Report No. 25343-BIH Bosnia and Herzegovina Poverty Assessment (In Two Volumes) Volume II: Data on Poverty November 21, 2003 Poverty Reduction and Economic Management Unit Europe and Central Asia Region Document of the World Bank TABLE OF CONTENTS VOLUME I1 INTRODUCTIONAND ACKNOWLEDGEMENTS ...................................................................... i IBUILDINGTHELSMS . ...................................................................................................................... A.Historical Background ................................................................................................................ 1 B.Survey Instrument ....................................................................................................................... 1 2 3 D.FieldWork................................................................................................................................... C. Sample Design and Weighting ................................................................................................... E.DataEntry.Cleaningand Variable Definitions........................................................................... 6 6 (i) Entry................................................................................................. Data 6 (ii) Cleaning......................................................................................................................... Data 7 (iii) Definitions: Urban. Rural and Mixed Localities ................................................................... 7 8 (vi) Definitions: IDPs and Refugees.......................................................................................... 11 (v) Definitions: Social Assistance and Pensions ...................................................................... (iv) Definitions: Labor Market Status.......................................................................................... 10 I1 MEASURINGWELFAREWITH LSMS DATA . ..................................................................... A. The Concept of Welfare ........................................................................................................... 13 13 B.Consumption or Income Aggregates for MeasuringWelfare.................................................... 13 C. Content of Consumption Aggregate & Adjustment .................................................................. 15 (i) 15 (ii) PriceAdjustments..................................................................................................... Conceptual Issues................................................................................................................... Spatial 15 (iii) CapitaorPerAdultEquivalent...................................................................................... 16 Per D.PovertyLines............................................................................................................................. 17 (i) PovertyLines........................................................................................................... Relative 17 (ii) Absolute Poverty Lines......................................................................................................... 17 E.Poverty and Inequality Measures............................................................................................... 18 (i) Poverty Index ......................................................................................................................... 18 (ii) -19 FData Needs andData Sources..................................................................................................... Inequality Measures ............................................................................................................. (i) Needs............................................................................................................................. 20 (ii) SourcesUsedinBiH.................................................................................................... 21 Data 20 Data I11 CONSTRUCTINGA CONSUMPTIONAGGREGATEFORBIH . ................................ 23 A.Food Consumption.................................................................................................................... B. Non-Food Consumption............................................................................................................ 23 24 (i)Education ............................................................................................................................... 24 (ii) .................................................................................................................................. 24 (111) ... Utilities Housing................................................................................................................................ 25 (iv) Durable Goods ..................................................................................................................... 26 C.(v)Total 27 (i) HouseholdConsumption.............................................................................................. 28 Adjusted Total Consumption..................................................................................................... Other Non-Food Consumption.............................................................................................. 27 (ii) PriceAdjustments..................................................................................................... (iii) CapitaorPerAdultEquivalents.................................................................................... Spatial 29 Per 30 IV CONSTRUCTIONOFPOVERTYLINESINBIH . .......................................................... A.Extreme or FoodPovertyLine.................................................................................................. 33 B. Constructionof the GeneralPovertyLine................................................................................. 33 34 V POVERTY.INEQUALITY AND THE CHARACTERISTICSOFTHE POOR . ...........37 A. Poverty ...................................................................................................................................... 37 B.Inequality................................................................................................................................... C.Poverty Incidence ...................................................................................................................... 39 40 40 (ii)Decomposition of Poverty within PopulationGroups............................................................... (i)Poverty Rates among Population groups.............................................................. 42 VI CHECKSFORROBUSTNESSOFPOVERTYFINDINGS . ............................................ 45 A.Robustness Checks with Respect to EquivalenceScales.......................................................... 45 (i) (ii) byDisplacementStatus........................................................................................... 47 Location and Poverty ............................................................................................................. 46 Poverty (iii) (iv) Employment Status .............................................................................................................. Education of the HouseholdHead ....................................................................................... 47 48 B.(v) 51 Robustness Checks Using Alternative Poverty Lines ............................................................... HouseholdSize...................................................................................................................... 52 (i)Location and Poverty ............................................................................................................ -53 (ii)Poverty by Displacement Status ........................................................................................... 54 (iii) (iv) Employment Status of Adults .............................................................................................. Education of the HouseholdHead ....................................................................................... 54 55 C.(v) 58 Robustness Checks Using Alternative Definitions of Well-Being ........................................... Household Size...................................................................................................................... 58 (i)Location and Poverty ............................................................................................................. 60 (ii) byDisplacementStatus........................................................................................... 61 Poverty (iii)Education of the HouseholdHead ....................................................................................... 61 (iv) Employment Status of Adults .............................................................................................. 62 D.(v) Household Size...................................................................................................................... 64 Conclusions ............................................................................................................................... 64 VI1 FROMFEATURESOFPOVERTYTO ITS CAUSES . ................................................... 67 ANNEXES Annex 1.Tests for Economies of ScaleinHousehold Consumption .................................................... 73 Annex 2 Food Poverty Line: Detailed Nutritional Assessment . ............................................................ 76 Annex 3 Constructing General Poverty Line . ..................................................................................... 79 References ............................................................................................................... 83 TABLES Table 1.1 Contents of BiH-LSMS Household Questionnaire 2 Table 1.2 Selection of Municipalities 4 Table 1.3 Sample Distribution by Municipalities 5 Table 3.1 Composition of Household Consumption 28 Table 4.1 Converting Food Consumption Patternsinto a FoodPoverty Line 34 Table 5.1 BiHPoverty Indices 37 Table 5.2 Inequality indices for BiHand Entities 40 Table 5.3 Poverty Profile: Poverty Rates by Groups 41 Table 5.4 Poverty Profile: Composition of the Poor Population by Groups 43 Table 6.1 Populationby Poverty Status Measured on the basis of Consumption or Expenditure 60 Table 6.2 Key characteristics of poverty and its robustness to Measurement Assumptions 65 Table 7.1 Regression of Log Consumption on Household Characteristics 67 Table 7.2 Household Characteristics, Simpleand Simulated Poverty Risksfor Per Capita Scale 71 Table A-1 Estimates for Equivalence Scale UsingEngel's Method 74 Table A-2 Tests of Equivalence Scales 75 Table B-1 Derivation of MinimumFoodRequirements for BiH 76 Table B-2 Nutritional Assessment of MinimumBaskets 77 Table B-3 Actual Reference and MinimumFoodBasket, Per Person, Kg/Lt/KM/ Month 78 Table C-1 Poverty Lines Based on Various Method of Estimating the FoodShare AndVarious EquivalenceScales 82 FIGURES Figure5.1 Poverty Incidence by Location inthe RS and FBiH,95 Percent Confidence Intervals 39 Figure 6.1 Poverty by Municipality Type, Comparisonof Equivalence Scales 46 Figure6.2 Poverty by Displacement Status, Comparison of Equivalence Scales 47 Figure6.3 Poverty by Education of Household Head, Comparison of Equivalence Scales 48 Figure 6.4 Poverty by Employment Status of Individuals ( L O Definition), Comparison of EquivalenceScales 49 Figure 6.5 Poverty by Employment Status of Individuals (BiHAdministrative Classification), Comparisonof Equivalence Scales 50 Figure 6.6 Poverty by Employment Status of Head of Household ( L O Definition), Comparison of Equivalence Scales 51 Figure 6.7 Poverty by Household Size, Comparison of Equivalence Scales 52 Figure 6.8 Poverty by Location Comparing Alternative Poverty Lines 53 Figure 6.9 Poverty by Displacement Status Comparing Alternative Poverty Lines 54 Figure 6.10 Poverty by the Level of Education of Household Head Comparing Alternative Poverty Lines 55 Figure 6.11 Poverty by the Employment Status of a Person Comparing Alternative Poverty Lines 56 Figure 6.12 Poverty by Registered (Official) Labor Force Status of Adults Comparing Alternative Poverty Lines 57 Figure 6.13 Poverty by the Employment Status of the HouseholdHead Comparing Alternative Poverty Lines 57 Figure 6.14 Poverty by Household Size Comparing Alternative Poverty Lines 58 Figure 6.15 Poverty by Location Comparing Alternative Definitions of Welfare 60 Figure 6.16 Poverty by Displacement Status Comparing Alternative Definitions of Welfare 61 Figure 6.17 Poverty by Household Head's Level of Education Comparing \ Alternative Definitions of Welfare 62 Figure 6.18 Poverty by Poverty by Labor Force Status of Adults ( L O definition), Comparing Alternative Definitions of Welfare 63 Figure 6.19 Poverty by Registered Employment Status of Adults, Comparing Alternative Definitions of Welfare 63 Figure 6.20 Poverty by Household Size, Comparing Alternative Definitions of Welfare 64 Figure 7.1 Simple versus partial correlation inpoverty analysis 68 Figure C-1 Actual relative food consumption, fitted relative food consumption line, and derivation of the Poverty Line using the per capita scale 81 INTRODUCTION AND ACKNOWLEDGEMENTS Monitoring their population's welfare or poverty levels helps countries around the world to determine the effectiveness of government economic and social policies. Also, as many government and non-governmental programs are specifically tailored to meeting the needs o f the poor, more accurate identification o f this group i s critical to the targeting o f such programs. And, finally, the impact of all social and economic policy on welfare can only be assessed by constant collection of data on well-being in a country. Measuring welfare requires good micro-level or household data. Such detailed data provide information not just on average levels o f welfare (as GDP per capita indicators do), but also on the distribution and characteristics of the rich and the poor. Additionally, such data sources provide insights on factors that affect the ability o f the poor to move out of poverty. In an attempt to answer basic questions about who are the poor in Bosnia-Herzegovina (BiH), what are their characteristics, what i s the distribution of wealth, and what are the factors that affect welfare, the Government of BiH, through its three statistical organizations (the State Agency for Statistics [BHAS], the Republika Srpska Institute of Statistics [RSIS] and the Federationof BiHInstitute of Statistics [FIS]), undertook the country's first representative multi- purpose household level survey in the latter part o f 2001. This survey, the BiH-Living Standards Measurement Study survey [BiH-LSMS], was designed to measure welfare in BiH and to provide information on how welfare levels are correlated with observed social variables such as unemployment, health and education. A fundamental use of the data is to inform the Poverty Reduction Strategy that the Government i s in the process of developing and implementing. This volume represents the results of joint efforts by the Poverty Assessment team and all those who contributedto the successful completion of LSMS: the three statistical organizations o f BiH, and especially members of the LSMS team who acted as counterparts, experts and advisors Edin Sabanovic (FIS), Fahrudin Memic (FIS), Mladen Radic (RSIS), Nada Jakovljevic (RSIS), Nora Selimovic (BHAS), Jovanka Vukovic (Public Fund for Child Protection RS), Fadil Suljic (Federal Employment Bureau), as well as Peter Lynn (University of Essex) who provided sampling advice, K.E. Vaidyanathan (UNDP chief advisor) who provided crucial help during the implementation o f the LSMS, the UNDP Sarajevo office, especially Armin Sirco and Goran Kurtic, and World Bank staff members and consultants, Ruslan Yemtsov (Poverty Assessment team leader, author of some chapters in this volume), Kinnon Scott (BiHLSMS team leader and a key contributor to many chapters in this volume), Irina Smirnov (World Bank Sarajevo Office), Thomas Mroz (University o f North Carolina), Paulette Caste1 (Consultant) and Jacob Tomse (Consultant). The purpose of this volume i s to describe the methods used to construct a poverty measure in Bosnia and Herzegovina using the BiH-LSMS data set, and to present a detailed poverty profile at the level of the country and o f each Entity (i.e., the Republica Srpska [RS] and the Federation of BiH.[m>iH]). All material presented in Chaptersl-7 i s closely linked to the analysis of poverty inVolume I, providing details necessary to understand its findings. The first chapter provides a 1 description of the BiH-LSMS sample design, the survey instruments, the organization of the survey and the definition of key categorical variables. The second chapter provides an overview of the general concepts and issues involved in poverty measurement. Chapters 3 and 4 describe how these concepts have been applied to measure well-being and poverty in BiH. A summary of the key findings on poverty inthe country and the two Entities i s given in Chapter 5 , and various checks of the robustness of these results are presented in Chapter 6. Chapter 7 discusses the use o f the LSMS data to study causes of poverty by isolating the impact o f a set of key causal factors on the risk of becoming poor. There are Annexes 1-3, which provide detailed description of how the poverty line was developed and key measurement assumptions made, and are intended for specialist readers. .. 11 1. BUILDINGTHELSMS This chapter provides essential facts about the BiHLSMS survey, focusing on three critical issues: why the sampling used for the survey is representative of the population of each Entity and the country as a whole; what information was collected; and how this information was used and adjusted to classify respondents by their socio-economic status. Among the variables created on the basis of the information collected, this chapter covers only "categorical" variables, leaving the description of welfare measurement to the next two chapters. A. HistoricalBackground 1.1 In 1992, Bosnia-Herzegovina, one of the six republics in former Yugoslavia, became an independent nation. A civil war started soon thereafter, lasting until 1995 and causing immeasurable suffering, widespread destruction, and substantial loss of life. Following the Dayton Accords, Bosnia-Herzegovina (BiH) emerged as an independent state comprising two Entities, namely, the Federation of Bosnia-Herzegovina (FBiH) and the Republika Srpska (RS), and the district of Brcko. 1.2 In the post-war process of rebuilding the economic and social base of the country, the govemment has faced problems created by a shortage o f relevant data on the welfare of citizens. The three statistical organizations in the country (the State Agency for Statistics for BiH, BHAS; the RS Institute of Statistics, RSIS, and the FBiHInstitute of Statistics, FIS)' have been active in working to improve the data available to policy makers at both the macro and the household level. One aspect o f their activities i s to design and implement a series of household surveys. The first of these surveys i s the Living Standards Measurement Study survey (LSMS). Later surveys will include a Household Budget Survey (HBS: an income and expenditure survey) and a Labor Force Survey (LFS). A subset o f LSMS households will be re-interviewed inthe two years following the LSMS to create a panel data set. 1.3 The three statistical organizations began work on the design of the LSMS in 1999. It was determined from the start that the survey should provide data at both the country and the Entity level, and allow valid comparisons between both Entities. It was also decided that the project would include not only the collection of data, but also their analysis and public dissemination. The present volume is one of the outputs of these activities. 'In principle, BHAS, i s the country level statistical office responsible for collating information from the two Entity- level statistical institutes (FIS and RSIS) and for setting country-wide standards inthe field of statistics. The two Entity-level statistical offices are responsible for data collection and collation within their respective Entities. At the time of the survey, the political status o f Brcko was still under discussion; it did not yet have a separate statistical office as it now does. 1 1.4 The survey, the analytical work and the dissemination of the data were carried out by the three statistical organizations with financial and technical support from the UK Department for International Development (DfID), the United Nations Development Program (UNDP), the Japanese Government, and the World Bank.2 1.5 Overall management of the project was undertaken by a Steering Board comprising the Directors of the RS and FBiHStatistical Institutes, the Management Boardof the State Agency for Statistics, and representatives from DfID, UNDP and the World Bank. Day-to-day project activities were carried out by a Survey Management Team, made up of two professionals from each of the three BiHstatistical organizations. B. Survey Instrument 1.6 The BiH LSMS i s a multi-topic household survey covering a wide range of factors that affect welfare: housing, education, health, labor, migration, credit, privatization vouchers, social assistance, consumption, agricultural and non-agricultural activities. The household questionnaire and the individual questions included in it were designed to address the specific situation of the country and the data needs of policy-makers. In addition, several sections of the questionnaire were based on draft questionnaires for future surveys (the HBS and the LFS) and/or older surveys to allow some tracking of indicators over time. The process of designing the questionnaire was lengthy and involved an inter-agency team from the three BiH statistical organizations. In addition, several ministries provided detailed comments and suggestions. Table 1.1 provides a summary description of the modules included in the questionnaire. Table 1.1 Contents of BiH-LSMS Household Questionnaire Module Description 1. Roster This module listed all the members of the householdand their characteristics such as relationshipto headof household, age, sex, marital status, andmembers absent from the household. 2. Housing This moduleincludedthe following sections: A. Descriptionof primaryresidence: type of dwelling, condition of dwelling, number of rooms, plinth area, andpresenceof utilities (electricity, water, sewerage, telephone, etc.); B1:Ownershipstatus and expenditureson electricity, water and sewerage; B2: Ownershipandpurpose of secondaryresidence; C: Possessionof durable goods, their date of purchaseandpresent value. 3. Education A. Child care and kindergarten attendance and monthly expenditure for child care (for children between0 and7 years of age); B. Generaleducation (for persons7 years old andover andfor childrenless than 7 years who attended school), literacy status, educational qualifications, informal payments and source of financial assistanceduring the academic year 2000-2001, distanceof the school from home etc. 4. Health A: Utilization of health care services: use of various types of health services-primary heath care centers (ambulanta or DZ), laboratory tests, pediatrician, gynecologist, dentist, other doctors, paramedics, alternative medical care, self medication-and expenditure on these services. This section also included questions on the prevalence of chronic ailments and the utilization of health insurance. B. Review of respondents' own health condition elicited information on perceptions about various healthconditionson athree or four point scale, focusing particularly on mentalhealthstatus. 5. Labor This module elicited information for persons 15 years old and over on their activity status during the reference week precedingthe survey. For employedpersons, information was sought on occupation, industry, employment status, place of work, previous employment, number of hours worked in the The creation of a master sample for the survey was supported by the Swedish Government through SIDA, the EuropeanCommission, UK DfID,and the World Bank. 2 week and monthly earnings. For unemployed persons, information was sought on duration of unemployment, last occupation, employment status and industry, method of seeking work, and whether registeredat an employment bureau. 6. Credit This module asked all persons 15 years and over the number of times the person had borrowedfrom different sources, amount borrowedduring the last 12 months, amount owedpresently, the monthand year of the last borrowing, andreasonsfor borrowing andrefusals of loans. 7.Vouchers: This module included questions on eligibility for privatization voucherslcertificates, value of voucherslcertificatesreceived, transactions made with them, sale value of voucherskertificatessold, and the nominalvalue of the vouchers or certificatesintheir possession. 8. Migration For persons 15 years and over, apart from current residence, information was sought on (i) municipality of birth, (ii) residence prior to April 1992, (iii) reason for migration and (iv) current residential status. 9.Social Assistance This module included questions on eligibility for old age pension, disability pension, survivor's pension, war veteran's pension, monthly pensionreceived, and the allowances and services received during the preceding12months. 10.End of First Visit This modulewas intendedto identify householdsto be coveredby Module 12 and Module 13. 