a lower net enrollment rate, both for primary school as a whole and for each primary school grade. The difference widens as we move from a coarse (ever attending school) to fine indicators (enrolled inthe grade corresponding to age). A similar picture emerges with regard to child health and immunization. Children with bothparents have fewer health complications during the last month, and for children less than 5 years o f age, all six variables indicate that higher immunization for children with fathers than without. No clear pattern emerges comparing children without fathers to children without mothers. ''We do not have a sufficient number o f observations on orphaned children (1.O percent of all children) to present reliable statistics. This result is confirmed insensitivity analysis with regardto equivalence scales. 81For example, assuming that the economies to scale are 25 percent instead o f zero, the poverty headcount for children without living mothers is 2 percent higher than for children with livingmothers. 185 Table 6.9: ChildWelfare and ParentalLivingStatus (%) Father and Father dead, Father alive, mother alive mother alive mother dead Education Schooling 63.6 (4.7) Enrolled inage-specific school 52.8 (3.9) Enrolled in age-specific grade 10.2 (2.2) Immunization BCG 27.8 50.6 (9.9) (19.0) Polio 36.6 50.6 (10.7) (19.0) DPT 47.2 39.9 (1 6.0) (16.9) DPT3 4.2 0.0 (3.0) (0.0) Measles 47.1 44.6 (15.5) (17.2) Vitamin A 0.0 1.8 (0.0) (1.8) Health No health complaints last month 40.5 31.6 34.3 (3.4) (5.1) (7.3) Note: Educationandhealth$gures consider childrenunder 15years.Immunization rates arefor children under 5years. Standarderrors inparentheses. Source: 2001 TLSS. 6.26 Our analysis shows that fatherless children experience more often, and deeper, poverty and lower welfare than children with living fathers.82 This examination is preliminary only and calls for more research to uncover the impact of child care arrangements on the welfare o f parentless and orphaned children. Nevertheless, these numbers suggest that the presence o f fathers lowers poverty. In addition, we find that with regard to education and immunization, parentless children, being either with a deceased mother or father, are consistentlyworse o f thanchildren with bothparents. ~~ 82This analysis does not establish whether parentless children suffer as a result o f the loss of parents or whether disadvantaged families were originally more affected by the death o f parents. 186 POLICY AND RESEARCH ISSUES 6.27 The analysis in this chapter confirms evidence from other countries. Female- headed households, widows, and parentless children experience severe hardship. More research is required to fully explore the complicated dynamics between family structure, community support, and welfare. Nevertheless, the results point to the need to develop a policy response. Possible interventions range from support to traditional community structures; transfers or income-generating activities to widows and households fostering fatherless children; targeted support for schooling and health care; and institutional care arrangements. 187 7. FOOD SECURITY INTRODUCTIONs3 7.1 Poverty means more than inadequate consumption, education, and health. It also means dreading the future. Living with the risk that a crisis may descend at any time, not knowing whether one will cope, i s part o f life for poor people. Poor people are often among the most vulnerable in society because they are the most exposed to a wide array o f risks. L o w income implies poor people are less able to save and accumulate assets, which inturn restricts their ability to deal with a crisis when it strikes. Poor people have developed elaborate mechanisms o f dealing with risk, but they often offer short-term protection at long-term cost, preventing any escape from poverty. 7.2 Risk is a pervasive characteristic o f life in developing countries. Different risks include natural and weather risks (for example, landslide, earthquake, drought), health risks (illness, disability, epidemic), economic risks (unemployment, resettlement), social risks (crime, civil conflict), environmental risks (pollution, deforestration), and political risks (coup d'etat). Some o f them affect an individual or household (illness, unemployment), others an entire village (drought), and yet others a nation as a whole (civil conflict). This distinction i s important, as, for example, a risk that affects an entire village cannot be insured solely within the village. While it is beyond the scope o f this report to discuss these aspects comprehensively, TLSS allows us to explore one issue o f vulnerability inmore detail: food security. PREVALENCE 7.3 Food security refers to assured access to enough food at all times for an active and healthy life. It includes the availability o f nutritionally adequate and safe foods, and a guaranteed ability to acquire acceptable foods in socially acceptable ways (without resorting to emergency food supplies, scavenging or stealing, for example). Perhaps the most important risk to food security for farming households in Timor-Leste is weather risk. Agriculture i s inherently dependent on the vagaries o f weather, like variation in rainfall. This leads to production (or yield) risk, and affects the farmers' ability to repay debt, to meet landrents, and, foremost, to provide adequate and sustainable food supplies. 7.4 Is the population exposed to food insecurity? At first sight, it would seem that TLSS has little to say on food security. Ideally, we would want to draw on nutritional and anthropometrical data collected over the entire course o f the year, covering the different stages o f the agricultural season. Yet, TLSS surveyed households only between late 83 This chapter was writtenby KasparRichter. 