Report No. 36358-BR Brazil Measuring Poverty Using Household Consumption January 10, 2007 Poverty Reduction and Economic Management Sector Unit Latin America and the Caribbean Region Document of the World Bank TABLEOF CONTENTS EXECUTIVESUMMARY i CHAPTER 1: BACKGROUND AND CONTEXT 1 SurveyData and Welfare Measures 1 Poverty Lines 3 Wide Range of Poverty and InequalityEstimates 5 CHAPTER 2: SETTINGPOVERTYLINES FOR BRAZIL 10 A brief description of the POF survey 11 The Cost of Basic Needs (CBN) Methodof SettingPoverty Lines 14 The FoodPoverty Line 14 Adjustingthe FoodPoverty Line for the Cost of NonFoodBasic Needs 18 The FoodEnergyIntake (FEI) Methodof SettingPovertyLines 22 Comparing the CBNand FEIPoverty Linesfor Brazil 26 Conclusions 29 CHAPTER3: SPATIAL PRICEINDICESFOR BRAZIL 31 A spatial price index based on the cost of food only 31 Spatialprice indices based on the cost of food and nonfood (housing) 34 Spatial price index based on the estimated poverty lines 36 The Sensitivityof InequalityMeasuresto Adjustments for RegionalDifferencesin the Cost of Living in Brazil39 Conclusions 41 CHAPTER 4: AN UPDATED REGIONALPOVERTYPROFILEFOR BRAZIL 42 The Headcountpovertyprofile of Brazil 42 The Poverty Gap profile of Brazil 47 A poverty profile based on housingand headof householdcharacteristics 48 Conclusions 51 CHAPTER 5: SOME POLICY IMPLICATIONS OF THE REGIONAL DISTRIBUTION OF POVERTY IN A comparisonof the povertyrates usingthe Minimum Livelihood Poverty Line and the AdministrativePoverty lineof R$100 52 A comparisonof the povertyrates obtainedusinghouseholdincome 53 The coverageofthe poor by SocialPrograms 54 Concludingremarksand next steps 56 REFERENCES 58 APPENDIX 1:CONSTRUCTINGA CONSUMPTIONAGGREGATEFOR THE PURPOSEOF WELFARE ANALYSIS USING POF2002-03 65 APPENDIX 2: TESTINGTHE SENSITIVITY OFTHE FOODPOVERTY LINES WITHTHE CBNMETHOD73 APPENDIX 3: USINGA NONPARAMETRIC APPROACHTO ESTIMATE THE LOWER AND UPPER POVERTYLINES 77 APPENDIX 4: INVESTIGATINGTHE SENSITIVITY OF THE POVERTYLINES DERIVEDFROMTHE FEI METHOD 79 APPENDIX 5: ESTIMATING SUBJECTIVEPOVERTYLINES 81 APPENDIX 6: THE REGIONAL PROFILEOF THE SEVERITY OFPOVERTYINDEX AND THE POVERTY PROFILEBASED ONTHE 2002-03 POF (POVERTYLINE:MINIMUMLIVELIHOODLINE) 85 APPENDIX 7: THE RELATIVE APPROACH TO MEASURING POVERTY 87 List of Fimres Figure 1 The Evolutionof Poverty (Headcount) and Inequality(Gini) inBrazil: 1995-2004............................. 7 Figure2 RegionalPatternsandTrends of Poverty inBrazil: 1995.2004 ............................................................ 8 Figure3 19 The FoodEnergyIntake (FEI) Methodfor SettingaPoverty Line........................................ Estimatingthe Cost of NonfoodBasic Need....................................................................................... Figure4 Figure5 Settinga PovertyLine with the FEImethodinUrbanand RuralAreas.............................................. 23 Figure6 Non-parametricestimatesof the calorie-expenditure curve in urbanandrural areas of the five regions of Brazil................... ................................................................................................................. 25 Figure7 Non-parametricestim f the priceof calorie-expenditurecurve inurbanandruralareasofthe five regions of Brazil................................................................................................................................. 26 Figure8 28 Spatial Price Indicesfor Foodonly: Laspeyresvs.Paascheprice indices The CBN and FEIRegionalPoverty lines........................................................................................... Figure9 33 Figure 10 Spatial PriceIndicesfor Foodand Housing: Laspeyres vs.Paaschepriceindices.............................. 36 Figure 11 Spatialprice indicesbased on the regionalCBN and FEIPovertyLines............................................ 38 Figure 12 Spatial priceindices basedon the regionalCBN-UpperPovertyLine and the LaspeyresFoodand Spatial priceindices based on the regionalFEIand the LaspeyresFoodand HousingIndex.............39 HousingIndex.................................................................................................................................... 38 Figure 13 Figure 14 The incidence of extreme poverty (Headcountpovertyindex) by region........................................... 45 Figure 15 The incidence of minimumlivelihood poverty (Headcountpoverty index) by region........................ Figure 16 The incidenceof poverty(Headcountpovertyindex) by regionusingthe upper poverty line............46 46 Figure 17 The Coverageof the Poor by SocialPrograms......................................................................... The incidence of poverty (Headcountpoverty rate) by regionusingthe FEIpoverty......................... Figure 18 ...47 56 ListofTables Table 1 3 Comparison of PovertyLinesand Rates in Brazil................................................................................. Poverty-RelevantSurveysinBrazil: A Comparisonof Basic Features................................................ Table 2 5 Table 3 Comparisonof InequalityIndices, Various Studies .............................................................................. 6 Table 4 Poverty Ratesby Regionfor DifferentData Sets andWelfareMeasures ............................................. 9 Table 5 Inequalityby Regionfor DifferentData Sets andWelfare Measures (GeneralEntropy Measure; 0.5)9 Table6 The regionaldistribution of the total populationand its share inthe 2002-03 POF............................ 12 Table7 MeanExpendituresper capita (PCE) and Incomeper capita (PCINC) by region(in R$per personper month)........................................................................................................................................ ............................................................................................................................................................ 13 Table 8 Regionalcomposition of the 20 to 40 percentilesrelativeto the regionalcompositionof total 15 The Composition of the Basic Needs FoodBasket............................................................................. Brazilianpopulation........................................................................................................................... Table 9 16 Table 10 17 EstimatedFoodshares basedonCBN methodA (Lower) and B (Upper).......................................... The FoodPoverty Line (inR$ perpersonpermonth)......................................................................... Table 11 20 Table 12 RegionalPovertyLines for Brazil based on the CBN method(inR$per personper month).............21 Table 13 CBN vs. FEIPoverty Lines................................................................................................................. 27 Table 14 33 Spatialprice indices basedon the cost of food and housing: Laspeyresvs.Paasche.......................... Spatialprice indicesbasedon the cost of food: Laspeyresvs.Paasche.............................................. Table 15 35 Table 16 Spatial priceindicesbasedon the regionalCBN and FEIPoverty Lines............................................ 37 Table 17 The Sensitivityof InequalityMeasuresto Adjustments for Regional Differencesin the Cost of LivinginBrazil .................................................................................................................................. 40 Table 18 The Headcountpoverty index (in %) for differentpovertylines......................................................... 42 Table 19 Increasesinthe Headcountpoverty index associatedwith increasesin the povertyline .................... 44 Table 20 A poverty profile based on the 2002-03 POF(Poverty Line: Minimum LivelihoodLine) .................48 The Poverty Gap index in (%) for differentpoverty lines................................................................... Table 21 50 Table 22 The regionalprofile of poverty:The Minimum Livelihood Poverty Line vs.the R$100 ComparingHeadcountPovertyRates P(0): Consumption vs.Income in POFand Income in PNAD 54 AdministrativePoverty Line.............................................................................................................. 52 Table 23 Table 24 The Coverage of the Poor by Social Programs ...................................................................... 55 ACKNOWLEDGEMENTS This report was prepared by a team that is composed of Emmanuel Skoufias (LCSPP, Task Manager), Kathy Lindert (LCSHD, co-task manager), Maria Victoria Fazio (LCSPP), Peter F. Lanjouw (DECRG), and Phillipe Leite (consultant). Fang Lai, Laura Ripani, and Alex Pagon provided research support in the very early stages of the project. Ane Perez Orsi Castro provided excellent administrative assistance. The peer reviewers of this report were Marcelo Neri (FGV), Ricardo Paes de Barros (IPEA), Fransisco Ferreira (DECRG), and Pedro Olinto (LCSPP). Menno Pradhan, and Steven Helfand, (UC Riverside) were peer reviewers for the Concept Note of this project. In addition to the guidance and advice of the peers' reviewers, the team is grateful for the useful comments of Martin Ravallion (DECRG), Ricardo Paes de Barros (IPEA), Jaime Saavedra (LCSPP), Gabriel Demombynes (LCSPP), and various participants in formal and informal seminars at IBGE. Special thanks are also due to the members of the IBGE team and especially to Elisa Caillaux, who coordinated the IBGE team with the support of Antonio JosC Ribeiro Dias. The members of the IBGE technical team included: Cristiano Boaventura Duarte, Danielle Carusi Machado, Debora Ferreira de Souza, Isabel Cristina Martins Santos, Jailson Mangueira Assis and Viviane Cirillo Carvalho Quintaes. Duringthe various stages of the process, the following people collaborated with the technical team: Andre Wallace Nery da Costa, Antony Teixeira Firmino, Cristiane Soares, Juliano Jos6 GuimarZes Junqueira, Lucian0 Tavares Duarte, Maria Beatriz AssunGZo Mendes da Cunha, Mauricio Franca Lila and Renata PachecoNogueira Duarte.Elisa. EXECUTIVE SUMMARY MEASURINGPOVERTYUSINGHOUSEHOLDCONSUMPTION:CONTEXT This report summarizes the work undertaken as part of the Brazil Poverty Measurement Study (BRAPOV) that supported a programof analytical work and technical support via an in-depth assessment of the measurement of poverty and inequality inBrazil. Past estimates of poverty and inequality have differed significantly depending on which welfare measures and poverty lines were used. The BRAPOV builds on the vast existing literature on the distribution of wealth in Brazil and takes advantage of a new and unique opportunity for constructing new poverty profile for Brazil arising from the recent completion of the Household Budget Survey (Pesquisa de Orcamentos Familiares or POF 2002-03). Unlike past surveys, the POF 2002-03 is the first nationally representative survey to include extensive questions on both consumption and income measures of welfare. For a variety of reasons, consumption tends to be a more accurate measure of welfare than income. Brazil's household surveys have traditionally collected data on income, though some smaller sub-national surveys have collected consumption data. The detailed POF 2002-03 thus presents a significant opportunity to analyze poverty and inequality usingboth consumption and income measures, as well as through qualitative measures via the survey's questions on perceptions of poverty. By virtue of timing, this POFalso presentsa baseline for poverty andlivingconditions.' The survey not only presentedan opportunity for in-depth analysis, but also for collaborationbetween the World Bank and partners in Brazil. Indeed, the BRAPOV was carried out through a collaborative process, working closely with the Brazilian Institute of Geography and Statistics (IBGE) and, occasionally, with other partners involved in distributional analysis in Brazil, such as the Institute of Research and Applied Economics (PEA). One objective of the BRAPOV Program has been to foster collaboration, institutional development, a transfer of technology, and capacity building in our counterpart agencies (primarily IBGE)for greater ownership, dissemination and sustainability of the analysis and results. Emphasis thus was placed also onprocess as a key inputfor impact, while at the same time balancingthis objective with the needto deliver quality and timely analytical work. Specifically, the analysis in this report builds on earlier studies on the methodologies for constructing consistent poverty profiles and poverty lines (e.g. Bidani and Ravallion, 1994, Ravallion, 1998 and Kakwani, 2003) and on the measurementof poverty in Brazil (such as Ferreira et al., 2003, 2000, Rocha 1997,2000, andBarros et al., 1995,2000) inorder to construct: poverty linesfor the different regions of Brazil; detailed spatial priceindicesto capture spatial variation inthe cost of living; an updatedpoverty profile;and, "micro-area"mapsof poverty andinequality for Brazil.2 Brazil does not yet have an official poverty line. Moreover, the more frequently collected data on income (Pesquisa Nacional por Amostra the Domicilios, PNAD, and the Monthly Employment Survey, PME) do not lend themselves to the construction of poverty lines (which i s usually based on consumption information and price indices). As a consequence there i s a wide range of poverty lines used in Brazil. ' conditions- as well as the impactof key programs (suchas BolsaFamilia). IBGE plans to repeat the POF in 2006-07. This would provide an opportunity to assess changes in poverty and living The World Bank is also supportingthe poverty map initiative under the HDTAL (which helps finance some of the consultant and equipment costs that would be incurredby IBGE in the exercise), and the BRASA (which expands on the issue of using the povertymapfor targeting socialassistance programs). The most commonly used set of poverty lines for policy are the "administrative poverty lines" that are typically set at arbitrary low levels of income such as fractions of the.minimumwage (e.g., ?hor `/4 of the minimumwage). Some remain fixed after their initial calculation as a share of the minimumwage (e.g., they were set at ?hor `/4 of the minimumwage in 2002, and remained fixed at those cutoffs even when the minimum wage was increased in subsequent years). These cut-offs have been widely used for determining eligibility for social programs. In fact, most social transfer programs use these cut-offs including: the Bolsa Familia Program and its predecessors (Bolsa Escola, Bolsa AlimentaqBo, Cart50 AlimentaqBo under Fome Zero, and Auxilio Gas); state and municipal safety net programs; as well as other constitutional social assistance programs such as the BPC-LOAS programs for poor elderly and disabled. These cut-offs are also widely used in the government's Multi-Year Plan (Pluno Pluri-Anuul, PPA). In2005 the Government formed a committee to establish an official poverty line. The Poverty Line committee consists of members from the Casa Civil, IBGE, PEA, and the Ministry of Social Development (MDS). As of June 2006, an official poverty line has not been made public yet. In the absence of an official poverty line it i s important to keep in mind that the poverty lines and poverty estimates presented in this study do not represent official measures. Also, the Bank's collaboration with IBGE does not in any way imply that IBGE endorses the poverty lines presented and discussed in this report. MEASURINGPOVERTYUSING HOUSEHOLD CONSUMPTION: REPORTSUMMARY Chapter 1provides the background and context of the measurement of poverty in Brazil. Chapter 2 provides some useful background information on the POF. For the construction of the regional poverty profile of Brazil the five geographical regions of the country are divided in a total of twenty one different areas (eleven metropolitan and ten urban and rural areas). Chapter 2 also presents two different methodologies for constructing poverty lines and compares and contrasts the advantagesand shortcoming of each. The first approach to setting a poverty line is the Cost of Basic Needs (CBN) method. The CBN method aims primarily at generating absolute regional poverty lines that are consistent. Consistency requires that differences in the nominal poverty lines across regions should be entirely (or as much as possible) attributed to regional differences in cost of living. The shortcoming of the CBN method i s that it strives for consistency at the expense of specificity. Specificity emphasizes that the nominal poverty lines should be sensitive to regional differences in tastes, perceptions about poverty, and in the standard of living. The second approach is the Food Energy Intake (FEI). The FEImethod, however, puts emphasis on specificity at the riskof yielding inconsistent poverty profiles. The evidence presented in chapter 2 demonstrates that the FEImethod yields poverty lines that embody differences in preferences (or tastes) between urban and rural areas, in addition to cost-of-living differences. Based on this and other international evidence, it i s determined that the CBN method is preferable since it generates poverty lines that reflect regional differences in cost of living and not differences in tastes, level of activity, relative prices, publicly provided goods and other determinants of affluence. The chapter concludes with the recommendation that the various dimensions of consumption- basedpoverty inBrazil can be best captured by three differentpoverty lines: A food or extreme poverty line (or indigence poverty line) that is determined by the cost of a basic needs food bundle that provides the recommended caloric requirements of 2,300 kcal per capita per day from a sufficiently diverse variety of food. Households with total consumption expenditures per capita less than or equal to the food poverty line may then be considered as householdsinextreme poverty, as they are unable to satisfy the basic food needs. The analysis in .. 11 chapter 2 determined that the food or extreme poverty line in Brazil is, on average, equal to R$61 per capita per month. Moreover, the extreme poverty line varies only a little from region to region, with the lowest value of R$55 in the rural South region and the highest value of R$65 in metropolitan Sao Paulo. An intermediate poverty line that is determined by the cost of satisfyingthe minimumlivelihood needs. This poverty line is the CBN-Lower poverty line discussed in chapter 2 that adjusts upwards the food poverty line for the cost of essentialnonfood needs. The upward adjustment for the cost of essentialnonfood needs i s determined by the nonfood expenditures of households that have total consumption expenditures equal to the value of the food poverty line but forego some spending on food in order to purchase these essential nonfood items. The analysis in chapter 2 determined that the minimumlivelihood or intermediate poverty line in Brazil is on average equal to R$103 per capita per month. As i s the case with the extreme poverty line, the minimum livelihood poverty line does not vary much from region to region, with the lowest value of R$90 inthe rural South regionandthe highestvalue of R$115 inmetropolitanSaoPaulo. 0 An upper povertylinethat sets a limit to the range of useful poverty lines. This more generous poverty line corresponds to the CBN-Upper poverty line that adjusts upwards the food poverty line for the cost of basic nonfoodneeds. In contrast to the minimumlivelihood needs poverty line (CBN-Lower), the adjustment for the cost of basic food needs i s determined by the nonfood expenditures of households who just satisfy the basic food needs (i.e., their food spending is already equal to the cost of the food poverty basket). The analysis in chapter 2 determined that the upper poverty line in Brazil is on average equal to R$220 per capita per month, more than two times the minimumlivelihood needs poverty line. This report takes the view that the extreme poverty line and the minimumlivelihood poverty line are the poverty lines most relevant for policy. This choice is based on two key reasons. First, these two poverty lines are the poverty thresholds most useful for identifying the households with the most pressing needs. The extreme poverty line is useful for identifying individuals who cannot even afford to satisfy the basic food needs, while the minimum livelihood poverty line i s useful for identifying individuals who cannot afford to satisfy the basic food & nonfood needs. The second reason i s based on more practical considerations. Both poverty lines are close to the "Administrative Poverty Lines" of R$50 and R$100 per capita per month that were used to determine eligibility for one of the major poverty alleviation programs of Brazil, the Bolsa F~rni`lia.~ The fact that on average the upper poverty line i s more than three times the average food poverty line suggests that the "basic nonfood needs of poor households" are overestimated to identify extreme poverty. To the extent that cost-of-living differences are substantial between regions, uses of nominal income or consumption expenditures to measure the inequality in the standard of living in Brazil may be quite misleading. A spatial price index i s especially useful for deriving more reliable measures of inequality in Brazil, which has one of the highest inequality rates in the world. Chapter 3 compares and contrasts a Laspeyres and Paasche index for food. The spatial variability in the cost of food i s found to be small. Expandingthese price indices to include the cost of housing reveals substantially larger differences in the cost of living across regions. A direct comparison of the implicit price index obtained from the regional variability in the nominal poverty line confirms that the cost of basic needsupper poverty line provides an adequate approximation to the regional variability in the cost of living differences across regions of Brazil, and thus a consistent poverty profile. Another important finding i s that consumption inequality i s considerably lower than income inequality based on either POF or the PNAD data. Adjusting nominal expenditures per capita by the proper spatial price index leads to a decrease inthe Gini inequality measure from 0.507 to 0.481. The BolsaFamiliaper capita income eligibility thresholds were just raised (in April 2006) from R$100 (upper threshold) and R$50(extremepovertythreshold) to R$120andR$60. 