11. H/HConsumption Each of the following sections elicited information on the quantity and value of purchaseditems and own production,andthe value of items receivedas gifts: A. Daily expensesduring the preceding 7 days includedquestionson frequently consumeditems such as tobacco, cigarettesand meals taken inrestaurants; B1 Food consumption: average monthly expenditures on items of daily consumption such as bread andcereals,meat, fish, etc; B2. Seasonalfood consumption-fruits, vegetables; C1. Monthly expenditureson non-foodproductssuch as transport, cosmetics, fuel ,cleaningproducts, etc; C2. Annual expenditureson other non-foodproducts, including clothing and footwear, furniture, etc; 12.Non-Agro Business This module elicited the following information from householdsengagedinnon-agriculturalactivity: A. Informationon establishments:natureof activity, personsengaged andnumberof activities; B. General information on the durationof enterprise operation,location, ownership,number of days a week operated, number of personsengaged. C. Labor: number of persons engaged, number of householdmembers and number of non-household members,number receivingwages incashor inkind. D. Revenues and inputs: number of months the business operated, gross earnings in an average month, expenses on inputs in an average month E.Capital asset: value of fixed capitalsuch as land, buildings, equipment and machinesetc., andmain problemsfacedby the establishment. 13.Agricultural Activities This module collected the following information on farming operations with special focus on farm management,inputs andearnings: Al: land used: area of different categories of land, extent used and not used, irrigated land, present value, nature of usellease,lease value during 2000-01, etc; A2. Landowned by householdbut not used: category of land, how obtained, present value, type of use contract,lease amount receivedduring2000-01, etc; B1. Use of forest land: age of forest, value of produce sold, value of produce usedby household; B2.Crop productionand use: area of landused, amount of crop harvested, amounts sold, lost, used as wages, usedas animal feed, processed, consumedby the householdand given away as agift; C1.Inputs andinvestments: total quantity of seeds or seedlings usedby the household, quantitybought C2. Inputsandinvestments: fertilizers - total quantity used, purchased, obtainedas gift; and its cost, quantity usedfromown production, whether obtainedas gift andfrom whom. C3 Inputsandinvestments: fuel andenergy - total quantity of fuel used, bought andits value, whether obtainedin any other way and from whom; C4. Inputs andinvestments: labor - number of paidworkers, numberof paid work days, average daily wage, whether payment was made inkindetc; C5: Inputs and investments: machinery - source of hire, number of hours machinery was hired, amount paidper hour andwhether payments were made in kind; D1 Livestock: number of various categories of livestock, value, livestock sold during the past 12 months, consumed, lost, gifted and bought, new born, number received as gift, whether any livestock product was sold andtheir value; D2. Animal feed: quantity of animal feed used during past 12 months, purchased and its value, own produced, whether receivedas gift andthe source of gift; E. Farmcapital assets: type of capital assets, their market value, age of the asset, whether the asset is rentedout, earningsduring 2000-01from renting out the capital assets. 3 C. SampleDesignandWeighting 1.7 Selecting a probability sample that would be representative of the country's population presented problems. The sample design for any survey depends upon the availability of information on the universe of households and individuals inthe country. Usually this comes from a Census or administrative records. In the case of BiHthe most recent Census was done in 1991. The data from this Census were rendered obsolete, not only by the simple passage of time but also because of the massive population displacements that occurred duringthe war. 1.8 At the initial stages of the project it was decided that a master sample should be constructed. Experts from Statistics Sweden developed the plan for the master sample and provided the procedures for its construction. The households for the LSMS were selected from this master sample. 1.9 The master sample was based on a selection of municipalities and a full enumeration of the selected municipalities. Optimally, one would prefer smaller units (geographic or administrative) than municipalities. Inthe case of BiH, however, it was determinedthat populationestimates were reasonably accurate for municipalities, but not for smaller geographic or administrative areas. 1.10 The first step in creating the master sample was to group the 146 municipalities in the country into three strata--urban, rural and mixed--within each of the two Entities. Brcko was excluded from the sampling frame. 1.11 Municipalities were selected with a probability proportional to estimated population size (PPES) within each stratum, so as to select approximately 50% of the mostly urban municipalities, 20% of the mixed and 10% of the mostly rural ones. Overall, 25 of the 146 municipalities were selected, 14 in FbiH and 11 in RS. The distribution o f selected municipalities over the sampling strata i s shown inTable 1.2. Table 1.2 Selectionof Municipalities Stratum i Total Sampled municipalities municipalities 1.Federation, mostly urban 10 5 2. Federation, mostly mixed 26 3. Federation, mostly rural 48 4. RS, mostly urban 4 5. RS, mostly mixed 29 6. RS, mostl; rural 29 4 Source: BHAS,RSIS, and FIS (2003) 1.12 In each of the. selected municipalities a full listing of households ("microcensus") was carried out. The work was carried out on a decentralized basis, whereby the FIS and the RSIS were responsible for carrying out the fieldwork under the general guidance of the BHAS. The municipalities cooperated by providing temporary office and storage space, along with recruitment of enumerators and controllers for the survey. The fieldwork was supervised by the staff of the two Entityinstitutes; all these groups were trainedby their respective institutes. 4 1.13 The municipalities were divided into geographic enumeration areas (EAs). In theory, each EA consisted of the number of households that could be interviewed by a census enumerator in one day, based on the 1991 Census. At the time the master sample listing operation was carried out, however, many of EAs actually contained many fewer households (in some cases, zero). As EAs were to be the primary sampling unit for the LSMS survey, the first step was to combine contiguous EAs until a new enumeration area with a minimum o f 50 households was formed. These newly constructed EAs were called groups of enumeration areas (GNDs) and replaced the original small EAs. 1.14 Based on the population figures from the master sample microcensus, 250 EAs were selected with probability proportionate to size (PPS) from FbiHmunicipalities, and 200 EAs with PPS from RS municipalities. Within each of the 450 selected EAs, 12 households were randomly selected, giving a total sample of 5,400 households made up of 2,400 in RS and 3,000 in FbiH.(see Table 1.3). Table 1.3 SampleDistributionby Municipalities Municipality Sample Statistical Sample Weighted - Weighted - households weight for proportion sample sample each households proportion household 1 Banja Luka 936 64.6 0.173 60485 0.055 2 Srpska Ilidza 84 542.5 0.016 45570 0.041 3 Cajnice 36 1586.5 0.007 57116 0.051 4 Mordica 132 285.4 0.024 37674 0.034 5 Novi Grad 156 263.3 0.029 41071 0.037 6 Prijedor 432 83.2 0.080 35964 0.032 7 Visegrad 84 422.7 0.016 35509 0.032 8 Knezevo 60 698.8 0.011 41930 0.038 9 Samac 108 369.2 0.020 39874 0.036 10 Srbac 120 345.2 0.022 41425 0.037 11 Zvornik 252 170.3 0.047 42922 0.039 12 Centar 276 160.7 0.05 1 44345 0.040 13 Nov Sarajevo 288 152.8 0.053 44013 0.040 14 Novi Grad 432 96.4 0.080 41642 0.038 15 Tuzla 528 78.7 0.098 41540 0.037 16 Zenica 468 83.2 0.087 38931 0.035 17 Breza 60 704.6 0.011 42275 0.038 18 Travnik 192 191.1 0.036 36702 0.033 19 Visoko 144 242.9 0.027 34980 0.032 20 Vogosca 84 511.6 0.016 42976 0.039 21 Gradacac 144 337.3 0.027 48572 0.044 22 Grude 48 1163.0 0.009 55826 0.050 23 Kakanj 144 359.6 0.027 5178l' 0.047 24 Posusje 48 974.2 0.009 46762 0.042 25 Zavidavici 144 413.3 ~. 0.027 59516 0.054 ~ Source: BHAS, RSIS, and FIS (2003) 1.15 An important point about the LSMS sample design is the fact that in each municipality each household had a different a priori probability o f being selected in the survey. Therefore, to project survey results onto the population, one needs to multiply the results from each household 5 in a municipality by its "weight" (given in Table 1.3). The sum of weights gives an estimated population of the two Entities. All results in the Poverty Assessment are weighted with these factors. D.FieldWork 1.16 A draft questionnaire was pilot tested during the period June 25-July 20, 2001 in the two Entities. After the test, the health and labor modules were cut back substantially. The credit module was also cut back, given concerns about questionnaire complexity. The non-agricultural enterprise module was also reduced substantially. 1.17 Fieldwork started on 26 September, 2001 and ended on 23 November 2001. Its timing was limited by the need to finalize all interviews before the start of Ramadan since household consumption patterns were expected to change during the fasting month. On average, interviewers took 1.5 hours per household to collect the data. Only in the case of households with over 5 members did the interview take longer. The interviewing was conducted at the convenience of respondents, which meant that interviews were conducted both during the day and during the evenings, and throughout the week, including weekends. 1.18 Each interviewer was assigned two clusters of households. (Each cluster consisted of 12 households in an enumeration area or group of enumeration areas). Interviewers often visited each household more than twice. All information was collected from direct informants, except in the case of children under 15 whose parents were asked to provide the information. Otherwise, the interviewer carried out a series of interviews in the household, one for each member. 1.19 Each Entity provided interviewers and supervisors with badges and letters of introduction. Communication with field staff was improved by recruiting interviewers and drivers who had cell- phones. 1.20 Overall, the response rate in the survey was 82 percent. For each enumeration area, four replacement households were selected prior to the field work. Using these replacement households as needed (a total of 938 households), the final sample size was 5,402 households interviewed. E.DataEntry,CleaningandVariablesDefinition (i)Datu Entry 1.21 An integrated approach to data entry and fieldwork was adopted. Data entry proceeded side by side with data gathering to ensure verification and correction in the field. Data entry stations were located in the regional offices of the Entity institutes and were equipped with computers, modems and dedicated telephone lines. Completed questionnaires were delivered to these stations each day for data entry. 1.22 Data were entered soon after questionnaires were completed, and a customized data entry program was used to identify errors (missing data, inconsistencies and the like) in the data. This enabled interviewers and supervisors to check each questionnaire, resolve any small difficulties andor decide that the interviewer needed to return to the household for clarification. The data 6 entry program was designed to detect many errors at the entry stage, thereby minimizing the need for ex-post facto data editing. Once all data were compiled in the Entity offices, a check was made to ensure the structural consistency of data files, i.e. that no records were duplicated or omitted. 1.23 Every efforts was made to make data sets from each Entity compatible so that a country- wide data set could be created. (ii)Data Cleaning 1.24 Itis important to note what is meant by `data cleaning' interms of the BiH-LSMS data set. In the sense that it is a faithful reflection of the responses of all interviewees, the data set can be considered `cleaned'. But as participation in the survey was voluntary, informants had the option to refuse to answer specific questions, and may have provided information that was not always consistent. 1.25 Some data sets are processed so that all missing values are imputed, all outliers revalued and all inconsistencies fixed based on a given set of assumptions. This was not done in the case of the BiHLSMS dataset available for public use. The survey team decided that there was no single "correct" way to resolve problems of missing data, outliers and inconsistencies. Each user would need to make his or her own decision on how to treat such problems based on the type of analysis being carried out. For some analyses, information on outlier values i s key; for others, outliers would distort findings and would need to be dropped or given an imputed value. The same point applies to missing values. Some analysts will chose to drop cases with missing values for the variables of interest to them, while others will impute such values, using medians, mean or complex multi-variate techniques. In order to ensure the usefulness of the BiH data set for all users, the LSMS data as disseminated didnot impute missingvalues, reconcile inconsistencies, re- value outliers, or in any way alter the responses provided by the respondents. 1.26 In most cases, respondents' direct answers were enough to consistently classify them into groups. For example, respondents' answers determined their grouping by education level (those with primary education, with secondary completed, etc.). But in some cases, classification required a combination of responses to various parts o f the questionnaire, or some adjustments, notably with respect to labor market status and receipt o f pensions. Note that no adjustments were made in LSMS raw data, and adjustments discussed below were only made for the Poverty Assessment. (iii)Definitions: Urban,Rural and Mixed Localities 1.27 The urban-rural definitions used in BiH (discussed above in section C) are unusual, with large administrative units such as municipalities classified as if they were completely homogeneous. Their classification into urban, rural, or mixed comes from the 1991Census, which used the predominant source o f income of households in the municipality to categorize it. Urban municipalities are those where incomes of 65 percent or more o f the households are considered to be "urban," and rural municipalities are those where the proportion o f households with "urban" 7 incomes i s below 35 percent. The remaining municipalities were classified as mixed (urban and rural). 1.28 This definition is imperfect in several ways. First, the distribution of income sources may have changed dramatically after the war: populations have shifted, large industries have closed and much agricultural land remains mined. Second, the definition i s not comparable to those used in other countries, where villages, towns and cities are classified as rural or urban by population size or by types o f services and infrastructure available. Third, municipalities can be made up of communities with substantially different characteristics. Nevertheless, it was not felt that these imperfections were substantially detrimental to the sample design. The urbadrural definition can be useful for analytical purposes, and a check shows that in fact it i s broadly consistent with the type o f economic environment that all households in a locality face.3 (iv) Definitions: Labor Market Status 1.29 The state o f the labor market i s one o f the key policy issues in BiH. This report uses the LSMS survey to construct and analyze unemployment and employment figures. The simplest and the most direct way i s to use respondent self-identification, such as "employed", "unemployed", or "retired". This approach, however, suffers from deficiencies. Actual employment status is often very complex, and respondents have difficulties in unambiguously identifying themselves into a single category (for example, working pensioners). Unemployment presents special problems. Registered unemployment in BiH i s huge: for every 100 registered employees there are 40 people registered as unemployed. This number was broadly confirmed by LSMS respondents. But in economic terms, unemployment i s not only defined by registration; it denotes a person who in a particular moment in time does not have ajob, i s actively searching for ajob, and i s ready to start working. Registration i s not an accurate labor market status definition, as people may want to register to get certain benefits (such as health insurance). Others may be working "in the gray sector" (without registration). Such arrangements are attractive to employers, who "economize" on payroll contributions to various social funds. This "economy" will allow them to offer higher wages than for comparable work in the formal economy. Thus economic definitions which rely on actual reported activities are preferable. 1.30 The next few paragraphs summarize the approach usedto construct key labor market status variables. The "universe" consists of all persons older than 15 years old (that is, they were past their 15`h birthday as o f 1" November 2001, i.e. at the time of the administration o f the survey). The working age population i s all persons o f "legal" working age (i.e. above 15 and below the normal retirement age, 55 for women and 60 years for men); if a person during the week preceding the interview i s employed by someone, works for her/his own benefit, or has a job to return to, then he or she will be counted as employed (even if this person i s outside the legal working age). A person i s classified as unemployed if he/she i s in the working age population, but not employed as defined above, has actively sought work in the last 4 weeks, and i s available for For example, the percentage of LSMS households ineach stratum reporting using agricultural land or having livestock is highest inthe "rural" municipalities and lowest inthe "urban" municipalities. However, the concentration of agricultural households i s higher inRS, so municipality types are not comparable across entities. The percentage reporting no land or livestock inRS was 74.7% in"urban" municipalities, 43.4% in"mixed" municipalities and 31.2% in "rural" municipalities. Respective figures for FBiHwere 88.7%, 60.4% and40.0%. 8 work (able to start to work in the next 2 weeks). True (LO) unemployed defined in this way are different from discouraged workers, who are non-employedpersons of working age who were not looking for job in the last 4 weeks, but wanted to work, were available for work, but believed that no suitable job was available. 1.31 All other persons in the universe, who are not employed nor unemployed,are considered out ofthe laborforce, or inactive. 1.32 The application of such a classification in the BiH LSMS reveals that the registered unemployed include many different sub-categories. Not all registered unemployed are ready to take ajob when one becomes available. Many people are registered as unemployed not in order to find ajob but for other reasons (such as obtaining benefits). People may be without ajob and not looking because they are "discouraged" as discussed above; or they may not be ready for ajob for personal reasons, or may be simply unwilling to work (i.e. be inactive). All of these diverse categories may end up as registeredunemployed. Specifically, according to the LSMS: 0 Out of 498,000 registered unemployed inBosnia and Herzegovinain2001, only 122,000 (24.4 percent) qualified as unemployed according to the standard definition of unemployment; 0 Among the remainder, 124,000 (25.0 percent of the registeredunemployed) were actually employed workers, and 252,000 were inactive (50.6 percent o f the total); 0 Not all of those who were actually unemployed hadregistered69,000 (36.3 percent o f the unemployed on the LSMS definition) didnot approach employment offices and register; 0 Amongthe inactive persons who were registered, slightly more than half qualify as discouraged inthe sense described above. 1.33 Another important distinction i s the definition of "formal" versus "informal" sector. According to the "Resolution Concerning Statistics of Employment in the Informal Sector" (The Fifteenth International Conference of Labor Statisticians, L O [1993]), the defining element of informal employment i s the fact the employer i s not an incorporatedbusiness. That is, a worker i s treated as informally employed if he (she) i s either self-employed (in an unincorporated business), or an employee working for an unincorporated employer. Note that by their nature, informal jobs are associated with different degrees o f risk and stability. The self-employed, by definition, take responsibility for both job risks and profits, and their status i s directly affected by their business performance; employees in the informal sector are usually more exposed to risk, because they enjoy less protection from joblessness. At the same time, however, formal sector workers may or may not be coveredby health and pension insurance. 1.34 To determine formal vs. informal work status in the BiH LSMS, the convention was adjusted. For some workers, it was possible to infer from the questionnaire whether or not their employers were incorporated businesses; those categorized as working in informal employment were all unpaid supporting family members, farmers working on their on own farms, and those engaged in similar activities (described as the sale of agricultural and other products, etc). In contrast, formally employed are all workers who worked in incorporated businesses, public enterprises and international organizations. However, for many workers, it was impossible to infer from the questionnaire whether or not their employers are incorporated businesses. These workers 9 were categorized into the formal sector if their pension contributions were paid, and into the informal sector if they were not. (v) Definitions: Social Assistance and Receipt of Pension 1.35 The LSMS questionnaire asked respondents whether they received a pension, type of pension (civil, military, or veterans'), and the amount received. Preliminary analysis of the results demonstrated that the survey did not match the official figures. The information was reorganized according to the information on age and level of benefit that appeared consistent with the publishedinformation. The final results were closer to the official figures. 1.36 Particular problems requiring adjustment included the following: 0 Some persons said that they were receiving 2 or more pensions, which i s legally impossible; 0 Many respondents said that they were receiving less than the minimumdisability and survivor benefits 1.37 The results by category of pensioners and pension were reorganizedto achieve consistency. In this reorganization, all values reported in the legal range were unchanged. The rest were changed, starting with the old-age pension (retaining only its value when it was declared and reclassifying others to nil), then followed by disability pensions, and finally by survivor pensions. The reported veterans' benefits were divided in two groups: invalids and family according to the reported percentage of disability. All disability pensions excluded from the category of "pensions" because of reported wrong amounts were moved into the category o f veterans' invalid benefits unless the recipient was under age 18 (veterans benefits do not have strict minima). After such adjustments total number of beneficiaries (receiving civil and military pensions and veterans' benefits) estimated on the basis of the survey i s close (within 10percent marginof error) to the total reported officially. 1.38 LSMS data also show that the average benefit received i s about 6% higher than the official average pension, so estimated overall expenditures on pensions i s very close to the published figures. 1.39 The survey also reports the overall amount received by individuals for social assistance in the year, and the corresponding type of allowances (one or several). There are three types of allowance--permanent, temporary and assistance for the care of a dependant person. The survey indicates an individual's entitlements for each type, and reports the total amount of cash benefit received in the last year. Survey results are in line with what we know about overall social transfers, but not the distribution by type of benefits. 1.40 There was a concern that the questions was not precise enough to ensure that only government social assistance would be reported, and that individuals might report non-state types of assistance. These other types o f assistance are assuming diminishingimportance, however: the Public Expenditures and Institutional Review (PER) reports that if "there are also many donors and non-governmental organizations (NGOs) active in the social welfare sector.. .. many of the 10 larger programs have now been terminated or have been significantly scaled back [leading to] ... greater dependence on SWCP spending on local budgetary sources, especially since 1999." (The World Bank [2002]). 1.41 The survey finds many fewer beneficiaries of permanent social assistance than the PER indicated. Nothing could be done with this problem, however. In the survey only about 30, 000 persons (23 % of the survey beneficiaries o f social services and cash benefits) declare they receive social services (but no cash benefits). The survey seems to miss the people that are beneficiaries of social services other than cash. (vi) Definitions: IDPs and Refugees 1.42 The LSMS asked about respondents' (i) current residence, (ii)municipality o f birth, (iii) residence prior to the war (April 1992), (iv) reason for migration and (v) current residential status. To define displacement status, reported answers to questions (iv) and (v) were used. The only problem in using the data collected in the LSMS was that the answers were collected only from individuals above the age of 15. To overcome this problem, a simple imputation procedure was used based on the household roster, assuming that children had the same status as their parents. 11 12 2. MEASURINGWELFARE This chapter discusses the rationale behind constructing a core welfare variable based on consumption per equivalent adult, reviews the requirements that a poverty line needs to meet to provide an accurate picture of poverty, and provides details on thepoverty and inequality indices used in the study. A. The ConceptofWelfare 2.1 To examine poverty and inequality, we need a measure of material well-being. Ideally, this measure should correspond as closely as possible to the way a person experiences his or her standard of living. It i s natural to think that a person's standard of living, or material well-being, as a function of all goods and services consumedby that person. 