188 August to early December, and did not measure dietary intake or m a l n ~ t r i t i o n . ~ ~ However, to compensate for these data gaps, the survey included a range o f questions on the perception o f food security. While these subjective indicators raise questions with regard to the comparability o f responses,85 they nevertheless give instructive pointers bothto the extent and pattern o f food insecurity. Table 7.1: Food Security: Summaryby Domain National Urban Rural Rural Major Other West Center East Flat Mid High ~~~~~ Not enoughfood (# months) 3.6 1.8 3.7 3.9 3.7 4.2 3.9 3.9 3.9 (0.1) (0.2) (0.2) (0.3) (0.1) (0.2) (0.7) (0.2) (0.1) Enough food (#months) 6.7 9.1 5.8 6.0 6.1 6.8 6.5 6.1 6.4 (0.1) (0.2) (0.3) (0.3) (0.3) (0.3) (0.3) (0.3) (0.3) More than enough food (# months) 1.7 0.4 2.6 2.1 2.1 1.0 1.7 2.0 1.6 (0.1) (0.1) (0.4) (0.4) (0.3) (0.2) (0.5) (0.3) (0.3) At leastone monthnot enough food (%) 86 39 94 92 93 92 81 93 95 (1.5) (3.6) (2.2) (4.8) (2.4) (2.1) (8.6) (2.7) (1.8) Never more than enoughfood ("3) 54 87 36 39 44 73 52 48 53 (3.6) (3.1) (9.1) (9.2) (6.6) (4.9) (11.1) (5.9) (7.8) Not enoughrice or maize (# months) 3.6 1.8 3.7 3.9 3.8 4.2 3.9 3.9 4.0 (0.3) (0.1) (0.2) (0.6) (0.2) (0.1) Monthly food insecurity index (FII) (1 - 3) (0.1) (0.2) (0.2) 2.2 2.1 2.1 2.1 2.1 2.3 2.2 2.2 2.2 (0.0) (0.0) (0.0) (0.1) (0.0) (0.0) (0.1) (0.0) (0.0) FII(coefficient ofvariation, %) 27 10 33 31 31 24 26 30 29 (0.9) (1.0) (2.2) (1.7) (1.6) (1.3) (1.9) (1.4) (1.8) Note: .Y~urd"I errors hi pareiifhe.ses. Soiirce: 2001 TLSS. 7.5 A summary of self-perceived food security is shown in Table 7.1. The first five variables summarize the responses to food adequacy (not enough, enough, more than enough) for each month o f the last year. On average, the population got through 3.6 months with inadequate food during the last year, compared to only 1.7 months with more than adequate food. Not having enough food is a common situation for the population. It affected almost nine in ten persons for at least one month during the last year. At the same time, more than half o f the population did not experience a single month with more than enough food. 7.6 We can verify the estimate on the number o f months with insufficient food using a second question. It asked about the total number o f months at which the household did not have enough rice or maize to eat. The statistics, both for the country as a whole and in the regional breakdown, are very close to our first estimate. This suggests that in the perception o f the population, food security i s closely associated with having enough rice or maize to eat. 84 This omission was deliberate for two reasons. First, accurate collection of nutritional and anthropometrical indicators requires intensive special training of the enumerators, which appeared infeasible in view of time and budgetary constraints. Second, such information will be provided in the Demographic and Health Survey, plannedto be fielded during 2003. 85Forexample, is not having "enough" food equivalent to not meeting the required dietary norm? I s the notion of having"enough" food the same inurbanand rural areas? 189 7.7 H o w does food security vary across regions? The breakdown reveals a strong divide between major urban centers and the rest o f the country. Dili and Baucau experienced only 1.8 months o f inadequate food, whereas from other urban centers to the rural east, this number ranged from 3.7 to 4.2. Similarly, only four inten dwellers inthe major urban centers went through at least one month o f inadequate food, while this share i s more than four fifths for the rest o f the country. However, households in Dili and Baucau were less likely than families in other parts o f the country to have more than enough food. Inhabitants o f the two main cities had surplus food for less than half a month, compared to an average o f 1.9 months for the rest o f the country. Just over one in ten persons inthe major urban centers had more than enough food for at least one month, while inthe rest o fthe country every other person shared inthis experience. 7.8 Another perspective on the difference between Dili and Baucau and other parts of the country i s provided by the variability o f food security. Table 7.1 shows the mean and the coefficient o f variation o f the monthly food insecurity index (FII).86This index is a summary statistics on food availability over the course o f the year, and its coefficient o f variation measures its variability. Its average percentage deviation from the average is 27 percent. Inmajor urban centers, the average deviation i s only 10 percent, compared to 30 percent in the rest o f the country. Inother words, the variability o f food security i s three times as highinrural areas and other urban centers than inDili and Baucau. 7.9 These findings point to the following split: major urban centers have typically constant access to just enough food throughout the year, while other parts o f the country face greater fluctuation in food availability, and experience food shortage about twice as often as food excess. One interpretation o f this evidence i s the difference in capacity to keep consumption constant over the year. The greater reliance on non-agricultural income sources allows households in Dili and Baucau to keep consumption constant at an adequate level across the year. Two factors could explain this ability o f consumption smoothing. First, urban incomes are likely to be less variable as they depends less on the agricultural seasons. Second, as they receive a higher share o f income in cash, city dwellers may be able to engage more insaving and dis-saving o f income. 