111 ... Based on the estimated poverty lines, chapter 4 of the report constructs an updated poverty profile from an entirely new source (consumption expenditures from the POF). The profile builds on the rich existing literature on poverty inequality, including the work conducted under the last Poverty Assessment (Report No. 20475-BR, 2001) and the work by Ferreira, Lanjouw and Neri (2003 and 2000). This poverty profile examines patterns inthe distribution of poverty usinga variety of poverty measures (absolute incidence of poverty, headcount index, PI index of poverty depth, and p2 measure of poverty severity). The profile examines these patterns across a variety of characteristics of the population, including, inter alia: (a) region and area of residence (metropolitan areas, non-metropolitan urban, and rural areas); (b) housing characteristics (e.g., housing status, water, sanitation, electricity, waste disposal, and access to paved road); (c) characteristics of the household head, such as gender, age, race, education, migration status, and occupational category. Overall the poverty profile based on consumption expenditures from the POF turns out to be in remarkable agreement with the existing poverty profiles generated by income measures in the PNAD surveys (e.g. Ferreira et al., 2003). According to the estimates, approximately 8.5% of the Brazilian population did not have total consumption expenditure sufficient to buy the basic needs food bundle. Given the total population of Brazil, the estimate of extreme poverty in Brazil implies that 14,903,203 individuals are inextreme poverty. The poverty estimatesincrease substantially when higher poverty lines are used to take into account basic nonfood expenditures. The minimumlivelihood poverty line implies poverty rate of 21.5% for Brazil which amounts to 37,696,336 individuals being unable to meet basic food and nonfoodneeds. The aggregate statistics for Brazil also conceal very large regional disparities. The six regions with the highest extreme poverty and minimum livelihood poverty are the rural areas in the Northeast, with an extreme poverty index of just under 3196, followed by the rural areas in the North, urban areas in the Northeast, rural areas in the Center-West region and Southeast region, and urban areas in the North, which has an extreme poverty of 9.5%. The aggregatefood poverty gap in Brazil, measuring the average distance below the food poverty line (as a proportion of the food poverty line), i s estimatedto be 2.5%. This food poverty gap represents0.45% of the country's aggregate consumption of all goods and services. Using the minimumlivelihood poverty line of R$103 per person per month, the value of the poverty gap index increasesto 7.3% which amounts to 0.767% of the country's aggregate consumption. These estimates suggest that the potential gains from targeting are quite large in Brazil. For example, the costs of assuring that everyone in the country can afford the poverty food bundle without targeting is about 40 times the cost of a perfect targeting scheme that transfers an amount equal to household-specific poverty gap to each poor household. It i s important to clarify that this estimate represents only the potential gains from perfect targeting, where "perfect" i s defined as providing tailor-made household-specific transfers that vary depending on the distance of household consumption from the poverty line. The extent to which such potential gains can be realized in practice depends on the constraints and the costs faced by policy makers in identifying the household- specific poverty gaps. Chapter 5 of the report explores some of the policy implications that can be derived from the regional distribution of poverty in Brazilbasedon the poverty analysis usinghousehold consumption. The poverty estimates from the consumption-based poverty lines are compared to the poverty estimates of the "administrative poverty line" (the R$100 and R$50 formerly used for Bolsa Familia and other programs). It is found that the enforcement of the same poverty line that was used as athreshold for eligibility to the Bolsa Familia program tends to result in some leakages (inclusion errors) in the rural and non- metropolitan urban areas and some undercoverage (exclusion error) of the poor in the metropolitan iv areas.4 Even though these estimates are not intended to provide an assessment of the targeting performance of the Bolsa Familia program, they do suggest that improvements of the targeting of the programcould be achievedeasily by employing poverty lines that vary from region to region. A comparison of the poverty and inequality rates using income from the POF survey, instead of consumption as a measure of household welfare is also presented. With the POF per capita consumption expenditures as the standard of comparison, the income per capita measure from POF tends to overestimatepoverty inthe non-metropolitan urban areas and underestimateit inthe ruralareas. Also, an analysis of the coverage of the poor by the social program contained in the POF confirms that social assistance programs, such as the Bolsa Escola andAuxilio Gas that are currently merged into Bolsa Familia, are much better targeted towards the poor in comparison to the social insurance programs. Although the objective of social insurance programs i s primarily protection from poverty rather than redistribution of income to the poor, these findings suggest that it is also important to reconsider the level of fiscal resources allocated to financing the deficits of social insurance programs especially in light of the fact that less room i s left in the government budget for spending on better targeted social assistance programs. One more comparison worthy of serious consideration in the future concerns the regional distribution of poverty based on the minimum livelihood poverty line and the regional distribution of federal funds for poverty alleviation. To the extent that the distribution of federal funds does not match the regional distribution of poverty, a re-alignment in the regionalallocation of federal funds may be called for. The BRAF'OV also provided the foundations for building a consumption-based poverty map. Poverty maps are especially useful for identifying the geographic variations in poverty within the twenty one different regions that are represented in POF, and they can be used for the design and better targeting of programs, budget allocation, and for monitoring and evaluation. Micro-area poverty maps are constructed usingeconometric techniques5that combine data from the 2000 census with data from the 2002-03 POF. By combining census and household survey data, the poverty maps benefit from the strengths of each data source: the complete coverage of households in the country with the census, and the more precise measures of household consumption and income from the POF. Statistical models are developed using "explanatory variables" in the household survey that are also included in the census. Once robust models have been identifiedto predict consumption (andor income) in the household survey (usingthis common set of explanatory variables), these models are applied to census data at the household level to predict per capita consumption (or income) in the census (including an error estimate). These household-level estimates are then aggregated to small statistical "micro areas" to obtain robust estimates of the percentage of householdslivingbelow the poverty line inthese areas. As of June 2006, the official poverty lines have not been made public yet and thus the work on poverty maps remains incomplete. It i s hoped that the completion and publication of official IBGEpoverty maps will be completed in the near future. For poverty maps, the Bank adopted a "transfer of technology" and "capacity-building" approach. Specifically, IBGE actually i s in the process of implementing the work It is important to note that some degree of leakage occurs in all programs, regardless of income thresholds. In fact, Brazil's recent PNAD 2004 shows that, on average, the Bolsa Familia programis quite well targeted (in terms of minimizing leakages), with 73% of all benefits going to the poorest quintile and 94% going to the poorest two quintiles. These results put Bolsa Familiaas the most accuratelytargetedpublic transfer programinLAC. SeeLindert, Skoufias,andShapiro (forthcoming). These methods were pioneeredby researchers at the World Bank in 1996 (Hentschel and Lanjouw, 1996). The techniques have been further refined, mostly under the leadership of researchers at the World Bank in collaboration with universities and in- country partner institutions (e.g., Hentschelet. al. 1998, Hetschel et. al. 2000 and Elbers, Lanjouw and Lanjouw, 2002). These maps have been appliedto numerous countries aroundthe world. Henninger and Snel (2002) summarizesexperiences with the development anduseof povertymapsin severalcountries. V ("learning-by-doing"), with the World Bank providing training and technical assistance (via formal seminars and workshops, on-going training and supervision, regular missions, continuous feedback via email, etc.). This collaboration has already resulted in a significant transfer of technology to build IBGE's capacity for carrying out such work, conducting further analysis, and implementing future updates. In fact, IBGE sees the "poverty map project" as an innovative chance to integrate its own data instruments (census, surveys, GIS, etc.) and staff across the institution. The IBGE "poverty map team" comprised of some 10-15 people with different professional backgrounds (e.g., statisticians, economists, etc.) and coming from different departments in IBGE (e.g., those responsible for household surveys, the census, the GIS unit, etc.). vi CHAPTER 1:BACKGROUNDAND CONTEXT Poverty and inequality have been the subjects of numerous studies in Brazil, reflecting their relatively highlevels and hence importantplace indebates about the country's development experience. Estimates of poverty and inequality vary widely, depending on data sources, welfare measures; and poverty lines used. SURVEY DATA AND WELFAREMEASURES Poverty i s traditionally measured through monetary measures of income and consumption through nationally representative household surveys. Brazil has a long-standing tradition of collecting household survey data, carried out by the National Statistics Office (IBGE).6 The more frequent surveys (PNAD, and the employment survey, PME) have generally focused on the collection of income and employment data, while intermittent surveys (POFs) have collected more complete information on income, consumption and expenditures (Table 1). Consumption i s generally viewed as a more accurate measure of welfare for poverty analysis since (a) consumption tends to fluctuate less than income in the short run; (b) income commonly suffers from numerous measurementerrors due to informal, seasonal and in-kind earnings; (c) there is a strong life-cycle pattern inincome since it typically rises and falls inthe course of one's lifetime; and (d) measures of income can suffer perceived incentives for under-reporting by respondents. For the purpose of analyzing poverty and inequality in Brazil, each survey has its advantages and disadvantages reflecting the cost trade-offs between frequency, depth of questionnaires and sample size (Table 1). The most frequent survey, the PME (employment survey), i s conducted on a monthly basis but is deficient for poverty analysis due to its sub-national sample and limited questionnaire (providing data mainly on labor earnings). Instead, most estimates of poverty and inequality have been based on the PNAD (household survey) due to its regular frequency (annual) and large coverage (almost national sample). Despite these advantages, the PNAD questionnaire has a number of shortcomings for the purposes of poverty analysis, including incompletemeasurementof income (particularly for income from transfers, housing, in-kindbenefits, self-employment, agricultural production for own-consumption), and a lack of data on consumption and expenditures. The use of the incomplete measure of income as an indicator of welfare may result in inaccurate measures of poverty for two important groups: self- employed informal sector workers and cultivating households? The 1996/97PPV (Living StandardsMeasurement Survey) was designed to fill some of the data gaps left by the PNAD. It provides a much more detailed picture of household expenditures and consumption (in addition to income), as well as the utilization of publicly subsidized services in education, health and transportation. However, the survey sample i s relatively small and not representative for the entire country. Moreover, the survey was only conducted once (as a pilot in 1996),andthe results are now quite out of date. Most recently, IBGE fielded a household budget survey (the POF 2002-03), primarily to generate information neededto update the consumption basket for price indices. This new POF i s both nationally representative and extensive in its questionnaire coverage of income, consumption and expenditures. It thus provides a new and unique opportunity for a thorough analysis of the measurement of poverty and inequality inBrazil, as discussedbelow. See Bianchini andAlbieri (1998 and 2002 (revision)) for acomparisonof many of IBGE's surveys, includingthe PNAD, PME, POFandPPV; see Paesde Barros, MendonqaandNeri (1995) for acomparisonofthe (former) POFsurveysandthePNAD. See recentwork by Ferreira, Lanjouw andNeri (2000 and 2003) for an analysis of these shortcomings. 1 Comparative studies of the various surveys suggest that differences in surveys and welfare concepts yield significant differences in welfare distributions (income, consumption, expenditures) and measures of poverty and inequality. Specifically, BarrosandMendonqa (1992) compare the information on income from the PNAD and PME and encounter considerable differences. Similarly, Rocha (1993) finds that poverty estimates based on the earlier ENDEF and POF surveys are quite different. More recently, two studies have compared the results emerging from (a) the PNADand the (old) POF; and (b) the PNAD and the PPV andfind important differences inthese cross-survey and cross-concept comparisons: PNAD vs. (old) POF: Differences in income measures. Barros et al. (1995) compared the income measures captured in the PNAD and POF for the overlapping metropolitanareas in 1996. First, they found considerable discrepancies in the cumulative distributions and means of income from the PNAD and the POF.* Second, they found even larger differences in distributions and means for total family income, which i s not surprising given the POF's relatively more in-depth treatment of non-labor sources of income. These differences yield significant differences in inequality and poverty estimates. The cross-survey estimates would likely diverge even more with coverage of rural populations for whom non-formal labor earnings carry more weight. Income vs. consumption and expenditure: previous POF surveys. Barros et al. (1995) also compared income, consumption, and expenditure measures from the 1995/96 POF itself (again for the metropolitan regions that it covered that year). They found considerable differences between income and consumption, but more consistency between income and expenditures. These differences applied to the cumulative distributions, means, and consequently, the estimates of poverty and inequality. Comparingand Combining the PPV and the PNAD. Elbers et. al. (2003) compare estimates of poverty and inequality for the Northeast and Southeast regions usingdata from the PNADand PPV surveys, both for 1996.9 First, they find that poverty and inequality are statistically significantly higher when measured using income from the PNAD versus consumption from the PPV. The patterns of poverty and relative rankings across regions, however, are quite similar between the two surveys. Second, they find very comparable estimates and patterns of poverty and inequality between the (observed) consumption measures from the PPV and an imputed measure of consumption in the PNAD.1° Poverty and inequality estimated on the basis of consumption in the PNAD (and PPV) tend to be much lower than estimates basedon the income measure of well-being. The study demonstrates that the differences inestimates of poverty in the PNAD and the PPV are not due to non-comparability of these surveys, but rather due to differences in the measurementandconcept of income and consumption. *Moreover, they found considerable discontinuities,with discrete shares of the populationlocatedat specific income levels, in the PNAD labor incomeestimates, as compared with those of the POF, which presented far more continuity in the cumulative distributionfunction. The questionnairesdiffer even intheir treatment of labor income, with POFestimatesbasedon an average of six retrospective responses given by each person for hiskier income over the past six months, and the PNAD referring to a singlepersonalresponseabout monthlylabor incomenormallyreceived. Usingthe same povertyline across the surveys. loThis consumption variablei s imputedusingeconometric consumptionmodels derivedfrom the PPV that are then pluggedinto the PNAD using comparable "explanatory" variables. See Elbers et. al. (2003) and Elbers et. al. (2001) for methodological details. 2 Table 1 Poverty-Relevant Surveys inBrazil: A Comparison of BasicFeatures Sample survey Periodicity Geographic Coverage Size Welfare Advantagesfor Disadvantagesfor HH Measures PovertyAnalysis Poverty Analysis PNAD Annual since Incomplete (house-hold 1967(exceptin @ $ Sample: National measurement of income income years of the ~ ~ t i ~ l , 105,000~ ~ ~Income coverage(almost) Lack comprehensive survey) . Census) the North) (2Cw Frequency(annual) consumption and expenditure data Smile: not-national PME Lack comprehensive (employment Monthly since Six Metropolitan 37,212 Labor Earnings Frequency(monthly) data onincome (only survey) 1980 Regions (latest) Partialpanelsample labor earnings) Lack consumption and expendituredata North-East and Income, Sample:not-national PPV (LSMS) 1996/97 South-East 4,944 Consumption questionnairefor Frequency: single Regions and income, consumption Exuenditures andexpenditure survey, out-of-date ENDEF National 55,000 POF-old Eh'DEF 1974-75 POF-old: 11 13,707 Consumption Income, questionnaire for Sample: not-national (houS-hold POF 1987/88 urban areas Frequency:infrequent budget survey) POF 1995/96 POF-old: 11 and income, consumption andexpenditure surveys, out-of-date urban areas 19,816 Expenditures Sample: nationally representative POF-new Income, Extensive Frequency: single (house-hold 2002/03 National 48,470 Consumption questionnairefor survey (with this budgetsurvey) Expenditures and income, consumption sample: though hopeto andexpenditure repeatin 1995). Recent Sources: World Bank staff analysis of survey questionnaires and existing literature; IBGE; Bianchini and Albieri (1998 and 2002). POVERTY LINES In addition to the variation in surveys and welfare measurement, there is a wide range of poverty lines used in Brazil. In fact, Brazil does not have an official poverty line. Moreover, the more frequently collected data on incomes (PNAD, PME) do not lend themselves to the construction of poverty lines (which i s usually based on consumption information and price indices). As a result, three categories of poverty lines can be found inthe literature (see Table 2): 0 "Administrative Poverty Lines." The first set is commonly used for policy (and will hence be called "Administrative Poverty Lines" in this paper) as well as by various researchers. These Administrative Poverty Lines are typically set at arbitrary low levels of income such as fractions of the minimumwage (e.g., % or VI of the minimumwage). Some remain fixed after their initial calculation as a share of the minimumwage (e.g., they were set at 95 or VI of the minimumwage in 2002, and remained fixed at those cutoffs even when the minimumwage was increased in subsequent years). These cut-offs have been widely used for determining eligibility for social programs. Infact, most social transfer programs use these cut-offs including: the Bolsa Familia Programand its predecessors(Bolsa Escola, Bolsa AlimentaGIo, Cart20 AlimentaGBo under Fome Zero, and Auxilio Gas); state and municipal safety net programs; as well as other constitutional social assistanceprograms for poor elderly and disabled. These cut-offs are also widely used in the Government's Multi-Year Plan(Plano Pluri-Anual, PPA). 3 0 "International Poverty Lines." The second set of poverty lines is generally found in the literature involving international comparisons and/or the Millennium Development Goals, (MDGs). This set is similarly "arbitrary" and involves converting the international extreme and full poverty lines of US$1 and US$2 per day into Brazilian currency with purchasing power parity (PPP) adjustments. "Consumption-Based Poverty Lines." Finally, the third set of poverty lines used in the literature has attempted to construct "meaningful" poverty lines using information on the structure and costs of consumption (food and nonfood). The 1974-75 ENDEF survey provided the first opportunity to create a consumption-based poverty line (with statistically-significant information for 23 regions in Brazil). Researchers then used price indices to update those ENDEF-based poverty lines, applying them to PNAD income data for subsequent years. The passage of time made the ENDEF consumption patterns obsolete. Consumption information for the 1986-87 and 1995-96 POF was used to generate poverty lines for those years, although such information was only available for 11 metropolitan areas. The 1996 PPV also included consumption information needed for the construction of consumption-based poverty lines. Researchers then update the poverty lines from the 1995-96 POF or 1996 PPV using price indices, and then apply them to the PNAD. Some example results are included in Table 2. On variant of the consumption-based poverty lines is the "misery line" presented by Marcel0 Neri of the Social Policy Center at FGV. The misery line i s basedon a food basket that guarantees 2,288 calories per day and the consumption patterns of the poorest 20-50% of the population. This results in a line of R$108 per person per month for the greater S ~ Paul0 area (October 2003), O which is quite close to the "Administrative PovertyLine" of R$100 per person per month. 4 Table 2 ComparisonofPovertyLinesandRatesinBrazil PovertyLine R$per month % of popbelow line Rate Year Survey "Administrative Poverty Lines" Thresholds for Transfer Programs* - ExtremePoverty(=% the 2002 Minimum 12.93%(Individuals) 2002 PNAD Wage) R$50per capita income 9.61% (Households) 2002 PNAD FullPoverty(=%the 2002 Minimum Wage) R$100per capita income 32.33% (Individuals) 2002 PNAD 25.65% (Households) 2002 PNAD International Povertv Lines US$l per day convertedPPP(indigence) R$62.58 (1999) 4.0% 1999 PNAD US$2per day convertedPPP(full poverty) R$125.16 (1999) n.a. Consumption-Based Poverty Lines ~ ____ ~ ~ ~ ~ _ _ _ _ _ _ _ ~_ ~ ~~ IPEA (poverty linesfromPPV) ' ExtremePoverty- 1999 R$76.36(Sep. 1997SBo Paulo) 14.5% (Income) 1999 PNAD FullPoverty 1999 - R$152.73 (Sep. 1997SBo Paulo) 34.1% (Income) 1999 PNAD ExtremePoverty- 1996 Same source, adjusted 15.0% (Income) 1996 PNAD FullPoverty- 1996 Same source, adjusted 33.5% (Income) 1996 PNAD FLN/ WorldBank (povertylinesfromPPV) Extreme Poverty R$65.07(1996 SHo Paulo; PPV) 22.59% (Income) 1996 PNAD Full Poverty R$131.97 (1996 Slo Paulo;PPV) 45.29% (Income) 1996 PNAD CEPAL (povertylinesfrom PPV) ExtremePoverty R$139.30 (1999) 13.8% (Income) 1996 PNAD Full Poverty n.a. n.a. 1999 PNAD These poverty lines are the existing thresholds for the Bolsa Famflia Program (launched in 2003). They are consistent with the thresholds used for the pre-reformprograms (Bolsa Escola, Bolsa Alimentaqiio, etc.) and had been establishedas equal to `/4 and !h of the minimumwage in 2002. The minimum wage has since been increased to 240 in 2003 and 260 in 2004, but the thresholds remain fixed at R$50 and R$100for the BolsaFamiliaProgram. Sources: CEPAL (2002a) * Barros et. Al. (2000) Ferreira et al. (2000) and World Bank (2001b) CEPAL (2002) WIDERANGEOFPOVERTY ANDINEQUALITY ESTIMATES PovertyRates. Given the different surveys, welfare measures, and poverty lines, it should come as little surprise that estimates of poverty in Brazil vary widely (Table 2). Estimates for extreme poverty in 1996 range from 15% (PEA) to close to 23% (FGVNorld Bank). For that same year, estimates for full poverty range from a third of the population (IPEA) to just under a half (FGVNorld Bank). Similar ranges are found for 1999, with a lower estimate ofjust 4% for extreme poverty ("$1 a day" measure) to close to three times that estimate (14.5%, IPEA). Inequality. Likewise, the estimates of inequality (Gini Coefficients) vary substantially, from 0.594 to 0.65 for 1999 (using the PNAD, Table 3). Across surveys and welfare measures, the differences in the levels of inequality are even more stark. 