2.2 But how can one compare different individuals consuming different quantities of various goods? Economic theory allows us to rank levels of well-being using the cost (monetary value) of the consumption bundle consumed in a given period. The intuition i s simple: the individual could have bought a cheaper bundle of goods, but he or she did not. Hence, he or she must get a higher level of well-being from the current bundle of goods than from any cheaper bundle of goods. The cost of the consumption bundle i s therefore a "money-metric utility"; it represents well-being expressedinConvertible Marks (KM)inthe case of BiH,4 Euros or any other currency. B. Consumptionor IncomeAggregatesfor MeasuringWelfare 2.3 In theory, any welfare measure should include all of the factors (including health, leisure, social capital and other desiderata) that contribute to welfare. In practice, however, due to measurement and valuation difficulties, the focus in micro-data analysis i s only on material well- being using information on consumption of goods and services by a household. Even such `simple' measures are, in practice, quite complicated to capture well, and there is debate as to whether income or consumption i s the preferable measure. 2.4 Income i s often considered to be the preferred measure, because it i s an indicator of the "potential" to enhance welfare (including non-material aspects such as leisure). But income suffers from several defects both in theory and in practice. First, income can be highly volatile, The BiHcurrency, the Convertible Mark (KM),was, at the time of the survey, pegged to the GermanDeutsch Mark. This value at the time of the survey (Nov. 2001) was approximately US$ 1.9. 13 whereas consumption can be, and is, more readily smoothed by individuals. This smoothing makes consumption a better indicator of welfare than income, because it more accurately represents the welfare level of an individual at any given time. In transition economies such as BiH, people are paidvery irregularly, with several months of wage arrears being common. Inthis context, relatively steady consumption-based welfare measures give a more accurate picture than often erratic income-based measures. 2.5 Second, regardless of the measure, it i s essential that it be comprehensive, that no aspect of income or consumption be ~ m i t t e d .Otherwise, erroneous conclusions may be drawn about the ~ numbers and characteristics of the poor. If, for example, the value of home-produced food were omitted from an income aggregate (total income measure), then rural populations would look much poorer than they actually are. Or if a consumption aggregate i s constructed using only monetary expenditures, those who receive in-kind benefits from employment would look poorer than they actually are. 2.6 Again, measurement problems are more severe in transition countries with respect to income than with respect to consumption. Income under-reporting i s marked for many reasons including sometimes because survey respondents are not willing to fully disclose illegal or semi- legal income sources. Experience in BiH showed that households were not willing to provide information on unregistered businesses and informal sector activities. Finally, produce from household plots has become a mainstay of food consumption, but i s not a standard component of money income. 2.7 Relatedly, practical experience suggests that the quality of consumption-based data obtained from households i s better than the quality of income-based data. At the top end of the income distribution, households tend to under-report their income, reflecting a lack of faith in the confidentiality of the survey, concerns about the tax authority, complexity of earnings that would lengthen an interview, and the like. At the other end o f the income distribution the problem i s less one of willingness to provide accurate data and more one of inability to do so. Households engaged in informal activities andor with household businesses often cannot separate out what i s `household' income and what i s `business income' thus undermining the reliability o f the data collected. 2.8 In summary, given the difficulties of defining a total welfare measure, a money-metric measure such as consumption or income i s typically used. And given the problems noted above with respect to income-based measurement, the remainder o f this discussion of measuring welfare here will focus on consumption. Total income includes: all labor income, all income from home production, all income from self-employment, household enterprises, public and private transfers, rents, the use value o f durable goods and housing. Total consumption requires data on food consumption (from purchased, home-produced and gift goods), together with non- food consumption, and the use value o f durable goods and services. 14 C. Contentof ConsumptionAggregateandAdjustments 2.9 Consumption needs to be properly measured and adjusted to constitute an appropriate welfare index. This section discusses measurementand adjustment issues. (i) ConceptualZssues 2.10 There are conceptual issues to be kept in mindwhen constructing a money-metric welfare measure based on consumption. Essentially, these relate to the distinction between expenditure and consumption. First, only consumption during the survey reference periodi s measured. Unlike food, consumer durables and housing are consumed over a longperiod of time. Hence, it would be inaccurate to attribute expenditure on such goods wholly to the reference period. Therefore such expenditures are excluded, and the imputed value of the consumption flow associated with the possessionof a consumer durable i s includedinstead. 2.11 Second, expenditures that reflect differences in need or tastes are excluded. When consumption is usedas a measure of well-being, higher consumption should indicate a higher level of well-being. For most consumption items this correspondence i s reasonable. However, for some categories, such as health expenditures, this correspondencei s questionable. 2.12 Third, since the mainpurpose of measuring welfare is to compare households to eachother, two key adjustments must be made in order to obtain proper rankings of households and individuals . These are adjustments for spatial price differences and for household composition.6 (ii)Spatial PriceAdjustments 2.13 Consumption (or income) i s only a valid measure of well-being if people who spend more actually consume more, or higher-quality, goods, and not if they merely spend more for the same goods owing to higher prices. Hence, we need to adjust for possible differences in prices across different geographical areas. A small example illustrates the point. A kilogram of carrots might cost 1.5 Euros in a capital city and only 1.1Euros in a small village. But the benefit of consuming a kilogram of carrots i s the same regardless of where they were bought or at what price. Thus, to compare the welfare levels of two households or individuals, we need to adjust the prices paid so that the welfare obtained i s the same in monetary terms. 2.14 Large differences typically exist in the cost of living between urban and rural areas and often within each area. In principle one could use regional components of the national Consumer Price Index (CPI) to adjust for spatial differences. In most countries, however, the CPI i s not calculated at the regional level. For this reason, it i s worthwhile to use the household survey data themselves to construct CPIs at the relevant geographic levels (See Chapter 111,Section C. ii). Note that, inthe case where data are collected over a long period of time, it would also be necessary to adjust for changes inprices over time. 15 (iii)Per Capita or Per Adult Equivalent 2.15 Consumption data from household surveys are collected at the level of the householdrather than the individual. But, in order to determine the welfare levels of people, total household consumption must be divided among household members. Consumption cannot, however, be explicitly assigned to individual household members using the data. Instead, an adjustment based on some allocation rule must be imposed to attribute their share of households resources to individuals within the household. One such allocation rule i s simply to divide total household consumption by the number of household members. This gives us per capita consumption. 2.16 This is the most commonly applied method and its implies that all family members receive an equal share o f household resources. Alternative allocation rules, known as equivalence scales, are often proposed. Although there exists little guidance for choosing among the wide range of possible scales, it i s important to examine the sensitivity of poverty comparisons to the particular allocation rule chosen. This issue is particularly important when one is concerned with the demographic characteristics of the poor. 2.17 Two alternative allocation rules are adult equivalents and economies of scale. They reflect the fact that one would, ideally, like to take into account the fact that children and adults do not consume at the same levels and, also, the fact that economies of scale exist in households. To do this one can use the following equation to adjust the actual number of household members to a number of `equivalent adults'. EA= (A+a K ) e Where: EA=number of equivalent adults A= number of adults K=number of children a =parameter for economies of scale 0 =parameter for shared goods consumed 2.18 Adults vs. Children: Typically children consume less than adults in a household. They have lower caloric needs, their clothes are often significantly cheaper and they consume a more limited list of items. The parameter alpha in the equation reflects the lower cost of children and can take on a value from 0 to 1. Assigning this parameter a value of 1essentially assumes that children consume the same as adults and i s equivalent to the per capita measure. 2.19 Economies o f scale: The parameter theta adjusts for shared and private goods in the household and can take on a value from 0 to 1. If all goods were private goods (such as food which only one person can consume) the parameter would be equal to 1 and again, would be equivalent to a per capita measure. 16 D.Poverty Lines 2.20 Once a consumption aggregate i s constructed and adjusted for prices and household composition, it is possible to rank all individuals by their welfare levels. Often, however, it i s important to be able to classify individuals into categories of poor and non-poor. For this purpose poverty lines, either relative or absolute are calculated. Individuals' consumption i s then compared to these lines: if consumption i s below the line, the individual i s categorized as poor. 2.21 A poverty line is set to a value of consumption (or income) below which one would be considered to be poor by the society in which one lives. A poverty line can be set in a variety of ways, depending on the purpose and needs. Below we first distinguish between relative and absolute poverty lines, and then consider two specific types o f absolute poverty lines. (i)RelativePoverty Lines 2.22 The value of a relative poverty line i s based on how one group in the society compares to the rest. The common practice in European countries (where relative lines are most frequently used) is to set the poverty line at a fraction of median income. Thus this line incorporates the overall wealth of the country and its average standard of living. A person who i s considered to have consumption below this line i s considered to be poor relative to other people in the country. Clearly, in very wealthy countries, the `poor' could have a standard o f living that would be considered more than adequate in other countries or at an earlier period. (ii)Absolute Poverty Lines 2.23 An absolute poverty line, as its name implies, does not measure poverty relative to others' welfare levels but instead attempts to establish the value o f consumption that any person needs, regardless of time and place. 2.24 Extreme or Food Poverty Line: The most commonly used absolute poverty line i s based on food consumption. Its use reflects the fact that every person needs a certain number o f calories per day to maintain life and the energy required to work and participate in his or her society. Nutritionists set minimumcaloric requirements by taking into account the age, gender and level of effort expended by individuals. Using this accepted minimum caloric requirement, the cost o f the absolute food poverty line i s set at the money value required to obtain this minimum level of calories. 2.25 It is important to note that even an absolute food poverty line has an element of `relative poverty' in it. In theory, the lowest cost of obtaining the minimum level o f calories would be the accurate value of the food poverty line. This lowest cost can be obtained through the solution o f an optimization problem. But while the solution might be accurate, it would in all likelihood consist of a diet that would be completely unacceptable for people in the country in question on cultural grounds. To avoid this difficulty, the costs o f obtaining the minimum level of calories are instead based on actual patterns of consumption observed in the country. This does not imply that everyone must eat a similar diet, and in fact, no one does actually eat the `average' diet. But it does mean that absolute poverty lines based on food consumption are quite specific to each country and would be inaccurate elsewhere. 17 2.26 General Poverty Line: A second absolute poverty line i s based on the concept that food i s not the only good required by an individual. For example, to survive winter, a person needs housing, and to work a person needs to be able to clothe himself or herself adequately. Unlike food consumption, however, where there are objective measures of what i s needed (calories), there are no accepted standards for non-food consumption of goods and services. Ten people would devise ten different lists of `needs': no criteria exist for determining which list i s most appropriate. Any attempt to create a set basket of non-food needs is, essentially, a very subjective effort and would be closer to a relative poverty line concept than an absolute one. 2.27 Instead, one can use the data and the patterns of consumption of the population to calculate an allowance for essential non-food spending that i s added to the value of the extreme food level. How this i s done differs: "... In thepoverty line developedby Orshanskyfor the United States, the basicfood poverty line was scaled up by afactor of three, based on the empirical observation that in the United States approximately 75percent of the average household's budget was spent on non-food items (Orshansky, 1963, 1965). As pointed out by Deaton (1997) the choice of scalar was quite arbitrary and is not terribly intuitive. Ravaillon (1994a, 1998)proposes two alternatives, both of which differ from the Orshansky approach in that the determination of required non-food expenditures is based on the expenditurepatterns of thepoorer members of thepopulation. Thefirst, "austere" approach entailsfinding the amount normally spent on non-food items by those households whose total expenditure [. .. ] isjust equal to thefood poverty line, and adding this amount to thefood poverty line. The idea is that because these households are sacrificing essentialfood consumption in order to acquire a certain number of non-food items, they must view these items as essential. The second, "upper-bound' approach is to scale up thefood poverty line by the amount spent on non-food by households whose actualfood expenditures equal thefood poverty line [..I" (Lanjouw and Lanjouw, 2001). E.Poverty and InequalityMeasures 2.28 The simplest and most common measure of poverty i s the headcount index which simply indicates the percentage of the population in households whose per capita consumption i s below the poverty line. This measure, however, says nothing about the how far below the poverty line, or how poor, these individuals are As this i s important, several other measures are used: the depth and the severity of poverty. (i) Poverty Index 2.29 Measures of dimensions of poverty are based on the Foster, Greer, and Thorbecke (1984) class of poverty measures. This class i s described by: 18 where a i s the parameter (explained below), z i s the poverty line, Ci is equivalent consumption of individual i,and n i s the total number of individuals. If we set a equal to 0, we obtain P(O), or the poverty headcount index. P(0) simply measures the fraction of individuals below the poverty line. Ifwe set a equal to 1, we obtain P(1), or the poverty gap. This characterizes how many resources are needed to bring the consumption of all of the poor to this poverty line. The poverty shortfall i s a poverty measure that takes into account how far the poor, on average, are below the poverty line. One can show that P(l) = P(0) x (Average Shortfall) where the average shortfall is the amount, measured as a percentage o f the poverty line, by which the mean consumption of the poor on average falls short of the poverty line. Finally, if we set a equal to 2, we obtain P(2), sometimes also called the severity o f poverty or FGT(2). This poverty measure captures difference in the severity of poverty, since it effectively gives more weight to the consumption of the poorest. (ii) Inequality Measures 2.30 Inequality matters because, unless a society i s highly mobile, the economic distance between the rich and the poor presents an important indicator of differences in values, aspirations, consumption patterns and lifestyles across groups. Inequality has many correlates: social exclusion, declining investment in human capital in low income areas, declining confidence in the government, increased economic insecurity, and impaired functioning of democracy. 2.31 There are many statistical measures o f inequality. Some are more sensitive to different parts o f the income or consumption distribution than others and some are more easily interpreted than others. Perhaps the easiest to interpret i s the 90/10 percentile (or decile) ratio. It shows how many times the poorest person in the top decile consumes more than the richest person in the bottom decile. The 90110 ratio i s the product o f the 90/50 ratio ("rich to middle" ratio) and the 50/10 ratio ("middle to poor"). These ratios, while of interest, are only sensitive to parts of the distribution. 2.32 Other common measures of inequality take into account the entire distribution. For example, the population can be divided into equal-sized groups based on consumption per capita. If we choose to use 5 groups, then for each of these groups, or quintiles, we can show their shares in total consumption. Since in a perfectly egalitarian world all groups would consume 20% of the total, deviations from this even distribution can measure consumption inequality. 2.33 There are also special indices that summarize the whole distribution in one number-the Gini coefficient, the Theil entropy index, and the Theil mean log deviation and standard deviation of logs. The Gini coefficient i s perhaps the best-known inequality statistic and i s given by: 19 where there are n individuals indexed by i, their consumption i s given by ci, mean consumption i s denoted by p, and where rj i s household's i rank inthe consumption ranking (i.e. for the household with lowest consumption rj equals 1 while for the household with the highest consumption ri equals n). The Gini coefficient ranges between 0 (perfect equality) and 1 (complete inequality). The Gini i s most sensitive to inequality inthe middle of the distribution. 2.34 The Theil entropy index i s given by: 2.35 The Theil entropy index i s most sensitive to inequality inthe top of the distribution while the Theil mean log deviation measure, given by: i s most sensitive to inequality in the bottom range of the distribution. Unlike the Gini, neither the Theil index nor the mean log deviation measure i s eas to interpret except in reference to other countries or the same country at different points in time. Y F.DataNeedsandDataSource 2.36 All of the measures of poverty described in this chapter impose very strict requirements on the data to be usedfor such measurement. I s the BiHLSMS up to the task? (i)Data Needs 2.37 To construct a total consumption measure requires micro-level (household level) data on total consumption as well as the composition of the households. The key components of consumption are: 1. Foodconsumption a. Purchased b. Home produced c. Gifts d. Consumedoutside the home For both measures, zero denotes perfect equality. For complete inequality (one person consumes everything), E(0) reaches infinity while E(1) reaches nln(n). 20 2. Education expenditures 3. Expenditures on utilities 4. Use value of housing 5. Use value of durable goods 6. Other non-food consumption a. Purchased b. Gifts 2.38 In addition to the value of consumption, data are also needed on the quantities of food consumed in order to construct the food poverty line. Additionally, price or unit value information i s needed to construct regional CPIs to adjust for spatial differences in the cost o f living (discussed further inChapter 111, Section C ii). And, finally, data on the composition o f households i s needed make adjustments for individual consumption (Chapter 111, Section C iii). (ii)Data Source Used in BiH 2.39 The BiH-LSMS collected data on consumption, as well as other aspects of households directly or indirectly affecting their living standards. 2.40 The BiH LSMS provides all of the data needed for measuring welfare levels. The next two chapters describe the procedures used in constructing a consumption aggregate (welfare measure) and the assumptions that underlie the procedures (Chapter 111); and the construction of two absolute poverty lines, a food or extreme line and a general poverty line (Chapter IV). 21 22 3. CONSTRUCTINGA CONSUMPTIONAGGREGATE FOR BIH Below, we outline how the LSMS data were used to construct total household consumptionfrom the six components specified in Chapter II,Section F (i). A. Food Consumption 3.1 The first step in constructing a consumption aggregate i s to value the quantity of food that i s consumed. The LSMS collected data on food purchases, home production of food and gifts of food received by households as well as food consumed outside the home. Basic data were collected on 66 food products, or groups of products consumed by the household. These covered fruits and vegetables, dairy, grains and cereals, meat, poultry and fish, beverages and condiments, and staples such as sugar, oils and the like. Households provided information on the quantity and value of each item purchased and produced at home. For all gifts o f food, the household provided information only on the value of the gift (as if it hadbeen purchased). 3.2 Data on food consumption was collected for a one month period. Total annual food consumption was calculated as the sum o f the value o f all purchases, home produced food and all gifts received, times twelve. 3.3 As in any data set, especially complex ones like the BiHLSMS, some data problems were found. As noted in Chapter I, E ii,the LSMS data were not "cleaned" to smooth data anomalies, so that individual users could decide whether and how to do so in light of their individual needs. For the purposes of the BiH Poverty Assessment, a number of adjustments were made to the LSMS data. In cases where there were outliers or missing values, these were replaced with median values to maximize the available information. Outliers were defined as any value greater or less than the median value, plus or minus three standard deviations at the country level. Missing values and outliers were replaced by the median value at the smallest geographic area possible' (group of enumeration areas, or municipality, or Entity, or country). 3.4 In addition to food consumed in the home, data were collected on food expenditures outside the home: meals eaten outside the home (breakfast, lunch and dinner) as well as snacks and other food consumption. Data were collected for a seven day period and annualized by multiplyingby the number of weeks inthe year. Ifthere were less than five prices available to calculate the medianprice at any given geographic level, the median for the next higher (larger) geographic area was substituted. 23 B. Non-Food Consumption 3.5 The components of consumption included here are education, utilities, housing, durables, and other non-food consumption (weekly and monthly purchases of personal care and household items, small appliances, transport and recreation). For most of these items, the value of consumption i s equal to the expenditure for the item or the monetary value of a gift. For two of these items, housing and durable goods, the actual expenditures does not represent consumption. As discussed in Chapter 11, Section C i,the value of the consumption flow associated with using these items i s calculated instead. (i)Education 3.6 Education can be seen as both a consumption item and an investment for future earnings. We have included expenditures on education as part o f consumption. Data were collected for all members of the household. For pre-school and kindergarten age children, households provided information on formal and informal payments related to these services. For school-age children, data were provided on costs of annual and special tuition, membership fees for parent associations, school uniforms and clothing, textbooks and school supplies and food and lodging, as well as on other expenditures for tutorials and the like. Households also provided information on informal payments to schools for repairs, maintenance, and classroom equipment. Although information was also collected on the transportation costs associated with schooling this was not included here as it was assumed that this cost was captured in the "other non-food" component of the questionnaire and to include it here would be double counting. 