86 For a given month, this index equals 1for more than enoughfood, 2 forjust enough food, and 3 for not enough food. FIIincreases infood insecurity. P 90 Figure 7.1: Household Food Security by Month 100 80 60 40 20 0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec High E3 Average 0 Low 1 Source: 2001 TLSS. 7.10 The previous discussion suggests that, at least outside Dili and Baucau, some month are characterized with plentiful food while in others there is a shortage. InFigure 7.1 we trace the three responses on food availability over the course o f the twelve months preceding the survey. The striking feature i s the strong seasonality. Food security is lowest from November to February, with over two thirds o f the persons not having enough food. Incontract, from April to August, less than one intwenty individuals suffer from insufficient food. Food shortages are linkedto the harvest cycle, as they are greatest at the end o f the rice harvest and before the maize harvest. Rice i s harvested from April to October, while the maize harvest starts only inFebruary and lasts to around April. 191 Figure 7.2: Not EnoughFood by Domain 90 80 70 e, 60 4 3 50 $ a 40 30 20 10 n " Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Source: 2001 TLSS. 7.11 Does the seasonal pattern o f food security carry over at the regional level? Figure 7.2 shows the percentage shares stating not to have enough food for each month o f the year. It reveals the remarkable homogeneity o f the food availability circle across regional domains. In urban and rural categories alike, food security is lowest from November to February. The figure also highlightsthe split between Dili and Baucau and the rest o f the country. While the vast majority o f households in the rural domains and other urban centers have not enough food inthe lean season, the percentage o f persons suffering from not enough food inthe major urban centers rises never about 30 percent. 7.12 Outside Dili and Baucau, the typical household is a subsistence farmer, with little market access and non-farm sources of income, dependent on its own crop production for food provision. This evidence illustrates that across regions, these farmers engage incrop production with broadly similar seasonality, leading to a comparable monthly food security profile. 7.13 To summarize, subjective assessments of food adequacy suggest that food insecurity is widespread. Close to nine in ten persons experience inadequate food provision at some point during the year, while fewer than one intwo have too much food during any month in the year. Food security is closely tied to having enough rice and maize. Food shortages are aligned with the harvest cycle at the national and regional level. They are greatest during November and February, at the end o f the rice harvest and before the maize harvest. Major urban centers have typically access to just enough food all throughout the year, while other parts o f the country face greater fluctuation in food availability, and experience food shortage about twice as often as food excess. 192 FOOD SECURITYAND POVERTY 7.14 Agriculture i s o f overwhelming importance for living standards. About seven in ten persons live with heads o f households who work on a household farm, and over three quarters are with heads whose main occupation i s farming. Given this dependence on the moods o f agricultural seasons, what i s the implication o f the intra-year cycle o f food security for poverty? Figure 7.3: NationalPovertyandInterview Date 60 g 40 h 0 15 40 65 90 115 Day of interviewperiod Source: 2001 TLSS. 7.15 Inorder to explore the impact of seasonality onpoverty properly, we would need to draw on consumption data covering both lean and harvest seasons. TLSS was fielded over a period o f about four month, lasting from about mid-August 2001to mid-December 2001. The subjective food security indicators showed that in 2001 food availability was closely aligned to the harvest cycle. August was the last month o f the plentiful season, and lack o f food became more severe from September until the end of the year, and had its peak inJanuary. On the basis o f this pattern, we would expect poverty to show broadly an increase from early inthe survey to the end o fthe survey. InFigure 7.3, we display the national pattern, linking the average poverty headcount to a count of the days in the survey. We find indeed a strong dependence o f the poverty headcount to the timingo f the interview. Fewer than one inten persons live below the poverty line at the beginning of the survey. The share o f the poor rises continuously untilabout three month after the start o f the survey, or about mid-November, peaking at about 45 percent. This share then remains fairly constant duringthe last month. 193 Figure7.4: RegionalPoverty and Interview Date 60 1 I Outside DiliBaucau __ DiliiBaucau _++- - 1- c - c + _ . . _ + + j + 0 15 40 65 90 I15 Day of interview period Source. 2001 TLSS 7.16 Does this national picture also hold at the regional level? In our analysis o f food security, we found a different food security profile for the major urban centers than for other parts o f the country. In Figure 7.4, we display separate plots for Dili and Baucau and the rest o f the country.87Again, these two parts o f the country show marked differences. Poverty in Dili and Baucau i s overall much lower, and starts rising from the second month of the survey period onwards, and increases right through until the end o f the survey period. It confirms that even Dili and Baucau are affected by the lean season. The delayed rise in poverty could point to a greater, even though imperfect, capacity to smooth consumption. 