5 Table 3 Comparison of InequalityIndices, Various Studies Study Year Survey Sample Measure Gini Coefficient CEPAL' 1999 PNAD National Income 0.64 Barros' 1998 PNAD National Income 0.60 World Bank' 1999 PNAD National Income 0.594 Sources: CEPAL (2002a) Barros et. Al. (2000) World Bank (2003), WB 24887-BR World Bank (2001), WB ' ' 20475-BR Moreover, it may be that Brazil appears to be more unequal compared to other countries and regions because of systematic differences between income and expenditure inequality (as discussed below)." Indeed, estimates of inequality basedon (imputed) consumption are far lower than those basedon income inthe PNAD(see Table 5 below). Patterns of Poverty and Inequality over Time. As discussed above, the PNAD provides the only (almost) nationwide estimates of welfare (income) over time. The top part of Figure 1 shows extreme poverty (indigence) and poverty rates for Brazil from 1995 to 2004 using income measures from the PNAD and IPEA poverty lines. Between 1995 and 2004 extreme poverty and poverty rates in Brazil declined 2 percentage points (from 14.5 to 12.2 and from 33.8 to 31.7, respectively). Duringthe period there were three peaks in 1997, 2001 and 2003, and declines in 1998, in 2002 and particularly in 2004. The increases were mostly related to slowdowns and falls in GDP growth rates. For instance, in 2003 the economy shrank slightly by about 0.2 percent (which represented a 1.7 fall in average per capita income), and extreme poverty and poverty rates rose by around one percentage point, accounting for a 2 percent increase. The fall in poverty between 2001 and 2002 was not accompanied by appreciable growth rate, which may suggest that the decrease was influenced by redistribution policies." The considerable 8 percent decrease inpoverty rates between 2003 and 2004 can be attributed both to a GDPgrowth rate of 5 percent in 2004 inBrazilianeconomy and also to improvements inincome distribution. Income inequality, as measured by the Gini coefficient, fell considerably between 2003 and 2004 from 0.585 to 0.574. In fact, the overall reduction in poverty between 2001 and 2004 has been almost exclusively attributed to the reduction in income inequality, since per capita income fell during that period. The bottom part of Figure 1shows the persistent decline in the value of the Gini coefficient used to measure income inequality in Brazil especially between 2001 and 2004. Thus the decrease in poverty has been accompanied by the decline in inequality along the period. Although it would be tempting to attribute much of the decline in inequality to recent social policies, a more definite answer to this issue can only be provided with further research.I3 l1See Elbers et. al. (2004) andde Ferrantiet. al. (2004). l2In2000the economy grew by 4.5 percent; in 2001this fell to 1.8 percent. 13Infact, inMay 2006, IPEA, on behalfof the Braziliangovernment, decidedto sponsor a HighLevel Committee and a network of specialistsandresearchinstitutions in Brazilto study the causesof the decline ininequality. The reportof the Committee i s to bereleasedto the public inearly August 2006. 6 Figure1 The Evolutionof Poverty(Headcount)andInequality(Gini)inBrazil:1995- 2004 20.0 , I36.0 s!! 18.0 17.0 30.0 0 u) 15.0 28.0 14.0 -- 26.0259 (u w 12.0 x2 E 13.0 -- 24.0 11.0 j 22.0 1 0 . 0 ! : : : : : : 20.0 ~ ~ : , ~ ~ ~ : : : : ! 1995 1996 1997 1998 1999 2001 2002 2003 2004 I +Extreme poverty rates+Povertyrates1 L I - I V 0.570 - 0.565 Source: Barros et al(2005). Estimates basedon PNAD. Note: The PNAD survey was not collected in 2000. RegionalPatterns and Trends. The regionalprofile of poverty suggests several importantpatterns and trends. First,poverty rates are higher inruralareas than urban, and inthe North and Northeast (absolute poverty numbers, however, show higher density in urban areas). Brazil can be divided into two "super" regions: the fxst, covering the North and North East, with very high poverty rates (over 50% using income measures in the PNAD), and the rest of the country with lower poverty rates (but higher absolute poverty density). Second, these regional patterns inpoverty rates are fairly robust regardlessof welfare- measure/data source c~mbination.'~Table 4 shows that, althoughconsiderabledifferences in the level of poverty across surveys and welfare measures, the patterns of poverty by region are fairly consistent despite these measurement differences. Regardless of the welfare measure or survey, both approaches find clear evidence that poverty rates are highest in the rural and urban Northeast and lowest in the metropolitan areas of the Southeast. Third, there are considerable regional differences with respect to l4Regionalpatterns compareestimatesfrom the PNAD (income and imputedconsumption) and the PPV (consumption). Elbers (2004) et. al. 7 relative ranking of inequality across regions between all three data-source/welfare-measurecombinations (Table 5). Fourth,interms of trends over time, although overall poverty rateshavefallen inrecentyears, this reduction i s not uniform nationally (Figure 2). While rural poverty declined over the period from 1995-2004,poverty rates inmetropolitan areas rose during that period. According to Thomas (2004), the fastest proportional poverty reduction in this period took place in the South, the Center-West, and the Northeast. For example, the share of the Northeast (of total poor) reducedfrom 55% in 1998 to 52% in 2001, while the share inthe Southeast rose from 24% in 1998 to 26% in 2001. Thomas (2004) notes that developments in the Northern Region and in S b Paulo state should be of particular concern on the grounds of rising poverty inthose areas inrecentyears. Figure2 RegionalPatternsand Trendsof PovertyinBrazil:1995-2004. 60.0 7 - I 55.0 - 50.0 - 45.0 1 40.0 - 35.0 - 30.0 - 25.0 - - = 20.0 - A - - 15.0 -. ' 10.0 1995 1996 1997 1998 1999 2001 2002 2003 2004 I -+-bktropoliian -Urban -Rural I Source: Centro Politicas Sociais (CPS)-FGV usingmicro data fromPNAD. Note: The PNAD survey was not collected in 2000. 8 Table 4 Poverty Rates by Regionfor Different Data Sets and Welfare Measures PNADIncome PPV Consumption PNADImputedConsumption RuralNorthEast 71% 50% 52% UrbanNorthEast 48% 38% 39% RuralSouthEast 38% 26% 27% Metro Salvador 36% 20% 21% Metro Fortaleza 35% 19% 18% Metro Recife 34% 22% 14% Metro BeloHorizonte 15% 8% 9% UrbanSouthEast 12% 5% 5% Metro Rio de Janeiro 11% 3% 4% Metro SFioPaul0 7% 4% 3% Source: Elbers, et. al. (2004). Data from the PNAD 1996 and PPV 1996. Poverty line of R$65.07 (in 1996 Sao Paulo reais). Table 5 Inequality by Region for Different Data Sets and Welfare Measures (General Entropy Measure; 0.5) PNADIncome PPV Consumption PNADImputedConsumption RuralNorthEast 0.72 0.36 0.40 UrbanNorthEast 0.65 0.40 0.32 RuralSouth East 0.62 0.42 0.35 Metro Salvador 0.62 0.37 0.50 MetroFortaleza 0.59 0.34 0.33 Metro Recife 0.58 0.38 0.33 Metro BeloHorizonte 0.50 0.38 0.32 UrbanSouthEast 0.49 0.24 0.44 Metro Rio deJaneiro 0.48 0.28 0.30 Metro S b Paulo 0.48 0.30 0.24 Source: Elbers, et al. (2004). Data from the PNAD 1996 and 1996. Poverty line of R$65.07 (in 1996 Sa0 Paulo reais). 9 CHAPTER2: SETTINGPOVERTYLINESFORBRAZIL Regional poverty profiles attempt to describe how a measure of poverty varies across different regions of the country. Their main purpose i s to inform policy makers about the distribution of poverty across regions, so it can facilitate the formulation, design and targeting of social programs aimed at alleviating poverty in the short runandor inthe longrun. The commonly used approach i s to classify households within any given region as poor if their welfare measure i s less than or equal to a poverty line specific to each region. The typical dilemmas faced in the construction of any regional poverty profile are (i) choosing an appropriate measure of household welfare (e.g. household consumption or income), and (ii) setting appropriate poverty lines for eachregion. This report adopts consumption as the preferred measure of household welfare. Defining loosely a poor household as having a "low level of resources over its lifetime,'' there i s strong theoretical support for household consumption as the preferred measure of the long-run level of resources available to a household (Deaton and Zaidi, 2002).15 Various theoretical results suggest that consumption is less susceptible to seasonal (or inter-temporal) variation and provide a strong basis for the use of cross sectional measures of household consumption at any point in time to target program resources towards households with lower lifetime wealth. Aside from the theoretical considerations for the use of consumption as the best available indicator of household welfare, there i s a variety of practical considerations (e.g., Deaton, 1997). From the household's perspective, information about consumption may be a less sensitive topic than information about income. Having chosen a measure of welfare at the household level, one now needs to make a conversion from a household to an individual basis. This report follows the common practice of making the conversion to the individual level by dividing total expenditures by the number of people in the household (e.g. Deaton and Zaidi, 2002). Implicitly the use of consumption per capita as a measure of welfare makes the following set of assumptions: (a) everyone in the household receives an equal allocation irrespective of age or gender; (b) everyone in the household has the same needs irrespective of age or gender; and (c) the cost for two (or three or more) people living together is the same as the cost of each person living separately. Although the first assumption could be easily defended based on the constraints imposed by lack of information on consumption or income at the individual level, the other two assumptions may be questionable. It i s possible that not everyone in the household has the same needs and in particular that needs vary based on gender and age. It i s also possible that there are "economies of scale" to living together, perhaps because family members benefit from each other's consumption, or because there are public goods that can be used by all family members at no additional costs. Under both of these circumstances, starting with a one-person household, the increase in the minimum cost of living associated with an extra person in the household may not be the same for a two-person or a three-person family. These implicit assumptions separately, and in combination, have important consequenceson the poverty status of large families. For example, it is often the case that the use of a per capita measure of welfare, typically results in larger households commonly classified as poor. Whether this i s correct or not depends on whether the marginal increase in the cost of living associatedwith an extra person inthe household i s equal to or lower than the cost of living increase assumed by the per capita measure. Inthe absence of no l5 This arguably capturesonly one of a number of importantdimensionsof welfare, namely the ability of householdsto purchase goods through markets. But it is an important dimensionthat is commonly focused on in both policy analysis and the relevant literature. For a more complete welfare analysis, one may wish to supplement such information with data on access to public goods that cannot be purchasedthrough markets (especially where access is not highly correlatedwith incomeor consumption), or evenwith indices of "capability" (Sen, 1992). 10 generally accepted methods for calculating either adult equivalent scales or for accounting for economies of scale within households, per capita consumption i s used in spite of its limitations and its consequences for welfare and poverty measurement. The setting of an appropriate poverty line for each region involves a number of considerations. Differences in the cost of living, as well as differences in preferences, food tastes, and average living standards (or affluence) across regions are among the factors that need to be taken into account. The "relative" approach to measuring poverty tends to define poverty lines relative to the average standardor living of a region (or a country). The "absolute" approach to measuring poverty, i s based on the principle that there i s a socially acceptableminimumstandard or set of basic needs. Households with a standardof living below the socially acceptable minimumstandard are those considered to be poor. Absolute poverty lines are usefulfor evaluating the effect of poverty alleviating policies over time. Directly related to the concepts of absolute and relative poverty are the frequently encountered concepts of consistency and specificity (Ravallion and Bidani, 1994). Consistency, which i s analogous to the concept of absolute poverty, requires that the real poverty line, defined as the nominal poverty line after adjusting for the cost of living differences between regions, be the same for all regions. In other words, consistency requires that differences in the nominal poverty lines across regions should be entirely (or as much as possible) attributed to differences in regional cost of living.I6 Specificity, on the other hand, emphasizes that the socially accepted minimum standard or set of basic needs should be sensitive to regional differences intastes, preferencesand perceptions. Thus, specificity i s more akinto the concept of relative poverty, which emphasizes that the poverty line should be anchored in relation to the average standard of living of the region rather than a socially acceptable standard that may appear to be alien for one or moreregions of the country. Acknowledging the preceding considerations involved in the setting of poverty lines, this chapter investigates and compares the poverty lines resultingfrom usingtwo fundamentally different approaches to setting poverty lines." The first approach to setting a poverty line i s the Cost of Basic Needs (CBN) method. The second approach i s the Food Energy Intake (FEI) method. Unfortunately, in Brazil, as in many other developing countries, there i s no satisfactory spatial cost-of-living index. Given this basic constraint, each method has its relative advantages and disadvantages. The CBN method, for example, i s aimed primarily at generating absolute regional poverty lines that are consistent, at the risk of sacrificing specificity. The FEImethod, on the other hand, emphasizes specificity at the risk of yielding inconsistent poverty profiles.18 A BRIEFDESCRIPTIONOFTHE POFSURVEY The POF survey aims to measure the structure of consumption, expenditures and income of the Brazilian population. The earlier versions of the survey, the 1987/1988 and the 199Y1996 POF were conceived to review the structure of the price indexes buildby IBGE, and covered only nine metropolitan regions plus Goiiinia and Brasilia (Distrito Federal). The 2002-03 POF survey i s the f r s t consumption survey since 1975 that allows representative statistics for the urban as well as the rural areas of all the five regions of l6Kakwani (2003) also argues that a desirablepoverty line should also be "horizontally equitable", meaning that poverty lines should differ not only acrossregions but also differ dependingon individual circumstances, such as age and gender. l7See appendix 5 for a summary discussion of the effort to derive subjective poverty lines basedon the POFdata. Unfortunately, the estimates obtained for the subjective poverty lines appear to be problematic and thus not very useful for investigating the complementaritiesbetweensubjective measuresof poverty andthe objectivepoverty estimates discussedinthis chapter. The regional distribution of poverty using the relative approachto measuring poverty i s explored in Appendix 7. It is shown that the relative approach to poverty measurement yields regional poverty line estimates and poverty rankings that are very similar to those obtained with the FEImethod. 11 Brazil. The survey is also representative at the state level, as well as for urban areas (though not rural areas) within each state. For the construction of the regionalpoverty profile of Brazil the five geographical regions were divided in twenty one different regions, eleven metropolitan and ten urbanand rural regions, as follows: Table 6 The regionaldistributionof the totalpopulationandits share inthe 2002-03POF Regions Total population Population share 1 MetroBelem 1,845,708.10 1.05% 2 North Urban 8,229,439.10 4.69% 3 Rural 3,533,712.70 2.02% 4 Metro Fortaleza 2,985,822.90 1.70% 5 Metro Recife 3,331,278.30 1.90% 6 Northeast Metro Salvador 3,088,893.00 1.76% 7 Urban 25,579,176.00 14.59% 8 RWal 13,940,461 .OO 7.95% 9 Metro Rio DeJaneiro 11,052,249.00 6.30% 10 MetroSa0 Paul0 17,696,179.00 10.09% 11 Southeast MetroBelo Horizonte 4,437,345.50 2.53% 12 Urban 35,016,773.00 19.97% 13 Rural 6,586,851.30 3.76% 14 Metro Curitiba 2,641,166.40 1.51% Metro PortoAlegre I 16,722,914.00 9.54% 16 l5 south Urban 5,937,284.00 3.39% 17 Rural 2,430,221.80 1.39% 18 Brasilia 1,333,65 1.00 0.76% Goianiamunicipality 6,392,352.60 3.65% 20 l9 CenterWest Urban 2,194,866.30 1.25% 21 Rural 355,452.52 0.20% Total 175,33 1,797 100% Source: WorldBankestimates usingthe 2002-03 POF. The POF sample i s defined to capture expenses from each family (consumption unit) living in the same household. For comparison, the PNAD 2002 represented 51,560,959 families while the 2002-03 POF represented48,534,638 families (based on the sample of 48,568 householdsinthe survey). The interviews took place from July 2002 to June 2003 covering the full diversity of items of the country (e.g. there 5,442 different codes for food products purchased). The consumption expenditure module collects information on household and individual food and nonfood expenditures using four recall or reference periods: last 7 days (for food expenditures), last 30 days, last 90 days and last 12 months. The reference period for all earnings and other income received i s the last 12 months. In the version of the POF data sets releasedto the public all nominal values are expressedinJanuary 15th2003 prices. The POF survey also collects information of both monetary andnon-monetary expenses as well monetary income and imputed household rent for owners. 19. The survey classifies as monetary expenditures the expenditures made in cash, credit card or check. Non-monetary expenditures correspond to all types of 19 Definedaccordingthe "Informe de la DBcimoSBptimaConferknciaInternationalde Estadisticosdel Trabajo (2003)" 12 auto-consumption or trade with no money involved. Individual respondents were asked to use current marketprices to value allnon-monetarytransactions. Table 7 provides the regional household consumption expenditure per capita (PCE) and the mean household income per capita (PCINC) obtained from the POF survey. The per capita income variable used here from the POF is the household income total made publicly available by IBGE excluding net withdrawals from savings. It i s important to note the sizeable difference between the PCE and PCINC variables in the POF. The difference between PCE and PCINC imply a saving rate of 33% which is considerably higher than the saving rate of 20% estimated from National Accounts. One possible explanation for this difference is the fact that many publicly provided goods are excluded from the consumption aggregate!' This implies that the difference between PCE and PCINC includes not only savings but savings plus components of public spending such as spending on public schools free of charge etc. Table 7 also reports the mean PCINC obtained from the 2004 PNAD survey (with nominal household income deflated to January 2003).It i s critical to keep in mind, however, that the household income variables in the POF and the PNAD surveys are not really comparablebecause they are collected based on very different reference periods (the PNADfor the last monthandthe POFfor the last year). Table 7 MeanExpenditurespercapita (PCE)andIncomeper capita (PCINC) by region(inR$ perpersonper month). Regions POF POF PNAD PCE PCINC PCINC 1 Metro Belem 299.0 387.3 322.4 2 North Urban 238.2 315.1 287.7 3 Rural 135.0 193.5 185.3 4 Metro Fortaleza 309.4 443.9 316.9 5 Metro Recife 331.3 434.3 348.1 6 Northeast Metro Salvador 386.8 555.5 350.0 7 Urban 207.6 281.9 239.6 8 Rural 111.9 125.3 111.3 9 Metro Rio De Janeiro 547.7 794.6 577.7 10 Metro Sa0Paul0 525.3 819.4 545.9 11 Southeast Metro Belo Horizonte 429.1 728.3 470.0 12 Urban 381.3 584.0 452.3 13 Rural 207.0 314.5 232.1 14 Metro Curitiba 522.8 802.3 609.6 Metro Port0Alegre 485.0 807.6 587.3 16 l5south Urban 368.3 574.7 487.1 17 Rural 236.9 346.6 300.2 18 Brasilia 596.2 940.5 733.6 GoianiaMunicipality 425.9 654.8 n.a*. 20 l9CenterWest Urban 268.5 411.7 413.2 21 Rural 217.7 303.6 242.2 National 335.9 502.1 389.1 Metropolitan 457.7 698.5 514.2 UrbanexcludingMetropolitan 310.7 461.8 381.4 All UrbanincludingMetropolitan 372.2 560.8 431.7 Rural 159.7 217.3 182.6 Source: World Bankestimates usingthe 2002-03 POF. andthe 2004 PNAD *Inthe PNAD survey it was notpossibleto identify the Goianiamunicipalityseparately so itis classified with urban areas. 2o The general guidelines for deriving a household consumption aggregate to measure household welfare based on Deaton and Zaidi (2002) are discussed in detail in Appendix 1 of this report. Appendix 1 also presents the kernel estimates of the density function of the POF income and expenditures (inper capita terms) as well as the density function of the per capita income from the 2004PNAD. 13 THECOST OFBASICNEEDS (CBN)METHODSETTINGPOVERTYLINES OF The CBN method in essence determines the consumption bundle considered adequate for basic consumption needs and then estimates the cost of this basic needs bundle in each of the regions of the country. A household is then considered poor if its consumption expenditures are less than or equal to the cost of this basic needs bundle. The CBN method can be best described as a two-step method, whereby in the first step, the food poverty line for each region i s determined, and in the second step, the food poverty line is adjusted upward by an allowance for basic nonfood needs. The Food Poverty Line The food poverty line is derived as follows: Firsta reference population group is chosen to determine the composition of the basic needs food basket. Second, the basic needs food basket i s constructed with three properties in mind: (i) the composition of the basket reflects the variety of food items consumed by a reference population close to the expected poverty threshold; (ii) it provides the recommended food energy requirement of 2300 kcal per capita per day; and (iii) the recommended caloric requirements are derived from a sufficiently diverse variety of foods (e.g. some meat and fruits and vegetables and notjust rice and other cereals). The reference population chosen to determine the basic needs food basket is the set of households in the 20 to 40 percentiles of the distribution of the total per capita expenditure (PCE). The levels of per capita consumption for households in the 20% to 40% are between R$94 and R$165?2 Table 8 below summarizes the regional distribution of the total population in Brazil, the regional distribution of the reference group of 20-40 percentiles of the distribution of PCE and the regional distribution of the bottom 20 % of the of the distribution of PCE. As it can bee seen, the distribution of the population among Brazilian regions in the 20 to 40 percentiles is similar to the pattern of the distribution of total population in thoseregions. Incontrast, the bottom20 percentiles were overrepresentedby the Northeast. According to the Food and Agriculture Organization (FAO) the average daily caloric requirement for Brazil is estimated at 2,300 kcal per capita per day. It i s important to keep in mind that setting the food energy requirement at 2,300 kcal per capita per day is rather arbitrary, since food energy requirements vary by age, gender, and level of physical a~tivity.'~Moreover, as it i s clearly noted by FAO, there i s no implication that exactly 2,300 kcalmust be consumedby every person duringeachand every day. ''**Thissectiondraws heavily from Bidani andRavallion(1993),RavallionandBidani (1994),andRavallion(1998). The sensitivityof the food poverty line by the CBN methodto the choice of the referencepopulationis examined in greater detail inAppendix 2. The foodpovertyline was not sensitiveat all to the choiceof the referencepopulation. 23 Infact, as Rocha(1997)notesinher study onpovertylines for Brazil, the FA0 recommendedcaloric requirementshavebeen decliningover time (seeannex 1of her paper). 