3.7 The information was collected for the school year prior to the implementation of the survey, in other words for the school year 2000-2001. In calculating annual expenditures for pre- school education, checks were made o f outliers (as in the other sections) and outliers and missing values were replaced by median values at the Entity level. For primary and secondary school data, outliers and missing values were replaced using municipal level medians. For tertiary education, outliers and missing values were replaced using the Entity level medians. Total household education expenditures are simply the sum of the annualized individual educational expenditures. (ii) Utilities 3.8 Households provided information on their monthly expenditures on utilities and related services: electricity, district heating, piped gas, gas in containers, oil (liquidfuels), coal, firewood, water and sewerage, central hot water, garbage disposal, land occupation fees, common area fees, radio and TV subscriptions and telephones. Data was collected from households for the month precedingthe survey. 3.9 To calculate annual expenditures we needed to take into account the fact that household expenditures on utilities are higher in winter months than in summer months. Thus, additional questions were asked about expenditures levels for oil, coal, firewood, water and sewerage, electricity and piped gas for winter months. The annual expenditures for each of these items were constructed usingsix months of summer expenditures and six months o f winter expenditures. 24 (iii)Housing 3.10 Calculating the value of housing for inclusion in the consumption aggregate is more complicated than for other consumption items. One consumes housing over a long period of time. Thus the value of housing for inclusion in an annual consumption aggregate must reflect the value of the housing that one receives during the year, not the total value of the housing. A simple example shows the logic of this approach. 3.11 Imaginethree households that are exactly alike in their composition and total consumption. The first household rents its flat, the second household owns its flat and the third household lives in a temporary shelter. The consumption aggregate would include the rental payment of the first household. If no value i s calculated for the second household's housing, the second household would look poorer than the first, when we know that they are exactly the same. Additionally, the second and third household would have the same consumption level but we know that the welfare of the third household i s actually lower than that of the second household as its members live in a temporary shelter. To avoid this type of mis-ranking of households, we need to estimate a value for the housing of the non-renters, i.e., the second and thirdtypes of household. 3.12 For households that rent their housing, it i s assumed that the monthly (annual) rental payment i s equal to the amount of housing `consumed' in that year. The difficulty arises for households that own their housing. To ensure that the comparison of welfare levels between householdsi s accurate, the value of ownedhousing must be calculated.' 3.13 A two-stage process was usedto do this. First, we ascertained the reported monthly rent payment for all households that rent their housing units. We then ran a rent regression for the group of rentin households to identify the determinants of rent based on a vector of housing characteristics.lF Separate regressions were estimated for each of the two Entities. In addition, in both regressions, variables identifying location (municipality) were also included. The characteristics of housing that were found to be significant in determining rental values were: the number of rooms, areas (sq. meters), having central heating, having a telephone, urban or rural location, overall assessment of housing quality, type of housing, and existence of extra rooms (such as a garage, separatebathrooms, cellars etc.). 3.14 Inthe secondstage, once the parameters of the two regressionswere estimated, information on the housing characteristics of non-rental households was put into the equations and an imputed value of housing was estimated. This was done separatelyfor FBiHand RS. This is typically what is done inNational Accounts calculations, although BiHdoes not yet incorporate such calculations for lack of data. loSuch a regression, where the rent is related to characteristics of housing (such as location, basic amenities etc.) is called "hedonic". It assumes that consumers value positively and pay more for attractive dwelling, and value negatively (and pay less for) a dwelling with unattractive characteristics. 25 (iv) Durable Goods 3.15 All consumption expenditures on durable items are excluded from the consumption aggregate. Instead we include the rental value of consumer durables for which we have ownership information. This value can be called a consumption flow from a durable good. It i s derived by estimating the cost of owning a durable good, which involves two elements (as suggested in Luttmer (2000): (i) Depreciation: the drop in value of the good duringthe course of the year; (ii) (Forgone) real interest: the interest one could have earned if one had invested in a financial asset instead of a consumer good, or the interest one has had to pay on a loan taken out to finance the consumer good 3.16 Expressedmathematically: Consumption flow = SV + r V = (S + Y) V, where Sis the depreciation rate, i s the interest rate, and Vis the current value o f the good. Y 3.17 Hence, to be able to estimate the consumption flow of a consumer durable, we need four pieces of information: (i) whether the householdowns the consumer durable; (ii) the value of the durable; (iii) the depreciation rate o f the durable; and (iv) the interest rate 3.18 The BiH LSMS sought information about possession of, estimated current value, and age of 23 categories of consumer durables. This information was used to calculate the consumption flow of each of these 23 categories. The calculation consisted of the three steps summarized below. Step 1: Estimation o f depreciation rate and median new value 3.19 We know the value and age o f a consumer durable for a subset of households who (i) report owning each consumer durable, (ii) report the estimated market value of that consumer durable (reported as the value that respondents think they can sell the durable for), and (iii) report its age (time since the purchase or acquisition of each durable). For this subset of households (excluding the outliers defined in Step 2 below), we run a regression of the form: where vaZuek,j i s the expenditure of household i on durable k, v0,k i s the log of the value of durable of type k at acquisition, agek,j i s the age o f durable k o f household i and&j,ii s an error term. 26 Step 2: Calculate for each household the current value of its durables 3.20 We can infer the value of consumer durables for households that report owning them. For households that report both the age and the value, we use the reported value if it falls within a factor of 6 standard deviations of the median value of durables of the same type in the country as a whole. For other households, which either do not report the value or whose reported value falls outside the range, we estimate the value of the consumer durables usingthe estimated depreciation schedule: y,k -S,age,,,) A (estimated value),,, =exp($,,, h where Go,, and 8, are the estimated log of value at acquisition and depreciation rate from the median regressions and c.,kis the estimated current value of consumer durable k for household i. Step 3: Calculate the consumption flow from the durables 3.21 Finally, we calculate the consumption flow from the possession of durable k in household i as: (ConsumptionJ1ow),,k= ( 8, + r )e,,, 3.22 The real interest rate, r, i s assumed to be constant at 10p.a.%. (v) Other Non-Food Consumption 3.23 Detailed information on other non-food expenditures were also collected from households in the following areas: daily expenditures (tobacco, newspapers, etc.), transportation (fares, fuel, maintenance and parking), household cleaning products, personal hygiene products, clothing and footwear, household furnishings and services, electronic and photographic equipment and small consumer appliances, recreation and leisure activities, and equipment, data on expenditures in each of these categories. Different reference financial services, and special events. 3.24 Households were asked to provide periods were used to help the household correctly recall expenditures. Each expenditure was annualized and the sum of these values was included in the consumption aggregate together with the value o f any gifts to the households. C. AdjustedTotal Consumption 3.25 As noted in Chapter 11, sections 11, C iiand iii,adjustment of households' consumption for the purpose of valid comparisons between them, consists of two steps: price adjustment and adjustment for the composition o f a household. Before making these adjustments, this section outlines the structure o f total consumption by its key components. 27 (i) TotalHousehold Consumption 3.26 Total household consumption i s the sum of all food and non-food consumption (including derived values as described above for housing and durable goods consumption). Table 3.1 gives an overview o f the components included in the consumption aggregate and their relative importance inhousehold consumption. 3.27 Two critical imputations affect the level and the distribution of this total consumption aggregate: (i) the use of self- reported prices (respondents' estimates of the cost of buying of the corresponding item) to compute the value of consumption in-kind from own agricultural production and gifts received; and (ii) the use of imputed values (based on reported market rents and housing characteristics) for owner-occupied housing, instead of declared implicit rents for this category o f housing. 3.28 Both of these imputations reflect the current best practice approach to evaluate consumption. l1 first The takes into account the difference between producer prices and consumer prices for food-producing households (most of the in-kind consumption). The second i s designed to avoid the, typically unrealistically high self-assessment of the implicit housing rents providedby homeowners. Table 3.1 Compositionof HouseholdConsumption Compositionof Household Consumption ConsumptionCategories Annual Household Percent of total Consumption, KM consumption Foodconsumedat home 3,766 32.5 Of which: own-production 911 7.9 Of which: receivedas gift 118 1.o Foodconsumedoutside of home 552 4.8 Housing 3,920 35.6 Of which: paidrent 68 0.6 Of which: imputedrent 2,3 16 20.0 Of which: utilities 1,536 13.3 Imputedconsumptionflow from durables 330 2.9 Other non-food 3,002 25.9 Of which: goods and servicespurchased 2,066 17.9 Of which: expenditureson education 246 2.1 Of which: daily non-food expenditures 544 4.7 Of which: gifts of non-food goods and services 145 1.3 Total 11,571 100.0 Source:LSMS (2001).Note: Amounts are expressed inKM. No adjustmenthas been made for regional price differences. The means are weighted by sample weights. The unit of observation is the household. 3.29 On average, consumption is valued at 11,571 KM per household per year. Food and housing are the largest budget items, with 37 (includingfood consumedoutside of home) and 36 Agnus Deatonand Zaidi Salmon Guidelines for Constructing Consumption Aggregate", LSMS Working Paper " Series, World Bank,2002. 28 percent of the total (note that the Croatia HBS data, based on the same data processing methods, give corresponding shares very close to those for BiH, i.e., 36 percent and 32 percent o f the total). The imputed consumption flow from consumer durables constitutes only 3 percent o f the consumption aggregate. Average actual expenditures on purchases of these types of goods are excluded from the consumption aggregate, but would also amount to 3 percent if included (331.91 KM).Health expenditures, which are excludedfor reasons explained earlier, would have amounted to 5 percent of the total (555.29 KM). 3.30 The LSMS data could also be used to construct an aggregate which i s closer to actual total personal consumption expenditure as measured by in the System of National Accounts (SNA). Such an aggregate would include health expenditures and all types of expenditures on goods and services, but exclude imputed rents and the imputation of the flow of services from durables. It would also use market prices (rather than self-reported prices) for food consumption from own production. It was noted in Chapter I1that such a measure presents problems for determining welfare. It is, however, appropriate for comparison with national macroeconomic data. The expenditure aggregate would amount for a total o f 9,616 KMper household per year - a slightly lower value than our consumption aggregate, but still very much in the same range. But it i s important to keep in mind this approximately 17 percent difference when comparing all results obtained from consumption-based figures to other sources of information or other surveys. 3.31 As discussed in Chapter 11, adjustments may need to be made in household consumption data to ensure comparability. One such adjustment might be to take into account changes in prices over time. In the case of the BiH LSMS, however, the reference period was similar for all households, and interviews were conducted within a short time period, so no adjustment i s needed for this purpose. Two other possible adjustments identified in Chapter I1are discussed in the BiH context below. (ii)SpatialPrice Adjustments 3.32 Spatial (i.e., geographic) differences inprices can cause the same bundle o f goods to be more expensive in one region than in another. But these differences do not reflect differences in material well-being. Hence, we need to correct for them. 3.33 We used the Paasche price index to deflate for regional price differences, which i s theoretically better than Laspeyres12 but requires knowledge about the quantities o f all goods consumed by each household. The Paascheindex for a household living in area Y i sgiven by: l2See Grosh, Margaret andPaulGlewwe, eds, (2000). DesigningHouseholdSurveyQuestionnairesfor Developing Countries: Lessons from 15 Years of the LivingStandards MeasurementStudy Surveys, TheWorld Bank, Washington,D.C. 29 where P, i s the price index for area Y, Qk,,i s the quantity purchasedof good k inarea Y,Pk,r i s the price of good k in area Y, andpk,Oi s the reference price of good k. To implement this formula we needed to make a number of choices: (i) pricedatatouseandhowtodefinethereferenceprices, what (ii) todefinetheregions how Y, and, (iii) todowhenpricedataaremissing. what 3.34 We calculated the quantities usin the LSMS for 66 food categories and we based our price deflator only on these food price data.I!? This assumes that regional variations in the non-food prices are similar to regional variations in observed food prices. As no information on unit prices for non-food items was available at the regionally disaggregated level at the time of calculation this was the only feasible way to proceed. The regional (group of enumeration areas) food price index i s the arithmetic weighted mean of the food price indices of all households in the area. The municipalities' food price index, PrFoodi s the arithmetic weighted mean of the food price indices , o f the enumeration areas in the given municipality. Therefore this food price index adjusts for urbadrural price differentials to the extent possible. 3.35 While the LSMS does not report prices, it does report expenditures and quantities for purchased food items. This allows us to calculate unit values for each food item as the ratio of the expenditure to the quantity bought. Though unit values are not as accurate as prices because they may also capture differences in the quality of the item bought, they represent the only data source for regional price differences. We use the unit values from the LSMS to calculate a separate food price index for each household. The reference price, pk,O, i s found by taking the national median unit value for item k. This ensures that the median unit value are based on a large number of observations and i s likely to accurately reflect the true price. 3.36 Regional price differences seem to be substantial. The most expensive areas are the municipalities around Sarajevo while the cheapest area i s the rural portion of Republika Srpska. The cost of living difference between these extremes i s 30 percent. Analysis of inequality and poverty needs to take into account these regional price variations. (iii)Per Capitaor Per Adult Equivalent 3.37 As previously noted, consumption data inthe BiH-LSMS were collected at the level of the household rather than the individual. This means that an adjustment based on some allocation rule must be introduced in order to attribute their share of household resources. This is an important task: as Lanjouw et a1 (2000) have shown, varyingthe economy of scale parameter may change the relative poverty risks of different demographic subgroups o f the population, notably the elderly and children. 3.38 There are, however, no set rules for calculating the exact adjustment for children and economies of scale (Deaton and Paxon 1996, Deaton 1997). A series of tests using BiH-LSMS l3We chose not to use special price questionnaire files in each enumeration area, but to use the actual purchase prices reported by households in the survey, and to aggregate them to the level of groups of enumeration areas. 30 data i s described in the Annex 1. The results clearly indicate that we do not have, on scientific grounds, any clear reason for selecting one equivalence scale over another. 3.39 Given this problem, it i s important that the chosen equivalence scale seems plausible to people familiar with the structure o f spending in the country. One way to judge the plausibility of equivalence scale i s to consider the implications of the chosen equivalence scale on the monthly expenditures that make households with different compositions equally well-off. A final consideration for the choice of equivalence scale i s comparability and ease of communication. The per capita scale i s easier to explain to the general public than the OECD scale ("the first adult counts as one, the other adults count as 0.7, and children count as OS"), which in turn i s much easier to explain than scales involving as and 8s. 3.40 Combining the results of the tests and the t judgment of the BiH statistics team on the plausibility of various scales, and taking into account issues of comparability and ease o f communication, we decided to take the per capita scale as our baseline, but also to use other scales to see to what extent different allocation rules produced different results. 31 32 4. CONSTRUCTIONOFPOVERTYLINESINBIH Construction of apoverty line involves a number of steps and choices, as already discussed. We now apply these to the BiH 2001 LSMS data to obtain an `extreme'' and a `general' poverty line for the country. A. Extreme or Food Poverty Line 4.1 The first step in constructing a poverty line i s to decide on a welfare measure at the level of the individual. The welfare measure used here is per capita annual consumption, as previously discussed. Individuals can be ranked on this basis by consumption level, from lowest to highest. The next step i s to determine food consumption patterns, as a basis for constructing an extreme or food poverty line. Because our interest i s in people at the lower end o f the distribution, the analysis focuses on the average consumption patterns of the bottom 30 percent only, excluding the richer component of the population. As a matter o f practice, we exclude the very poorest lowest 10 percent, for two reasons: (i)these people may be so poor that their consumption patterns are unrepresentative of any normal pattern and, (ii)their observed low levels o f consumption may reflect measurement error. To avoid these potential difficulties, we use the data for individuals at the 10-30percent consumption levels. This part o f the sample is called the "reference" group. 4.2 Having obtained the reference group's basic food consumption patterns (average quantities of all food items purchased, received as gifts or consumed from own production), these averages were expressed in terms of calories. This value reflects the patterns o f consumption o f this segment of the population and, hence, i s the basis for the food poverty line. 4.3 The required level of calories per person, per day i s set at 2100 kilocalories (K Cal). For the reference individuals, on average, caloric consumption was well above this level (amounting to 3431 K Cal per day per capita on average), so the reference group average amounts were adjusted downward to create a `food basket' that provides 2100 K Cal, with the proportion of calories from each food being the same as for reference individuals. The final step was to convert these food quantities into a monetary value. 4.4 Table 4.1 shows how this was done, including monetary conversion based on multiplying adjusted quantities by observed prices, to yield the food or extreme poverty line. The value o f this poverty line was calculated to be 760 KMper person, per year. 4.5 Annex 2 reports the detailed results o f nutritional analysis of the food basket derived with this simple method. It shows that in itself this basket does not meet detailed nutritional norm relevant for a population with the demographic structure of BiH. But it also shows that one can 33 propose a basket with different quantities by item that would cost exactly 760 KMand would meet the most critical food needs to at least the 100% level. Table 4.1 Converting Food Consumption Patterns into a Food Poverty Line (perperson per day) I 1 Goods Average KCalories KCalories Adjusted Adjusted Average Cost Cost per Quantity I per obtained KCalories quantities Price per Per year (gramsj kilogram Rice 11.6 4150 48 29 7 1.63 0.01 4.2 Bread 189.0 2410 455 275 114 0.87 0.10 36.2 Poultry ... ... 27.0 820 22 13 16 4.77 0.08 27.9 ... ... ... ... ... TOTAL I ... I 3431 I 2100 I 759.86 4.6 This process illustrates a general point, that the composition of a specific food basket may and should differ across assumptions and methods. Determining a minimumfood basket i s not an exact science: all minima are based on certain assumptions about the activity levels of individuals, and the cooking methods and dietary habits of the population. What i s important i s that the minimumas defined approximates to the purchase of a basket that provides consumption of basic nutrients at least at the level of minimumnorms, or better. The exact composition of such a basket may differ depending on the methods and assumptions used, but represents a reasonable and consistent approximation of the extreme poverty line. B. Constructionof the General Poverty Line 4.7 As already noted, individuals also have non-food needs. Taking into account the need for non-food consumption requires adding an allowance for non-food goods and services to the food poverty line. The `upper-bound' method was used here to determine the value of the general poverty 1ine.l4 4.8 To determine the allowance for non-food consumption based on LSMS data, the first step was to select those individuals whose food consumption i s equal (plus or minus 5 percent) to the value of the food poverty line, in order to provide a basis for determining the general (food plus non-food) povcrty line. The share o f this group's total consumption that goes to non-food consumption i s then calculated. This share i s the `allowance' for non food consumption that i s added to the value o f the extreme poverty line to get the general poverty line. 4.9 InBiH,the share of non-food consumption among those whose food consumption equals the value of the food poverty line i s 65.5 percent: food consumption represents 34.5 percent. The value of the general poverty line i s thus: l4For details see: Martin Ravallion (1994), Poverty Comparisons Chur Switzerland, Harwood Academic Press. 34 General Poverty Line= Value of food consumption +Value of Non-Food Consumption Where: FoodConsumption=Value of FoodPoverty Line =760 KM=34.5 % of GPL Non-foodConsumption=65.5 % of GPL General Poverty Line = 760/0.345 = 2198 KM=760 + 1438. 4.10 This method of deriving the general poverty line is the simplest way to determine a minimum monetary value consistent with the consumption patterns of the poor. Other, more complicated methods are described in Annex 3. An application of these methods yields a wide array o f choices for setting an absolute poverty line based on LSMS data. This wide spectrum again highlights the basic point that recurs throughout this analysis: setting a poverty line i s not an exact science. It needs to include latitude for value judgments and (sometimes differing) expert opinions at every stage. Setting the value of the non-food component i s no exception to this rule. We have chosen the simplest method described above because it i s the most transparent, easily replicable and intuitive. But it may not be the most accurate method. For example, Annex 3 shows that a poverty line of KM 1840 might be considered as a more robust alternative poverty line. On the other hand, if a particular methodology for setting a general poverty line i s not commonly understood, its use will not help national poverty diagnostics efforts. Given the fact that any poverty line is a matter of compromise and convention, and includes value judgments, the team considered the poverty line of around KM 2,200 per capita as the most appropriate for use with the BiHLSMS dataset to analyze poverty in the country. 