7.17 This strong evidence for seasonality o f poverty raises an immediate question. In the analysis o f the poverty profile, we argue that about two fifth o f the population live below the poverty line. In view o f the intra-year fluctuations o f living standards, this estimate is specific to the survey period. H o w representative is therefore this poverty rate o f 40 percent for the year as a whole? In the absence o f information o f consumption behaviour throughout the year, we have to rely on subjective food security for a rough assessment. We compare the average value o f food security for the survey period with the annual average. Taking as weights the percentage shares o f interviews conducted in August, September, November, and December, we calculate that the share o f not having enough food for the survey period is 30 percent. The annual average for this variable is 34 percent. Overall, this comparison suggests that the "survey" poverty rate is fairly close, and possibly slightly higher, to the "annual" poverty rate. 87 Separating out domains also serves as a cross-check on the finding of an increase in poverty over the survey period. For example, a sequencing of interviews first in urban areas (with low poverty) and then rural areas (with high poverty) could have produced such a spurious relationship between poverty and interview date. The sequencing of TLLS interviews across the domains was designed to be broadly representativeat differentmonths. 194 COPINGWITH FOOD SHORTAGE 7.18 What happens when a family i s faced with a risk o f food shortages? And how does a householdrespond to a food crisis? Farmers have always been exposed to weather risks, and for a long time have developed ways o f reducing, mitigating, and coping with these risks (Besley 1995, Dercon 2002). Traditional risk management covers actions taken both before ("ex-ante") and after ("ex-post") the risky event occurs (Siege1 and Alwang 1999). These strategies are often costly, as they lower vulnerability in the short term at the expense o f higher vulnerability over the longer term: a farmer's decision to undertake, or not to undertake, a certain activity is not just dependent on achieving the highest expected return but also on the variance o f the returns. In particular, to limit vulnerability, farmers give up higher income from specialization in return for a lower variability o f income. We can draw on the survey to explore the relevance o f various actions inthe context of Timor-Leste. Ex-Ante Strategies 7.19 Farmers reduce and mitigate the risk o f a food shortage before it occurs. At the household level, such ex-ante strategies range from the accumulation o f buffer stocks as precautionary savings, to varying cropping practices (planting different crops, or in different fields, staggered over time, inter-cropping, and relying on low risk inputs), and to the diversification o f income-generating activities (working in farm and non-farm small businesses, and seasonal migration). At the community level, villages mitigate food insecurity with irrigation projects and conservation tillage that protects soil andmoisture. 7.20 InTable 7.2, we pull together key variables that characterize ex-ante household coping strategies. It provides information on the main assets (savings, livestock holdings, and land), cropping patterns, andjob holdings. We distinguish DiWBaucau from the rest o f the country. We also separate households who experienced in the last year food shortages for more than a third o f the year ("food insecure") from those who suffered no more than four months o f lack o f food. About one in five persons were food insecure in DiWBaucau over the last year, compared to one inthree inother parts o f the country. 7.21 I s food security associated with more ex-ante coping? In DiWBaucau, the distinguishing feature o f households in terms o f food security is being employed in the non-agricultural sectors. By contrast, outside Major Urban Centers, dependence on agriculture is almost universal, and food security is related not just to and being more diversified, but also having more assets and outputs, interms o f savings, livestock, crops, andjobs. 195 Table 7.2: Food Security and Ex-AnteCopingStrategies Sample DilirSaucau Other domains Secure Insecure Secure Insecure Savings Y O All 22 10 54 44 (4) (4) (5 1 (6) USDollars per capita Assets holders 87.9 11.0 35.0 18.8 (115,514) (48,391) (31,820) (21,982) Livestock holding Y O All 68 90 89 (8) (2) (3) # per capita Livestock holders 1.4 2.3 2.0 (0) (0) (0) USDollars per capita Livestock holders 0.070 0.114 0.088 (152) (131) (114) Land Y O All 19 57 95 94 (3) (9) (1) (1) Ha per capita Land holders 0.21 0.29 0.31 0.32 (0.04) (0.06) (0.02) (0.02) US Dollars per capita Land holders 0.505 0.326 0.727 0.829 (1,235) (532) (406) (675) Crops YO All 19 51 95 93 (3) (9) (1) (2) # per capita Crop holders 0.68 0.65 0.85 0.78 (0.07) (0.05) (0.03) (0.04) YOselling Crop holders 46 56 72 55 (8) (9) (4) (6) Jobs # per capita All 0.23 0.27 0.37 0.33 (0.01) (0.02) (0.01) (0.02) %agriculture Job holders 17 53 84 89 (3) (11) (2) (1) %secondaryjobs Job holders 4 2 8 7 (1) (2) (1) (1) Shares 82 18 67 32 Note: Securerefers to households who experiencedfood shortagesfor at mostfour months during the lastyear, and znsecure to all other households. Standard errors inparentheses. All Rupiah valuesfrom the survey were convertedto USDollars using an exchange rate of 10,000RupiaWUSDollar. Source: 2001 TLSS. 196 Ex-Post Strategies 7.22 The survey provides also information on actions that households undertook in response to a food shortage. Table 7.3 shows the ex-post actions taken by families when faced with lack o f food. Household heads were asked to give up to three responses, ranked by degree o f importance. Almost all families (99 percent) reported two actions, and close to 90 percent three actions. The need to resort to multiple strategies i s in itself an indication o f vulnerability. The number o f coping strategies is linked to poverty: o f those engaged in at most two actions, only one infour are poor, compared to almost one intwo for those reportingthree strategies. Table 7.3: Coping StrategiesWhen Not Enough Food By relevance Overall First Second Third Dili/Baucau Other domains Ate less food Changeddiet Sold livestock or assets Borrowedmoney Got food aid Note: The responses by relevance do not sum to 100 due to the omission of the "Others" category. Standard errors inparentheses. Source: 2001 TLSS. 7.23 Separating out the coping strategies suggests a sequencing o f responses. At first, the household head experiences anxiety about food insufficiency, leading to decisions to reduce the household's food budget by altering the quality or variety o f food consumed by the family. Overall, almost all households either change their diet or skip meals when faced with insufficient food (see Table 7.3). These two actions were not just most widespread, but also took priority over other responses. 