14 Table 8 Regionalcompositionof the 20 to 40 percentiles relativeto the regional composition of total Brazilianpopulation Reference Regions population Bottom20 Dif with Population (20.40%of Dif Total Percentiles ~ o t a l (in %) PCE~ %\ (in 1 Metro Belem 1.05 1.21 0.16 0.74 -0.31 2 North Urban 4.69 5.98 1.29 5.83 1.14 3 RWal 2.02 2.94 0.92 4.8 2.78 4 Metro Fortaleza 1.70 2.12 0.42 1.71 0.01 5 Metro Recife 1.90 2.04 0.14 1.48 -0.42 6 Northeast Metro Salvador 1.76 1.58 -0.18 1.06 -0.7 7 Urban 14.56 18.52 3.93 25.84 11.25 8 RWal 7.95 9.97 2.02 22.81 14.86 9 Metro Rio De Janeiro 6.30 3.65 -2.65 2.21 -4.09 10 Metro Sa0 Paul0 10.09 6.6 -3.49 2.12 -7.97 11 Southeast Metro BeloHorizonte 2.53 2.06 -0.47 0.67 -1.86 12 Urban 19.97 17.1 -2.87 11.54 -8.43 13 RWal 3.76 5.51 1.75 5.27 1.51 14 Metro Curitiba 1.51 1 -0.51 0.31 -1.2 l5 south Metro Port0Alegre 9.54 1.39 -0.7 0.57 -1.52 16 Urban 3.38 8.06 -0.54 5 -3.6 17 RWal 1.39 2.88 0.35 2.31 -0.22 18 Brasilia 0.76 0.75 -0.48 0.34 -0.89 '9 Center West GoianiaMunicipality 3.65 0.49 -0.15 0.12 -0.52 - 20 Urban 1.25 5.19 0.9 4.14 -0.15 21 RWal 0.20 0.96 0.17 1.14 0.35 Source: World Bankestimates usingthe 2002-03POF. Given the large number of food items (5,442) in the POF survey, the selection of the specific food items composing the basic needs food basket was based on the following steps. First, the 5,442 different food items were f r s t grouped in41 food groups (i.e., cereals, beans, vegetables, etc.). Second, the specific food items was performedby selecting food items with the highest frequency (most frequently purchased) from food groups with an average weighted share of greater than 1 percent. Among the 41 different food groups only 24 had a frequency greater than one percent. The resulting basket of basic food consumption contained 26 specific food items: 1 from each of the 24 different groups with two extra items in the vegetables food group.24 Next, the quantities were expressed in per capita and per day terms by dividing by the number of household members residing in the household (see column b). The average quantity of each of the 26 items in food basket was rescaled to ensure that the food basket yields 2,300 kcal per capita per day. This was done by multiplying the average quantity of each of food item with a conversion factor of 0.59 obtained from the ratio of the recommended daily caloric requirement per capita and the total calories yielded by the average quantities per capita per day inthe poverty basket (see columns d & e). Table 9 below presents the composition and adjusted quantity of each food item in the basic needs food basket. It is importantto notethat we havealsotestedthe sensitivity of the food poverty line by increasing the number of food items in the basket and/or by allowing the composition of the poverty basket to be sensitive to regional differences in food tastes and preferences. For details see appendix 2. 15 tfl Product Next, the food poverty fine (or ~ l calf e ~ ~ poverty or ~i ~~ ~e ~ tine)eforn ~eache region~is ~~ ~a~ ~ ~ ~ ~ ~ ~ by valuingthe basic needs fmd ~ ~ ~ ~ baskets s~ ~ ~ ea ~eachd~region. Specifi~~lly~food poverty it1 ~ e ~ y the h e ineach region is e s ~ i ~ usingethe e x p ~ e ~ s ~ ~ ~ ~ ~ : a ~ ~ 26 FPLR=cp,!(q: " k ), i=l where the superscript R denotes the twenty one different regions, p,! i s the average unit value of food item i in region R,25 4: i s the average quantity of food item i in the basket, and k i s the conversion factor?6 The unit values p,! are defined as the total expenditure reported for a specific itemdivided by the total quantitiespurchased of each item. Ideally, it would be better to use the actual market prices for eachfood itemin eachregion. Table 10 The FoodPoverty Line (inR$per person per month) Regions FoodPoverty Line 1 Metro Belem 63 2 North Urban 60 3 Rural 59 4 Metro Fortaleza 59 5 Metro Recife 62 6 Northeast Metro Salvador 63 7 Urban 60 9 Metro Rio DeJaneiro 62 10 Metro SaoPaul0 65 11 Southeast Metro Belo Horizonte 59 12 Urban 64 13 RWal 58 14 Metro Curitiba 60 l5south Metro Port0Alegre 64 16 Urban 57 17 Rural 55 18 Brasilia 62 l9CenterWest GoianiaMunicipality 59 20 Urban 61 National 61 Metropolitan 62 UrbanexcludingMetropolitan 61 Rural 58 Source: WorldBank estimatesusingthe 2002-03POF. Povertylines are expressedinR$, January 2003. The numbers in the last four rows are simple (unweighted) averages of the region-specific poverty lines 25 In Appendix 2, we tested the sensitivity of the estimated food poverty lines by using the median unit values and median quantitiesineachregion; and found no significant changesinthe estimates for the food poverty line. 26 Infact k= (2,300/3,865)= 0.595084. 17 Since the composition of the food basket i s held fixed andthe quantity of eachfood item is not allowed to differ from region to region, it is implicitly assumed that households do not respond to differences in relativeprices.27 Table 10presents the regional food poverty lines obtained usingthe preceding procedures. Ingeneral, the food poverty lines are very similar across the twenty one regions, suggesting very small differences inthe cost of living across regions.28Nevertheless, the food poverty lines are generally lower in rural areas in comparison to metropolitan and urban areas within each region. Also, in metropolitan areas prices are even higher than in urban non metropolitan areas. The highest food poverty line appears in the metropolitanregionof Sao Paulo where the value of the basket reachesR$65 per capita per month. Adjustingthe FoodPovertyLinefor the Costof NonFoodBasicNeeds Inprinciple, one could apply the same general approach inconstructing an index for nonfooditems. One could determine a bundle of essential nonfood items that enter the basket, and then cost that bundle separately in each region. Unfortunately, there are a number of factors preventing the application of this approach. For example, to determine the composition of the food basket for the poverty line, one can use the recommendedfood energy requirement as an anchor for food consumption. It is practically impossible to devise an analogous method for determining the specific requirements of each nonfood item (such as housing, transportation needs, utilities and clothing). Moreover, even in the event that one manages to determine the specific requirements of nonfood items, it i s difficult to monitor the prices of nonfood items, since the prices of the most nonfooditems arerarely available. Ravallion (1998) proposes two ways of estimating the upward adjustment (allowance) to the food poverty line to account for basic nonfood needs. Each method i s based on an intuitive criterion for defining the basic non food consumption and separately applied they yield a lower and an upper estimate of the (total) poverty line. The first method (method A) i s based on households whose total per capita expenditures are equal to the food poverty line. Provided that households with this level of total expenditures spend something on nonfoods, it follows that they are willing to forego some food spending to satisfy some of their basic non food needs. Graphically, line segment A infigure 3, representsthe allowance for basic nonfoodneeds that should be added to the food poverty line. The line segment A represents the amount of expenditures that households with expenditures equal to the food poverty line (Zf) forego in order to purchase basic nonfood items. The second method (method B) is based on households whose food expenditures per capita are equal to the food poverty line. As figure 3 displays, these households end up spending an additional amount representedby the line segment B for nonfood items. Clearly, since the adjustment for basic nonfood expenditures with method A is lower than the adjustment with method B, the former adjustment yields a lower poverty line, while the later yields an upper poverty line. 27Thus, the food povertyline is analogousto a Laspeyrespriceindex. Note, that the food poverty line can also be interpretedas the cost of achievinga minimumlevel of utility, if one were willing to assume that utility compensatedsubstitutioneffects are zero (Ravallion, 1998). 28The spatialvariability ofthe food andthe overallpoverty line are examinedinmore detailinthe next chapter. 18 Figure3 Estimatingthe Cost of NonfoodBasicNeed FoodPoverty Line, zf Zf Z Following Ravallion (1998), the ad'ustment to the food poverty line for basic nonfoods was performedby estimating an Engel curve such as2 d w; =ao+ x20a j R j +,Oh( PCE zpL)+flh+ j=1 where w; i s the share of food expenditures of household h, a o i s a constant term, R is set of binary dummies for 20 of the 21 regions of Brazil (the Sao Paul0 metropolitan area was included in the constant term), PCE i s per capita expenditure, FPL i s the Food Poverty Line, and X summarizes a set of demographic characteristics. Specifically, X included the number of males and females of different age groups, and the gender, years of education, and employment status of the household head. The lower adjustment to the food poverty line for basic nonfoods (line segment A in figure 3) for each region Rj i s obtained by first estimating the food share of households with per capita expenditures equal to the food poverty line (ie., PCE=FPL) basedon the expression: where' X h,denotes the average household characteristics of households in the reference population (the 20 to 40 percentile of the PCE distribution). Armed with an estimate of the food share of households the lower poverty line may be estimated as: PLR(lower)= FPLR (1-w; )FPLR=2FPLR-W; FPLR= FPLR(2-w; . + ) (2.4) 29We have also estimated the lower and upper estimates of the food shares using a version of the nonparametric approach suggested by Ravallion (1998). Estimating the food shares nonparmetrically did not introduce any major changes in the level and variability of the regionalpoverty lines. Estimatesof the regional poverty lines using anonparametricapproach are presented inAppendix 3. 19 The alternative approach for estimating the adjustment to the food poverty line for basic nonfoods estimates is also based on the estimates of equation (2). The upper poverty line i s based on the estimation of the food share at which the food per capita expenditure equals the food poverty line. Let w: denote food share in region R at which per capita expenditure equals the food poverty line. Then w i may be identified with the value of w: that satisfies the equation3': (2.5) where bR=bo +bjRj +#fh.Then the upper estimate of the poverty line may be obtained as: P L ~(upper)= FPL~ w i (upper) Table 11 below presents the estimated food shares using methods A (lower) and B (upper). Consistent with Engel's Law, that predicts that food shares decline with total spending, the food share associated with method B i s lower than the food share obtained with methodA. Table 12presents the corresponding estimates of the region-specific lower and upper poverty lines for Brazil. Table 11 EstimatedFoodshares basedon CBNmethodA (Lower) andB (Upper). Regions Predictedfood sharesfor the: Lower Upper 1 Metro Belem 0.343 0.326 2 North Urban 0.306 0.288 3 Rural 0.414 0.401 4 Metro Fortaleza 0.316 0.299 5 Metro Recife 0.314 0.296 6 Northeast Metro Salvador 0.301 0.283 7 Urban 0.352 0.336 8 Rural 0.433 0.420 9 Metro Rio De Janeiro 0.266 0.247 10 Metro Sa0 Paul0 0.239 0.219 11 Southeast Metro Belo Horizonte 0.266 0.247 12 Urban 0.285 0.266 13 Rural 0.333 0.316 14 Metro Curitiba 0.252 0.232 l5South Metro PortoAlegre 0.274 0.255 16 Urban 0.278 0.260 17 Rural 0.372 0.357 18 Brasilia 0.229 0.208 Goianiamunicipality 0.265 0.245 - l9CenterWest 20 Urban 0.279 0.261 21 Rural 0.333 0.316 Source: World Bank estimates using the 2002-03 POF. ~~ 30Equation(2.5) is solvedby numericallyby an iterativemethod.For furtherdetails see Ravallion(1998). 20 Table 12 Regional Poverty Lines for Brazil based on the CBNmethod (in R$per person per month) 2 North Urban 102 211 3 Rural 93 147 4 Metro Fortaleza 99 199 5 Metro Recife 104 210 6 Northeast Metro Salvador 108 227 7 Urban 100 181 8 RWal 92 140 9 Metro Rio De Janeiro 107 253 10 Metro Sa0Paul0 115 304 11 Southeast Metro BeloHorizonte 103 244 12 Urban 109 242 13 Rural 97 185 14 Metro Curitiba 105 263 l5south Metro Port0Alegre 111 256 16 Urban 99 224 17 Rural 90 156 18 Center West Brasilia 109 303 19 Goiania municipality 103 246 20 Urban 105 238 21 RWal 100 191 National 103 220 Metropolitan 106 246 Urban excludingMetropolitan 103 219 Rural 94 164 Source:WorldBankestimatesusingthe 2002-03POF.. Povertylines are expressedin R$,January 2003. The numbersin the last four rows are simple (unweighted)averages of the region-specificpoverty lines One rather surprising finding i s the fact that the lower estimate of the poverty line for Brazil i s practically identical to the "Administrative Poverty Line" of R$100 per capita per month used widely for determining eligibility for social programs, and discussed in more detail in the previous chapter of this report. Table 12 also reveals that there i s a considerable gap between the lower and the upper poverty lines obtained with the CBN method. The CBN-Upper poverty lines are on average two times as large as the lower poverty lines, and more than three times as large as the food poverty lines (reported in table 10). That i s especially the case in metropolitan areas, where the predicted share of food expenditure i s lower than innonmetropolitanareas.31 31 Given the considerable gap betweenboth lower andupper bounds, andthe two different poverty figures that these two extreme lines would entail, an alternative practice found inthe literature is to average the lower and the upper estimates of the poverty lines. This would give an intermediate point of estimate for the non food allowance betweenthe two extremes commented inthe graph. 21 THEFOOD ENERGY INTAKE (mI)METHODSETTINGPOVERTYLINES3' OF As mentioned in the introductory part of this chapter, another desired feature of a poverty line is "specificity". Specificity i s the term used by Ravallion and Bidani (1994) to summarize the extent to which a poverty line i s able to reflect the local customs, food tastes and preferences and perceptions of what constitutes poverty ineachregion of the country. Inthis section we describe the FoodEnergy Intake (FEI) method as an alternative approach to setting poverty lines. The emphasis of the FEIapproach is on specificity rather than consistency. It is argued that the particular features of the method that allow poverty lines to reflect the region-specific food tastes and preferences also tend to result in poverty lines that are inconsistent across regions. Thus, there is a natural conflict between deriving poverty lines that satisfy the properties of consistency and specificity simultaneously. The remainder of this section describes the FEImethod and the main advantagesand disadvantagesassociatedwith it. Incontrast to the CBNapproach, which determines a consumptionbundle that is considered adequatefor basic consumption needs and then estimates the cost of this basic needs bundle in each region, the FEI method consists of identifying the total consumption expenditure or income at which a person's typical daily food energy intake is just sufficient to meet a predetermined food energy req~irement.~~Figure 4 below displays graphically how the poverty line can be determined using the FEI method. The upward sloping line in this figure depicts the line summarizing the relationship between total daily caloric availability per capita and total consumption expenditure. Given the recommended energy requirement of 2,300 kcal per day per capita, the "calorie-expenditure'' line may then be used to find the total consumption expenditure that corresponds to it, in this case point Z inthe horizontal axis of figure 4. The advantagesof the FEImethod inderiving a poverty line are numerous. First, it is simple and it can be applied easily within each of the twenty one regions of Brazil to derive a region-specific total poverty line. Second, the method does not require an adjustment for the consumption of nonfood, since it yields automatically the level of total (food and nonfood) expenditures that are associated with attainment of the 2300 daily caloric requirement per capita. Third, the method can be easily adjusted to derive either a region-specific food poverty line or determine the composition of the region-specific food poverty basket. For example, a food poverty line may be obtained by identifying the level of food expenditures of households with 2,300 kcal per capita, available per day and total consumption expenditures equal to Z (or plus or minus a small amount around Z). Along similar lines, the composition of the food poverty basket could be determined by including the food items purchased most frequently by the same group of households. Thus, even though the FEI method per se does not necessarily require price data, it can also be used to come up with a list of items of a region-specific poverty basket whose prices could be measuredover time to update the poverty line at regular time intervals. 32Strictly speaking, in the case of Brazil, the correct description for the method discussed in this section is Food Energy Availability (or FEA) rather than FEI. This is because the POF survey collects information on household expenditures for consumption (Le. availability of calories) ratherthan actualcaloric intakeor actual consumption of food. 33The methodhas beenappliedin numerouscountries such as Indonesia, (Ravallion and Bidani, 1994), India and Pakistan(see Kakwani2003, for asummary). 22 Figure 4 The Food Energy Intake (FEI)Methodfor Setting a Poverty Line DailyFood-Energy Intake(or availability) of calories per capita 2300 z! Total ConsumptionExpenditure However, the simplicity of the FEImethod comes at a cost. The primary disadvantage associatedwith the FEImethod is that it yields poverty lines that are not consistent. The lack of consitency of the poverty lines arises from the fact that differences in the poverty lines between two regions are attributable to differences in the cost of living between these two regions as well as other factors. These other factors include differences in tastes, levels of activity, relative prices, publicly provided goods and other determinants of affluence. Figure 5 Setting a Poverty Line with the FEImethodinUrbanand RuralAreas Daily Food-EnergyIntake(or availability) of calories per capita I 2300 Y I I ! I I I I I I ZR ZU TotalHouseholdExpenditure To get a better sense of the shortcomings associated with setting poverty lines with the FEI method consider figure 5. This figure illustrates the FEI method for two regions: a rural and an urban region. 23 According to the FEI method the poverty line for rural areas ZRi s lower than the the poverty line for urban areas denoted here by ZU.To the extent that other factors, in addition to differences in the cost of living between urban and rural areas, affect the difference between the two poverty lines, then these two poverty lines are not likely to be consistent. The two poverty lines derived by the FEImethod infigure 5 depend critically on the positions of the two calorie-expenditure lines for urban and rural areas and the relative positions of these calorie-expenditure lines, depend on a variety of other factors besides differences incost of living. For example, the activity level of jobs in urban and rural areas may differ. Activities in the typical urban job may require fewer calories than activities in the typical rural job (e.g. agricultural labor). Thus, regional differences in activity levels may result in caloric intake or availability being lower for urban households at any given level of real expenditure. This, in turn may affect the relative position of the calorie-expenditure line and ultimately the difference between the rural and urban povery lines Z, and ZU. It is also likely that households in urban areas may have more expensive food tastes. Thus even in the unlikely case that the cost of food i s the same between between rural and urban areas, households in urban areas may prefer to consume food items of higher quality and thus of higher price (e.g. buy organically grown fruits and vegetables, instead of vegetables from the regular grocery store). As a consequence, urban households may spend more per calorie consumed, or equivalently, the caloric availability obtained for any given level of real expenditure i s likely to be lower. Another factor affecting the relative location of the calorie-expenditure lines for urban and rural areas i s differences inthe relative prices of food and nonfood. As Ravallion andBidanistate: "To the extent that prices differ between urban and rural areas (say, becauseof transport costs for food produced in rural areas), different nominal poverty lines should be used. However, relatice prices can also differ, and (in general) this will alter demand behavior at given real expenditure levels (nominal expenditures deflated by a suitable cost-of-living index)." For example, the prices of of some nonfood goods tend to be lower in relation to foods in urban areas than in rural areas, and retail outlets for nonfood goods also tend to be more accessible (so the full cost, including time, i s even lower) in urban areas. This may mean that the demand for food and (hence) food energy intake will be lower in urban areas than in rural areas at any given real expenditure level.. .." The preceding arguments imply that the difference between the two poverty lines ZRand Zu derived by the FEImethod is likely to embody more differences than just differences in the cost-of-living between urban andrual areas. As a consequence, these households whose total consumption expenditures equal to these nominal poverty lines may not have exactly the same standard of living (or welfare) as i s required by a poverty profile that is consistent. As a means of investigating the issues for the case of Brazil, figure 6 presents non-parametric estimatesof the calorie-expenditure curve for urban and rural areas in each of the five main regions of Brazil: North, Northeast, Southeast, South, and Center-We~t.~~The vertical axis ineach of these graphs is the logarithm of daily caloric availability per capita (InPCK), estimated from the POF survey, while the horizontal axis i s the logarithm of per capita expenditures (1nPCE). The horizontal line in each of these figures depicts the recommended daily caloric requirement of 2,300 kcal per capita. Preliminary estimates of the poverty 34 Thus, the metropolitanareas ineach of these regions are combined with the other urban areas, in order to economize on the number of graphs. For more technical details on how to derive the non-parametric graphs presented hereinthe reader i s referred to Subramanian andDeaton(1996) andDeaton (1997). 24 lines for the urban and rural areas in each of the five regions can be obtained by tracing down a vertical line from the point of intersection of the recommended daily caloric requirement with the corresponding calorie-expenditure curve. Since the calorie-expenditure curves for urban areas are further to the right of the calorie-expenditure curve for rural areas, it follows that the poverty lines for urban areas will be higher than the poverty lines for rural areas. The consistency of these poverty lines relates to the question of whether the gap between urban and rural poverty lines of the FEImethod i s capturing cost of living differences as well as other additional confoundingfactors. Figure 7 presents the corresponding non-parametric estimates of the calorie price-expenditure curve for urbanandruralareas of the same regions. Inthese graphs the vertical axis is the logarithm of the cost per calorie (1nPKAL) while the horizontal axis is the same as before (i.e. lnPCE). The calorie price i s derived by dividing total food expenditures for consumption at home by the caloric content of the food items purchased. The positive slope of these curves implies that increases in household living standards (measured by total per capita consumption expenditure) are associated with a higher price per calorie. Thus, to the extent that the price of calories captures quality, in both urban and rural areas, households with a higher standard of living seem to prefer food of higher quality.In addition, the higher position of the calorie price-expenditure curve for urban areas relative to that for rural areas suggests that urban households pay a higher price per calorie for any given level of real expenditure. Thus, figure 7 provides some strong indications that the poverty lines obtained with the FEI method in figure 6 are likely to embody differences inpreferences (or tastes) between urban and rural areas, in addition to cost-of-living differences. Figure 6 Non-parametric estimatesof the calorie-expenditure curve inurbanand rural areas of the five regionsof Brazil - URMN g). 