35 36 5. POVERTYAND THE CHARACTERISTICSOFTHE POOR INBIH Having constructed consumption-based extreme and general poverty lines, for BiH, we can now use LSMS data to determinepoverty levels in the country. Thekeyfinding is that, in 2001, there was no extreme orfood poverty, but almost one-fijlh of thepopulation had consumption levels below the general poverty line and, thus, are classified as poor. A second majorfinding is that consumption-based inequality is low in BiH. This chapter presents detailed information about the extent of poverty and inequality, and about key characteristics of thepoor. A. Poverty 5.1 The first estimate o f poverty i s based on the headcount index, or the percentage o f the population in households whose per capita consumption i s below the extreme or general poverty lines. In BiH, there i s no measurable extreme poverty: all households in the BiH-LSMS had per capita consumption levels above 760 KM per year. This does not mean that no household anywhere in the country suffers from food poverty: only that such cases are so rare that they are not captured in a sample survey. (It i s also important to note that 18.9 percent of the population spends less on food than the value o f the minimum food basket. While not all of these people can be called food poor - some may have lower physiological needs than impliedby average norms or have deliberately selected low food consumption - this finding supports the judgment that there i s definitely a measurable extent of deprivation inthe country.) 5.2 Table 5.1 presents a set of poverty indicators for BiHbased on the general poverty line. On this basis, 19.5 percent of the population is classified as poor. As the table shoes, average poverty rates vary substantially by type of location: urban municipalities have the lowest levels (13.8 percent) while mixed (urban andrural) municipalities have the highest (23.6 percent). Table 5.1 BiHPoverty Indices (Percent) Poverty Measures BiH Urban Mixed Rural 6. Head Count 19.5 13.8 23.6 19.9 95% confidence interval k p.p. k3.6 k2.8 k6.7 k6.7 Poverty Gap 4.6 2.8 5.7 4.9 Severity of Poverty 1.6 0.9 2.1 1.6 Shortfall 23.5 20.5 24.1 24.4 Source: BiH-LSMS 2001, basedon per capita consumption and general poverty line . The type of location is defined according to the 1991Census classification. Standard errors computed with the stratified sampling design (Kisheffect) correction. 37 5.3 To highlight some of the distributional aspects of poverty, Table 5.1 also provides estimates o f the depth (measured as the poverty shortfall or poverty gap) and the severity of poverty (see Chapter I1 (i)). The poverty gap is equal to 4.6 per~ent.'~Poverty severity is a measure closely related to the poverty gap but gives those further below the poverty line-the poorest-a higher `weight' in aggregation than those closer to the poverty line. Its level in BiH i s found to be 1.6 percent. The final variable, which shows a national average shortfall o f 24 percent, means that the average consumption of the poor falls 24 percent short of the general poverty line. 5.4 These data suggest that the depth and severity of poverty are not extreme, consistent with the moderate level of inequality observed in BiH (discussed in Section B below). They also suggest that both distributionally sensitive measures and poverty incidence move closely intandem by type of location: locations with high poverty incidence are also characterized by high poverty gap and severity. 5.5 Statistical estimates obtained on the basis o f any sample survey are inevitably subject to some margin of error. Surveys are designed to reveal the characteristics of the "universe" (or all households in the country) by studying only a limited number of cases. As the selection o f such cases - i.e., the respondents to be surveyed - i s a random process, each result obtained in a survey has a certain probability o f matching the corresponding value for the "universe". Previous surveys conducted in BiH by other agencies used non-probability sampling, and estimates of their precision could not be calculated. The BiH-LSMS sample allows us to calculate the level of precision of our estimates: Table 5.1 re orts the 95% confidence intervals for the point estimates of poverty incidence by locality type!6 The confidence intervals around the poverty point estimates are quite broad, as illustratedin Figure 5.1. 5.6 Figure 5.1 illustrates the caution needed when comparing poverty rates for different groups within a country. Intersecting confidence intervals for groups being compared mean that one cannot reject the null hypothesis that their poverty rates are the same. This does not, however, mean that we have to accept that the two rates are the same. What we learn from failing to reject a zero difference when the point estimate of the difference i s substantial i s that the BiH-LSMS sample i s not large enough to be able to discern a substantively large difference in rates. And, i f other sources o f information also suggest large differences in point estimates - for example, between Entities - then the observed pattern i s evidence of real differences. This underscores the needto draw on other data sources along with the BiH-LSMS data when drawing conclusions. l5A poveity gap of 4.6 percent means that if a country could mobilize resources equal to 4.6 percent of the poverty line for every individual and distribute these resources to the poor in the amounts exactly needed to bring each poor individual up to the poverty line, then, in theory, poverty could be eliminated when the transfer was made. l6Reported survey results should be accompanied by the standard errors for each value. With almost perfect (95%) certainty (called 95 percent confidence in the text) the true value lies within two standard errors of the mean estimate from the survey. This i s the format in.which Table 6 reports the LSMS data. For example, a poverty rate of 19.5% for the country has a standard error o f 0.018, therefore the Table reports the statistical range for the poverty rate as 19.5% & 3.6 percentage points (=0.018.2), meaning that, with 95% probability, the actual value for this variable inBiH lies between 15.9% and 23.1%. 38 Figure 5.1 PovertyIncidenceby Locationinthe RS andFBiH, 95 Percent ConfidenceIntervals 45% -6- lower 40% bound R! 35% -+-upper boundR! 30% 25% 20% 15% 10% 5% 0% Mixed Rural Source:BiH-LSMS 2001 Note: The graph shows poverty rates based on the general poverty line by locality type and by Entity (solid lines). The broken lines show upper and lower bounds for poverty rates at 95 percent confidence intervals (Le., rates are within upper and lower bounds with 95 percent confidence). B. Inequality 5.7 Table 5.2 presents a set of consumption inequality indices for Bih and the two Entities. As noted in Chapter 11, the decile ratio (90/10 percentile ratio) shows what multiple of the consumption of the richest person in the bottom decile is consumed by the poorest person in the top decile. The 90/10 ratio i s the product of the 90/50 ratio ("rich to middle" ratio) and the 50/10 ratio ("middle to poor"). The distance between the middle and the poor i s noticeably wider in RS than in FBiH, while the rich are above the middle by similar ratio in the two Entities. Another way of comparing groups i s to look the share of total consumption going to the top and bottom quintiles. InBiH, the poorest 20% of the population (interms of consumption) commanded about 9.5% of total equivalent consumption, while the richest 20% usedabout 35.8%. 5.8 The Gini coefficient, Theil and log mean deviation show inequality to be relatively low in the country. We also report the Gini index adjusted for scale economies according to OECD methodology, which gives a slightly lower inequality level. Finally, the table demonstratesthat the distributionof consumption inthe two Entities i s similar. 39 Table 5.2 Inequality indices for BiHand Entities InequalityIndex BiH RS FbiH Decile ratios of the consumptionper capita (ratio of consumption of the rich to thepoor) 90/10percentileratio 3.29 3.49 3.13 Middleto poor (50/10) 1.82 2.00 1.74 Richto middle (90/50) 1.81 1.74 1.80 Quintile shares of total national (entity) consumption,percent Poorest 20% of the population 9.5 9.2 9.9 Lower middle20% 14.2 14.3 14.2 Middle 20% 17.9 18.3 17.7 Uppermiddle20% 22.7 23.1 22.5 Richest20% of the population 35.8 35.1 35.8 Other inequality indices Gini index 0.26 0.26 0.26 Meanlog deviation(Theil) 0.11 0.11 0.11 Entropyindex 0.12 0.11 0.12 Gini index: usingOECD scale 0.24 0.24 0.23 Source: LSMS 2001primary data. Note: Unless otherwise stated, all measures use the general poverty line and per capita consumption. C. Poverty Incidence 5.9 This section uses LSMS data to develop poverty profiles for BiH. Two profiles are presented. The first offers poverty rates for different population groups. The second decomposes the incidence of poverty within each o f these groups. (i)Poverty Rates among Population Groups 5.10 The key findings on poverty rates by population groups are shown in Table 5.3. 5.11 The data presented provide insights into factors associated with poverty and which population groups are most at risk. This information i s essential for the development of an efficient poverty reduction strategy. 5.12 The BiH-LSMS reveals several groups whose poverty rates are above the national average. For example, children (especially in the RS) are at a strong disadvantage. IDPs and refugees have a significantly higher poverty rate than other groups. Returnees have a very highpoverty rate in RS, but their poverty rate in FBiH is significantly below the BiH average, highlighting different conditions for return in the two Entities and perhaps helping to explain observed return patterns. Another factor differentiating RS from FBiHi s that living in mixed (semi-urban) localities in the former i s associated with poverty rates twice those prevalent in the latter. 40 Table 5.3 Poverty Profile: Poverty Rates by Groups Characteristics (personal and Poverty Standard Poverty Incidence Standard Incidence, Poverty Standard household) Incidence, BiH Error' Rs Error' FRiU Error' Location Urban 13.8% 0.014 12.5% 0.027 14.3% 0.016 Mixed 23.6% 0.034 30.8% 0.059 15.6% 0.025 Rural 19.9% 0.034 23.9% 0.061 17.9% 0.039 War displacement status' Placeof residenceunaffectedby war 19.4% 0.026 23.2% 0.050 17.5% 0.028 Movedduringthe war 12.1% 0.016 14.5% 0.035 10.8% 0.015 Retumees (DPs andrefugees) 16.7% 0.033 28.4% 0.061 11.0% 0.037 RemainDPs or refugees 34.3% 0.041 38.0% 0.063 29.2% 0.037 Age of a person Children (18 andbelow) 27.2% 0.025 32.2% 0.042 24.6% 0.030 Youth(19-24) 18.1% 0.026 23.9% 0.053 14.4% 0.025 Prime working age (25-49) 19.8% 0.020 25.0% 0.040 16.8% 0.019 Preretirementage (50-55F160M) 14.4% 0.023 21.8% 0.052 9.2% 0.014 Retirement (>55 women, >60 men) 13.0% 0.015 19.1% 0.030 8.6% 0.014 Educationof the household head None 28.4% 0.052 34.5% 0.092 21.5% 0.049 Primary 25.6% 0.030 31.1% 0.057 21.6% 0.028 Secondary 9.9% 0.017 11.2% 0.026 9.4% 0.022 SecondaryVocational 18.4% 0.023 21.8% 0.040 16.6% 0.027 Junior College 9.8% 0.024 12.2% 0.043 7.8% 0.028 University 2.2% 0.008 2.7% 0.014 2.0% 0.009 General employment status of the household head Elderly' not working 18.7% 0.022 27.8% 0.042 12.6% 0.022 Workingage, not employed 28.1% 0.031 36.7% 0.054 24.7% 0.037 Employed 16.0% 0.021 19.7% 0.043 13.5% 0.020 Detailed employment status of adults4 Economicallyinactive(ILO) 20.5% 0.021 25.7% 0.042 17.9% 0.024 Student 8.0% 0.024 12.6% 0.062 5.8% 0.018 Unemployed(ILO) 29.1% 0.043 39.0% 0.082 22.0% 0.033 Employedinthe informalsector 16.7% 0.028 19.0% 0.044 14.3% 0.037 Employedinthe formal sector 12.4% 0.023 18.4% 0.051 8.2% 0.011 Registeredlabor market status of adults Dependentfamily member5 21.6% 0.028 27.8% 0.054 17.8% 0.027 Pensioner,disabled, student 13.4% 0.013 17.3% 0.026 10.8% 0.013 Registeredunemployed 24.0% 0.022 29.1% 0.038 20.8% 0.027 Registeredemployed 12.4% 0.024 18.7% 0.054 8.2% 0.011 Size of the household 1Person 2.8% 0.009 2.9% 0.010 2.7% 0.013 2 Persons 6.3% 0.010 9.4% 0.019 4.6% 0.011 3 Persons 10.6% 0.020 15.9% 0.044 7.2% 0.014 4 Persons 17.4% 0.023 21.2% 0.049 15.2% 0.021 5 andmore 33.8% 0.036 40.7% 0.063 29.1% 0.040 Total 19.5% 0.018 24.8% 0.038 16.3% 0.018 Source: BiH-LSMS 2001. Poverty i s basedon consumptionaggregate per capitawith adjustment for spatialpricevariation, andgeneralnational povertyline. 'Note: DPs standfor Intemally DisplacedPersons.' Standarderrorscorrectedfor stratified sample design.95 percentconfidence intervalis approx. k 2 st. errors aroundthe mean.'Based on migrationmodule,for childrenbelow 15 basedonhouseholdheadstatus. Pensionage 4The employmentstatus i s definedaccording to the ILOcriteria (see Chapter Ifor details of the definition). Not working, not a pensioner or disabled, andnotregisteredas unemployed 5.13 Another group with a higher than average risk of poverty are the unemployed and discouraged workers. The unemployed have poverty rates that are nearly twice as high as those among the employed. And, finally, education, or the lack of it, i s also associated with elevated poverty rates; individuals living in households with a head of household with only a primary education or less are about three times more likely to live in poverty than those where the head of household has ajunior college education. 41 5.14 The data also identify groups that, contrary to common belief, are not among the poorest, and actually do not fare much worse than average. One of these i s the elderly. Elderly individuals o f pension age are less likely to be poor than an average person in the country. Perhaps not surprisingly, another such group is those employed, especially in the formal sector. Those working inthe informal sector have a somewhat higher poverty rate, but their rate is still below the national average. 5.15 It is important to note that some groups perceived by the public as especially vulnerable and requiring specific, well-targeted help (Roma women, for example, or IDPs in collective centers) are not covered adequately by the LSMS sample. These groups are either too small or do not fall into the household sample frame. Qualitative data or additional targeted sampling of people inthese groups will be needed to assess their situation adequately. Decomposition of Poverty within Population Groups 5.16 Table 5.4 decomposes the incidence o f poverty within each o f the broad categories identified in Table 5.3, showing the percentage breakdown of each category's poverty incidence by subcategory. A number of features of the poor in BiH emerge from the analysis. For example, poverty in Bosnia and Herzegovina tends to have a young face: around a third o f all poor people are below 18. Therefore issues related to child welfare will be central for developing a strategy to reduce poverty in the long-term. 5.17 The importance of education i s reinforced by the finding that just under 60% of the poor live in households where the household head has only primary education or less. Thus, an emphasis on education reform - helping to offer the young a better education, and to provide educational opportunities for older people with little schooling - could improve the prospects o f a large share of the poor. 5.18 A third important finding is that the data reject the common perception that poverty in BiH i s fundamentally the result of unemployment. The unemployed actively searching for work (ILO definition) account for only 12.7 percent of the poor. If we look at the household as an economic unit, the conclusions are even stronger. Only one-third of the poor live in households where the head i s of working age and not employed. By contrast, over 40 percent o f the poor live in households where the household head i s working. It should be noted, however, that the unemployed do have a higher poverty rate than the employed or inactive persons. But, because this group i s not large relative to others, the fact that the unemployed have a higher poverty rate does not mean that they represent a large fraction o f the poor. Over 20 percent o f the poor are retired or incapable of working. If we add the disabled to this group, then close to a quarter o f the poor in the country are not able to work. Thus labor market programs will affect them only indirectly (through increased incomes of other household members). 42 Table5.4 Poverty Profile: Composition of the Poor Populationby Groups (percent) Characteristics (personal and Share in Share among Share in Share among Share in Share among household) Population, the Poor, Population, the Poor, Population, the Poor, BiH - BiH Rs Rs FBiH FBiH Location . Urban 25.6 18.2 18.6 9.4 29.9 26.3 Mixed 31.6 38.2 43.6 54.2 24.3 23.3 Rural 42.8 43.7 37.8 36.4 45.9 50.5 War displacement status' Placeofresidenceunaffectedby war 46.2 45.9 40.1 37.6 49.9 53.6 Movedduringthe war 29.6 18.4 28.4 16.6 30.3 20.2 Retumees(DPs andrefugees) 7.6 6.5 6.6 7.5 8.3 5.6 RemainDPs andrefugees 16.6 29.2 24.9 38.3 11.5 20.7 Age of a person Children(18 andbelow) 24.7 34.4 22.2 28.9 26.2 39.6 Youth(19-24) 9.1 8.5 9.2 8.9 9.0 8.0 Prime working age (25-49) 35.1 35.7 34.4 34.7 35.6 36.6 Re-retirementage (50-55F/60M) 9.9 7.3 10.9 9.5 9.3 5.2 Retirement (>55 women, >60 men) 21.3 14.2 23.4 18.0 19.9 10.5 Educationof the household head None 7.9 11.5 11.0 15.3 6.0 7.9 Primary 35.7 46.9 39.4 49.5 33.4 44.6 Secondary 13.0 6.6 9.5 4.3 15.1 8.7 SecondaryVocational 34.1 32.2 31.7 28.0 35.5 36.2 Junior College 4.6 2.3 5.3 2.6 4.1 2.0 University 4.8 0.5 3.0 0.3 5.9 0.7 General employment status of the household head Elder13notworking 25.5 24.5 27.2 30.6 24.4 18.9 Workingage, not employed 23.0 33.2 16.9 25.1 26.8 40.8 Employed 51.5 42.3 55.8 44.3 48.8 40.4 Detailedemployment status of adults' Economicallyinactive(ILO) 47.3 53.8 40.9 45.5 51.3 62.0 Student 3.6 1.6 3.0 1.6 4.0 1.6 Unemployed(ILO) 7.9 12.7 8.5 14.3 7.5 11.1 Employedinthe informalsector 15.0 13.9 19.9 16.4 11.9 11.4 Employedinthe formalsector 26.3 18.0 27.8 22.1 25.4 14.0 Registeredlabor market status of adults Dependent family member4 31.5 39.0 30.5 37.4 32.1 40.5 Pensioner,disabled, student 30.9 23.7 31.5 24.0 30.6 23.3 Registeredunemployed 16.0 21.9 15.5 19.9 16.3 23.9 Registeredemployed 21.6 15.4 22.6 18.6 21.0 12.1 Size of the household 1person 4.9 0.7 5.2 0.6 4.8 0.8 2 persons 13.6 4.3 12.5 4.7 14.2 4.0 3 persons 18.1 9.8 18.7 12.0 17.7 7.8 4 persons 29.5 26.3 28.0 24.0 30.4 28.4 5 personsandmore 34.0 58.8 35.7 58.6 32.9 59.0 Total 100 100 100 100 100 100 Source: BiH-LSMS 2001. Poverty i s basedon consumption aggregateper capitawith adjustment for spatialpricevariation, andgeneralnational poverty line. Notes: Basedon migrationmodule, for childrenbelow 15, informationimputedfrom the migrationstatus of the householdhead * Ofpensionage The employmentstatus is defined accordingto the ILOcriteria (seeChapter Ifor details of the definition) Not working, nota ' pensioneror disabled, andnotregisteredas unemployed Finally, while poverty rates do vary across locations, poor individualscan be found throughout BiH. Half of the poor inBiHare locatedinthe RS, half inthe FBiH Less than 20 percentof the poor live inpredominantly urban locations; hence, poverty predominates inmixed or rural communities across the country. 43 44 6. CHECKSFORROBUSTNESSOFPOVERTYFINDINGS As indicated in previous chapters, the measurement of welfare using household data is based on a range of decisions and assumptions. It is therefore important to test the results of the analysis against major altemative assumptions to determine how robust the findings are. If a change of assumptions drastically altersfindings, then caution will be needed in using the results reported in Chapter V. The poverty profile has suggested that poverty is importantly correlated with factors such as the type of municipality in which people live, displacement status, education level of the household head, employment status and household size. In this chapter, we test the robustness of these findings to changes in assumptions with respect to (i) household economies of scale (section A below) and (ii) altemative poverty lines (section B), in order to determine whether they remain valid. Section C discusses a third possibility - using a possible altemative measure of welfare based on either income or expenditure rather tan on consumption. A. Robustness Checks with Respect to EquivalenceScales 6.1 We start by 'seeing whether or not incorporating economies o f scale in the analysis would change the basic profile of the poor in BiH. As noted earlier, our analysis did not take account of economies of scale, but instead used a simple per capita allocation rule that assigned household consumption equally among all household members. W e need to check for the robustness of our results to this choice. The use of a per capita measure o f individual welfare assumes that there are no economies of scale inhousehold consumption, in the sense that the per capita cost of reaching a specific welfare level does not fall as household size increases. If this assumption i s relaxed, it could affect comparisons o f poverty between large and small households, and in turn the rankings of different household groups: for example, households made up o f the elderly are typically small, while those with many children are definitionally relatively large. 6.2 The LSMS data for BiH do not offer evidence o f statistically significant economies of scale. Nevertheless, if we were to find that deviations from the zero economies o f size assumption result in sharp re-rankings between groups, then there would clearly be reason for caution in interpreting the baseline poverty profile results. Thus we need to check whether some groups are systematically re-ranked if we run a poverty profile with a different equivalence scale assumption. If we do not find significant changes, we can conclude that our poverty profile is robust to the choice of equivalence scale. 45 6.3 To simplif the test we construct a set of poverty measures using the OECD I(old) equivalence scalel'and the OECD I1(new) equivalence scale.'* The OECD I1equivalence scale implies substantially greater scale economies than the OECD Iscale. 6.4 As any assumption about scale economies affects household level welfare for all households with more than one member, we should not apply the poverty line derived for per capita values to welfare indices derived on the basis of a per adult equivalence scale to get poverty status. Instead, we choose the poverty line so that overall poverty incidence remains the same but allows the economies o f scale to re-rank households That gives us results which are easily comparable. 6.5 The goal of this comparison i s to determine the extent to which taking into account economies of scale would affect the overall profile of poverty that we have constructed. Therefore, we test a set of key poverty profile results for the measurement assumptions, looking at the changes in incidence by location, displacement status, labor market status, education and household size. (i)Location and Poverty 6.6 Figure 6.1 shows poverty rates under different equivalence scales by municipality type. Though there i s some slight variation inpoverty incidence according to the assumptions used, with the OECD I1scale leading to a slightly higher poverty incidence in urban areas, the overall results and relative ranking are broadly consistent across all scale assumptions, and therefore are robust to these adjustments. Figure 6.1 Poverty by MunicipalityType, Comparison of Equivalence Scales Poverly by urbanlrural 35% 30% 25% 10% 50% 0% Conwmplon p r Conwmplion. pc CoMvmpPon per Cmsumpliaper Consumpton p Conrumpllonp r a a -03 II,cas e a r s d Iscde e e cecd II =ab e a 0.03 I sal* Source: BiH-LSMS2001. Nore: The two panels from left to right show results for RS and FBiH. Each vertical bar representsthe levelof poverty risk for a type of municipality (see legend onthe right). Onthe horizontalaxis are the methodologies usedto measurepoverty: per capita is inthe centre of eachpanel, the OECD I1scale i s to the left and the OECD Ito the right. Comparing the height of the bars inside eachpanel, we can see to what extent poverty rates are robust to economies of scale adjustments. ~~ "Wherethefirstadultcountsas1.0,everysubsequentadultas0.7,everychildas0.5ofan"equivalent adult". Where the first adult counts as 1, every subsequent adult as .5, every child i s 0.3 of an "equivalent adult". 46 6.7 Note that the mixed municipalities keep their rank as poorest in RS. Thus the poverty profile results are reasonably robust to changes in the methodology regarding the scale economies and the equivalence scale, in a sense that poorest and richest regions preserve their rank across methods. (ii) Poverty by Displacement Status. 6.8 Figure 6.2 provides a comparison by residence status (returnee, displaced, refugee, etc), using the different equivalence scales. The figure shows how remarkably robust the conclusion about the relative poverty o f IDPs and returnees are to different measurement assumptions. Figure 6.2 Poverty by Displacement Status, Comparison of Equivalence Scales 40% 35% 30% 25% 20% .Moved dunngthe war OReturnees 15% 10% 5% 0% Source: BIH-LSMS 2001. Note: The three panelsfrom left to right show results for RS,FBiHand BiH. Each vertical bar representsthe level of poverty risk for a group (see legend on the right). On the horizontal axis are the methodologies usedto measurepoverty: per capita i s in the centre of eachpanel, the OECD I1scale to the left and the OECD Iscale to the right. Comparing the height of bars inside eachpanel, we can see whether the poverty rate is robust to economies of scale adjustments and comparing the same type of bar across panels one sees whether rates for a group are the same across Entities. (iii)Education of the Household Head 6.9 Figure 6.3 shows that education levels, which are a key determinants o f poverty, maintain their importance regardless of the equivalence scale adopted. The per capita scale (inthe center of each panel) makes the profile somewhat "flatter" (i.e., it reduces differences between the least 47 educated and the other education categories) than the two OECD scales, suggesting a possible relationship between the level of education of the household head and household size, but the ranking i s always preserved and shows remarkable stability with respect to equivalence scale assumptions. Figure 6.3 Poverty by Education of Household Head, Comparison of Equivalence Scales [ONone mPrimary 0Secondar d 6 faSecondVo 0JunCoile Source: BiH-LSMS2001. Note: The three panelsfrom left to right show results for RS,FBiHand BiH. Each vertical bar representsthe level of poverty risk for an educational category (see legend on the right). On the horizontal axis are the methodologies usedto measure poverty: per capita is in the centre of each panel, the OECD I1scale to the left and OECD Ito the right. Comparing the height of bars inside each panel, we can see the extent to which the poverty rate i s robust to economies of scale adjustments. Comparing the same type of bar across panels one sees whether rates for a group are the same between Entities. (iv) Employment Status 6.10 We use three different measures of employment status to address the effects of including economies of scale inthe poverty numbers: the standard (ILO) definition of labor force status; the definition of labor force status using the BiH administrative classification; and status of the household head, using the ILO definition. The standard, or ILO definition classifies a person as employed if he or she i s presently working or on leave from a job. The unemployed are all those who do not presently have ajob, are actively searching for employment and are able to take a job at once if it were offered to them. All others of working age are classified as inactive. The BiH administrative classification considers those of working age and registered as not working at the Employment Service Office as unemployed, regardless o f whether they are actually inactive or, even in many cases, employed. 