7.24 Only if the situation required further adjustment, then households also undertook distress sales o f livestock and other farm assets. Every other household reported this response, most o f them as third action. Selling productive assets i s clearly a last resort. It make ends meet today at the cost o f lowering the future income stream. Furthermore, it requires having marketable assets in the first place. For example, only one quarter of those without livestock holdings reported asset sales, compared to over half for those owning animals. 197 Table 7.4: CopingStrategies: IntrahouseholdTransfers Percentage Number % % same us %of oftransfers relatives posto Dollars expenditure Donors Grants Loans Recipients Grants 8.9 3.4 97 61 2.09 7.6 (0.9) (0.4) (7) (5) (3,273) (1.2) Loans 12.1 1.4 64 n.a. 2.42 11.3 (1.3) (0.1) (5) n.a. (8,792) (0.4) Donors or recipients Total 30.8 (1.9) Non-poor 34.2 . (2.0) Poor 25.7 (2.6) Note: Standard errors inparentheses.AI1 Rupiah valuesfrotit the survey wereconverted to USDollars using an exchange rate of 10,000 Rupiah/US Dollar. Source: 2001 TLSS. 7.25 Other strategies played a minor little role. Private transfers are informal ways in which individuals exchange cash, food, and clothing, informal loans and assistance with work and child-care. Only about one in fifth families obtained resources from friends, relatives, and neighbors. Over half o f the households receiving private transfers state this only as the third line o f response. The limitedrole o f private transfers, especially for poor and vulnerable families, is confirmed by the evidence presented in Table 7.4. It gives summary statistics on grants and loans received and given over the last twelve months. Overall, three in ten persons live in households that were engaged in either given or receiving transfers. The amount o f monthly net transfers (grants and loan received minus those given) was minor, totaling on average less than 3 percent o f householdexpenditures among those giving and/or receiving. Net transfers accounted for more than 10 percent o f household expenditures for only about one in thirty persons. The vast majority o f transactions i s among relatives, and transfers occur predominately among households livinginthe same posto. Private transfers are more widespread and frequent, and larger in both absolute amount and relative to expenditure among the non-poor than among the poor. Food aid, either from government, NGOs, or the international community, was irrelevant- only one ina hundred persons benefited from such relief. 7.26 The overwhelming importance o f dietary adjustments compared to reliance on asset sales and support from others or is also related to the nature o f the risk. Food insecurity is related both to the agricultural cycle and weather-related production risks, 198 and i s a "covariate" risk. It concerns many households in a community or region at the same time. Under great stress, informal arrangements tend to break down, as the members o f the community, or "risk pool", are commonly affected. The income o f the village as a whole i s reduced, triggering a collapse o f community-based informal insurance arrangements (Morduch 1998). For example, as farmers attempt to sell livestock to make ends meet after a drought, livestock prices will fall as supply outstrips demands. Similarly, when farmers seek off-farm employment in response to a natural disaster, the suddenrise in labor supply will drive down market wages. Furthermore, the family's neighbors and friends are faced with the same negative income shock, and are likely to be reluctant or incapable to provide loans or grants to them. Table 7.5: Coping Strategies When Not EnoughFood: Who Suffers? By relevance Overall Population First Second Third Head Wifelhusband Children Grandchild Niecelnephew Fatherlmother Sisterlbrother Soddaughterinlaw Brotherlsister inlaw Others Note: Standard errors inparentheses. Source: 2001 TLSS. 7.27 When households cut back on meals or change nutrition, who suffers the most? The survey asked families to identify up to three household members, who are affect most incase o f a food shortage. The responses are shown inTable 7.5. The striking result i s that children appear to take the brunto f the adjustment. They account for between three fifth to three quarters of the three most affected individual, even though they represent just over half o f all household members. Since malnutrition at young age can lead to long-term health problems, this points to a potentially permanent detrimental consequence o f even occasional food shortages. 199 POLICY AND RESEARCH ISSUES 7.28 Subjective assessments o f food adequacy suggest that food insecurity is widespread. Food availability is aligned with the harvest cycle at the national and regional level. Major urbancenters typically have access to just enough food throughout the year, while other parts o f the country face greater fluctuation infood availability, and experience food shortage about twice as often as food excess. Food insecurity duringthe lean seasons i s also associated with higher poverty. Households have multiple ways of dealing with food insecurity, which often lower vulnerability in the short term at the expense o f higher vulnerability over the longer term. Almost all households either change their diet or skip meals when faced with insufficient food -to the detriment o f especially children. 