7 8 7 4 7 2 8 8 8 8 g 8 4 58 8 2 8 2 5 8 5 8 3 5 4 5 5InFCE 5 6 5 7 5 8 5 3 51 45 REGIOCF~ REGION3E 5l5" 8 5 7 5 8 5 REGION1 -RUPAL - URMN -PUPAL - E URWN REGIOCY~ REGIONInPCE 5 Note: Region l=North; Region2= Northeast; Region 3=Southeast; Region4= South; Region5=Center-West 25 Figure 7 Non-parametric estimates of the price of calorie-expenditure curve inurbanand rural - areasof the five regionsof Brazil - RUWL URMN - RUPAL - URMN -RURAL -URBAN -RURAL -URPAN - - RUPAL URBAN i -.2 .2I -.3 ..4 -.3 ..4 3:5 4 4 k b a 4'5 k 5:5 B 5:s 7:s B 8% REGIONl 4E 5:5" 6 8:5t 1:5 8:5 E 3:5 REGIOJYt Note: Region l=North; Region 2= Northeast; Region 3=Southeast; Region4= South; Region5=Center-West COMPARINGTHE CBNAND FEIPOVERTY LINESFORBRAZIL The first step towards applying the FEImethod on the POFdata set i s to calculate caloric availability for each food item purchased by each household in the POF survey. To do this, we used quantity-to-calorie conversion factors made available by IBGE.The total calories purchased for consumption at home at the household level were then estimated by aggregating across all the food items purchased.The total caloric availability per capita (PCK) was estimated by adding to the calories purchasedfor consumption at home, an estimate of the household calories consumed outside the home. The survey collects information only on expenditures for food consumed outside the home (such as restaurant meals, street vendors etc) but not quantities nor prices. Under these circumstances, the estimate of the calories consumed outside the home was derived by dividing the expenditures of food consumed outside the home by 1.5 times the calorie price paidfor items purchased for consumption at home.35Thus, it i s assumedthat the cost of calories for food consumed outside the household is 1.5 times higher, in accordance with internationalevidence (e.g. Subramanianand Deaton, 1996). Second, for eachof the twenty one regions of Brazil we estimated the followingregression model: In(PCE)=a +p*PCK + E , (2.7) where PCE denotes total (food + nonfood) expenditures per capita, and PCK denotes daily caloric availability per capita. The poverty line in each region was estimated using the region-specific estimates of the parameters a, and p, and the recommended daily caloric availability per capita Le., POVLINE=exp( & +B*2300). Thus, POVLINE represents the per capita expenditure in any given region at which the per capita energy requirement of 2,300 kcalis met.36 35 The calorie price is derivedby dividing total food expenditures for consumption at home by the caloric content of the food items purchased. 36 An analogous approach was used for the estimation of the food poverty line using the FEI method. Equation (1) was re- estimated by replacing the dependent variable PCE by per capita food expenditures (PCFE). Note that, in principle, it is also possibleto apply the FEImethodusingdifferent minimum daily caloric requirementsfor different regions (e.g., urbanvs. rural). Appendix4 providesamoredetailedinvestigationofthe sensitivityof the povertylinesderivedby the FEImethod. 26 Table 13 CBNvs. FEIPoverty Lines Regions F E Z * CBNUpper CBNLower 1 Metro Belem 140 195 105 2 North Urban 126 211 102 3 RUral 75 147 93 4 MetroFortaleza 141 199 99 5 Metro Recife 154 210 104 6 Northeast Metro Salvador 187 227 108 7 Urban 99 181 100 8 RUrd 60 140 92 9 Metro Rio De Janeiro 255 253 107 10 Metro Sa0Paul0 285 304 115 11 Southeast Metro Belo Horizonte 235 244 103 12 Urban 213 242 109 13 RUral 132 185 97 14 Metro Curitiba 275 263 105 l5south Metro Port0Alegre 287 256 111 16 Urban 200 224 99 17 RUrd 134 156 90 18 Brasilia 288 303 109 GoianiaMunicipality 266 246 103 20 l9Center-West Urban 155 238 105 21 RUrd 120 191 100 National 170 220 103 Metropolitan 229 246 106 Urban excludingMetropolitan 159 219 103 Rural 104 164 94 Source: World Bank estimates using the 2002-03 POF. Poverty lines are expressed in R$, January 2003. The numbers inthe last four rows are simple (un-weighted) averagesof the region-specific poverty lines The regionalpoverty lines obtained from the FEImethodare shown inTable 13. To facilitate comparison, the lower and upper poverty lines obtained by the CBN method are presented next to the FEIpoverty lines. Figure6 presents the poverty lines of table 13 ingraphical form. The graph makes it more apparent that the upper CBNpoverty line is quite similar inboth level and variability to the FEIpoverty line. Table 13 and figure 8 suggest that the adjustment of the food poverty line for the cost of basic nonfood needs i s very sensitive to the methodused. 27 Figure 8 The CBNandFEIRegionalPoverty lines 300 275 .--8c250 225 200 175 $ 150 125 0 100 75 I 50 Y ! ! 25 Northeast Sodhast 1 South 1 Center-West I +FEI ~ +CBNUppe! ~ -CBN-Lowe! Source: WorldBank estimates usingthe 2002-03 POF. A close examinationof table 13 andthe corresponding figure 8 leads to the following observations: On average, the FEIpoverty lines suggest that the difference between the cost of living between metropolitan and rural areas i s much higher than the cost of living differences obtained with the CBN-Upper and especially the CBN-Lower poverty line. Expressed as a fraction of the value of the poverty line in rural areas, the CBN-Lower poverty lines suggest that on average the cost of living in metropolitan areas i s 12.8% higher than in rural areas, whereas the CBN-Upper poverty line yields a cost of living in metropolitan areas that i s 50% higher. Incontrast, the FEIpoverty lines suggest that the cost of living i s 120% higher. This finding i s consistent with the argument inthe earlier part of this chapter that the FEImethodis likely to yield poverty lines that embody differences in preferences (or tastes) between urban and rural areas, in addition to cost-of-living differences. Ineachof the fiveregions of Brazil, andinallsubregionsof the Northandthe Northeast, the FEI poverty line in rural areas is lower than the CBN-Upper poverty line. Thus the FEIapproach is likely to yield poverty rates that are "too low" in rural areas and "too high" for urban areas relative to the poverty rates obtained with the CBN-Upper. In fact, in the rural North and Northeast regions, the FEI method i s likely to yield the lowest poverty rates, since it results in poverty lines that are even lower thanthe poverty linesobtained with the CBN methods. There is a large difference in the level of the CBN-Lower and the CBN-Upper poverty lines. The explanation behind these large differences is based on the assumptions used to adjust the food poverty line for basic nonfood needs. The starting point of either the lower or the upper adjustment to the food poverty line i s the simple fact that there is no explicit information on the basic nonfoodneeds of poor households.The lower poverty line is obtained under the assumption that the basic nonfood needs of a poor household i s best determined by the nonfood expenditures of households with total consumption expenditures equal to the value of the food poverty line. The nonfood expenditures of these households may be considered as absolutely necessary for sustaining the minimumliving standards, simply becausehouseholds are willing to forego some spending on what is required to attain the minimum caloric requirement (the basic needs food 28 basket) in order to purchase these nonfood items. In contrast, the upper poverty line i s derived under the assumption that the essential nonfood needs of a poor household can be determined by the nonfood expenditures of households who are satisfying the basic food needs (i.e., their food spending equals the cost of the basic needs food basket). Even though the reference group of households used by the upper poverty line may sound more appealing, it i s important to keep in mind that the nonfood expenditures of households satisfying the basic food needs do not necessarily represent the "basic nonfoodneeds of poor households". Inreality, the "basic nonfood needs of poor households" may be lower than the nonfood expenditures of these households. It is for this reason that the upper poverty line i s best considered as setting a limit to the range of admissible poverty lines. CONCLUSIONS The analysis of this chapter suggests that the maincomponents of poverty inBrazil can be best captured by three different poverty lines: A food or extreme poverty line that is determined by cost of the basic needs food bundle that yields the recommended energy (caloric) requirements of 2,300 kcal per capita per day from a sufficiently diverse variety of food. Households with total consumption expenditures per capita less than or equal to the food poverty line may then be considered as households in extreme poverty, as they are unable to satisfy the basic food needs. The analysis in chapter 2 determined that the food or extreme poverty line inBrazil is, on average, equal to R$61per capita per month. Moreover, the extreme poverty line varies only a little from region to region, with the lowest value of R$55 in the rural South region and the highest value of R$65 inmetropolitan Sao Paulo (see table 10). An intermediate poverty line that is determinedby the cost of satisfyingthe minimumlivelihood needs. This poverty line i s the CBN-Lower poverty line discussed in chapter 2 that adjusts upwards the food poverty line for the cost of essential nonfood needs. In this poverty line, the adjustment for the cost of essential nonfood needs i s determined by the nonfood expenditures of households that spend as much as the value of the food poverty line but forego some spending on food in order to purchase these essential nonfood items. The analysis inchapter 2 determined that the minimum livelihood or intermediate poverty line in Brazil i s on average equal to R$103 per capita per month, which i s very close to the "Administrative Poverty Line" of R$100 per person per month used for eligibility in the Bolsa Familia program, and the "misery line" of R$108 estimated by Social Policy Center of the Getulio Vargas Foundation (FGV). As i s the case with the extreme poverty line, the minimumlivelihood poverty line does not vary much from region to region, with the lowest value of R$90 in the ruralSouth region and the highest value of R$115 in metropolitan Sao Paulo (see table 12). An upper poverty line that sets a limit to the range of admissible poverty lines. This generous poverty line corresponds to the CBN-Upper poverty line that adjusts upwards the food poverty line for the cost of basic nonfoodneeds. Incontrast to the minimumlivelihood needs poverty line, the adjustment for the cost of basic food needs i s determined by the nonfood expenditures of households who are able to satisfy the basic food (i.e. their food spending equals the cost of the food poverty basket). The analysis inchapter 2 determined that the upper poverty line in Brazil i s on average equal to R$220per capitaper month, almost two times the minimumlivelihood needs poverty line. This report takes the view that the extreme poverty line and the minimumlivelihood poverty line are the poverty lines most relevant for policy. This choice i s based on two key reasons. First, these two poverty lines are the poverty thresholds most useful for identifying the households with the most pressing needs. 29 The extreme poverty line is useful for identifying individuals who cannot even afford to satisfy the basic food needs, while the minimumlivelihood poverty line i s useful for identifying individuals who cannot afford to satisfy the basic food nonfood needs. The second reason i s based on more practical considerations. Both poverty lines are close to the "Administrative Poverty Lines" of R$50and R$100 per capita per month that are already used to determine eligibility for one of the major poverty alleviation programs of Brazil, the Bolsa The fact that on average the upper poverty line i s more than three times the average food poverty line suggests that the "basic nonfood needs of poor households" are overestimated. Based on the observations above it is also quite apparent that the choice of one or more poverty line for Brazil would benefit from further investigationon the regional cost-of living in Brazil. The spatial price indices presented in the next chapter provide some additional considerations useful for choosing among the poverty lines discussedso far. 37The BolsaFamilia per capita incomeeligibility thresholds were just raised (in April 2006) from R$100 (upper threshold) and R$50(extremepovertythreshold) to R$120andR$60. 30 CHAPTER 3: SPATIAL PRICE INDICES FOR BRAZIL Ideally, differences between the economic welfare of households living in different regions can be determined by adjusting for the cost-of-living differences between regions. Cost-of-living differences could be measuredby a spatial price index, analogous to the Consumer PriceIndices (CPI) usedroutinely in adjusting for inflation over time. Unfortunately, in Brazil, as in most developing countries, a suitable spatial price index, especially for nonfood items i s not available. A spatial price index i s especially useful for deriving more reliable measures of inequality in Brazil. To the extent that cost-of-living differences are substantial between regions, uses of nominal income or consumption measures to measure the inequality in the standard of living in Brazil may be quite misleading. Various expenditure surveys, notably the PPV of 1996, suggest that price variations across this continent-sized nation are substantial. Brazil's earlier detailed expenditure survey of metropolitan areas, the POF 1996, broadly confirms the importance of these differences, even though, by construction, it cannot measure cost-of-living disparities between the metropolitan and rural areas of the country. A spatial price index i s also useful for the estimation of consistent poverty profiles. Ferreira et al. (1998, 2000, 2003), for example, use the PPV to construct a spatial price index for Brazil. This index may then used to derive "real" expenditures. Comparing real expenditures with the poverty line for the reference region of the metropolitan area of Sao Paulo, then yields a consistent regional profile of poverty. Inthis chapter we conduct a close examination of the spatial price indices that can be constructed from the POF survey. We first present a Laspeyres and a Paasche spatial price index based on the cost of food only. Next, we expand these indices to include the cost of housing as a measure of nonfood costs. Finally we examine the spatial price indices that can be derived from the estimated poverty lines of the previous chapter. Using the ratio of the nominal poverty line of each region to the poverty line of a reference region, one can derive a spatial price index summarizing the cost-of-living in the region relative to the cost of living in the reference region. Provided they represent sufficiently accurately the cost-of living differences between regions, the nominal regional poverty lines derived either by the CBN or by the FEI methods can be usedto derive a spatial price index. A SPATIALPRICEINDEXBASEDONTHE COSTOFFOODONLY We begin with a discussion of the spatial food price indices that can be constructed from the food poverty line using the CBN.38The CBN food poverty line (FPL)expressed as a ratio to food poverty line in the reference region of the Sa0 Paulo metropolitan area, may be interpreted as a Laspeyres price index denoted by FPI, .Usingmore formal notation: where the superscript R denoted the region and the superscript 0 denotes the reference region. Expression (1) implies that FPI,, is a weighted sum of the relative price of each of the twenty six items in the 38 Iti s importantto keep inmindthat the basic needs food basket i s basedon the consumption patternsof the 20-40 percentiles of the nationaldistribution of PCE. As it i s documentedin Appendix 2, the cost of the food basket was not sensitive to changes in the referencepopulation. 31 poverty basket, using as a weight the expenditure (or budget) share of the each item in the reference region. A well known limitation of the Laspeyres food price index is that it is basedon the assumption that there are no substitution effects among foods even though their relative prices may differ significantly between regions. Inorder to examine the sensitivity of the food price index to the weights used, a Paasche price index was also estimated for the same twenty-six food items entering the food poverty basket.39The expression for the food Paascheprice index FPI: is / .1 FPI,~= i=l 26 . P 9 d i=l Contrary to the Laspeyres index, the Paasche index takes into account household spending patterns of each household in the country (e.g., see Deaton and Zaidi, 2002). Inthe Paasche price index, the relative price of each food item in the basket i s weighted by the share of household h's budget devoted to that particular food item. As a consequence the Paasche index i s household-specific. The regional Paasche price index for cost-of-living, i s derivedby taking an averageof the household-specific values of the price indices ineachregion?' Table 14and figure 9 present the two spatial price indices for food. Figure 9 reveals that the variability in the cost of food across regions in Brazil is quite small. Moreover, there are some discrepancies in the relative costs of food between the two indices. For example, the Paasche index suggests that the cost of food i s practically identical between Rio de Janeiro and Sa0 Paulo (and Brasilia), whereas the Laspeyres index suggests that food costs less inboth Rio and Brasilia. Similar discrepanciesappear in the Northeast and in the South. Overall, however, the small variability observed for both the Laspeyres and Paasche indices suggests that the low variability in the cost of food with the CBN method i s not driven by the fixed quantities of the food items inthe basket. 39The Paascheindex inexpression (2) is based onthe full sampleof householdsinthe POF. Usingonly householdsinthe 20-40 percentilesofthe nationalPCEyieldedavery index values. As Deatonand Zaidi (2002) discuss in greater detail, deflating consumption expenditures by a Paasche regionalprice index yields an approximation of "money metric utility", whereas deflating by a Laspeyres regional price index yields an approximationof the "welfare ratio." The latter approachto measuringstandard of living across householdsis better suited for measuringthe redistributiveaffects of policies. 32 Table 14 Spatial priceindicesbasedonthe cost of food: Laspeyresvs. Paasche Region PaascheFood Price Laspeyres Food Price Index Index 1 MetroBelem 0.955 0.968 2 North Urban 0.955 0.924 3 Rural 0.905 0.901 4 MetroFortaleza 0.934 0.906 5 Metro Recife 0.844 0.945 6 Northeast Metro Salvador 0.928 0.974 7 Urban 0.910 0.927 8 RWal 0.916 0.901 9 10 11 Southeast Metro Belo Horizonte 0.928 0.908 12 Urban 0.981 0.977 13 Rural 0.945 0.889 14 Metro Curitiba 0.885 0.919 l5South MetroPort0Alegre 0.905 0.989 16 Urban 0.856 0.879 17 Rural 0.779 0.849 18 Brasilia 1.011 0.947 l9CenterWest Goianiamunicipality 0.979 0.913 20 Urban 0.962 0.939 21 Rural 0.952 0.919 Sao Paulo 1.000 1.000 Metropolitan 0.929 0.941 Urban excluding Metropolitan 0.933 0.929 Source:World Bank estimatesusing the 2002-03 POF. The numbers in the last four rows are simple (un-weighted)averages of the region-specificpovertylines Figure9 SpatialPriceIndicesfor Foodonly: Laspeyresvs. Paaschepriceindices +.Pwche F d Prloe Index +Lespyrs F d Prioe Index Source:World Bankestimatesusingthe 2002-03 POF. 33 SPATIAL PRICE INDICES BASEDONTHE COST OFFOODAND NONFOOD(HOUSING) The same approach used to derive the Laspeyres spatial price index for food can also be used to derive a spatial price index for both food and nonfood. The more complete Laspeyres index is constructed from the weighted sumof the Laspeyres index for food ,presentedabove, and the Laspeyres index for housing rent, i.e., (3.3) The weight assigned to the food index, w i , is the average share of food among households in the 20-40 of the nationaldistributionof PCE, residinginthe Sao Paulo metropolitan area (the reference region). The estimate of the housing rental rate in region R, denoted by iR, with iodenoting the housing rent in metropolitan Sa0 Paulo, i s obtained from a regression of the form: 20 lnr =ao+ZajRj +m+ E , (3.4) j=l where InY i s the logarithmof the (actual and imputed) rental rates contained inthe POF survey41,a,i s a constant term, R i s set of binary dummies for 20 of the 21 regions of Brazil (the Sao Paulo metropolitan area being the excluded reference region), and X summarizes a set of characteristics of the household residence. Specifically, X includes the number of rooms in the residence, whether the residence has no electricity, and a number of binary variables indicating whether the household has no access to paved road, no electricity, the type of dwelling (4 binary variables), the type of water system (3 binary variables), the type of sanitary service (3 binary variables), and the type of floor (4 binary variables). After estimating the regression equation using the full sample of household sin the POF survey, an estimate of the logarithm of the housingrentalrate inregionj may be obtained from the expression: where x2,-,,, denotes the average characteristics of the residences of households inthe 20-40 percent of the nationalPCE distribution. The corresponding Paasche price index for the cost of food and nonfood i s constructed based on the formula 42 41 The rent variable in the POF i s composedof the actual rent reportedby householdsrenting a houseandreported implicit rent for householdswho owntheir house. 42 The Paascheindex inexpression (6) i s basedon the full sample of householdsinthe POF. Usingonly householdsinthe 20-40 percentilesof the nationalPCEyieldeda very index values. 