48 6.11 Figure 6.4, which compares the relative risk of poverty by employment status across methodologies using the L O standard definition, offers a robust picture: the unemployed have much higher risk of poverty regardless of the method used to measure poverty. In all variants the unemployed clearly stand out, while employment in all cases i s clearly associated with lower poverty rates. There i s also a clear distinction between the two Entities, whichever equivalence scale i s used. Figure 6.4 Poverty by Employment Status of Individuals(ILO Definition), Comparison of Equivalence Scales 45% 40% 35% J 30% BEcon. inactive 25% 0Student E! Unemployed(ILO) 20% OInfrml Sct Wrk tZ4FormalSct Wrk 15% 10% 5% 0% Source: BiH-LSMS 2001. Note: The three panels from left to right show results for RS, FBiHand BiH. Eachvertical bar representsthe level of poverty risk for an educational category (see legend on the right). On the horizontal axis are the methodologies used to measurepoverty: per capita is inthe centre of each panel, the OECD I1scale to the left and OECD Ito the right. Comparing the height of bars inside each panel, we can see the extent to which the poverty rate is robust to economies of scale adjustments. Comparing the same type of bar across panels one sees whether rates for a group are the same between Entities. 6.12 Figure 6.4 presents the same comparison o f equivalence scales with respect to employment status, but uses the BiH administrative classification of joblessness rather than the ILO definition used in Figure 6.4. It confirms the robustness of robustness the findings across different equivalence scales. Inspection o f Figures 6.4 and 6.5 also shows that the poverty rate for the unemployed i s much higher under the ILO definition than under the administrative (registered unemployed) definition, as would be expected from the discussion o f this variable in Chapter I. 49 6.13 Finally, Figure 6.6 looks at the effect of using different equivalence scales when the employment variable used i s the employment status of the household head, using the ILO definitions. Here, we start to see some differences across groups depending on the equivalence scale used, because demographic factors influence this set o f outcomes more than the previously listed results. For example, both OECD scales give relatively smaller differences between poverty rates for households headed by the elderly and households headed by jobless, working age adults than does the per capita scale. This result reflects differences in household size rather than deep underlying labor market determinants. Taking Figures 6.4, 6.5 and 6.6 together leads to the conclusion that thus the poverty profile results are more robust across equivalence scales in the case of individual labor market status, especially usingthe ILO definition of unemployment. Figure 6.5 Poverty by Employment Status of Individuals(BiHAdministrative Classification), Comparison of Equivalence Scales 35% 30% 0Inactivewrk capablf 25% 8Students PensianerlOisabledr 20% 15% ORegistr Unemp 10% IllRegistr Employed 5% 0% Source: BiH-LSMS 2001. Note: The three panels from left to right show results for RS, FBiHand BiH. Each vertical bar represents the levelof poverty risk for an educational category (see legend on the right). Onthe horizontalaxis are the methodologies used to measurepoverty. 50 Figure 6.6 Poverty by Employment Status of Head of Household (ILO Definition), Comparison of Equivalence Scales Source: BiH-LSMS 2001. Note: The three panelsfrom left to right show results for RS, FBiHand BiH. Each vertical bar representsthe level of poverty riskfor an educational category (see legend on the right). On the horizontal axis are the methodologies usedto measure poverty. (v) Household Size 6.14 Poverty by household size i s the variable that i s expected to be most sensitive to economies of scale assumptions. And this i s indeed what we find (see Figure 6.7). The poverty profile has a very clear stepwise shape using the per capita scale: larger households are poorer than smaller ones. Controlling for economies o f scale and the differential cost of children, however, reduces the differences in poverty rates by household size (i.e., the profile becomes much flatter). Note, however, that large households remain the poorest even controlling for scale economies using the OECD Iscale. The OECD I1scale produces a profile that reverses the relative position of one- member households, making them the poorest category by household size. Thus population sub- groups likely to fall into this category (i.e. the elderly) have much higher poverty rates using this equivalence scale. Nevertheless, even the OECD I1 scale maintains the picture o f an elevated poverty rate for the largest households (5 and more members). 51 Figure 6.7 Poverty by Household Size, Comparison of Equivalence Scales 45% c 40% 35% 30% 0Householdsize=l 25% W Householdsize=2 0Householdsize3 20% W Householdsize=4 Household 15% 10% 5% 0% Source: BiH-LSMS 2001. Note: The three panels from left to right show results for RS, FBiHand BiH. Each vertical bar representsthe level of poverty risk for an educational category (see legend on the right). On the horizontalaxis are the methodologies usedto measure poverty. 6.15 The finding that emerges from this section i s that, almost without exception, the key determinants of poverty and economic vulnerability remain essentially the same regardless of the equivalence scale used. B. Robustness Checks usingAlternative Poverty Lines 6.16 A second way of checking the robustness of the findings i s to look at the possible impact of changing the value or level o f the poverty line itself on key factors associated with poverty in BiH. To do this, we construct two alternative poverty lines to test whether the profiles o f those just above our general poverty line and those well below this general (or baseline) poverty line differ significantly from the characteristics of the baseline poor. The first alternative sets the poverty line 50% higher. This produces an overall poverty incidence of 53.1 percent. The second alternative is lower (as i s the line that i s often used in BiHto trace the evolution of poverty over time): it i s set at KM200 per month per family of four. Applying per capita conversion to this line, we get a value that i s one third of the general poverty line that we have calculated. Applying this lower line to the per capita consumption measure produces a poverty incidence o f just 5.4 percent. Obviously, changing the value o f the poverty line changes the overall incidence of poverty. But what we are concerned about i s whether alternative poverty lines re-rank households in such a way as to change the profile of poverty. To focus on this and to make results across different poverty lines 52 comparable, the incidence of poverty for each group i s expressed in relative terms, i.e. divided by the national poverty rate for each line. 6.17 To test whether key poverty profile results are sensitive to the level of the poverty line, we use the same key variables as in the previous section - location, displacement status, labor market status, education and household size. (i)Location and Poverty 6.18 InFigure 6.8 we look at povertybytype of municipality. Figure 6.8 Poverty by Location ComparingAlternative Poverty Lines 3 0 0 0 Consumptionp c Consumplionpe , Consumptonpc, Conrumptonp o Consumptionpc , Consumptionpc, bwer line general line higher line bwer line general line higherline Source: BiH-LSMS 2001. Note: The two panels from left to right show results for RS and FBiH. Each vertical bar representsthe relative poverty status of a group (see legend on the right). The horizontalaxis shows the methodologies used to measurepoverty: the baseline inthe center, the 33% lower poverty line to the left and the 50% higher line to the right. Comparing the height of bars inside eachpanel shows whether poverty incidence i s robust to the poverty line chosen. 6.19 There are significant changes in poverty incidence by type of municipality. For example, the lower alternative line produces a much sharper poverty profile in RS (the lower line gives a value for mixed municipalities that i s twice the value using the higher line, and about 25% higher than that under the "general" baseline). Using the higher alternative poverty line reduces differences by municipality types, but the overall profile i s preserved with mixed municipalities in the RS having the highest poverty values. In the Federation, differences between municipality types by alternative poverty lines are only slightly more pronounced than for the baseline; the main exception i s the significantly different pattern derived from using the higher poverty line, with mixedmunicipalities scoring higher than both rural and urban localities. 53 6.20 Nevertheless these results show some sensitivity to changes in the poverty line, in the sense that the poorest and the richest localities do not always preserve their rank across sets of lines and Entities. (ii) Poverty by DisplacementStatus 6.21 Figure 6.9 compares poverty by displacement status using different poverty lines. It shows that the conclusion about the poverty of IDPs and refugees and returnees i s extremely robust to different measurement assumptions. Figure 6.9 Poverty by Displacement Status Comparing Alternative Poverty Lines 3.0 2.0 .Moved dunngthe war ORetumees tB 1.o P E 0.0 Source: BiH-LSMS 2001. Note: The two panels from left to right show results for RS and FBIH. Eachvertical bar represents the relative poverty status of a group (see legend on the right). The horizontalaxis shows the methodologies used to measurepoverty: the baseline in the center, the 33% lower poverty line to the left and the 50% higher line to the right. Comparing the height of bars inside eachpanel shows whether poverty incidence i s robust to the poverty line chosen. (iii) Education of the HouseholdHead 6.22 Figure 6.10 shows the consistent ranking o f poverty incidence by different levels of household head's education level regardless o f the poverty line used Usingthe lower poverty line (on the left of each panel) makes the profile "sharper" (i.e. increasing differences between 54 education categories) compared to the baseline, suggestinga concentration of household headedby persons with low education at the bottom of the consumption distribution, especially inthe RS. Figure6.10 Poverty by the Levelof Educationof Household Head ComparingAlternative Poverty Lines Source: BiH-LSMS 2001. Note: The two panels from left to right show results for RS and FBiH. Eachvertical bar represents the relative poverty status of a group (see legend on the right). The horizontal axis shows the methodologies used to measure poverty: the baseline inthe center, the 33% lower poverty line to the left and the 50% higher line to the right. Comparing the height of bars inside each panel shows whether poverty incidence i s robust to the poverty line chosen. (iv) Employment Status of Adults 6.23 Again, using the three ways of looking at labor market status-the ILO definition, the BiH administrative definition, and the labor market status of the head of household (on the ILO definition)-the results are quite robust to the choice of poverty line. The conclusions are stable and rebust, regardless of the poverty line used. Inall variants, the unemployed have higher poverty rates, while those with jobs have lower rates. Rankings by labor market status are also similar across the two Entities, regardless of the poverty line used. Figure 6.11 provides the information for individuals, usingthe L O definition. 55 Figure 6.11 Poverty by the Employment Status of a Person Comparing Alternative Poverty Lines 3.0 f 2.5 e fX c 2.0 eL 0Student BUnemployed (ILO) f 1.5 glnfrml Sct Wrk fsXz 1.o n 0.5 0.0 Source: BiH-LSMS 2001. Note: The two panels from left to right show results for RS and FBiH. Each vertical bar representsthe relative poverty status of a group (see legend on the right). The horizontal axis shows the methodologies used to measurepoverty: the baseline in the center, the 33% lower poverty line to the left and the 50% higher line to the right. Comparing the height of bars inside each panel shows whether poverty incidence i s robust to the poverty line chosen. 6.24 Figure 6.12 also provides information on poverty levels by labor market status based on different poverty lines, but here usingthe BiH administrative definition of labor market status. As in the case of the comparison using different equivalence scales, the Figure 6.12 emphasizes the robustness of poverty rates by labor market characteristics, and a comparison of Figures 6.11and 6.12 shows poverty rates to be higher using the economically based ILO definition than with the administrative one. 6.25 Finally Figure 6.13 looks at relative poverty rates based on the labor market status o f the household head. Unlike the earlier robustness check for economies o f scale usingthis variable, the pattern of poverty incidence by employment status remains similar regardless of the poverty line used. 56 Figure 6.12 Poverty by Registered (Official) Labor Force Status of Adults Comparing Alternative Poverty Lines 0Inactivewrk capable PensionerlDi sableCt'Studc nts 0Registr unemp IXEmployed Registr Source: BiH-LSMS 2001. Note: The two panels from left to right show results for RS and FBiH. Each vertical bar representsthe relative poverty status of a group (see legend on the right). The horizontal axis shows the methodologies used to measurepoverty: the baseline inthe center, the 33% lower poverty line to the left and the 50% higher line to the right. Comparing the height of bars inside each panel shows whether poverty incidence is robust to the poverty line chosen. Figure 6.13 Poverty by the EmploymentStatus of the Household Head Comparing Alternative Poverty Lines Source: BiH-LSMS 2001. Note: The two panels from left to right show results for RS and FBiH. Each vertical bar represents the relative poverty status of a group (see legend on the right). The horizontal axis shows the methodologies used to measurepoverty: the baseline in the center, the 33% lower poverty line to the left and the 50% higher line to the right. Comparing the height of bars inside eachpanel shows whether poverty incidence i s robust to the poverty line chosen. 57 (v) Household Size 6.26 Poverty by household size i s the demographic variable that was shown to be the most sensitive to different equivalence scales. This i s not the case with respect to different poverty lines. Indeed, as Figure 6.14 shows, there i s near complete stability of patterns o f poverty incidence by household size for the different poverty lines tested here. The poverty profile has a clear stepwise shape in virtually every case. Thus this variable i s robust to changes in the level of the poverty line (butnot, as already noted, to adjustment for economies of scale). Figure 6.14 Poverty by Household Size Comparing Alternative Poverty Lines 3.0 2.0 BHousehold size=2 OHousehold size=3 BHousehold size=4 OHousehold s i z e d 1.o 0.0 Source: BiH-LSMS 2001. Note: The two panels from left to right show results for RS and FE3iH. Eachvertical bar representsthe relative poverty statusof a group (see legend on the right). The horizontal axis shows the methodologies used to measure poverty: the baseline in the center, the 33% lower poverty line to the left and the 50% higher line to the right. Comparing the height of bars inside eachpanel shows whether poverty incidence is robust to the poverty line chosen. C. Robustness ChecksUsing Alternative Definitionsof Well-Being 6.27 As discussed in earlier chapters, the data on poverty in this study have been developed usingconsumption as the basic measure of welfare. 6.28 In principle the welfare aggregate could be defined in terms of income or expenditure, rather than in terms of consumption. In theory, therefore, one such test might be to compare the poverty rate obtained using consumption-based and income-based measures. As previously discussed, however, income i s often measured with a large degree of error in household surveys. In the BiH case, the experience of both the pilot and the full surveys showed a large degree of under-reporting of self-employment or informal labor. On average, reported cash income of households in the LSMS i s just M of reported cash expenditures, meaning that respondents did not 58 report their true income levels. Thus the total income aggregate i s probably not a good measure to use for comparative purposes, because it i s not possible to determine whether differences are due to under-reporting or to the different concepts embodied in income and consumption. We therefore do not include an income-based test inthe sets of robustness checks described inthis chapter. 6.29 A second possibility is to measure poverty on the basis of expenditure. While we have chosen a consumption-based measure as the most appropriate for the purposes of this study, there may be elements of the process used to construct it - for example, using imputations to value owner occupied housing and consumer durables - that could drive possible alternative results. An expenditure-based welfare measure differs from a consumption-based one in three major respects: (i)it does not include any imputationsfor rental values of housing or flow of services from durables, (ii)it includes expenditures that may not necessarily be considered as welfare-increasing on technical grounds; (iii)it uses market prices, rather than own valuations reported by respondents, to value food items that are own-produced or received as gifts and consumed by households. In this sense, this aggregate is very much aligned with household expenditures collected and constructedin householdbudget surveys. 6.30 It is possible to check the robustness of these data by constructing an alternative welfare aggregate from the BiHLSMS which i s closer to the standard statistical definition of total personal expenditure as measured by the national accounts. An expenditure-based aggregate includes health expenditures and all types o f expenditures on goods and services, but excludes imputed rents for housing and flows o f services from durables. It also uses market prices (rather than self assessment evaluations) for food consumed from own production. While we consider an expenditure-based measure to be inappropriate for welfare measurement as explained earlier, it is appropriate for comparison with macroeconomic data. This measure yields a welfare aggregate of 9616 KM per household per year - a slightly lower value than our consumption aggregate. The roughly 20% difference should be kept in mind while comparing results obtained from consumption based figures to other sources of information or other surveys. 6.31 That having been said, the total value for the expenditure aggregate i s reasonably close to the consumption aggregate, and the measures for inequality are also similar. Using the expenditure aggregate to develop the poverty line needed to calculate poverty rates gives a value that differs from the poverty line used in this report, since the structure o f expenditures with respect to food and non-food items differs from a structure based on consumption. Running the same procedure as we did for consumption ,we obtain a value of 1777 KMper capita per year as a general poverty line. Applying this line to expenditure per capita, we get a poverty incidence o f 19.1 percent. This estimate, on a very different basis &e., with respect to expenditure rather than consumption) gives a nearly identical value for poverty incidence, suggesting that our prior estimate of poverty i s robust even to such a radical change in approach. 6.32 But what about the poverty profile? Are the same households classified as poor using consumption andexpenditure, respectively, as welfare indicators? Table 6.1 below shows how the division of the population covered by the LSMS into poor and non-poor using the two different methodologies. 59 Consumption based Non- Poor Poor Total Expenditure Based Non- Poor 76.68 4.22 Poor 4.03 15.07 19.1 Total 80.71 19.29 lOO.Od Source: LSMS data 2001, general poverty line, per capita. Inaverage prices. Sampling weights. 6.33 The Table shows that there i s a very significant overlap between the two methodologies: 15.07 percent o f the population i s classified as poor under both measurement assumptions and only around 1/5 o f all the poor are classified differently under the different methodologies. 6.34 We next need to do a robustness check by determining whether key vectors of poverty show substantial variation under the two alternative (expenditure vs. consumption) measurement methodologies, by running a poverty profile using different assumptions. We use the same variables used in the previous sets o f robustness checks based on equivalence scales and alternative poverty lines. (i)Location and Poverty 6.35 Figure 6.15 looks at the poverty profile by type of municipalities, using the two different welfare measures. We would expect to see quite a substantial re-ranking using expenditures as opposed to versus consumption: after all, the key differences between the two are imputed rents, which are clearly higher in urban areas, and the value of own agricultural consumption, which attracts a higher valuation at market price rather than at self-reported values. Figure 6.15 Poverty by Location Comparing Alternative Definitions of Welfare 0 35 0 3 025 g 0.2 % Durban E ORuml h: t 0.15 0 1 0 05 0 EXpendtWe Consumptm. Expendllure Consumption. pc pc w pc Source: BiH-LSMS 2001. Note: The two panels from left to right show results for RS and FJ3iH. Each vertical bar represent the level of poverty incidence for a category of municipality (see legend on the right). The horizontal axis lists a set o f methodologies used to measurepoverty i s listed: consumption for expenditure. Comparing height o f bars inside 0 each panel, we test whether poverty risk are robust to measurement assumptions. 60 6.36 Using an expenditure-based measure does affect the relative position of different types of locality: urban municipalities become poorer, rural areas a little richer. Note, however, that mixed municipalities keep their rank as poorest in RS and least poor in the Federation regardless o f the measurement assumption used. Thus the outcomes are reasonably robust to changes in the methodology, in the sense that the poorest and richest areas preserve their rank across methods. (ii) Poverty by Displacement Status 6.37 Figure 6.16 provides a comparison by displacement status using different measurement methodologies. The conclusion about the poverty of IDPs and refugees and returnees i s very robust to the two different methodologies. Figure 6.16 Poverty by Displacement Status Comparing Alternative Definitions of Welfare 0 4 0 3 025 E 0 2 d t 0 15 0 1 0 05 0 Source:BIH-LSMS2001. Note: three panels from left to right show results for RS, FBiHand BiH. Each vertical bar represent the level of poverty incidence for a category (see legend on the right). The horizontal axis lists a set of methodologies used to measure poverty is listed: consumption for expenditure. Comparing height of bars inside each panel, we test whether poverty risk are robust to measurement assumptions. (iii) Education of the Household Head 6.38 Figure 6.17 shows that the education level o f the household head-one of the key determinants of poverty-retains its importance regardless o f which welfare measure i s used. Using expenditures makes the profile "flatter", i.e. it reduces the differences between education categories, but the ranking i s almost always preserved. Recall that this variable was also remarkably stable with respect to equivalence scale assumptions. 61 Figure 6.17 Poverty by Household Head's Level of Education Comparing Alternative Definitions of Welfare 0 4 0 35 0 3 0 25 0 15 0 1 0 05 0 Source: BiH-LSMS 2001. Note: three panels from left to right show results for RS, FBiHand BiH. Each vertical bar represent the level of poverty incidence for a category (see legend on the right). The horizontal axis lists a set o f methodologies used to measurepoverty i s listed: consumption for expenditure. Comparing height of bars inside each panel, we test whether poverty risk are robust to measurement assumptions. (iv) Employment Status of Adults 6.39 Figures 6.18 and 6.19 compare the relative incidence of poverty by two definitions of employment status usingexpenditure- and consumption-based welfare measures. 6.40 The Figures show the robustness of the findings under different welfare measures. The unemployed, however defined, are poorer, and employment clearly reduces poverty incidence. There i s also a consistent pattern between the Entities, which i s preserved under any measurement methodology. 6.41 Figure6.18 defines employment status in line with the L O definition. Figure 6.19 uses the official "registered" or "administrative" classification used inBiH. 6.42 Comparing Figures 6.18 and 6.19 shows not only the robustness of the poverty profile across methodologies, but also - as in previous comparisons in this chapter - that the L O ("economic") definition of unemployment i s associated with higher poverty levels than the BiH official ("administrative") definition. 