7.29 These findings call for more survey work explicitly designed to capture the temporal dimension o f food security and poverty, and to investigate household coping strategies. Understanding the underlying causes o f food security (lack o f cash incomes which allow households to purchase food during periods o f shortfall, lack o f availability of food in markets, or lack o f storage) would help design appropriate policies. Overall, policies should be aimed at helping poor people manage risk better by reducing and mitigating risk and lessening the impact or shocks. They comprises multiple measures, rangingfrom developing human resources, improving access to productive resources and remunerative employment, expanding markets, infrastructure, and institutions, to sound governance and trade and macroeconomic policies. 200 DETERMINANTSOFPOVERTY INTRODUCTION" 8.1 Poverty has many causes - economic, demographic, social and cultural. As the analysis in this report has confirmed, this i s true also for Timor-Leste. H o w can we disentangle the impact of these factors on poverty? Although two-way tables are informative about associations between factors, they cannot answer the key question whether these relationships hold up when other influences are held constant. For example, there i s a clear correlation between the education o f the household head and poverty. But this link could be due to third factors related to both education and poverty, like occupation or household assets.89 8.2 The standard tool to address this issue i s to conduct a multivariate analysis of the determinants o f living standards. Such examination can be helpful in identifying correlations between variables, such as those between consumption, characteristics of the household head, household demographics and assets, and community features. In this section, we analyze the determinants of one particular dimension of living standards: household consumption per capita and the implied probability o f being consumption- poor. 8.3 For the interpretation o f these results, it is important to distinguish between the characteristics o f the poor and the roots o f poverty. For example, the analysis shows that the poor live in rural areas and with household heads whose human capital is low. This finding as such does not yet identify a cause ofpoverty, but explains part o f the variation inper capita consumption, taking as givenpast household demographics, human capital, physical assets, and community characteristics. The estimates do not account for the process by which households or communities acquired these features. It i s dangerous to conclude those characteristics are causes andthen draw policy conclusions. MODEL 8.4 Inthis section, we describe the basic approach to modeling the determinants of poverty. We adopt a two-step procedure." First, we regress the log o f real per capita consumption on a range of determinants: ''''Thischapter was written by Kaspar Richter. However, insome cases, like geographical targeting, simple profiles without controllingfor other factors can be more useful (Ravallion 1996). This approach follows Chaudhuri (2000), Datt andJolliffe (200l), Hentschel et a1(2000), IFPRI(1998) and Ravallion(1996). 201 Inc,i = a p'x,; + +...+p'x,; + E / where c , ~i s real per capita consumption o f household j, x k i s a set o f K (k=I, .,.,K) household and community characteristics, and &,ji s a normally distributed random error term with mean zero and constant variance, capturing unobserved variables. Inthe second step, we derive fromthis regressionthe predictedpoverty headcount: where z denotes the poverty line equal to Rp154,374, or US$0.51 in current exchange rates.91 92 We allow for regional differences by estimating the regression separately for the five core analytical domains Dili/Baucau, Other Urban Centers, Rural West, Rural Center, and Rural East.93 8.5 In order to estimate the regression, we have to specify the determinants of consumption. The selection o f variables i s driven by five considerations. First, the empirical analysis is obviously limited to factors that are observed and measured in the TLHS and the Suco Survey. As such, it cannot identify all of the various determinants and correlates o f poverty. In particular, the role o f exclusion and social capital in promoting poverty cannot be adequately analyzed due to gaps inthe available data sets. Second, the bivariate analysis on the welfare profile suggested a number o f key drivers for consumption and poverty that we should take account o f in the analysis. Thirdly, we can only include variables that are arguably exogenous to current c o n ~ u m p t i o nIn~ . ~ particular, we do not include detailed housing characteristics, as they determine actual or imputedrents which are one component of the consumption aggregate. Fourth, we also include a set o f community level determinants, both at the Aldeia (12 variables) and Suco level (10 variables). This not only ensures that the household level factors are purged from observed community-level determinants, but it also allows us later to simulate the impact o f community level variables on household con~umption.~' Fifth, we also allow for interactions between factors, but in most cases the parameter estimates become more imprecise due to collinearity with other variables, so we limit the number o f interactions included. 8.6 The determinants can be grouped into the following categorie~:~~ 91 An alternative approach would be to directly link the poverty headcount, a binary variable, to the explanatory factors, but this procedure does not exploit fully the information contained inconsumption. 92 The same approach can be used to derive alternative poverty measures, like the poverty gap or the severity o fpoverty. 93We test for equality o f parameter estimates across the three rural and two urban groupings, and strongly reject this homogeneity hypothesis. 94Correlation betweenthe explanatory variables and error terms leads to inconsistent parameter estimates. 95 An alternative approach is to include indicator variables (fixed-effects) at the community level. While this would control for both observed and unobserved community time-invariant effects, we would no longer be able to identify the impact of specific community level factors on consumption. 