34 Table 15 Spatial price indices basedtsn the cost of food and ~o~~~~~~Lispeyres YS. Paasche Keginn LaspeyresPrice PattschePrice lndev 2 North trlhan 0 7t3 0.617 3 Rural 0 654 0.589 4 MetroFortslem 0.626 0.61i 5 Metro R m f c 0.770 0.642 h w or the^^^ bfetro Snhadoc 0.780 0 755 7 Urban 0 515 0 541 8 Rurd 0 437 0.524 9 MetroRio Uc Janeirts 0.9ib L 0.957 11 Urbsn 0.634 0.68f3 13 KLtral 0.581 (3.h36 14 Metro Cur1tha 0.809 0.865 15 Metlo PorpuAlcgrc 0.847 0.79s 16 Souih Clrt.an 0.620 0.633 i 7 Rural 0.516 0.5') 1 18 BrStaIIa I.078 i252 0.723 0.67' Table IS snd figure I O present the Laspeyres and Paasche price indices. As in the cabe of food. the regional Paasche price index is derived by taking an aberage of the h ~ ) u ~ ~ hvalttes~ of~rhe - ~ ~ e ~ ~ ~ ~ Paascheprice indices in each region. Figure 10 Spatial Price Indices for Foodand Housing: Laspeyres vs. Paasche price indices 1.4 1.2 t 1.1 0.9 x 0.8 = 0 0.6 0.5 0.3 - 0.2 - Northeast I Southeast South 1 Centerwest -.iap-Paasche Price Index +Laspeyres Price Index Source: World Bankestimates usingthe 2002-03 POF. Even though, there i s no particular reason for the values of the Laspeyres and Paasche indices to be similar (e.g. Deaton and Zaidi, 2002), figure 10 reveals that there is a close relationship between the two indices. This suggests that the weights applied to the relative prices are very similar between the two indices. It also implies that differences in household consumption patterns across regions do not play an important role in the estimated cost of living differences across regions. A simple comparison between figures 9 and 10also reveals that the rental cost of housing i s the primary cause of the large cost of living differences across regions. SPATIALPRICEINDEXBASEDONTHE ESTIMATEDPOVERTYLINES The last set of spatial price indices that can be constructed i s using the estimated poverty lines of chapter 2. Expressing the poverty lines of the CBN or the FEI methods in terms of the poverty line in the reference region of metropolitanSa0 Paulo. For example, PIPL =-PLR PLO (3.3) It is important to keep in mind that the spatial price indices, obtained from the ratio of the CBN poverty lines are not necessarily equivalent to a Laspeyres price index. Even though the food component of the poverty line holds fixed the compositionandthe quantities of the items in the basic needs food basket, the adjustment of the food poverty line for basic nonfood needs does not hold constant neither the composition nor the quantity of nonfoodgoods (see chapter 2 for more details). These spatial price indices are presentedinTable 16 and in the corresponding figure 11. As already noted inchapter 2, the FEI spatial price index displays the greatest regional variability in comparison to the CBN poverty lines. According to the FEIpoverty lines, the cost of living in the rural Northeast i s 21.2% 36 of the cost of living inmetropolitan Sa0 Paulo. Incontrast the CBN-Upper (Lower) poverty line suggests that the cost of living in the same region isjust over half (51.3%) of the cost of living inmetropolitan Sao Paulo. Taking as given that the Laspeyres spatial price index presents a fairly accurate picture of the regional variation in the cost-of-living, one can compare the regional variability of the CBN poverty lines and the FEIpoverty lines against it. This can help determine the extent to which the regional differences in the nominal poverty lines reflect cost-of living differences. Figures 12 and 13 facilitate the comparison of the regional variability of the CBN and FEIpoverty lines with the regional variability in the cost of food and housingmeasuredby the Laspeyres index calculated above. As it i s apparent, the CBN-Upper poverty line is able to approximate the regional variability in the cost- of-living better than the FEIpoverty line. Thus the CBN poverty line i s more likely to yield a consistent poverty profile than the FEIpoverty line. Table 16 Spatialpriceindicesbasedon the regionalCBNandFEIPoverty Lines CBN CBN Region Lower Upper FEIindex 1 Metro Belem 0.911 0.640 0.490 2 North Urban 0.889 0.695 0.441 3 Rural 0.811 0.484 0.264 4 Metro Fortaleza 0.866 0.655 0.495 5 Metro Recife 0.905 0.689 0.540 6 Northeast Metro Salvador 0.940 0.746 0.657 7 Urban 0.868 0.596 0.349 8 Rural 0.802 0.461 0.212 9 Metro Rio De Janeiro 0.932 0.833 0.896 10 11 Southeast Metro Belo Horizonte 0.895 0.802 0.823 12 Urban 0.952 0.796 0.746 13 Rural 0.842 0.607 0.465 14 Metro Curitiba 0.912 0.865 0.964 l5South Metro Port0Alegre 0.970 0.843 1.006 16 Urban 0.859 0.735 0.701 17 Rural 0.785 0.513 0.471 18 Brasilia 0.952 0.997 1.011 GoianiaMunicipality 0.900 0.810 0.934 20 l9CenterWest Urban 0.917 0.782 0.545 21 Rural 0.871 0.628 0.419 Sao Paulo 1.000 1.000 1.000 Metropolitan 0.918 0.788 0.782 Urbanexcluding Metropolitan 0.897 0.721 0.556 Rural 0.822 0.539 0.366 Source:World Bankestimatesusingthe 2002-03 POF. The numbersinthe last four rows are simple (un-weighted) averages of the region-specificvalues. 37 Figure 11 Spatial price indicesbasedon the regionalCBNand FEIPoverty Lines 1.2 1 1 ... 0.9 0.8 - 0.6 0.5 0.3 I Northeast 1 Southeast south I Centerwest +CBN Lower +CBN Upper -4- FEI Source: World Bankestimatesusingthe 2002-03 POF. Figure 12 Spatial price indices basedon the regional CBN-Upper Poverty Line and the Laspeyres Food and Housing Index 1.2 1.I 0.9 0.8 - 0.6 0.5 0.3 f 0.2 I 1 North 1 Northeast I Southeast 1 South 1 Center West 1 &CBN Upper +Laspeyres Source: World Bankestimatesusingthe 2002-03 POF. 38 Figure13 Spatial priceindicesbasedon the regionalFEIand the LaspeyresFoodand Housing Index 1 1 0 9 0 8 - 0 6 0 5 0 3 0 7 1 I II I I Nsrtheast I Southeast I South 1 Center West 1 +Laspeyres .-b-FEI Source: World Bankestimatesusingthe 2002-03 POF. THESENSITIVITYOFINEQUALITYMEASURES TO ADJUSTMENTSFORREGIONALDIFFERENCES INTHE COSTOFLIVINGINBRAZIL It is a well known fact that Brazil has one of the highest inequality rates inthe world. Usingincome from the PNAD surveys of Brazil the Gini coefficient i s estimated to be between 0.594 and 0.640 in 1999. All of the inequality indices obtained from the PNAD typically do not adjust for differences in the cost of living between regions and instead use nominal household income. In addition, inequality indices based on consumption are typically lower than inequality indices based on income. One important advantage offered by the availability of the regional cost of living indices i s the opportunity to investigate how the measurement of inequality in Brazil i s affected when one uses real household expenditures or real household income. Real expenditures are a better measure of household welfare especially if one intends to compare individual welfare across space (or time) (e.g. see Deaton and Zaidi, 2002). However, before we embark with the discussion of how the estimates of inequality are affected one caveat i s in order. The consumption aggregate constructed for the purpose of measuring household welfare (discussed in detail appendix 1) with the POF data did not include the flow of services from a number of durable items such as furniture and appliances. Although it could be reasonably argued that he omission of the flows of services from a number durable items from the consumption aggregate matters relatively little to poverty calculations, the omission of services from durable items i s likely to be more important for inequality calculations. Thus the PCE from POF Ilikely to provide an underestimate of poverty, since it excludes items likely to be important in the consumption expenditures of wealthier families. With this caveat in mind, Table 17 presents the Gini coefficients and the General Entropy (GE( )) class of inequality indices using PCE and PCINC from POF and PCINC from the 2004 39 PNAD.43At this point it suffices to say that the GE inequality indices can be allowed to differ in their parameter a .The more positive a is, the more sensitive GE(a ) is to income differences at the top of sensitivities to differences in different parts of the distribution, by assigning different values to the the distribution; the more negative a i s the more sensitive it i s to differences at the bottom of the distribution. GE(0) i s the mean logarithmic deviation, GE(1) i s the Theil index, and GE(2) i s half the square of the coefficient of variation. The Gini coefficient i s most sensitive to income differences about the middle (more precisely the mode). Rows (l), (4) and (7) present inequality measures using the nominal values of PCE and PCINC. As anticipated, consumption inequality (using PCE) i s considerably lower than income inequality (using PCINC from POF or from PNAD). Using real PCE, that i s deflating nominal PCE by the spatial price index based on the upper poverty line in table 9, results ina decrease of the Gini inequality measure from 0.507 to 0.479. Similar declines occur when nominal expenditures are deflated usingthe Laspeyres cost of living index in Table 17. Given that the spatial price index obtained from CBN upper poverty line is very similar to the Laspeyres price index it i s no surprise that practically identical Gmi coefficients emerge (e.g. compare row 2 with row 3). Table 17 The Sensitivity of InequalityMeasuresto Adjustmentsfor RegionalDifferencesin the CostofLivinginBrazil Welfaremeasure Gini GWO) GE(1) GW2) (1) NominalPCE-POF 0.507 0.459 0.477 0.802 (2) PCE deflated by UPPER Poverty LinePriceIndex o.479 0.398 0.422 0.688 (3) PCE deflated by Laspeyres Price Index o.481 0.404 0.424 0.683 (4) Nominal PCINC-POF 0.593 0.659 0.721 1.778 (5) PCI deflated by UPPER Poverty Line PriceIndex o.568 0.588 0.666 1.682 (6) PCI deflated by Laspeyres Price Index 0.571 0.596 0.670 1.703 (7)Nominal PCINC-PNAD 0.564 0.591 0.644 1.588 (8) PNAD PCI deflated by UPPER Poverty Line PriceIndex o.543 0.534 0.597 1.423 (9) PNAD PCI deflated by Laspeyres PriceIndex o.547 0.545 0.603 1.425 Source: World Bankestimates usingthe 2002-03 POF. Even larger reductions are observed for the GE inequality measures that can weight different parts of the distribution differently. For example, with the GE(2) that is more sensitive to income differences at the top of the distribution of consumption, the estimate of inequality drops from 0.802 to 0.688. Declines in the inequality measures based on income also occur, but they are less dramatic than the decline in consumption inequality. 43The formula for the Gini coefficient is 2- 1 n n Iyi - yIwhile the formula for the generalizedentropy measure is 2n yi=l j=1 40 CONCLUSIONS The analysis inthis chapterreveals the following: Irrespective of whether one uses a Laspeyres or a Paasche price index, the variability in the cost of food across regions inBrazil i s quite small. This suggests that the low variability in the cost of food with the CBNmethodi s not driven by the fixed quantities of the food items inthe basket. Expandingthe Laspeyres andPaascheprice indices to include the rental cost of housingleads to a greater variability inthe cost of living across regions. This however, does not alter the close relationshipbetween the two indices across the twenty one regions of Brazil. Thus, differences in household consumption patterns across regions do not seem to have an importantrole inthe estimated cost of living differences across regions. The CBN-Upper poverty line i s able to approximate the regional variability in the cost-of-living better than the FEIpoverty line. Thus the CBN poverty line i s more likely to yield a consistent poverty profile than the FEIpoverty line. The proper measurement of consumption and income inequality in Brazil must definitely take into consideration the cost of living differences between regions. The spatial price indices constructed inthis chapter provide a good start inthis direction. 41 CHAPTER4: AN UPDATEDREGIONALPOVERTYPROFILEFORBRAZIL THEHEADCOUNTPOVERTYPROFILEOFBRAZIL Using the three poverty lines based on the cost of basic needs method, this chapter derives poverty estimates along the lines suggestedby Foster, Greer Thorbecke (1984)? Table 18 The Headcount poverty index (in %) for different poverty lines FEI Regions Extreme Minimum Poverty Upper Poverty Livelihood Poverty Line Line 1 Metro Belem 3.6 16.7 45.3 28.3 2 North Urban 9.5 27.4 63.4 37.1 3 Rural 22.1 45.7 71.1 33.1 4 Metro Fortaleza 7.6 20.8 53.1 38.5 5 Metro Recife 5.4 17.7 47.5 33.4 6 Northeast Metro Salvador 3.6 14.4 44.3 36.1 7 Urban 15.3 36.8 64.9 36.4 8 Rural 30.6 55.0 76.3 31.1 9 Metro Rio De Janeiro 2.4 8.8 40.7 41.O 10 Metro Sao Paulo 1.5 7.2 44.3 41.7 11 Southeast Metro Belo Horizonte 0.7. 7.1 40.3 39.5 12 Urban 5.0 14.4 46.2 40.7 13 Rural 11.0 28.2 61.9 41.8 14 Metro Curitiba 0.9 4.7 38.3 41.1 15 Metro Port0Alegre 0.6 7.2 36.6 42.0 16 South Urban 3.4 12.1 45.1 39.3 17 Rural 5.5 15.9 39.0 30.4 18 Brasilia 0.6 8.1 45.3 41.2 2o 19 Center West GoianiaMunicipality 1.2 5.1 40.2 42.7 Urban 7.1 22.9 62.1 40.2 21 Rural 11.3 30.9 61 n 39.1 National 8.5 215 52.8 38.4 Metropolitan 3.1 11.4 44.0 37.9 Urban excluding Metropolitan 8.1 21.9 543 40.2 All UrbanincludingMetropolitan 6.0 175 50.0 393 Rural 20.6 41.0 663 34.1 Source: World Bank estimates using the 2002-03 POF. Table 18 presents the regional profile of poverty based on the Headcount poverty index, which i s equivalent to the percentage of the population that i s below the poverty line in each region. According to these estimates, approximately 8.5% of the Brazilian population did not have a total consumption 44 Poverty indexes belongingto the FGT class are defined as: FGT(Cl)=- (` :yi)]asi,where z is the value of [ n i=l the povertyline andSi i s the indicative variable that is equal to 1if the i-th individual is below the poverty line and 0, if not. For a=O, 1and 2, FGT(a) represents, respectively,the incidence(proportionof poor), the intensity or depth (poverty gap) and the severity of poverty (squaredpovertygap). The larger the coefficienta,the greater the weight attributedto the povertygaps. 42 expenditure sufficient to buy the basic needs food bundle. Given the total population of Brazil of 175,331,797 the estimate of extreme poverty inBrazil implies that 14,903,203 individuals are in extreme poverty.45 The poverty rates increase substantially when higher poverty lines are used to take into account basic nonfood expenditures. The minimum livelihood poverty line implies poverty rate of 21.5% for Brazil which amounts to 37,696,336 individuals being unable to meet basic food and nonfood expenditures. Clearly, the distribution of consumption i s very steep between these two poverty lines. An increasein the average poverty line from R$61 to R$103 increase the poverty rate from 8.5% to 21.5%. Similarly, an increase in the average poverty line from R$103 to R$220, the upper poverty line, increases the poverty rate from 21.5% to 52.8%, resultinginmore than 92.5 millionindividualsbeing classifiedas poor. One key result obtained from the poverty rates associated with the upper poverty line i s that the adjustment of the food poverty line for "basic" non food needs makes a tremendous difference on the estimated poverty rate for The aggregate statistics for Brazil conceals very large regional disparities. One of the striking characteristics of poverty in Brazil i s that it varies dramatically across geographic regions and areas. According to Ferreira, Lanjouw and Neri (2003), poverty incidence is higher in small and medium towns than inmetropolitanregions. Table 18 confirms that this i s also the case in the 2002-03 POF survey. Extreme poverty i s concentrated in the rural areas of Brazil which have a headcount poverty index of 20.6%. Urban areas (excluding metropolitan areas) have an extreme poverty rate of 8.1%, followed by metropolitan areas with an extreme poverty rate of 3.1%. The same pattern holds for the intermediate poverty line and the upper poverty lines. Table 19 displays how the poverty rates increaseas the poverty line i s increased from the extreme poverty line to the minimumlivelihood poverty line and then to the upper poverty line. The columns of this table are estimated by differencing the poverty rates associatedwith the different poverty lines in table 10. For example, column (1) of table 11 displays the increase in the poverty rate when the poverty line i s increased from the extreme poverty line to the minimumlivelihood poverty line. This table i s useful for determining how steep i s the distribution of per capita expenditures (PCE) ineachregion. Table 19 makes more apparent the source of the increasedpoverty rates associatedwith an increase inthe poverty line. For example, at the national level an increase in the poverty line from the extreme poverty line to the minimum livelihood poverty line increase the national poverty rate by 13 percentage point (21.5-8.5=13). The large increase in the poverty rate of ruralareas, i.e. by 20.4 percentage points implies that the distribution of consumption in rural areas i s very steep between these two poverty lines and less steep in the metropolitan and urban areas where the poverty rate increases by 15.9 percentage points. However, the increase inurban areas i s considerably greater (36.9 percentage points) than the increase in the poverty rate in rural areas (25.3 percentage points) when the poverty line i s increased from the minimum livelihood poverty line to the upper poverty line. Thus, the potential increases in the official poverty line are likely to entail a shift in scope of poverty alleviation programs as more of the new poor are likely to be located inurban areas. 45This estimate i s remarkably close to the number of people estimated by the Ministry of Planning and the Ministry of Social Developmentof Brazilto experiencefood insecurity (Brazil: SegurancaAlimentar 2004). 46Even though the poverty rates with the upper poverty line may appear to be "too" high, they are not unique. Ferreira et al (2003) report anationalpoverty rateof 45.29% using a similar methodon the 1996PNAD. 43 Table 19 Increases inthe Headcount poverty index associatedwith increasesinthe poverty line Regions (I)* w** (3)*** 1 Metro Belem 13.1 28.6 41.7 2 North Urban 17.9 36.1 54.0 3 Rural 23.6 25.4 49.O 4 Metro Fortaleza 13.1 32.4 45.5 5 Metro Recife 12.2 29.8 42.1 6 Northeast Metro Salvador 10.8 29.9 40.6 7 Urban 21.5 28.1 49.6 8 Rural 24.4 21.3 45.7 9 Metro Rio DeJaneiro 6.3 31.9 38.2 10 Metro Sa0 Paul0 5.7 37.1 42.8 11 Southeast Metro Belo Horizonte 6.4 33.2 39.6 12 Urban 9.4 31.8 41.2 13 Rural 17.3 33.7 50.9 14 Metro Curitiba 3.9 33.5 37.4 l5south MetroPort0Alegre 6.6 29.4 36.0 16 Urban 8.7 33.0 41.6 17 Rural 10.3 23.1 33.4 18 Brasilia 7.5 37.2 44.7 GoianiaMunicipality 3.9 35.0 39.0 20 l9CenterWest Urban 15.8 39.2 55.0 21 Rural 19.7 30.1 49.7 National 13.0 31.3 44.3 Metropolitan 8.2 32.6 40.9 Urbanexcluding Metropolitan 9.4 36.9 46.2 All Urban including Metropolitan 15.9 28.1 44.0 Rural 20.4 25.3 45.7 *Source: World Bank estimates usingthe 2002-03 POF. Column (1) is the poverty rate with the minimumlivelihood poverty line intable 10minusthe extreme poverty rate intable 10. ** Column (2) i s the poverty rate with the upper poverty line minusthe poverty rate with the minimum livelihood poverty line intable 10. *** Column (3) is the poverty rate with the upper poverty line in table 10minus the extreme poverty rate in table 10. Figures 14 through 17, basedon table 19, present the rankingsof the different regions of Brazil resulting from the four different poverty lines. As it is apparent, the six regions with the highest extreme poverty and minimumlivelihood poverty are the rural areas in the Northeast, with an extreme poverty index of just under 31%, followed by the ruralNorth, urbanareas in the Northeast, ruralareas inthe Center-West region and Southeast region, and urban areas in the North, which has an extreme poverty of 9.5%. The same six regions rank at the top using the more generous upper poverty line with the only difference being that urbanareas inthe North switchrank with rural areas inthe Center-West (see figure 16). However, the poverty ranking of regions changes considerably when one applies the FEI poverty lines that emphasize specificity. As figure 17 shows, the FEIpoverty line results in poverty rates that are very similar between regions, rangingbetween 35 and 40% in the majority of the regions. More importantly, the rural Northand Northeast regions, the two regions ranked as the poorest regions by the other poverty lines emphasizing consistency in terms of command over basic consumption needs, end up being ranked among the regions with the lowest poverty rates inBrazil. 44 The observed differencesin the regional poverty rankings of regions suggest that there is a considerable conflict between consistency and specificity. Although, in principle, a basic needs bundle should reflect the regional specificity of preferences, the regional standard of living, and the regional perceptions of what constitutes poverty, it turns out that an effort to do so, yields a regional poverty profile that may not be as usefulfor policy. Thus, ultimately, the choice of the methodusedto set poverty lines and measure poverty dependson the purpose of the poverty profile. The remainder of this report, assumes that the purpose of the poverty profile is to inform policy makers about the regional distribution of poverty so it can facilitate the formulation, designand targetingof social programs aimed at alleviating poverty. For this reason, the poverty profile obtained by using the FEI poverty line is not discussed further. Figure 14 The incidence of extreme poverty (Headcount poverty index) by region 40.0 1 W." 30.0 25.0 20.0 15.0 10.0 5.0 0.0 I Extreme Poverty rates -National average I Source: World Bank estimates usingthe 2002-03 POF. 45 Figure 15 The incidence of minimumlivelihood poverty (Headcount poverty index) by region 80.0 75.0 70.0 65.0 60.0 55.0 50.0 45.0 40.0 35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0 Source: World Bank estimates using the 2002-03 POF. Figure 16 The incidence of poverty (Headcount poverty index) by region using the upper poverty line 80.0 75.0 70.0 65.0 60.0 55.0 50.0 45.0 40.0 35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0 Source: World Bank estimates using the 2002-03 POF. 46 Figure 17 The incidence of poverty (Headcount poverty rate) by region usingthe FEIpoverty line 1-.- 45.04 40.0 35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0 I Povertvrates - Nationalaveraae I Source: World Bankestimates using the 2002-03 POF. THEPOVERTYGAPPROFILEOFBRAZIL The poverty gap or depth of poverty index, measures the average distance of consumption of poor households from the food poverty line as a proportion of the food poverty line. The aggregate food poverty gap in Brazil, is estimated to be 2.5%. Using the minimumlivelihood poverty line of R$103per personper month, the value of the poverty gap index increases to 7.3%. Table 20 reveals that the poverty gap profile is similar to the incidenceof poverty profile of Brazil4' The food poverty gap index i s higher in rural areas, followed by urban areas and metropolitan areas. Specifically, the ruralareas in the northeastregion are areas with the highestpoverty gap index of 10.6%, followed by the ruralareas inthe North that register a food poverty overt gap index of 5.9%. The poverty gap index, can also be interpreted as an indicator of the potential for eliminating poverty by targeting transfers to the poor (Ravallion, 1994). The minimum cost of eliminating poverty using targeted transfers i s simply the sum of all the household-specificpoverty gaps in the population. A government deeply concerned about eliminating poverty would have to spend at least this amount if it were to eliminate poverty. To be able to spend this minimumcost, however, requires that the government have a large amount of information such as the distance (poverty gap) of each poor household from the poverty line. At the other extreme, one can consider the maximum cost of eliminating poverty, which can be derived by assumingthat the government knows nothing about who is poor and who is not. In this latter case, the government would have to give a transfer equal to the value of the poverty line to all the households in the country so as to ensure that the poor, whoever these are, can afford the cost of the basic needs basket. Then, it can be easily shown that the ratio of the minimumcost of eliminating poverty with perfect targetingto the maximum cost with no targetingi s simply the poverty gap. 47The poverty gap measuresthe average distance below the poverty line and i s expressed as a proportion of the poverty line. The correspondingseventy ofpoverty profile is presentedinthe Appendix 6. 47 The aggregate food poverty gap in Brazil represents0.45% of the country's aggregate consumption of all goods and services!