62 Figure 6.18 Poverty by Poverty by Labor Force Status of Adults (ILO definition), ComparingAlternative Definitions of Welfare mDlscrgdWrkr td B OStudent llUnempbyd DlnfrmlSct Wrk Source: BiH-LSMS 2001. Note: three panelsfrom left to right show results for RS,FBiHand BiH. Eachvertical bar representthe level of poverty incidence for a category (see legendon the right). The horizontalaxis lists a set of methodologiesusedto measure povertyis listed: consumptionfor expenditure. Comparingheight of bars insideeach panel, we test whether povertyrisk are robust to measurementassumptions. Figure 6.19 Poverty by Registered EmploymentStatus of Adults, Comparing Alternative Definitions of Welfare 03 0 25 0 2 d c 0 15 01 005 0 Source: BiH-LSMS 2001. Note: three panels from left to right show results for RS,FBiH and BiH. Eachvertical bar representthe levelof poverty incidencefor a category(see legendon the right). The horizontalaxis lists a set of methodologiesusedto measurepoverty is listed: consumptionfor expenditure. Comparingheight of bars inside each panel, we test whether povertyrisk are robust to measurementassumptions. 63 (v) Household Size 6.43 Poverty by household size i s the demographic variable that i s expected to be most sensitive to economies of scale assumptions. And this i s indeedwhat we find as Figure 6.20 suggests. Figure 6.20 Poverty by HouseholdSize, Comparing Alternative Definitions of Welfare 0.45 0.4 0.35 0.3 0.25 HHotsheold size2 OHotsheold sizes HHousheold size4 0.2 HHousheoid size=>5 0.15 0.1 0.05 0 Source: BiH-LSMS2001. Note: three panelsfrom left to right show results for RS, FBiHand BiH. Eachvertical bar representthe levelof poverty incidencefor a category (see legendon the right). The horizontalaxis lists a set of methodologiesusedto measurepovertyis listed: consumptionfor expenditure. Comparingheightof bars insideeachpanel, we test whether povertyrisk are robustto measurementassumptions. 6.44 Finally, Figure 6.20 also shows that the poverty profile by household size i s robust to the two different welfare measures. The profile has a very clear stepwise shape for both expenditure and consumption: larger households are poorer than smaller ones. D. Conclusions 6.45 The BiH-LSMS allows us to study the poverty profile in BiH in significant detail. As discussed in earlier chapters, the methodology chosen uses most o f the best practice approaches to poverty measurement. It i s based on using a comprehensive consumption measure as the welfare indicator, and on using survey-generated prices to value the poverty basket. The robustness analysis i s summarized in Table 6.2. Table lists key correlates of poverty reviewed in the pervious sections in rows. Columns represent baseline definition o f poverty (first column) and its variations in terms of scale economies, poverty line sand definition o f well being. Cells in Table 6.2 say "yes" when a listed characteristic i s strongly associated with poverty under a given measurement assumption, and "no" when such association i s only weak or reverse (the characteristic i s associated with lower poverty rate). Robustness can be assessed by comparing across columns: i s the same characteristic i s listed consistently as "yes" or "no", it i s a robust 64 correlate of poverty ("yes"), or wealth ("no"). If there is a switch between "yes" and "no", the characteristic is not robustly associated with poverty under the tested measurement assumptions. Table 6.2 Key characteristics of poverty and its robustness to measurement assumptions. Characteristics of poverty Baseline, OECD I OECD I1 Higher Lower Expenditure consumption scale scale poverty poverty per capita per capita line line Mixed(semi-urban) municipalitiesin RS Rural municipalitiesinFBiH IDPsandRefugees Householdsheadedby personswith low education(primary or less) Householdsheadedby personswith no no no no no no educationabove secondary Unemployed(ILO) andinactiveadults Yes Employedaccordingto registration no Registeredunemployed Yes Householdheadedby elderly no Largerhouseholds no Source: Staff estimates based on BiH-LSMS 2001. 6.46 This chapter has shown that the poverty profile by economic characteristics thus derived is robust to three kinds o f alternative measures - different equivalence scales (a per capita scale versus OECD measures that incorporate economies o f scale: Section A); higher or lower poverty lines (section B); and using expenditure instead of consumption as the welfare measure (Section c>. 65 66 7. FROMFEATURESOFPOVERTYTO ITS CAUSES 7.1 Having discussed constructed a poverty profile for BiH and having determined that the profile was robust to alternative assumptions and measures, we can now move a step further and discuss the causes of poverty. 7.2 Identifying the key characteristics of the poor i s an important first step in designing effective social policy to reduce poverty among currently poor households and individuals, and to prevent others from becoming poor. Since poverty in BiH i s a multi-faceted phenomenon, involving the interaction of many characteristics, it i s important to try to isolate the impact of each factor, controlling for the impact of others. Such a statistical decomposition i s also necessary given the degree of imprecisionof simple cross tabulations. Finally, this type of analysis i s helpful for predicting the poverty impact of a particular Government program. To analyze the importance of each factor in determining the poverty status of a household we conducted a multivariate analysis of poverty risk. 7.3 Figure 7.1 summarizes the rationale for this approach. The poverty profile shows simple correlations between household characteristics and poverty. For example we learned that the unemployed are more likely to be poor than the employed, indicating a correlation between poverty and unemployment. At the same time, we also know that poverty i s correlated with low education: those with primary education or less have a high incidence of poverty. But these two factors - unemployment and education - are not independent from one another. What i s the true ("net" or "partial" as economists call it) link between employment and poverty: are the unemployed poor purely because they are without work, or also because they have low education? If the latter is true, then even if poorly educated individuals find work, they are likely to get low paying jobs, so that their risk of being poor remains. If the partial correlation between unemployment and poverty i s much smaller than the corresponding simple correlation, this would imply that much of the relationship between unemployment and poverty can be accounted for by lower education, and not by the lack of employment per se. 7.4 This may seem evident in the particular case of education and employment: we know that limitededucation often means low productivity and low earnings, so that employment alone may not "solve" poverty problem for this group. The value added in multivariate analysis is that it can estimate the magnitude of these effects, and can be used to estimate not just three, but multiple links and correlations at the same time. It helpsto provide insights about which policies will have the highest impact on poverty. In addition, we can simulate the impact on policy of a particular intervention - for example, ifthe unemployed are helped to get work, by how much will their poverty risk fall? 67 Partial correlations: p Z & E q . Partial correlation Simple (controllingfo correlation Partial correlation (controllingfor unempl.) Figure 7.1 Simple versus partial correlation inpoverty analysis. 7.5 There i s no straightforward way to estimate these partial effects using micro data. For the purpose of this analysis, we have adopted a two-step approach that yields easy-to-interpret results and that uses all available information on the relationship between household characteristics and consumption. First, we estimate the relation between per capita consumption and key sets o f householdcharacteristics using an OLS regression: where eqconsi denotes (dependingon the specification) per capita or per equivalent adult basis. we calculate o f individual i,educi denotes a set of dummiesfor own education and education of the householdhead, Zfiarti i s a measure of labor force participation of a household and individual labor market status ,regioni denotes a set of regional dummy variables, dependencyiconsists of 2 control variables for dependency of children andthe elderly, sizei are control variables for household size, and vuZni are control variables for specific factors o f vulnerability (displacement status and disability in a household). The error term i s denoted by andthe a, are vectors o f ~i coefficients to be estimated. The estimates are reported inTable 7.1. 7.6 The second step i s to use this regression to simulate per capita consumption holding one o f these key characteristics constant across the population and calculate the relative poverty risks using this simulated measure of per capita or adult equivalent consumption. For example, if the regression indicates that, holding other factors constant, having a university education i s associated with a 16% increase in per capita consumption (compared to the average), we create the simulated consumption measure corrected for education by subtracting 16% from the consumption o f those with university education (and similarly for other educational categories). Next, we predict the poverty risk for each household in the dataset (depending on its particular configuration o f characteristics other than education) based on this simulated consumption, and we tabulate the relative poverty risk by education. The resulting relative poverty rates are purged from any effect runningthrough education, andthey therefore show a partial relationship. 68 Table 7.1 Regression of Log Consumption on Household Characteristics Dependent variable: Log household Dependent variable: Log household consumption per cauita consumption uer euuivalentadult Coefficient Standarderror Coefficient Standarderror l a ) Own education Unfinishedprimary, or continuing [omitted] education Primary -0.001 0.013 -0.001 0.013 Vocational Secondary 0.065 0.016 0.077 0.019 GeneralSecondary 0.076 0.018 0.065 0.016 Post Secondary 0.153 0.024 0.154 0.025 Ib)Own employment Inactive,not capable or unwilling 0.002 0.014 0.002 0.014 Discouragedworker -0.085 0.025 -0.085 0.025 Student [omitted] Unemployed -0.075 0.025 -0.075 0.025 Informalsector worker -0.077 0.027 -0.077 0.027 Formalsector worker -0.039 0.014 -0.039 0.014 2) Education of HH head Unfinishedprimary -0.392 0.050 -0.391 0.050 primary -0.186 0.033 -0.185 0.033 Vocational Secondary [omitted] GeneralSecondary -0.104 0.028 -0.104 0.028 Post Secondary 0.159 0.041 0.159 0.041 3) Laborforce participation # of employed/# of adults 0.345 0.046 0.345 0.045 4)Location RS -0.106 0.036 -0.106 0.036 FbiH [omitted] Urban [omitted] Mixed -0.054 0.034 -0.054 0.034 Rural 0.104 0.033 0.104 0.033 5)Displacementstatus Havenotmovedduringthe war Movedduringthe war 0.004 0.019 0.004 0.019 Returnees 0.054 0.040 0.054 0.040 RemainIDP/Refugee -0.119 0.032 -0.120 0.032 6)Dependency ratios # of small childhouseholdsize -0.498 0.080 -0.220 0.080 # of schoolage childrenhhd size -0.443 0.064 -0.166 0.064 # of elderlyhouseholdsize -0.085 0.043 -0.084 0.043 Disabledis present ina household -0.005 0.026 -0.005 0.026 7)Gender of household head Female 0.087 0.023 0.086 0.023 8)Household size Effectof 2"dmember -0.310 0.041 -0.148 0.041 Additional effectof 3rdmember -0.482 0.047 -0.261 0.047 Additional effectof 4" member -0.598 0.048 -0.344 0.048 Additional effect of 5" member -0.665 0.055 -0.391 0.055 Additional effect of 6", 7", 8" etc. member -0.769 0.053 -0.474 0.053 R2 0.3503 0.2689 Number of observations 16967 16967 Notes. Robust standarderrors are adjustedfor clusteringonhouseholdid. Source: 2001LSMS Survey.: per equivalent adult uses the OECD scale 7.7 Based on the regression using the log of per capita consumption as the dependent variable, the coefficients of the regression can be interpreted as partial effects measuredin percentageterms. For example, the coefficient for RS in the section of Table 7.1 covering "location" i s -.106. This therefore means that, holding all other variables constant, someone who lives inRS on average has 10.6% less consumption than a similar person in FBiH. We report here only results from a pooled 69 regression both data from RS and FBiH, where location i s controlled for. Note that running two separate regressions and then applying a Chow test on the equality of coefficients does yield negative results: the coefficients are statistically different. This, however, does not imply qualitatively different results about the determination of welfare in the two Entities. On closer inspection almost all the coefficients look alike, with only two notable exceptions: returnees and semi-urban municipalities in RS have significantly lower consumption than similar groups in FBiH,other factors beingcontrolled for. 7.8 Education turns out to be the strongest predictor of consumption, and therefore of poverty. For example, holding all other variables constant, households with a head with post secondary education on average consume 15.9% more than those with a head with secondary education, and more than 55% more (15.9% -(-39.2%)) than those with a head with an unfinished primary education. The regression includes both own education and the household head's education because both matter empirically. 7.9 Households with higher labor force participation have significantly higher consumption levels. Rural areas have significantly higher consumption per capita than urban areas, other things being equal. Finally, holding other variables constant, larger families have lower levels of consumption. 7.10 Surprisingly, this result i s also true controlling for economies of scale, as the two right- hand columns o f Table 7.1 demonstrate. Applying the per adult equivalent scale produces the same result as the per capita scale even for the effects of household size: controlling for other characteristics, larger households tend to have lower equivalent consumption than smaller ones. 7.11 Table 7.2 reports the poverty correlations in two ways. First, it shows observed poverty rates (in percentage points) by demographic subgroup, (measured as the poverty rate for that subgroup minus the overall poverty rate). Hence, a subgroup with an average poverty rate has an excess risk of 0 percentage points, but one with a rate of 25 percent as opposed to the 19 percent average rate i s 6 percentage points more likely to be poor than average, while a subgroup with a poverty rate of 5 percent i s 14percentage points less likely to be poor than average. 7.12 Second, Table 7.2 shows the partial effect of a characteristic on poverty-in other words, if all other factors included in the analysis are controlled for, or remain unchanged, the percentage shown in this column measures the effect of a characteristic on an above or below-average poverty rate. The table uses our baseline poverty measurement assumptions (per capita consumption and the "general" poverty line calculated in Chapter IV). For comparative purposes, it also provides partial risk data on a per adult equivalent basis for BiHonly. 7.13 We find that all the above or below-average (positive and negative) percentages work in the expected direction. Thus, the "partial risk" column for BiH shows that an additional dependant (be it a child, an elderly person or an adult) in a household generally increases the household's chances of being poor, other factors being controlled for. Also as expected, an additional employed person in a householddecreases the household's poverty risk. 7.14 Note that "net" risks for education remain almost as big as the simple (observed) risk presented in column 1 and dominate other factors: when the household head (usually its only 70 breadwinner in BiH) completes post secondary education, the poverty risk of the household i s reduced by 25 percentage points compared to a household headed by a person with unfinished primaryeducation, or by 40 percent comparedto a basehousehold. Table 7.2 Household Characteristics, Simple and Simulated Poverty Risks for Per Capita Scale Memo:partial Partialrisk for Equivalent Partialrisk PartialRisk Observed BiH adult, BiH Rs FBiH l a ) Own education (compared to unjinishedprimary) Primary -l%p.p +O%p.p +O%p.p +O%p.p +O%p.p Vocational Secondary -870p.p -4%p.p -470p.p -2%p.p -5%p.p GeneralSecondary -12%p.p -4%p.p -470p.p -l%p.p -5%p.p Post Secondary -21%p.p -4%p.p -5%p.p -3%p.p -470p.p I b ) Own employment(c0mpared to student) Inactive,notcapableor unwilling +lO%p.p -O%p.p -O%p.p +l%p.p -O%p.p Discouragedworker +21%p.p +6%p.p +6%p.p +2%p.p +8%p.p Unemployed +21%p.p +5%p.p +5%p.p +8%p.p +2%p.p Informalsector worker +9%p.p +4%p.p +4%p.p +l%p.p +6%p.p Formalsector worker +5%p.p +2%p.p +2%p.p +l%p.p +2%p.p 2 ) Education of HH head(compared to vocational) Unfinishedprimary +20%p.p +21%p.p +26%p.p +26%p.p +17%p.p Primary +15%p.p +ll%p.p +12%p.p +1470p.p +lO%p.p GeneralSecondary +8%p.p +6%p.p +6%p.p +5%p.p +6%p.p Post Secondary -8%p.p -4%p.p -4%p.p -8%p.p -3%p.p 3) Laborforce participation(compared to no one) Effect of 1" worker -8%p.p -7%p.p -9%p.p -9%p.p -6%p.p Additional effect of Zndworker -2%p.p -5%p.p -6%p.p -5%p.p -570p.p 4) Location (compared to urban in FBiH) RS +9%p.p +7%p.p +7%p.p Mixed +8%p.p +4%p.p +4%p.p +9%p.p +l%p.p Rural +5%p.p -7%p.p -7%p.p -l%p.p -lO%p.p 5)Displacement status (compared to" domestic") Movedduringthe war -770p.p -O%p.p -O%p.p +3%p.p -l%p.p Retumees -5%p.p -3%p.p -3%p.p +4%p.p -5%p.p RemainIDP/Refugee +13%p.p +8%p.p +9%p.p +15%p.p +2%p.p 6)Dependency ratios(compared to no dependants) Effectof 1" smallchild +7%p.p +6%p.p +3%p.p +ll%p.p +4%p.p Additional effect of 2"dsmall child +3%p.p +6%p.p +5%p.p +ll%p.p +5%p.p Effect of 1" schoolage child +3%p.p +5%p.p +2%p.p +5%p.p +5%p.p Additional effectof 2"dchild +1O%p.p +4%p.p +3%p.p +5%p.p +3%p.p Effect of 1" elderly -O%p.p +3%p.p +3%p.p +3%p.p +3%p.p Additional effectof 2"d +3%p.p +2%p.p +3%p.p +2%p.p +2%p.p 7) Gender of household head (compared to male) Female -8%p.p -5%p.p -5%p.p -6%p.p -3%p.p 8) Household size (compared to single person) Effectof Zndmember +4%p.p +7%p.p +6%p.p +1 O%p.p +5%p.p Additional effect of 3rdmember +2%p.p +5%p.p +lO%p.p +5%p.p +4%p.p Additional effect of 4" member +5%p.p +6%p.p +13%p.p +7%p.p +6%p.p Additional effect of 5" member +7%p.p +9%p.p +19%p.p +6%p.p +9%p.p Additional effect of 6", 7", 8* etc. member +12% . +28% .p +17% . +9% . Source: estimates basedonregressionresultsreportedinTable A1, 71 7.15 The interpretation of the results for own education i s not straightforward, as in this regression we control for the level of household consumption per capita (or equivalent adult), i.e. forcing all individual consumption level to be equal within each household regardless o f the education level of each member. This may in fact explain the flat profile of partial risks for own education. Note also that the poverty risks associated with pre-school age children are clearly elevated, especially in RS. 7.16 Two key findings emerge from the simulations described inthis chapter. First, the strongest determinant of poverty in BiH i s education. Thus, successful poverty reduction requires paying special attention to long-term efforts to improve the access of the poor to educational opportunities. That having been said, however, the second critical conclusion to be drawn from the simulations i s that they confirm the now widely accepted notion that there i s no single root cause of poverty. Rather, what makes people poor in BiH is a combination of misfortunes. This clearly signals the need for a multifaceted strategy to fight poverty. 72 Annex 1. Tests for Economies of Scale inHousehold Consumption To determine an appropriate equivalence scale for households inBiH, we start with the equation: discussed in Section 1I.C. (iii) above. By comparing the results using a reasonable range of values for the two parameters we test the robustness of the data. Some commonly used scales do not fall in the category of equivalence scales described by that formula, however. The OECD 1 formulation, for example, has used the following equivalence scale: Equivalent size = 1(first adult)+ (0.7 * additional adults) +( 0.5 children) * Different methods are used to set equivalence scales, but each has drawbacks. As a result, a wide variety of equivalence scales i s used in various countries. Estimates usingEngel's Method : The crucial assumption o f the Engel method i s that there i s an inverse and monotonic relationship between a household's well-being and the share of expenditure spent on food. Hence, this assumption implies that two households are equally well-off if and only ifthefoodshareintheirexpenditureisequal. Thisassumptionisquestionable, andconsequently, experts have advised against usingthis method." Hence, any estimates by this method should not be taken as definitive, but rather as one piece o f information that can aid in the selection of an equivalence scale. We estimate a semi-log formulation for Engel's relationship usingnon-linear least squares: Foodshare, = a,,+PIIn Expenditure, (Adults, +aChildren,)' +Ei 3 where Foodsharei i s the foodshare o f household i, Expenditurei i s its total household consumption The error term i s denoted by 4 while PO,PI, a and 0 are parameters to be estimated. expenditure, and Adultsi and Childreni are the number of adults and children in the household. This equation is estimated using non-linear least squares for the full sample of households in the LSMS as well as separately for both Entities. The estimates are shown in Table A-1.20 The estimates for BiH as a whole and both entities are quite consistent. The magnitude and sign of all coefficients also fit expectations: beta 0 i s positive and beta 1i s negative.The value of the intercept (close to one) i s also what one would expect, since the poorest families will spend everything on food. However, the explanatory power of the regression i s very weak, indicating significant "noise" in the data, which makes it difficult to calibrate the exact relationship. l9See Deaton,Angus, 1997, TheAnalysis of Household Surveys: A Microeconometric Approach to Development Policy, Baltimore,MD:Johns Hopkins University Press. 2o We also regressedthe food share on aquadratic polynomial in equivalenthouseholdexpenditure. Adding the quadratic term hardly improvedthe fit andyielded essentially the same estimatesfor a and 8. 73 Results for the RS are particularly "noisy", with the point estimate of 8 being 1.84, but the 90% confidence interval comprises 1. The point estimates of a in the Entity samples are strikingly different indicating that children require somewhat less resources than adults in RS, but more than adults in the FbiH. Other things being equal, the "cost" o f a child inFBiHi s double that inthe RS. This effect is so large as to drive the cost of children above that of adults for the entire country. But in both estimates the standard errors large enough to make it impossible to reject the null hypothesis that both 8 and a are equal to one. This supports what the estimate using the whole sample suggests: a simplest linear per capita equivalence scale i s our preferred estimate. Table A-1 Estimatesfor Equivalence Scale Using Engel's Method BiH (whole sample) RS Fl3iH a ,768 .326 1.433 e (.029450) (-0528318) (.0590385) .986 1.841 .862 (.019808) (.1779523) (.0207698) P O 1.080 .730 1.053 (.0314201) (.0430747) (..0380474) P1 -.OS3 -.039 -.OS3 (.0037963) (.005211) (.0045443) Adjusted R2 0.0878 0.0413 0.1087 No. of Observations 5189 2294 2895 Source: BiHLSMS (2001). Note: Standarderrors in parentheses The estimates in the previous table tell us the preferred equivalence scales for Engel methodology, but as we saw, the confidence intervals were relatively large. Hence, other equivalence scales may also be consistent with the Engel assumption. The extent to which an equivalence scale i s consistent with the Engel assumption can be tested by runningthe following auxiliary regression: Foodshare, =Do +P,In(Consumption,) +y ln(Size,) +S-+E, Kids , EqScale,,, Size, where EqSCUkkj i s the equivalent size of household i usingequivalence scale k, Sizei i s the number of members of household i and the other variables are as defined above. If equivalence scale k i s correct (and the Engel assumption holds), then the food share should be fully explained by equivalent consumption. In this case, the coefficient on household size ( f ~and the coefficient on the share o f children in the household (4should both be indistinguishable from zero in a statistical sense. This i s implemented by performing an F-test on the joint hypothesis that y =O and &O. The result of this test i s a p-value indicating the likelihood of obtaining the current estimates if the true value of yand dare zero. A higher p-value for equivalence scale k indicates that equivalence scale k i s more consistent with the Engel assumption (or a different assumption, such as subjective welfare data). 74 Such tests were carried out using BiH-LSMS data. The p-values of these tests for a number o f equivalence scales are reported in Table A-2. Unfortunately the results were rather discouraging, as the table below illustrates. TableA-2 Tests of Eauivalence Scales Engel Methodology P-valueon test of y =O and 6 =O EquivalenceScale a 0 RS FBiH (n=2294) (n=2895) Engel(whole sample) 0.8 1.0 0.0607 0.2540 OECD-I(current) 0.5 0.84* 0.0003 0.0741 ECA poverty 1.00 0.75 0.0002 0.1879 Per Capita(PC) 1.00 1.00 0.0007 0.4261 LuxembourgIncomeStudy (LIS) 1.00 0.50 0.0003 0.0022 Source: BiH LSMS (2001). Note: The equivalence scales are defined above. The LIS scale is the square rmt of the household size. The test statistics shown in Table A-2 do not accept any of the five equivalence scales when we define acceptance at p-values above .90 (for both RS and FBiH).Therefore we do not have, a scientific basis for selecting one equivalence scale over another. 