96The community level determinants were taken from the Suco Survey. 202 a. Household demographics: household size (number o f persons) and number of persons inthese age groups (under 6, 7 - 14, 15 - 49, and 50 plus). b. Head characteristics: gender, age, age squared, five education categories (no schooling, lower primary (year 1 - 3), upper primary (year 4 - 6), lower secondary, andpost-lower secondary (including university)), and six occupation categories (housework, farmer, non-farm worker, trader, teachedcivil servant, and other). C. Spouse characteristics: indicator variable for spouse present, age and age squared, and the five education categorie~.~~ d. Agriculture and assets: value o f total crop production, livestock holdings, and savings, all in Rupiah per capita; land holding per capita (hectare); and three indicators for crop mix (coffee, rice, andmaize). e. Housing: indicator variable for house ownership, and number o f years lived inthis dwelling. J: Infrastructure: three indicator variables on household access to safe drinkingwater, sanitation, and electricity. g. Access: minutes from dwelling to paved road, indicator whether this road i s accessible during the rainy season, and distance inkilometer from aldeia to suco center (from Suco Survey). h, Aldeia: twelve indicator variables on community facilities (primary school, secondary school, health center, church, kiosk, shop, everyday market, periodic market, bank, mill, vehicle passable road, pavedroad). i. Suco: indicator variable on irrigation, also interactedwhether household is rice producing; indicator variable on presence o f major private employer (more than five employees); ratios of number o f teachers per student and number of classrooms per teacher; ratios o f number of midwives and traditional birth attendants per population and days in month o f operating health service per population. j. Community Leaders: average characteristics o f respondents in Suco Survey in terms of years o f age, years o f education, and years lived in suco. ESTIMATION RESULTS including the omitted categories for the categorical variable^,^' are shown in Table 8.1. 8.7 The parameter estimates and t-statistics o f the 60 regressors for the five regions, The fit of the overall regression equals to 0.61, implying that the model explains three 97No spouse gender indicator is included as there are only three households with female heads and male spouses. Spouse occupation characteristics were generally insignificant and therefore omitted. Allowing for spouse characteristics is very similar to interacting head characteristics with the head's gender indicator. 98Coefficients of categorical variables have to be interpreted relative to the omitted group. 203 fifth of the observed variation in con~umption.~~The model has largest explanatory power in Rural West, and lowest accuracy in Rural East."' The statistical significance andcoefficients differ across regions andvariables, butmost signs are consistent withthe regularities emphasized inthe bivariate analysis. With the dependent variable specified in natural logarithm, the coefficients measure the percentage change in real per capita consumption from a unit change in the right-hand side variable, keeping other factors constant. We now turnto a brief discussion o fthe empirical findings. 8.8 Demographic factors have an important influence on consumption in both rural and urbanareas. Larger families are worse off than smaller families."' As expected, for a given household size, consumption declines with more children. In terms o f head characteristics, male gender has inthree o f the five regions a positive sign but is only in Other Urban Centers statistically significant.lo2 The picture on the impact o f age also varies from region to region. Education has the expected effect, with especially primary educationboosting consumption. Farmers tend to have lower living standards thantraders and civil servants, but again the picture differs between regions. Spouse characteristics have a similar impact as head characteristics, although the coefficients, including for education, are generally less significant. 8.9 Agriculture and assets are clearly essential in a subsistence economy. Crop production and land holdings have typically the expected signs, but are not always significant. The indicator variables on coffee, rice, and maize matter, suggesting that, for given asset endowment, coffee production boosts consumption, while maize, as a low value crop, tends to be associated with lower living standards. For most regions, consumption rises with more livestock and savings. Ownership o f housing is a better predictor for consumption inrural than inurban areas, while the number o f years lived in the same dwelling is mostly important in Other Urban Areas. Regarding infiastructure, sanitation and electricity have a more consistent and larger impact than safe drinking water. Close access to a paved road i s in four o f the five cases associated with higher consumption, but it is significant only in Rural West. Inaddition, there i s evidence that, inRural Center andRuralEast, aldeias far removedfrom suco centers are worse off. 8.10 Finally, community variables determine consumption. The coefficients and signs o f the aldeia facility indicators, capturing school, health, church, and economic infrastructure, depend on the region, and are volatile due to collinearity, but they are jointly highly significant. The picture i s similar for suco infrastructure, which accounts for irrigation, presence o f private employers, and education and health service variables. There i s also evidence that older community leader are related to higher living standards inruralareas. 