* This suggests that the potential gains from targeting are quite large in Brazil. For example, the costs of assuring that everyone in the country can afford the poverty food bundle without targeting i s about 40 times the cost with a perfect targeting scheme that transfers an amount equal to household-specific poverty gap to each poor household. It i s important to clarify that this estimate represents only the potential gains from "perfect" targeting, where perfect i s defined as providing "tailor- made" household-specific transfers. The extent to which such a potential can be realized in practice depends on the constraints and costs faced by policy makers in identifying the household-specific poverty gaps. Table 20 The Poverty Gap indexin(%) for different poverty lines Extreme Minimum. Upper Regions Poverty Livelihood Poverty Line 1 Metro Belem 0.8 4.2 15.5 2 North Urban 2.3 8.9 21.0 3 Rural 5.9 16.3 22.0 4 Metro Fortaleza 1.3 6.4 20.4 5 Metro Recife 1.6 5.4 16.8 6 Northeast Metro Salvador 1.5 4.4 17.0 7 Urban 4.4 13.3 24.4 8 RUal 10.6 22.3 27.8 9 Metro Rio DeJaneiro 0.4 2.3 13.3 10 Metro Sao Paul0 0.3 1.7 14.9 11 Southeast Metro Belo Horizonte 0.1 1.4 15.2 12 Urban 1.4 4.7 17.1 13 Rural 2.8 9.4 19.5 14 Metro Curitiba 0.0 1.2 11.2 l5South Metro Port0Alegre 0.1 1.6 13.0 16 Urban 0.9 3.6 15.6 17 RWal 1.3 4.7 12.7 18 Brasilia 0.1 1.5 15.7 GoianiaMunicipality 0.4 1.4 12.2 20 l9CenterWest Urban 1.8 7.2 21.2 21 Rural 4.3 10.9 19.7 National 2.5 7.3 21.1 Metropolitan 0.7 3.5 14.9 UrbanexcludingMetropolitan 2.2 7.4 19.8 All UrbanincludingMetropolitan 1.6 5.6 17.8 Rural 6.6 15.6 22.7 Source: World Bank estimates usingthe 2002-03 POF. A POVERTY PROFILE BASEDONHOUSING AND HEADOFHOUSEHOLD CHARACTERISTICS Table 21 presents a poverty profile based on some key characteristics of the household residence and of the head of the household. This poverty profile i s constructed suing the minimumlivelihood poverty line. Similar profiles using the extreme poverty line and the upper poverty line yielded very similar patterns 48This number i s obtainedby the multiplying the value of the food povertygap index for Brazil by the ratio of the average food povertyline to the meanPCEinBrazil (Le. 0.025*(61/335.9) = 0.025*0.1816 = 0.454%). Usingthe minimumlivelihoodpoverty line of R$103 the aggregatepoverty gap in Brazil represents0.767% of the country's aggregate consumption of all goods and services. 48 (see appendix 5). Irrespective of the measure of poverty used, the Northeast i s the poorest region, followed by the North, the Center-West, the South, and the Southeast in that order. This finding is consistent with all of the existing studies on poverty in Brazil. The Northeast region accounts for only 27.9% of the population of Brazil but close to 50% of the poor in Brazil. Combined with the North region, the Northeastand the Northregion accountfor over 60% of the poor inBrazil. Households in non-metropolitanurban areas make up 48.2% of the population and 49.1% of the poor in Brazil. Thus the number of urbanpoor persons in Brazil i s greater than the number of the poor persons living inrural areas. However, the poverty gap and severity of poverty indices are higher inthe ruralareas than inthe non-metropolitanurban areas. The housingstatus suggests that individuals with or in "ceded" housing, an arrangement predominant in some types o f agricultural contracts and among domestic servants, have the highest incidence of poverty (30.3%) and the second highest contribution to the nationalpoverty (15.9% of the poor). As for accessto services, 13.9% of the Brazilianpopulationdoes not have accessto pipedwater and4.4% has no electricity. More than half of these two groups of individuals are poor. Specifically, among those who have no access to piped water 57.2% are poor and among those who have no access to electricity 59.8% are poor. Along similar lines, 36.9 of the individuals classified as poor have live in a residence without piped water, and 12.1% of the poor live in a residencewithout electricity. Just less than 78% of the poor has no access to the regular sewerage systemand have to rely on alternativemeans for sanitation, such as cesspits, rivers or lakes. Infact, over 21% of the poor have no sanitation means at all. Finally, 40.5% of the Brazilian populationresides in houses located on unpaved streets, while the fraction of the poor individuals with homes on unpavedstreets i s over 63%. It is quite apparent that increasingaccess to basic services would go a long way towards increasingthe living conditionsof the poor. Turning to the partitions based on characteristics of the householdhead, one finds that individuals from female-headedhouseholds are almost as likely to be poor as individuals from male-headedhouseholds, (20.7% vs. 21.7%). Thus, gender of the household head does not appear to be a good predictor of the poverty status of an individual. Incontrast to gender, race does seem to have a stronger correlation with the poverty status of an individual. The incidence of poverty i s the highest among individuals residing in householdswhere the head is an indigenousperson (with a headcountof 38.3%), followed by those who reside in households where the head i s a Parda (30.8%) and then black (26.5%). More than half of the poor (58.3%) are Parda even though individuals from Parda-headed households are only 40.7% of the Brazilian population. The age of the householdhead appears to have a significant and negativecorrelationwith the incidence of poverty. Individuals from households with older (60+) household heads have the lowest incidence of poverty (18.8%) while individuals from householdswith younger (less than 25 year old) householdheads have the highest incidence of poverty (27.1%), with the incidence of poverty declining monotonically with the age group of the head. As is common in most countries, the years of education of the household head are strongly associated with the incidence of poverty. The incidence of poverty among individuals from households with a head that has 0 to 8 years of education is 27.2%, whereas the incidence of poverty among individuals from householdswith a head that has 9 to 13 (14+) years of educationis 5.3% (0.2%). Moreover, more than 95% of the poor are from householdswith a headthat has less than 9 years of schooling. The immigration status of the householdhead is weakly associated with poverty status. The incidence of poverty i s the lowest among households where the head is not a migrant, and the highest among households where the head was a migrant last year. However, the contributions of the three migrant categories to the nationalpoverty are inline with their respectiveshares inthe population. 49 Domestic servants and self-employed workers appear as having the highest incidence of poverty, 29.8% and 25.7%, respectively. Individuals from households where the head is self-employed comprise 39.8% of the populationof the poor, which is greater than their shareof 33.5% inthe total population. Table 21 A poverty profile based on the 2002-03POF(Poverty Line: MinimumLivelihood Line) Household Incidence Poverty Poverty Nzoqfal Av. Pc Subgroups to national consumption characteristics of poverty PO gap P1 severity P2 population poverty of subgroup Total 21.5 7.3 3.5 100.0 100.0 335.9 North 30.7 10.2 4.7 7.8 11.1 219.7 Northeast 38.3 14.3 1.3 27.9 49.7 206.2 Region Southeast 12.7 3.9 1.7 42.7 25.1 427.4 south 11.3 3.2 1.4 14.7 7.7 378.0 Center-West 19.6 6.1 2.8 7.0 6.3 335.1 Metropolitan 11.4 3.2 1.4 34.7 18.3 457.7 Area of residence metropolitan urban 21.9 7.4 3.5 48.2 49.1 3 10.7 Rural- 41.0 15.6 8.0 17.1 32.6 159.7 Housing Ownandalreadypaid 21.5 7.4 3.5 68.3 68.4 337.1 Own, still paying 8.5 2.5 1.1 5.4 2.2 486.0 Housingstatus Rented 18.3 5.9 2.8 13.5 11.5 357.0 Ceded 30.3 10.6 5.2 11.3 15.9 244.8 Other 28.7 10.6 5.2 1.6 2.1 233.7 Water Piped 15.7 4.8 2.1 86.1 63.1 372.2 Not piped 57.2 23.1 12.2 13.9 36.9 110.6 Sewerage system - . 10.2 2.9 1.3 46.7 22.1 465.7 Cesspit 15.6 4.6 2.0 16.7 12.2 317.4 RudimentalCesspit 33.2 11.5 5.4 23.5 36.2 197.0 Sanitation Drain 35.8 12.5 5.9 2.4 4.0 176.5 River or Lake 24.9 7.8 3.4 2.8 3.3 228.0 Other 36.3 12.5 6.0 0.5 0.9 172.8 None 62.5 26.2 14.3 7.4 21.4 102.2 Electricity Yes 19.8 6.5 3.O 95.7 87.9 346.4 No 59.8 25.8 14.4 4.4 12.1 105.8 PavedRoad E 13.1 4.1 1.8 59.5 36.4 429.9 33.7 12.1 6.O 40.5 63.6 1979 Household head Gender Female 20.7 6.7 3.2 22.7 21.9 366.0 Male 21.7 7.5 3.6 77.3 78.1 327.1 White 13.1 4.1 1.8 49.8 30.3 440.7 Black 26.5 9.3 4.5 8.6 10.5 240.4 Race Asian 10.5 4.2 2.4 0.6 0.3 677.1 Parda 30.8 10.9 5.3 40.7 58.3 223.9 Indigenous 38.3 17.0 9.6 0.4 0.7 261.9 0-24 27.1 9.1 4.2 4.4 5.5 242.8 Age group 45-59 2 5 4 23.2 7.9 3.8 49.2 53.0 310.7 19.4 6.7 3.3 30.7 27.7 378.0 60+ 18.8 6.2 2.9 15.7 13.8 358.1 0-8 27.2 9.4 4.5 75.6 95.6 234.5 Education 9-13 5.3 1.3 0.5 17.9 4.4 480.1 14+ 0.2 0.0 0.0 6.5 0.1 1124.4 Migrant- last year 24.5 8.4 4.0 13.1 14.9 291.4 Immigration Migrant- last 5 years 22.2 7.8 3.8 28.9 29.8 349.5 Not migrant 20.5 6.9 3.3 58.1 55.4 339.1 Domesticservant 29.8 8.3 3.4 3.5 4.9 201.1 Employed 19.7 6.7 3.2 55.6 50.6 342.7 Occupational Employer 2.7 0.7 0.3 5.O 0.6 739.1 category Self-employed 25.7 9.1 4.4 33.5 39.8 277.3 Apprentice 12.9 5.1 2.3 0.5 0.3 478.8 Worker for self- consumption 45.1 16.5 8.3 1.8 3.8 146.8 Source: WorldBankestimates usingthe 2002-03 POF. 50 CONCLUSIONS 0 According to the estimates, approximately 8.5% of the Brazilian population does not have a total consumption expenditure sufficient to buy the basic needs food bundle. Given the total population of Brazil, the estimate of extreme poverty in Brazil implies that just under 15 million (14,903,203) individuals live inextreme poverty. The poverty estimates increase substantially when higher poverty lines are used to take into account basic nonfood expenditures. The minimumlivelihood poverty line implies poverty rate of 21.5% for Brazil which amounts to 37,696,336 individuals being unable to meet basic food and nonfood expenditures. Similarly, an increase in the average poverty line from R$103 to R$220, the upper poverty line, increases the poverty rate from 21.5% to %, resulting in more than 83 million individuals being classified as poor. Clearly, the adjustment of the food poverty line for "basic" non food needs makes a tremendous difference on the estimated poverty rate for Brazil. Even though the poverty rates with the upper poverty line may appear "too" high, they are not unique. Ferreira et al (2003), for example, report a national poverty rate of 45.29% and a similar regionalpoverty profile using equivalent methods on the 1996 PNAD. The aggregate statistics for Brazil also conceal very large regional disparities. The six regions with the highest extreme poverty and minimum livelihood poverty are the rural areas in the Northeast, with an extreme poverty index of just under 31%, followed by the rural areas in the North, urban areas in the Northeast, rural areas in the Center-West region and Southeast region, andurbanareas inthe North, which has an extreme poverty of 9.5%. 0 The national estimate of the poverty gap index suggests that the potential gains from targetingare large in Brazil. The costs of assuring that everyone can afford the poverty food bundle without targeting i s about 40 times the cost with perfect targeting. 51 CHAPTER5: SOME POLICY IMPLICATIONS OFTHE REGIONALDISTRIBUTION OF POVERTYINBRAZIL The regional poverty lines estimated inthis report and the regional distributionof poverty rates basedon householdconsumption analyzed inthe previous chapter are particularly relevant for investigatingthe coverage andpotential impact of the poverty alleviationpolicies inBrazil. Inthis chapter, some of these policy implications are presentedand discussed inmore detail. ACOMPARISONOFTHEPOVERTYRATESUSINGTHEMINIMUM LIVELIHOOD POVERTYLINE AND THE ADMINISTRATIVEPOVERTYLINEOFR$100 As indicated earlier, the MinimumLivelihood poverty line is remarkably close to the Administrative PovertyLine of R$100per person per month. One key difference is that the MinimumLivelihood poverty line varies from region to region to reflect differences in the cost-of-living of poor households. Table 22 compares the regional profile of poverty obtained with the minimumpoverty line that varies from region to region with the poverty profile obtained by applying the fixed nominal poverty line of R$100, which is equal to half the minimumwage. The differences in the poverty rates associated with these two poverty lines provide an estimate of the errors of targeting associatedwith the Bolsa Familia program. Table 22 The regional profile of poverty: The MinimumLivelihood Poverty Line vs. the R$100 Administrative Poverty Line. Administrative Poverty Line Min Regions of R$100 for allBrazil Livelihood line Difference (1) (2) (1142) 1 Metro Belem 14.7 16.7 -2.0 2 North Urban 26.3 27.4 -1.1 3 RWal 50.2 45.7 4.5 4 Metro Fortaleza 21.3 20.8 0.5 5 Metro Recife 16.5 17.7 -1.1 6 Northeast Metro Salvador 12.5 14.4 -1.9 7 Urban 36.8 36.8 0.0 8 Rural 59.6 55.0 4.6 9 Metro Rio DeJaneiro 8.0 8.8 -0.7 10 Metro Sa0Paul0 4.4 7.2 -2.8 11 Southeast Metro Belo Horizonte 6.0 7.1 -1.2 12 Urban 12.1 14.4 -2.4 13 Rural 29.2 28.2 1.o 14 Metro Curitiba 4.4 4.7 -0.3 Metro Port0Alegre 5.4 7.2 -1.7 16 l5south Urban 12.4 12.1 0.3 17 Rural 20.1 15.9 4.2 18 Brasilia 6.2 8.1 -1.9 GoianiaMunicipality 4.1 5.1 -1.1 20 l9Centerwest Urban 20.7 22.9 -2.2 21 Rural 30.9 30.9 0.0 National 21.0 21.5 -0.5 Metropolitan 9.9 11.4 -1.5 Urban excludingMetropolitan 20.7 17.5 3.2 All Urban including Metropolitan 16.2 21.9 -5.7 Rural 44.5 41.0 3.5 Source: World Bank estimates using the 2002-03 POF. 52 As it is apparent, the enforcement of the same absolute poverty line as a threshold for eligibility to the Bolsa Familia program tends to result in certain leakages (inclusion errors) in the rural and non- metropolitan urban areas and some undercoverage (exclusion error) in the metropolitan areas. Even though these estimates are not intended as an assessment of the targeting performance of the Bolsa Familia program, they do suggest that improvements in the targeting performance of the program, however good or bad it is, could be accomplished by employing poverty lines that vary from region to region. A COMPARISONOFTHEPOVERTY RATESOBTAINED USINGHOUSEHOLDINCOME The POF 2002-03 i s the first nationally representative survey to include extensive questions on both consumption and income measuresof welfare. Although for a variety of reasons, consumption tends to be a more accurate measure of welfare than income, Brazil's household surveys have traditionally collected data on income. As discussed indetail inchapter 1,past estimatesof poverty and inequality have differed significantly depending on which welfare measuresandpoverty lines were used. The detailed POF 2002-03 thus presents a significant opportunity to analyze poverty and inequality using both consumption and income measure. As table 7 of chapter 2 shows, in the POF survey the mean value of per capita income (PCINC) i s greater than the mean value of per capita expenditures (PCE) in each region. Thus, all else equal, one would expect that using the same poverty line, poverty rates would be higher usingPCEthanusingPCINC. Table 23 compares the Headcount poverty rates obtained usingthe per capita income aggregate available inthe POFsurvey (PCINC). The poverty lineusedis the minimumlivelihoodpoverty line, as this poverty line i s more in agreement with the current policies of the Brazilian administration. Contrary to the expectation, poverty rates are higher using PCINC in a number of regions. At the national level, the poverty rate based on POF PCINC i s 19.3% which i s slightly lower than the poverty rate of 21.5% based on PCE. However this masks large differences between regions. For example, using PCE of POF as the standard of comparison, it seems that usingthe POF income variable tends to overestimate poverty inthe non-metropolitan urban areas and underestimate it inthe rural areas. For further comparison, the poverty estimates using the per capita income measure from the 2004 PNAD survey are also presented. Unfortunately, aside from being able to match income by type or source, the PNAD and the POF are not really comparable for income measures due to very different reference periods (the PNADfor the last month and the POF for the last year). Inaddition, it i s important to keep in mind that the PNAD survey was collected one year later than the POF survey so differences inpoverty could be present for other reasons (e.g. economic growth) besides differences in the variable used to measurehousehold welfare. UsingPCE of POF as the standard of comparison, the PNAD seems to overestimate the poverty rate at the national level and in most regions. The overestimation of poverty i s the highest in metropolitan areas by 5.7 percentagepoints, then in the urban areas by 3.4 percentagepints and then finally inthe rural area by 1.4 percentagepoints. 53 Table 23 ComparingHeadcountPovertyRatesP(0): Consumptionvs. IncomeinPOFand IncomeinPNAD Regions POF POF PNAD PCE PCINC PCINC 1 Metro Belem 16.7 18.0 25.0 2 North Urban 27.4 30.8 27.1 3 Rural 45.7 42.6 42.8 4 Metro Fortaleza 20.8 21.4 30.5 5 Metro Recife 17.7 15.9 34.7 6 Northeast Metro Salvador 14.4 14.3 30.0 7 Urban 36.8 33.9 37.5 8 RUral 55.0 54.8 57.4 9 Metro Rio DeJaneiro 8.8 10.8 12.1 10 Metro Sa0 Paul0 7.2 5.4 14.1 11 Southeast Metro Belo Horizonte 7.1 4.1 14.9 12 Urban 14.4 11.0 13.8 13 RUral 28.2 19.5 29.4 14 Metro Curitiba 4.7 3.9 10.3 l5south Metro Port0Alegre 7.2 6.2 11.3 16 Urban 12.1 9.3 9.5 17 Rural 15.9 9.4 17.9 18 Brasilia 8.1 7.4 16.6 GoianiaMunicipality 5.1 7.2 n.a. 20 l9CenterWest Urban 22.9 19.2 14.8 21 RUral 30.9 18.9 31.4 National 21.5 193 23.3 Metropolitan 11.4 10.8 17.1 Urbanexcluding Metropolitan 17.5 19.2 20.9 All Urban includingMetropolitan 21.9 15.7 19.4 Rural 41.0 37.1 42.4 Source: WorldBankestimates usingthe 2002-03 POF. Notes: *Inthe PNAD survey it was not possibleto identify the Goianiamunicipality separatelyso it is classifiedwith urbanareas. The PNAD poverty estimates include households with zero reportedincome. The poverty estimates did not change significantly when householdswith zero incomewere excluded. THECOVERAGEOFTHEPOORBYSOCIAL PROGRAMS The regional poverty lines and the corresponding regionalpoverty rates basedon household consumption offer the rare opportunity of conducting a preliminary investigation of the extent to which some of the social protectionprograms of the Brazilian government (those that are included inthe POF) are successful at coveringthe households classifiedas poor. The POF 2002-03 includes several key social insurance (SI) programs in its questionnaire, including: (a) publicly-funded pension benefits, which correspond with the RGPS and MU pension regimes depending on which sector the worker was employed in;49(b) public leave benefits; (c) the salary bonus (abono salurial PISPASEP); and (d) unemployment in~urance.~'Together, these programs account for 100%of totalfederal spending on social insurance (Lindert, Skoufias, andShapiro, forthcoming). 49The POFalso includesinformationon public pensioncontributions,which we are usingto analyze "net" public pensionbenefit receipts. A more detaileddescriptionof these programs may be found inAnnex 2 of Lindert, Skoufias and Shapiro (forthcoming). It shouldbenotedthat the FGTSis not includedin"public socialinsurancetransfers." 54 The POF 2002-03 also included several important social assistance (SA) programs in its questionnaire, including: two of the main pre-Bolsa Familia conditional cash transfers (Auxilio Gas and Bolsa Escola), the child labor eradication program (PETI) and Renda Minima, which refers to sub-national programs offered in some localities. Together, these programs account for 22.7% of total federal spending on social assistance. Some notable federal social assistance programs that were not directly covered by the POF 2002-03 questionnaire include: the BPC-LOAS benefits for the elderly and disabled,51Bolsa Fumaia (which was introduced after the survey was conducted), and school feeding. Table 24 presents the percentage of the population classified as poor using the three main poverty lines that reports receiving benefits from social insurance and social assistanceprograms. For example, column (1) of table 23, reveals that more than half (56.14%) of the population classified inextreme poverty @e., 14.9 mil persons) reports receiving some type of social protection (SP) benefit. Social protection is defined here as receiving either SA or S I benefits or both. Column (2) of the table reports the coverage of the population with per capita consumption expendituresjust above the food poverty line and less than or equal to the Minimum Livelihood Needs Poverty line (i.e. FPLcPCE<=MLPL), whereas column (3) reports the coverage of the populations incolumns (1) and (2) combined. Table 24 The Coverage of the Poor by SocialPrograms FPL&CE<=MLP MLPL&CE<=U PCE<=FPL L PCE<=MLPL PL PCE<=WL (1) (2) (3) (4) (5) All socialprotection(SP) 56.14% 51.70% 53.46% 44.74% 48.29% All socialinsurance(SI) 26.68% 30.13% 28.77% 34.15% 31.96% All social assistance (SA) 37.70% 29.68% 32.85% 15.02% 22.28% Auxilio Gas (SA) 16.99% 13.08% 14.63% 6.50% 9.81% Bolsa Escola(SA) 30.20% 23.38% 26.08% 11.39% 17.37% PETI (SA) 2.17% 1.13% 1.54% 0.47% 0.91% Abono salarial PISPASEP (SI) 3.07% 4.53% 3.95% 8.25% 6.50% PublicLeaveBenefits (SI) 0.60% 1.12% 0.91% 1.57% 1.30% PublicPensionreceipts (SI) 22.94% 23.86% 23.50% 24.55% 24.12% Rendaminima(SA) 7.33% 5.82% 6.42% 3.24% 4.53% Seguro desemprego (SI) 1.39% 2.08% 130% 3.11% 2.58% HouseholdObservations 3690 6257 9947 13691 23638 Population 14,9 16,224 22,770,260 37,686,485 54,860,957 92,547,442 % of Totalpopulation 8.5% 13.0% 21.5% 31.3% 52.8% Source:World Bankestimatesusingthe 2002-03POF. Notes:PCE: Per Capita Expenditure FPL region-specific FoodPoverty LinepresentedinTable 10inChapter 2. MLPL: region-specific MinimumLivelihoodNeedsPoverty Line (CBN-Lower) presented inTable 12inChapter 2. UPL: region-specific Upper Poverty Line (CBN-Upper) presented inTable 12in Chapter 2. S: Social programsinclude the set of social insuranceand social assistanceprograms containedinthe POF survey. SI: Social insurance includes: Abono salarial PISPASEP, Public Leave Benefits, Public Pension receipts and Seguro desemprego. SA: Social assistanceincludes: Auxilio Gas,BolsaEscola, PETI, and RendaMinima. Figure 18 offers a more practical way of assessing the coverage rate of the poor in Brazil by social assistance and social insurance programs. Figure 18 simply graphs the coverage rates of the different 51Some respondents did indicate receiving BPC benefits in response to a question regarding receipt of any "other" benefits but the sample was deemedtoo smallfor our analysis. 55 degrees of poor individuals by the different social programs included inPOF. Specifically, the set of three columns for each type of programi s obtained from the numbers incolumns (1) (2) and (4) of table 24. As the first set of bars in the left side of figure 18 reveals, social protection (Le., the combination of social assistance and social insurance) programs covers a large fraction of the household in extreme poverty with the coverage of the less poor decreasing steadily. However, the aggregate statistics on the coverage rates of the poor by social protection appear to conceal large disparities between the coverage rates of the poor by social assistance and social insurance programs. The coverage rates of the poor by the social insurance system are lower among the extreme poor and higher among the less poor households. In contrast, the coverage rates of the poor by social assistanceprograms are higher among the extreme poor (37.7%) and lower among the less poor households (15.02%). These patterns confiim that social assistance programs, such as the Bolsa Escola and Auxilio Gas that are currently merged into Bolsa Familia, are muchbetter targeted towards the poor incomparison to the social insurance programs. Although the objective of social insurance programs i s more protection from poverty rather than redistribution of income to the poor, these findings suggest that it is also important to reconsider the level of fiscal resources allocated towards social insurance programs especially in light of the fact that they leave lessroominthe government budget for spending inbetter targeted social assistanceprograms. Figure 18 The Coverage of the Poor by Social Programs 60.00% , I Source: World Bank estimates usingthe 2002-03 POF. CONCLUDINGREMARKSAND NEXTSTEPS A comparison of the poverty estimates that result from these consumption-based poverty lines to the "Administrative Poverty Line" (the R$100and R$50formerly used by the Bolsa Familia and other programs) reveals that the enforcement of the same poverty line as was used as a threshold for eligibility to the Bolsa Familia program tends to result in some leakages (inclusion errors) in the rural and non-metropolitan urban areas and some undercoverage (exclusion error) of the poor inthe metropolitanareas. One more comparison worthy of serious consideration in the near future concerns the regional distribution of poverty based on the minimum livelihood poverty line and the regional 56 REFERENCES Banerjee, A. V., and E. Duflo. 