75 Annex 2. Food Poverty Line: Detailed Nutritional Assessment The method for constructing a food poverty line described in Chapter IV i s focused on a certain target value of caloric intake, which i s set at 2100 KCal per day per capita. However, the composition of the basket obtained in this way may be such that several key nutrient requirements are not properly met. We thus did a further analysis to check the results against more detailed set of nutritional requirements, in addition to calories. First, to determine the caloric and nutritional needs of an average person in BiH, we divided the population (using the BiH - LSMS data) into 18 different demographic groups. Then, using the key nutritional norms by demographic group recommended by the World Health Organization (WHO) and the Food and Agricultural Organization (FAO), and the share of each demographic group in the population, the minimal amounts by nutrient needed for an average person in BiH were calculated (see Table B-1). For example, given the demographic composition of the population inBiH, the minimumfood basket i s 2239 KCal, as was assumed inthe basic derivation. Table B-1 Derivation of MinimumFoodRequirementsfor BiH Energy Protein Fat Fat Iron IodineVitamin Ribo- Niacin Folate Vitamin Thiam- Demo- (c) Min Max A flavin a,b C ine graphic Sex and Age shares (kcal) (g) (g) (g) (mg) 0%) h g h g ) (4 (pg) h g ) (md % retinal) below 1yr 950 14 11 50 350 0.5 5.4 32 20 0.2 2.73 I- ,- Boys 12-14years 2250 59 38 88 18 150 600 1.7 19.1 170 30 1.2 1.89 Boys 14-16years 2650 70 44 103 18 150 600 1.8 19.7 170 30 1.2 1.53 -Rnvs _ 16-18years 2770 81 46 108 11 150 600 1.8 20.3 200 30 1.2 1.64 i- Girls Girls Girls Girls Male Male ~~~~l~ or lactating Female >60years 1835 49 31 71 9 150 500 1.3 14.5 170 30 1.1 8.36 Female Lactating 2710 69 45 105 38 200 850 1.7 18.2 270 30 1.4 1.59 AverageBa 2239 48.9 36.9 86.19 12.27 139.5 527.3 1.4 16.0 160.4 27.7 1.0 100% _. - FAO, 1988.For riboflavin, niacin andvitamin C figures: FAO, 1982FoodandNutritionSeries - No. 29 - Given that caloric needs will not be met with a baseline basket, it i s legitimate to ask whether the food poverty line expressed in monetary terms o f KM 760 derived from it i s adequate. This i s highlighted by analysis o f the nutritional value of such a basket reported in Table B-2. A linear 76 optimization program was applied to find a composition of the food basket that could be achieved at the total cost equal to the one obtained in the baseline approach given BiH prices and the key nutritional requirements for an average citizen. Table B-2 compares the constructed basket to the primary baseline obtained by a simple arithmetical rule; it is clear a basket which meets nutritional standards. TableB-2 NutritionalAssessment of MinimumBaskets Initial Optimized Unitof Minimum Percentof Minimum Percentof NutritionItem Measurement IFood basketIRequiremendFoodBasket Requirement I Energy-. Kilocalorie 2100 95 2240 100 Protein Gram 56 115 65 134 Fat (maximum safe intake) Gram 76 205 61 71 Iron milligram 9 70 12 101 Vitamin A pgretinol 437 83 528 100 Vitamin C milligram 44 159 73 262 Thiamine milligram 1 113 1.3 126 Riboflavin milligram 1 58 1.09 75 Niacin milligram 9 58 13 78 Folate (kg1 138 86 263 164 To avoid unrealistic solutions in this linear programming exercise (in other words, to avoid constructing a diet, that while nutritionally sound, i s contrary to the consumption patterns o f the population), the actual reference group values o f consumption for most food items were taken as lower bounds. The final composition of this optimized food basket that meets nutritional standards i s shown in Table B-2. Thus our simplest method yields a total cost that, as best we can ascertain, can purchase a basket at BiHprices which meet strict nutritional tests. It should be e4mphasized that nutritional data here are only used as a tool to help construct the general poverty line. Nutritional status i s not itself the welfare indicator. One should not be surprised to find that someone living at the poverty line does not reach the nutritional requirement. The human body requires an absolute minimumfood-energy intake to maintain bodily functions at rest. These needs must take precedence over all else if one i s to survive for more than a relatively short period. Beyond that, food-energy intakes will determine what activity levels can be sustained biologically; the greater the intake, the greater the energy expenditure available, i.e., the greater one's potential activity level. Setting the food component o f a poverty line i s thus a matter of a normative judgment about what activity levels should be attainable. What is important i s that the method chosen provides a total estimate that has an adequate nutritional value; its specific composition is much less important issue. 77 PePoduct donth Unit INITIAL OptimisedMIN KCAL Price, KM Cost, MINIMUM BASKET Composition KM/Month ntiles. 78 Annex 3. Constructing A General Poverty Line There i s clearly a notion of absolute need in setting the minimum non-food requirement. For example, health i s essential for survival, and being healthy requires spending on clothing, shelter etc. Also, many activities normally deemed essential for escaping from poverty cannot be performed without participation in activities such as employment, education, etc. Hence, even people who are well short o f meeting food energy requirements spend money on non-food goods. As noted inthe recent overview by M.Ravallion2*"Of all the data that go into measuring poverty, setting the non-food component of thepoverty line is probably the most contentious". The basis for choosing a non-food requirement i s rarely transparent, and very different poverty lines can result, depending on the choices made. Questions that have to be answered to make these choices more transparent are: 0 Which group's food share should be used (that of the poor or the non-poor)? 0 Inwhat sense does the resulting line assure "basic non-food needs"? Ravallion demonstrates that there are objectively defined bounds on poverty lines. Specifically, he shows that under realistic assumptions, the poverty line cannot exceed the total spending of those whose actual food spending achieves basic food needs. This group i s therefore the key reference population group for defining the absolute poverty line. As explained in section IV.B, it i s unlikely that the data will uncover a substantial number of households whose actual food spending i s equal to the poverty line, therefore one has to set an interval around the poverty line to define this reference group. We have chosen a bound o f f 5 percent around the extreme (food) poverty line, and have obtained an average food share of 35 percent. But we observe a very large variation of the food share within this interval: the median, which i s a plausible estimate given the amount of variation, gives a 34 percent estimate. Changing the interval also shifts the estimate o f the food share. One can use non-parametric methods which impose the interval or the statistics to be estimated. To give a simple example, one can calculate the mean total expenditure of the sampled households whose food spending lies within a small interval (A 1percent ) around the food poverty line, then do the same for a larger interval (+ 2 percent ), then for a still larger interval (+ 3 percent ), etc. Then one takes an average of all these mean total expenditures and derives the food share. This gives a weighted non-parametric estimate with the highest weight on the sample points closest to the food or extreme poverty line (with weights declining linearly around this point). Applying this approach with BiHLSMS data produces a result that i s quite different from a simple calculation of a median: 32 percent with a very large 95 percent confidence interval between 22 and42 percent. Will imposing a specific form for the relationship between food consumption and welfare help? As suggested by Ravallion (1994), the food share can be estimated usinga food-share Engel curve of the form: *'RavallionM.Setting Poverty Lines: Economic Foundations of Current Practices. World Bank. 2001. 79 for sampled household i, where di i s a vector of demographic variables, yi i s the level of basket (extreme poverty line) and a , PI, consumption, ,f(yi ) i s the food consumption of a household i, bf i s the cost of the minimum food P2, y are parameters to be estimated. The value of estimate a obtained from the regression gives the average food share of those households that can just afford basic food needs. The poverty line is then given by di a*, where a* is: l/ar1 "+ pzf log(1Ja"132 This can be readily solved numerically. The application of this methodto the full sample leads to a regression result in which none of the parameters are statistically different from zero. To obtain any precision we need to clean the data, removing all observations with zero food share and those with food share 3 standard deviations below the national average. Once this cleaning is done, we obtain an estimate of the food share for the poverty line of 38 percent. This method of estimating the Engel curve produces a result which is inherently dependent on the choice of per capita equivalence scale. As proposed by Luttmer (2000), this approach can be extended and generalized to any equivalence scale. It aims at directly estimating what level of equivalent consumption corresponds to a sufficient food intake. We specify the following key variables: variable ReZFood i s the ratio of household food consumption to the cost of the minimumfood basket for the household. Hence, if ReZFoodequals one, the household spends exactly as much on food as i s required to purchase the minimum food basket for the household. The variable EqCons i s equivalent consumption for all modifications (as described in Annex 1). Our baseline poverty line i s estimated using our baseline equivalence scale, per capita, but we will also use alternative equivalence scales to test the sensitivity o f the results. We specify a log quadratic relationship between ReZFoodand EqCons, and estimate this relationship by non-linear least squares: ln(ReZFood,) = ab+alln(EqCons,) +a2ln(EqCons,)2 +gi where i indexes households, as are coefficients to be estimated, and Ei denotes the error term. After estimating this equation, we solve it for the level of equivalent consumption at which the householdjust attains the minimumfood intake (Le. ReZFood=l): where the carets on the as indicate the regression estimates, and PovZine i s the our estimate for the poverty line. Solving this equation yields: 80 The advantage o f this method i s that it i s inherently more robust to outliers, and we do not need to discard as many observations, as we did for the direct fitted Engel curve regression, to obtain statistically significant results. This approach i s becoming more intuitive as presented on the graph. The vertical axis on Figure C-1 gives actual food consumption of a household relative to the cost of the minimumbasket (in logs): it i s equal to zero when the household spends exactly as much as needed. The horizontal axis i s the consumption per equivalent unit (in our case, per capita). Each dot on the Figure i s a household in the LSMS. The upward sloping curve i s the estimated relationship between food consumption (expressed as the number of minimum food bundles consumed) and total consumption. Intersections of this Engel curve line with the horizontal line representing minimum food requirements gives the level of consumption at which basic food needs are met, i.e. the poverty line (vertical line). Figure C-1 Actual relative food consumption, fitted relative food consumption line, and derivation of the Poverty Line usingthe per capita scale. Poverty line pflOO100 : 1839.545093160685 4 i -4 i , 6 Ln8Equivalent Consumption 'P, 12 I Source: BiHLSMS and regressionresults Table C-1 shows the estimates for the poverty lines obtained using all listed methods for various equivalence scales. We start with our baseline approach, and show what differences emerge when one applies the median food share approach, or a set o f parametric approximations, and we then show the results of applying alternative equivalence scales. As the table shows, the estimates of the poverty rate are very sensitive to the method chosen for estimating the food share but, once set, are relatively insensitive to the choices of equivalence scale. Our baseline case, using the per 81 capita equivalence scale, which estimates the poverty rate at 19.5% of the population, lies toward the upper bound of the estimates presented. Table C-1 Poverty Lines Based on Various Methodof Estimatingthe Food Share and Various Equivalence Scales Methodsto derivefood shareand Poverty Characteristicsof the PovertyLine equivalencescales Rate (% or KM per household per year) Food Value Value Value Value Method and scale a e Non for for Couple Couple Food ** Single Single wlo with2 Adult Parent children children Per capita, average reference 1.00 1.00 19.5 35/65 2198 4396 4396 8792 Per capita, median reference 1.OO 1.OO 20.7 34/66 2243 4485 4485 8970 Per capita, fitted Engelcurve 1.00 1.00 15.4 38/62 2010 4020 4020 8040 Per capita, fitted relative food 1.OO 1.OO 12.3 41/59 1840 3679 3679 7358 OECD-I, median reference 0.5 0.84* 22.5 37/63 3850 5005 5775 8084 OECD-11, median reference 0.3 0.68* 22.8 36/64 3081 4622 5238 8320 ECA poverty, median reference 1.OO 0.75 21.5 33/67 3205 5390 5390 9065 Source: BiH LSMS (2001). Notes: The OECD equivalence scales are definedin Chapter V1.B. The ECA povertyvalues are from World Bankreport entitled"Making TransitionWork for Everyone: PovertyandInequality inEurope andCentralAsia." (World Bank 2000b) *Implied value, estimatebasedon BiHdemographic structure; ** Percentageon average. It i s important to note that each of the methods presented here is defensible on technical grounds. The most simple and unsophisticated approach gives results that are as plausible as much more demanding methods. 82 REFERENCES BHAS, RSIS, andFIS (2003). Living Standards Measurement Survey inBosnia and Herzegovina", Sarajevo, 2003. BHAS,RSIS, FIS andthe World Bank (2002). Welfare inBosniaandHerzegovina, 2001: Measurement andFindings, Sarajevo, 2002 (LSMS CD ROM). Deaton, Angus (1997). The Analysis of HouseholdSurveys: A Microeconometric Approach to Development Policy, Baltimore, MD:Johns Hopkins University Press. Deaton, Angus andChristinaH.Paxson(1998). Economiesof Scale, HouseholdSize, andthe Demandfor Food. Journalof PoliticalEconomy 106(October): 897-930. Deaton, Angus andSalmonZaidi (2002). Guidelines for ConstructingConsumption Aggregate, LSMS Working Paper Series, World Bank. Foster, James, J. Greer, andE.Thorbecke (1984). A Class of DecomposablePovertyMeasures. EconometricaVol. 52: 761-765 Grosh, Margaret andPaulGlewwe, eds. (2000). Designing HouseholdSurvey Questionnaires for Developing Countries: Lessons from 15 Years of the Living StandardsMeasurement Study Surveys, The World Bank,Washington, D.C. ILO (1993).Resolutionconcerning statistics of employment inthe informal sector. The.Fifteenth International Conference of Labor Statisticians, January 1993. OlsonLanjouw, Jean andPeterLanjouw (2001). How to CompareApples and Oranges: Poverty Measurement Based on DifferentDefinitions of Consumption. Review of Income andWealth, v. 47, no. 1(March): 25-42. Luttmer,Erzo (2001). PovertyandInequality inCroatia, Backgroundpaperfor Croatia-Economic Vulnerability andWelfare Study, Report No. 22079-HR, PovertyReductionandEconomic ManagementUnit,Europe and Central Asia Region, World Bank, Washington, D.C. PeterLynn (2002). LSMS-BiH SampleDesign andWeighting, PersonalCommunication,March 2002. Orshansky, Mollie (1963). "Children of the Poor." Social SecurityBulletin, v. 26 (July): 3-13. Orshansky,Mollie (1965). "Counting the Poor." Social Security Bulletin,v. 28 (January): 3-29. Ravallion, Martin (1994). Poverty ComparisonsChur Switzerland,Harwood Academic Press. Ravallion, Martin. (2001). SettingPovertyLines: Economic Foundationsof Current Practices, World Bank,WashingtonD.C. 83 The World Bank (1996). A Manual for PlanningandImplementationof the Living Standards Measurement Study by Margaret Grosh and Juan Munoz, LSMS, Working Paper No 126, WashingtonD.C, 1996. The World Bank (2000a). DesigningHousehold Survey Questionnaires for Developing Countries: Lessonsfrom 15 years of LSMS by Margaret Grosh andPaul Glewwe, 3 Volumes, Washington D.C. 2000. World Bank (2000b). MakingTransition Work for Everyone:Poverty andInequality inEurope and Central Asia. Washington, D.C. The World Bank (2002) BosniaandHerzegovina: FromAid Dependencyto Fiscal Self-Reliance. A Public Expenditureand InstitutionalReview The World Bank, Washington DC. UNDP, World Bank andDFID (2000). Building a Basis for a Statistical SysteminBiH,Project Document, Dec 2000. 84 Regional Forum for Gender Focal Points ECA Region November 24,2003 Background information and talkingpointsfor welcoming message and introduction preparedfor Annette Dixon This note contains background information and talking points for the introductory remarks to the Regional Forum for Gender Focal Points. The Forum will take place on Nov 24,2003 in MC 4-W150. You are scheduled to give he introductory remarks 9:OO- 9:15. This note i s organized into two sections: 1. General informationabout the meeting 2. Talking Points 1. GENERAL INFORMATIONABOUT THE MEETING Over the last couple o f years all country offices o f the E C A region - with the exception o f Bosnia - has appointed a member of staff to act as Gender Focal Point. None o f the appointed people is a gender specialist and most o f them to date have had very little exposure to gender analysis and gender related issues. However, in the context o f increasing decentralization o f World Bank activity and the growing emphasis on gender as an important issue o f development, the role if the Gender Focal Points in mainstreaming gender in WB activities and monitoring the implementation of the WB gender strategy is acquiring increasing importance. Last year we held in Warsaw the first learning programs designed to support the Gender Focal Points in taking on increasing responsibilities in this area. At the end o f the event the Gender Focal Point expressed an interest in meeting annually to exchange experiences and share concerns. This forum i s designed as a follow up to the Warsaw meeting. The focus o f the event i s on (i) monitoring the progress towards gender mainstreaming - at country, regional and global level -; (ii) preparing and action plan for gender mainstreaming in ECA; (iii) finalizing the TORSo f the GFPs; and (iv) establishing a regional network o f Gender Focal Point. Inaddition to exchanging experiences and sharing concerns, the objective o fthe meeting i s to set up a sustainable long distance network and. An agenda i s attached. Your introductory remarks are expected to last for about 10minutes. 2. TALKINGPOINTS During your remarks, w ewould appreciateit if you could convey the following main messages: 1. Attention to gender is essential to the Bank's and the ECA region's mission o f poverty reduction; and 2. Payingattention to gender issues is the responsibility o f all Bank staff. 3. Over the last years the ECA region has gone a long way inmainstreaming gender inWB operation and the WB has become much more visible in supporting ECA countries inimproving gender equality. 4. In the context o f increasing decentralization o f World Bank activity and the growing emphasis on gender as an important issue o f development, the role ifthe Gender Focal Points inmainstreaming gender inWB activities and monitoring the implementation o f the WB gender strategy i s acquiring increasing importance. 5. This focus event is designed to support the Gender Focal Points in taking on increasing responsibilities inthis area. I.Introduction andthanks: .. Iamdelightedtobegiventhe opportunitytowelcomeyoutothe first learning program for gender focal points. Iwouldliketothankallofyouforagreeingtoparticipateinthisevent. II.Attention to gender is essential to the Bank's and the ECA region's mission of poverty reduction: Gender inequality retards economic growth and poverty reduction. This i s a key conclusion o f the World Bank report Engendering Development-Through Gender Equality i n R ights, R esources, a nd Voice. A s the report makes clear, there is growing evidence that s everal aspects o f gender relations - the gender-based division o f 1abor; disparities between males and females inpower and resources; and gender bias in rights and entitlements - contribute to poverty and reduce the well-being o f men, women and children. Gender-responsive development actions, therefore, are critical to effective poverty reduction, Gender is essential to the ECA region's poverty reduction mission - there are important gender issues in the region: As the recent report Gender in Transition points out, the transition has had different impacts on women and men. These differences are essential to the economic and social development of the region. You will be discussing these issues in more details over the next couple o f days. III. Paying attention to gender issues is the responsibility of all Bank staff - implications for ECA staffi Gender inequalities have important policy implications for the Bank and need to be integrated into our work. We know that development policies that do not take gender relations into account and do not address such disparities will have limited effectiveness. Let me highlight a couple of areas where the policy implications o f gender inequality in the ECA region are particularly pronounced and give some examples o f how we can conduct our work ina more gender sensitive fashion: 1. Integrating gender issues in pension design: The case o f pension reform in Poland illustrates the fact that more equal labor market incentives will only provide more equal outcomes ifequality exists inlabor market opportunities. o Simulations show that after reform, average yearly female pension benefits will decrease from around 80 percent o f male benefits in the old system to 73 percent o f male benefits in the new system. By linking pension benefits more closely to contributions, reform has made labor market incentives between men and women more equal. However, labor market opportunities seem to become more unequal in Poland: the gender pay gap among highest paid workers increased by 13 percent between 1985 and 1997. o Theexperience in Poland indicates that minor reforms such as changes in retirement age can have major impacts on gender outcomes. Thispoints to the need for gender analysis prior to reform. Gender sensitive modeling of pension reforms prior to introduction can prevent unexpected increases in old-agepoverty and reduce reliance on social assistance. 2. Integrating gender into public sector down-sizing programs: Public sector downsizing often have gender-differentiated effects, for example: o Layoffs are rarely distributedevenly among menand women. Inpart this can be because women are often concentrated in the public sector (due to higher relative wages, benefits such as maternity leave, affirmative action programs, discrimination inthe private sector, etc). o Women who are laid-off from public sector jobs often suffer greater relative losses in both income (due to the fact that the public sector wage premium is generally larger for women than men) and non-income benefits (flexible hours, maternity leave, etc) than their male counterparts. o Gender analysis is an important component in understanding the negative distributional consequences of downsizing programs and in providing input to the design of retrenchment packages. I n Vietnam,for example, the Poverty Reduction Support Credit recommended the minimization of gender differentiated impacts. As in most countries, individual separation packages were to be based on two components: a lump sum and a multiple of earnings. After performing simulations of the impacts of different separation packages on various groups, the specijk weights placed on the two components were selected so as to minimize systematic over or under compensation based on gender. Apresentation on this issue is scheduledfor Tuesdaymorning Integrating gender into our work in the ECA region is a challenge. Moving forward Ithink it i s particularly important that we work to: o Mainstream gender into our main instruments: such as Poverty Reduction Support Credits (as our colleagues did in Vietnam); Structural Adjustment Credits, Poverty Assessments, Public Expenditure Reviews, and also Country Assistance Strategies. In a region where gender inequalities remain relatively small but future trends in a number o f areas are matter for concern, it i s important to recognize that economic policies and structural reforms that are gender neutral inprinciple may in practice have different effect on menand women. o Initiative new analytical work on gender: Ground breaking work is already being initiated inthe region, for example, on the economic cost o f alcoholism, and on gender issues in the informal labor market. Other potential areas include analytical work on intra-house hold allocation and time-use. o Keepin mindthat gender appropriateactionsmeans attentionto both women and men: Several emerging trends in the ECA region argue for greater attention to male-specific gender issues. The trend o f relative decline inmale life expectancy for example points to gender-specific risks -relatedtorisingunemployment andgrowingalcoholismanddepression among men - for which gender-targeted social programs may be warranted. Inthe ECA region in particular, Ibelieve we need to remind ourselves that gender issues include both male and female issues. 0 But important steps in the right direction have been taken over the last years. These progresses will be reviewed extensively over the day. IV.Conclusion: . Let me reiterate three main messages: i)we are aware that there are important gender issues in the E C A region - attention to these issues is essential to ECA's mission o f poverty reduction; ii)paying attention to gender issues i s the responsibility o f all Bank . staff, but iii)the role o f the Gender Focal Points in this context is paramount in both implementingthe gender strategy and monitoring its implementation. Development policies that do not take gender relations into account and do not address such disparities will have limited effectiveness. The ECA VP stands ready to assist and support the Gender Focal Points intheir attempt to mainstream gender into our work inthe region.