99 This R2 applies to the pooled regression, where the xi variables are interacted with the regional indicators. In order to fully replicate the results o f the separate regional regressions, we adjust the variance o fthe residual inthe pooled regression to differ by region. looThe regional R2varies from 0.64 in Rural West, to 0.61 in Other Urban Centers, to 0.59 in Rural Center, to 0.48 inDiWBaucau, and to 0.46 inRural East. This result holds for plausible alternative values ofequivalence scales. This confirms the results ofthe bivariate analysis on gender ofheadship. Itis important to remember that there is evidence for large differences on other welfare dimensions. 204 Table 8.1: OLS Regressionson LogPer CapitaConsumption DiliBaucau Other Urban Rural West Rural Center RuralEast Coef. t-stat Coef. t-stat Coef. t-stat Coef. t-stat Coef. t-stat Householddemographics Size (#) -0.073 -4.9 -0.061 -1.9 -0.083 -3.7 -0.010 -0.5 -0.052 -2.0 Age groups # 15-49 (omitted category) # 0-6 -0.004 -0.1 -0.080 -1.8 -0.053 -1.4 -0.142 -5.1 -0.090 -2.3 # 7-14 -0.091 -3.1 -0.037 -0.8 -0.037 -1.0 -0.082 -2.9 0.015 0.4 # 50 plus 0.014 0.2 -0.163 -2.0 -0.012 -0.2 -0.079 -1.6 0.031 0.5 Head Male 0.018 0.1 0.319 1.7 -0.038 -0.3 0.113 1.2 -0.190 -1.3 Age (years) 0.037 2.0 -0.019 -1.1 0.009 0.6 0.013 1.1 0.024 1.5 Age square 0.000 -2.2 0.000 1.0 0.000 -0.5 0.000 -1.1 0.000 -1.2 Head's education No schooling (omitted category) Lower primary 0.286 2.1 0.354 2.8 0.195 1.8 0.187 2.4 0.073 0.8 Upper primary 0.445 4.7 0.234 2.2 0.118 1.5 0.104 1.5 0.143 1.4 Lower secondary 0.461 3.6 -0.202 -1.2 0.040 0.3 0.145 1.7 -0.028 -0.2 Post lower secondary 0.433 3.7 0.102 0.7 0,110 0.8 0.254 2.3 -0.098 -0.6 Heads occupation Farmer (omitted category) Housework -0.094 -0.4 0.216 0.8 -0.250 -1.4 0.340 1.4 -0.022 -0.1 Non-farm worker -0,037 -0.3 -0.153 -0.7 -0.247 -0.9 0.434 3.2 0.353 1.6 Trader 0.096 0.7 0.719 3.3 0.249 1.2 0.099 0.5 0.292 0.6 TeacheriCivil servant -0,011 -0.1 -0.051 -0.3 0,142 0.8 0.178 1.4 0.314 1.4 Other 0.105 0.9 0.160 1.4 0.129 0.8 -0.027 -0.3 -0.132 -1.1 Spouse Spouse 0.282 0.6 0.340 0.5 -0.330 -1.0 -0.398 -1.1 -1.169 -2.0 Age (years) -0,013 -0.5 -0,011 -0.3 -0.015 -0.8 -0.030 -1.7 -0.058 -2.1 Age squared 0,000 1.0 0,000 0.9 0,000 0.7 0.000 1.9 0,001 1.9 Spouse's education No schooling (omitted category) Lower primary 0.061 0.4 0.184 1.0 0.002 0.0 -0.058 -0.6 0.021 0.1 Upper primary 0.053 0.5 0,176 1.5 0.184 1.7 0.095 1.2 0.149 1.3 Lower secondary 0.102 0.9 0.571 3.9 0.050 0.4 0.136 1.4 0.407 2.9 Post lower secondary 0.418 3.7 0.444 2.6 -0.180 -1.2 -0.018 -0.1 0.597 3.5 Agriculture andAssets Coffee? 0.894 2.7 0.117 0.8 0.187 1.2 0.239 4.1 0.036 0.1 Rice? 0.247 1.1 -0.125 -0.9 -0.166 -1.3 -0.112 -1.1 -0,029 -0.4 Maize? -0,145 -1.3 0.043 0.3 -0.001 0.0 -0.241 -3.8 -0.063 -0.7 Crop value per capita (Rp) 0.000 -0.2 0.000 0.9 0,000 1.2 0.000 2.6 0,000 0.2 Land per capita (has) 0.001 0.4 0.027 0.3 0.140 2.0 0.005 1.4 0.003 0.5 Animal value per capita (Rp) 0,000 2.6 0,000 -0.3 0,000 0.9 0.000 2.8 0.000 1.4 Savingsper capita (Rp) 0.032 0.4 0.022 0.2 0.102 1.3 0.299 6.0 -0.031 -0.4 Housing Owned? -0,001 0.0 -0.072 -0.6 -0.235 -2.0 -0.169 -1.7 -0.171 -1.0 Years lived -0.002 -0.4 0.022 4.1 0.019 3.1 0,005 1.6 -0.004 -0.8 205 DiliiBaucau Other Urban Rural West Rural Center Rural East Coef. t-stat Coef. t-stat Coef. t-stat Coef. t-stat Coef. t-stat Infrastructure Safe drinkingwater? -0.056 -0.7 -0.042 -0.5 -0.075 -1.4 0.038 0.9 -0.075 -1.0 Sanitation? 0.250 2.7 0.107 1.3 0.159 2.5 -0.100 -2.1 0.207 2.8 Electricity? 0.478 3.6 0,175 2.0 0.088 0.7 0.278 3.3 0.076 0.8 Access Pavedroad (mins) -0.002 -0.7 0.000 -0.1 -0.007 -1.8 0.000 0.1 0.000 -0.3 Road accessible during rainy season 0.008 0.1 -0.009 -0.1 -0.023 -0.2 0,008 0.1 0.079 0.8 Suco center (kms) -0.003 -0.1 -0.018 -1.1 0.001 0.1 -0.0 I5 -2.5 -0.066 -4.1 Aldeia Primary school? 0.127 1.2 0.171 1.4 0.046 0.3 -0.076 -1.2 0.230 2.5 Secondary school? -0.146 -1.4 0.011 0.1 0.133 0.9 -0.192 -1.9 0.396 3.0 Health center7 -0.174 -1.7 0.244 1.5 0.223 1.6 -0.110 -1.4 -0.496 -3.2 Church? 0,053 0.7 -0.010 -0.1 0.091 1.o -0.086 -1.2 -0.050 -0.5 Kiosk? 0.225 0.8 0.367 1.3 0.268 1.o -0.065 -1.0 0.117 1.3 Shop? 0,128 1.4 -0.055 -0.3 0.022 0.1 0.336 1.1 n,a. n.a. Everyday market? 0.019 0.2 -0.030 -0.1 n.a. n.a. -0.178 -0.8 -0.598 -3.2 Periodic market? 0.310 2.1 0.028 0.2 0.012 0.1 0.469 6.0 -0.023 -0.2 Bank? -0.134 -0.5 0.390 1.3 -0,259 -1.1 n.a. n.a. 0.474 1.9 Mill? -0.174 -1.5 -0.075 -0.5 0.146 1.2 -0.040 -0.5 -0.030 -0.3 Vehicle passableroad? -0.573 -1.3 -0.339 -1.0 n.a. n.a. -0.175 -1.9 -0.237 -1.3 Paved road? 0.325 1.5 -0.03 1 -0.2 -0.724 -3.1 0.030 0.5 -0.061 -0.6 suco Irrigation? 0.083 0.7 0.282 2.1 -0.096 -0.8 -0.023 -0.3 0.105 0.9 Irrigationifrice-producing household 0.583 1.3 -0.056 -0.3 -0.149 -0.9 0.272 2.0 0.391 2.4 Private employer? -0.018 -0.2 0.332 1.5 -0.581 -2.5 0.261 3.6 0.325 2.0 Teacher-student ratio 0.000 0.0 0.000 0.0 0.013 1.5 0.005 2.7 0.000 0.2 Classroom-teacher ratio -0,019 -0.I 0.661 2.0 -0.701 -2.0 -0.130 -2.3 0.119 1.3 Birthattendants per population -38.2 -0.7 114.3 1.2 483.6 1.3 -165.6 -1.9 11.7 0.3 Healthservice per population (daysimonth) 72.1 1.9 -124.4 -1.5 -94.8 -1.5 21.3 1.5 220.9 3.9 Community leader characteristics Age (years) -0.006 -0.5 -0.039 -1.5 0.042 1.8 0.015 I.9 0.048 4.3 Education (years) 0.019 0.8 -0.075 -1.0 -0.056 -1.2 0.050 2.5 -0.016 -0.5 Livinginsuco (years) 0.011 1.6 -0.005 -0.4 -0.024 -1.6 -0.001 -0.4 -0.013 -2.8 Constant 11.0 13.4 13.8 6.3 12.8 15.0 11.7 19.7 11.9 14.7 Observations 450 252 252 504 342 R-Square 0.48 0.61 0.64 0.59 0.46 SOVWC.2001 TLSS. SIMULATION METHODOLOGY 8.1 1 We now use the estimated model to predict the impact of changes on explanatory variables on poverty. These simulations are obtained in three steps. First, we use the A A A +...+,B A 1 A k estimated parameters p to generate predicted consumption 1nc.j =a+,B x,; x; where A indicates an estimated parameter. Then, we derive under the assumption of standard normally distributederror terms the probability of householdj to be poor: 206 A HC pl =Pr(lnc,,