2000. "Inequality and Growth: What can the Data Say?" NBER Working Paper7793. July. Barros, R. Paes de. 1996. Descrig6o da metodologia utilizada nu revis60 das estimativas de indigzncia de 1990. Rio de Janeiro: IPEA. Barros, Ricardo Paes de. 2004. "Pobreza Rural e Trabalho Agricola no Brasil ao Longo da Dkcada de Noventa." PEA. Barros, R. Paes de, L. Fox, and R. Mendonqa. 1993. "Poverty Among Female Headed Households in Brazil." Rio de Janeiro: IPEA, Texto para Discuss20 310, August. Barros, R. Paes de, R. Henriques, and R. Mendonqa. 2000. "A estabilidade inaceitAve1: desigualdade e pobreza no Brasil." In:ed. Henriques, R. Desigualdade e Pobreza no Brasil. Rio de Janeiro: IPEA,pp.21-48. Barros, R. Paes de and R. Mendonqa. 1992. Evolu@o do Bem-Estar e da Desigualdadeno Brasil desde 1960. Rio de Janeiro: IPEA,Texto para Discusdo 286. Barros, R. Paes de and R.Mendonqa. 1997. 0 impact0 do crescimentoeconBmico e de redup7es no grau de desigualdadesobre apobreza. IPEA,Texto para Discuss20 528. Barros, R. Paes de, R. Mendonqa and M.Neri. 1995. "AnEvaluationof the Measurement of Income and Expenditure inHousehold Surveys: POF vs. PNAD." Rio de Janeiro: PEA, mimeo. Barros, R. Paes de, R. Mendonqa and D. Santos. 1999. Incid2ncia e Natureza de Pobreza entre Idosos o Brasil. IPEA,Texto ParaDiscuss20 686, December. Bianchini, Z. M. and S. Albieri. 1998. "A Review of Major Household Sample Survey Designs Used in Brazil," Proceedings of the Joint IASS/IAOS Conference, Statistics for Economic and Social Development, September. Bidani, Benu and Martin Ravallion (1993) " Aregional Poverty Profile for Indonesia," Bulletin of IndonesianEconomic Studies,Vol. 29, no. 3 (December) pp. 37-68. Bianchini, Z. M. and S. Albieri. 2002. "Principais Aspectos de Amostragem das Pesquisas Domiciliares do IBGE-Revis20 2002." IBGE-Diretoria de Pesquisas. Working Paper No. 8. Bonelli, R. and L. Ramos. 1993. Distribuipio de renda no Brasil: avaliaG6o das tendgncias de longo prazo e mudangas nu desigualdade desde os meados dos anos 70. IPEA, Texto para Discuss80 288, janeiro. Bourguignon, F., F. H. G. Ferreira, and P. G. Leite. 2003. "Ex-ante Evaluation of Conditional Cash Transfer Programs: The Case of Bolsa Escola." Inequality and EconomicDevelopment in Brazil. Report No. 24487-BR. Washington: World Bank, October. distribution of federal funds for poverty alleviation. To the extent that the distribution of federal funds does not match regional distribution of poverty, a re-alignment in the regional distribution of federal funds may be called for. Usingthe POF per capita consumption expenditures as the standard of comparison, reveals that the income per capita measure from POF tends to overestimatepoverty in the non-metropolitan urban areas and underestimateit inthe rural areas. Lastly, an analysis of the coverage of the poor by the social program contained in the POF confirms that social assistance programs, such as the Bolsa Escola and AuxiZio Gas that are currently merged into Bolsa Familia, are much better targeted towards the poor in comparison to the social insurance programs. Although the objective of social insurance programs i s more protection from poverty rather than redistribution of income to the poor, these findings suggest that it i s also important to reconsider the level of fiscal resources allocated to financing the deficits of social insurance programs especially in light of the fact that less room is left in the government budget for spending on better targeted social assistanceprograms. The analysis inthis project also providedthe foundations for buildinga consumption-based poverty map. Poverty maps are especially useful for identifying the geographic variations in poverty within the twenty one different regions that are representedinPOF, and they can be used for the design and better targeting of programs, budget allocation, and for monitoring and evaluation. Micro-area poverty maps are constructed using econometric techniques52that combine data from the 2000 census with data from the 2002-03 POF. By combining census and household survey data, the poverty maps benefit from the strengths of eachdata source: the complete coverage of households inthe country with the census, and the more precise measures of household consumption and income from the POF. Statistical models are developed using "explanatory variables" in the household survey that are also included in the census. Once robust models have been identified to predict consumption (andor income) inthe household survey (using this common set of explanatory variables), these models are applied to census data at the household level to predict per capita consumption (or income) inthe census (including an error estimate). These household-level estimates are then aggregated to small statistical "micro areas" to obtain robust estimates of the percentageof households living below the poverty line inthese areas. For the work on poverty maps, the Bank adopted a "transfer of technology" and "capacity-building'' approach. Specifically, IBGE actually i s implementing the work ("learning-by-doing"), with the World Bank providing training and technical assistance (via formal seminars and workshops, on-going training and supervision, regular missions, continuous feedback via email, etc.). This collaboration has already resulted in a significant transfer of technology to build IBGE's capacity for carrying out such work, conducting further analysis, and implementing future updates. In fact, IBGE sees the "poverty map project" as a chance to integrate its own data instruments (census, surveys, GIs, etc.) and staff across the institution. As such, IBGEhas adopted the "poverty map project" as an innovative tool for integrating its own staff and management: the IBGE "poverty map team" comprised of some 10-15 people with different professional backgrounds (e.g., statisticians, economists, etc.) and coming from different departments in IBGE (e.g., those responsible for household surveys, the census, the GIs unit, etc.). It i s hoped that the completion and publication of official IBGE poverty maps obtained with these methods will be completed inthe near future. '*These methods were pioneeredby researchersat the World Bankin 1996(Hentschel and Lanjouw, 1996).The techniques have been further refined, mostly under the leadership of researchers at the World Bank in collaboration with universities and in- country partner institutions (e.g., Hentschel et. al. 1998, Hetschel et. al. 2000 and Elbers, Lanjouw and Lanjouw, 2002). These maps have been applied to numerous countries around the world. Henninger and Snel (2002) summarizes experiences with the developmentanduse of poverty mapsin several countries. 57 Camargo, J. M. and F. H. G. Ferreira. 1999. "A Poverty Reduction Strategy of the Government of Brazil: A Rapid Appraisal." Department of Economics, Catholic University of Rio de Janeiro, mimeo. Camargo, J. M. and F. H. G. Ferreira. 2001. "0 Beneficio Social Unico: uma proposta de reforma da politca social no Brasil." Departamento de Economia, Pontificia Universidade Catblica, Rio de Janeiro, Discussion PaperNo. 443. Centro de Politicas Sociais (2005) "Miseria em Queda" www.fgv.br/cps Chomitz, Kenneth M. and Edinaldo Tebaldi. September 2004. "Geography and Development in Northeast Brazil: Equity, Efficiency and Environment." Draft mimeo. CEPAL. 2000. PanoramaSocial de AmCrica Latina 1999-2000. Santiago. CEPAL. 2002. Meeting the Millennium Poverty Reduction Targets in Latin America and the Caribbean. Santiago: December. CEPAL. 2002. PanoramaSocial de AmCrica Latina2001-2002. Santiago. CEPAL. 2003. PanoramaSocial de AmCrica Latina 2002-2003. Santiago. Coady, D. 2003. Alleviating Structural Poverty in Developing Countries: TheApproach of Progresa in Mexico. Background paper for the World Development Report 2004. Washington: The World Bank. Deaton, A. 1997. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Johns Hopkins University Press. Deaton A., and S. Zaidi, 2002, Guidelines for Constructing Consumption Aggregates for Welfare Analysis, no. 135 in Living Standards Measurement Study Working Paper, The World Bank, Washington, DC Elbers, C., J. 0.Lanjouw, P. Lanjouw, and P. G. Leite. 2003. "Poverty and Inequality in Brazil: New Estimates from Combined PPV-PNADData." Inequality and Economic Development in Brazil. Report No. 24487-BR. Washington: World Bank, October. Elbers, C., J. 0. Lanjouw, P. Lanjouw. 2002. "Micro-Level Estimation of Poverty and Inequality." Econometrica, v.71, pp. 355-364. Ellwanger, R. 1991. Consumo alimentar por classe de renda nas regi6es metropolitanas, em Brasilia e Goidnia. Rio de Janeiro: IBGE. mimeo. Ellwanger R. 1992. Participapio na subcomisscio tdcnica sobre linhas de pobreza. Projeto politica nacional de salArios. Rio de Janeiro: IBGE, mimeo. FAOIOMS. 1985. Energy and Protein Requirements. Geneva. 59 Ferrer-i-Carbonell, A; Van Praag, B.M.S. (2001) "Poverty in the Russian Federation" Discussion Paper series IZA DPNo. 25 Ferez, J. 1996. Una estimacidn de las necesidades de energia y proteinas de la poblacidn. Santiago: CEPAL. Ferreira, F. and P. Lanjouw. 2001. "Rural Nonfarm Activities and Poverty in the Brazilian Northeast." World Development v.29, no.3, pp.509-528. Ferreira, F. H. G., P. Lanjouw, and M. Neri. 1998. The urban poor in Brazil in 1996: a new poverty profile using PPV, PNAD, and CensusData. World Bank, mimeo. Ferreira, F. H. G., P. Lanjouw, and M. Neri. 2000. "A New Poverty Profile for Brazil Using PPV, PNAD, and Census Data." Departamento de Economia PUC-Rio, TD #418, March. Ferreira, F., P. Lanjouw, and M. Neri. 2003. "A Robust Poverty Profile for Brazil UsingMultiple Data Sources." Review of Brazilian Economics, v.57, no.1, pp.59-92. Ferreira, F. H. G. and J. Litchfield. 1996. "Growing Apart: Inequality and Poverty Trends in Brazil in the 1980s." LSE-STICERD - DAW Discussion Paper No. 23, London, August. Ferreira, F. H. G. and J. Litchfield. 2000. "Desigualdade, pobreza e bem-estar social no Brasil - 1981195." In: ed. Henriques, R. Desigualdade e Pobreza no Brasil. Rio de Janeiro: PEA, pp.49-80. Ferreira, F. H. G. and R. P. de Barros. 1999. "The Slippery Slope: Explaining Increases in Extreme Poverty in UrbanBrazil, 1976-1996." Brazilian Review of Econometrics, v.19, no.2. Fiess, N. and D. Verner. 2004. "The Dynamics of Poverty and its Determinants: The case of the Northeast of Brazil and its States." Washington: The World Bank, World Bank Policy Research Working PaperNo. 3259, April. Fiszbein, A. and G. Psacharopoulos. 1995. "Income Inequality Trends in Latin America in the 1980s." In:ed. Lustig, N. Coping withAusterity: Poverty and Inequality in Latin America. Washington: The Brookings Institution, pp.71-100. Forbes, Kristin J. 2000. "A Reassessment of the Relationship between Inequality and Growth." American Economic Review, September, v.90, no.4, pp.869-887. Foster, J. Greer, J.; Thorbecke E., (1984) "A class of decomposablepoverty measures," Econometrica, no. 52, 1984. Goedhart, Th., Halberstadt, V., Kapteyn, A., and van Praag, B.M.S. (1977) "The poverty line: concept and measurement" Journal of HumanResources, 12: 503-520. Guerreiro Osorio, Rafael. November 2003. "0 Sistema Classificat6rio de `Cor ou Raqa'do IBGE." PEA. Texto ParaDiscuss20No. 996. 60 Hagenaars, A. J. M. and van Praag, B.M. S. 1985. "A Synthesis of PovertyLine Definitions." Review of IncomeandWealth, v.31, no.2, pp.139-54. Henninger, Norbert and Mathilde Snel. 2002. "Where are the Poor: Experiences with the Development andUse of PovertyMaps." Washington: World ResourcesInstitute. Henriques, R. ed. 2000. Desigualdade e Pobrezano Brasil. Riode Janeiro: PEA. Hentschel, Jesko and Peter Lanjouw. 1996. "Poverty Profile." pp. 53-91 in Ecuador Poverty Report. Washington, DC: The World Bank. Hentschel, J., J. Lanjouw, P. Lanjouw, and J. Poggi. 1998. Combining Census and Survey Data to Study Spatial Dimensionsof Poverty. Policy ResearchWorkingPaperNo. 1928. Washington, DC: The World Bank. Hentschel, J., J. Lanjouw, P. Lanjouw, and J. Poggi. 2000. "Combining Census and Survey Data to Trace the Spatial Dimensions of Poverty:A Case Study of Ecuador." The WorldBank Economic Review 14(1): 147-65. Hoffmann, R. 2000. "Mensura@o da desigualdade e da pobreza no Brasil." In: ed. Henriques, R. Desigualdade e Pobrezano Brasil. Rio de Janeiro: PEA, pp.81-108. Jovanovic, B. and Milanovic, B. 1999. "Change in the Perception of the Poverty Line during Times of Depression: Russia 1993-96." Washington: The World Bank, paper 2077, Country Economic Department. Kakwani, Nanak (2003) "Issues in Setting Absolute Poverty Lines," Poverty and Social Development Papers, No. 3 (June), Asian DevelopmentBank. Lanjouw, Peter F. "Constructing a ConsumptionAggregate for the N o s e of Welfare Analysis: Issues andRecommendationsConcerningthe POF2002/3 inBrazil." The World Bank, 2005 (mimeo). Lavinas, L. and R. Varsano. 1997. Programs de garantia de renda minima e apio coordenuda de combate apobreza. Rio de Janeiro: IPEADPES, TextoparaDiscusslo 534, setembro. Lindert K., E. Skoufias and J. Shapiro. forthcoming. "Redistributing Income to the Poor and the Rich: Public Transfers in Latin America and the Caribbean," Report prepared for the Office of the Chief Economistof the Latin American and Caribbean Region of the World Bank. Washington D.C: The World Bank. Litchfield, J.A (1999) "Inequality: MethodsandTools". Lokshin, M. and M. Ravallion. "Identifying Welfare Effects from Subjective Questions." Economica, August, v.68, pp.335-357. Lokshin, M., L. Umapathi, and S. Paternostro. 2004. "Robustness of SubjectiveWelfare Analysis in a Poor Developing Country: Madagascar 2001." World Bank Policy Research Working Paper 3191, January. Washington D.C: The World Bank, 61 Lustosa, T. and M. Landen. 1999. Cdlculo das necessidades energkticas da popula$o brasileira visandoa constru@o da linha depobreza. Comiss50de Estudos sobre PobrezaIPEAD3GE. Rio de Janeiro: IBGE/DPE/DEPIS, June. Maletta, H. 1998. Rural poverty in Brazil. Rome: FAO. Medeiros, Marcelo. March 2004. "As Fontes de Rendimentosdos Ricos No Brasil." IPEA: Texto Para DiscussgoNo. 1014. Medeiros, Marcelo. July 2004. "A Geografia dos Ricos No Brasil." IPEA: Texto para Discussgo No. 1029. Neri, M. 1999. "Uma Fotografia recentede pobrezabrasileira,PNAD data 1981-98." IBGE. Neri, M. 2000. As mudanGus da pobreza e da desigualdade cariocas nu dkcada de 90. Rio de Janeiro: IPEA, Texto paraDiscussgo709, February. Pessoa, D.G.C; and Silva, P.L.N. Antilise de Dados Amostrais Complexos. 13" Sinape, Associa@o Brasileira de Estatistica, 1998. Pradhan, M. and M. Ravallion. 2002. "Measuring Poverty Using Qualitative Perceptions of ConsumptionAdequacy." Review of Economics and Statistics, August, v.82, no.3, pp.462-471. Ravallion, Martin(1994) "Poverty Comparisons" Harwood Academic Publishers Ravallion, Martin (1998) "Poverty Lines in Theory and in Practice," LSMS Working Paper No. 133. Washington D.C.: The World Bank. Ravallion Martin and Benu Bidani (1994) "How Robust is a Poverty Profile? World Bank Economic Review, vol. 8, pg. 75-102. Rawlings, L. and G. M. Rubio. 2003. Evaluating the Impact of Conditional Cash Transfer Programs: Lessonsfrom Latin America. WashingtonD.C.: The World Bank. Rocha, Sonia. 1993. "Linhas de Pobrezapara as RegidesMetropolitanasna PrimeraMetade da DCcada de 80." Belo Horizonte: ANPEC v.N. Rocha, S. 1996. Poverty Studies in Brazil - A Review. Rio de Janeiro: IPEA, Texto para Discussgo398, January. Rocha, S. 1996. Renda e pobreza: os impactos do plano real. Rio de Janeiro: IPEA, Texto para Discussgo439, December. Rocha, S. 1996. "Poverty under inflation." In:Oyen, E. Poverty -a global review. Oslo: Scandinavian University Press. Rocha, S. 1997. "Crise, estabilizaggoe pobreza." Conjuntura EconBmica,janeiro, pp.22-26. 62 Rocha, S. 1997. "Poverty inBrazil inthe Eighties:A Review." Santiago: Seminar on Poverty Statistics, 7-9May. Rocha, S. 1997. "Do Consumo Observado h Linha de Pobreza." Pesquisa e Planejamento EconGmico, August, v.27,no.2, pp.313-352. Rocha, S. 1998. Renda e pobreza - medidas per capita versus adulto-equivalente. Rio de Janeiro: IPEA,TextoparaDiscussiio609. Rocha, S. 1998. "Pobreza no Brasil - PrincipaisTendencias da Espacializagiio." Anais XXVZ Econtro Nacional de Economia, December, pp.1665-1682. Rocha, S. 2000. "Estimagiio de linhas de indigenciae de pobreza: opgdes metodoldgicas no Brasil." In: ed. Henriques,R. Desigualdade e Pobrezano Brasil. Rio de Janeiro: PEA, pp.109-127. Rocha, S. 2000. Pobreza e Desigualdade no Brasil: 0 Esgolamentodos Efeitos Distributivos do Plano Real. Rio de Janeiro: PEA, TextoparaDiscussiio 721. Rocha, S. 2001. "The Use of Poverty Lines in Brazil." Rio de Janeiro: FourthMeetingof the Expert Group onPovertyStatistics, 15-17October. Sabdia, J. and S. Rocha. 1998. Programas de Renda Minima: Linhas Gerais de uma Metadologia de Avaliagfio a Partir da Experigncia Pioneira do Paranoa`. Rio de Janeiro: IPEA, Texto para Discussiio 582. Sant'Ana, S. R. and A. Moraes. 1997. "Avaliagiio do Programa Bolsa Escola do GDF." Brasilia: Fundagiio GrupoEsquelBrasil. Sedlacek, G., N. Ilahi, and E. Gustafsson-Wright. 2000. Targeted Conditional Transfer Programs in Latin America: an Early Survey. Washington: The World Bank. Skoufias, Emmmauel; M.V. Fazio. "A Revised Consumption Aggregate for Brazil Based In POF 2002- 2003." The WorldBank, 2005 (mimeo). Subramanian, S. and A. Deaton. (1996). "The Demand for Food and Calories." Journal of Political Economy vol. 104,pp. 133-162. Thomas, M. R. 2002. "Social Welfare and Poverty in Ceara, Brazil: Measurement and Progress, 1993- 1999." Washington: The World Bank, DiscussionNote, February. Thomas, M. R. 2004. "Poverty Update: Brazil, Its Regions, and Its States." Washington: The World Bank, World BankInternalDiscussionNote, May. United Nations Development Programme: International Poverty Center (September 2004). "Dollar a Day, How MuchDoesit Say?" In Focus. Urani, A. 1996. Zmpactosfiscais e distributivos das propostas de renda minima em debate no Brasil. Santiago: CEPAL. 63 Van Praag, B.M.S., and Frijters, P. (1999) "The measurementof welfare and well-being; The Leyden approach" In: Kahneman, D., Diener, E., and Schwarz (eds.). Well-Being: The Foundations of HedonicPsychology. RussellSage Foundation, New York. WorldBank. 1995. Brazil - A Poverty Assessment. ReportNo. 14323-BR. Washington. World Bank. 1999. Consultationswith the Poor: Brazil - National Synthesis Report. Poverty Reduction and EconomicManagementNetwork. Washington: September. WorldBank. 2001.An Assessment of the Bolsa Escola Programs. ReportNo. 20208-BR. Washington. WorldBank. 2001.Attacking Brazil's Poverty. ReportNo. 20475-BR. Washington: 31March. World Bank. 2001. Rural Poverty Alleviation in Brazil. Report No. 21790-BR. Washington: 27 December. World Bank. 2003. Country Assistance Strategy 2003-2007for the Federative Republic of Brazil. Washington. World Bank. 2003. Inequality and Economic Development. Report No. 24487-BR. Washington: October. World Bank. 2003. Inequality in Latin America and the Caribbean: Breaking with History? Washington. WorldBank. 2004. Brazil:Equitable, Competitive,Sustainable-Contributionsfor Debate. Washington. 64 APPENDIX-1 APPENDIX 1:CONSTRUCTINGA CONSUMPTIONAGGREGATEFORTHE PURPOSEOF WELFAREANALYSISUSINGPOF2002-03 The initial step in computing poverty and inequality is to choose an appropriate measure of household welfare. There are both conceptual and pragmatic reasons why consumption expenditures available from household surveys might be preferred for the purpose of poverty and inequality analysis to an indicator such as household income. It is argued, for example, that consumption expenditures reflect not only what a household i s able to command based on its current income, but also whether that household can access credit markets or household savings at times when current incomes are low or even negative (due perhaps to seasonal variation or a harvest failure). In this way, consumption is thought to provide a better picture of a household's longer run standard of living than a measure of current income. Further, consumption expenditures for the poor are often better captured than household incomes. While poor households are probably purchasing and consuming only a relatively narrow range of goods and services, their total income may derive from multiple different activities with strong seasonal variation and with associated costs that are not always easily assigned. The POF has collected expenditures in a very comprehensive list of items. The derivation of a consumption aggregate for the purpose of welfare analysis may not include all of these items. As discussed in detail by Deaton and Zaidi (2002), the reason for the exclusion of expenditures on some categories of items i s that some of these expenditures are better considered as investments, or inputs for production, rather than as welfare-enhancing expenditures. Moreover, for correct welfare comparability across households/individuals, it is important not to include in the consumption aggregate the acquisition value of lumpy expenditures on durable items, but rather a measure of the value of the services that these goods provide to the families. The inclusion of such expenditures on durables could drive a large wedge in recorded consumption between those who purchaseda durable good, and those who own it but purchased it in a preceding period. This would be inappropriate from the perspective of comparing welfare across such households because in both cases the services of the durable are being consumed. In sum, the process of constructing a consumption aggregatefor welfare analysis i s guided by a number of considerations. We went over these considerations presentedinDeaton and Zaidi (2002), and scrutinized each specific item in POF in collaboration with IBGE, in order to construct a consumption aggregate that could capture the economic dimensions of well-being in Brazil. In deriving the "preferred" consumption measure from the POF data, there are multiple objectives of comprehensiveness, transparency and credibility that must be retained as central focus. This appendix summarizes the steps and considerations in the procedure for constructing the consumption aggregate used in the process of setting the regional poverty lines and the poverty map for Brazil. Selection of items from POFfor constructingthe consumptionaggregate The POF collected information on household acquisitions of goods (purchased for own use or for other households, received as gift, and self-produced) in the previous periods of 7, 30 and 90 days, and 12 months. The 7 days recall includes acquisitions of food, both inside and outside the home, and transport expenses. The 30 days recall was applied to a range of nonfood consumption goods, such as pharmaceutical products, and also leisure and entertainment. The 90 days reference period include clothing and a variety of services, among others. The expenses made throughout the 12 month period comprise the acquisition of durable goods (like houses, cars andelectronic appliances). Inorder to aggregatethe consumption measure, we addedthe items purchasedandacquired (from gifts or self production) that could reflect the standard of living of the household in a longer run than the sum of current expenditures. The exercised consisted of appending 10 groups of items: "Food consumption (including consumption inside and outside the home)"; "Housing"; Health"; "SchoolingEducation"; 65 APPENDIX-1 "Transport"; "Clothes"; "CultureLeisure"; "Personal Services"; "Hygiene and personal care"; and "Others". Among these groups, there are items which deserve a special consideration. In general, the procedure for scrutinizing the items to include in the consumption aggregate followed the guidelines of not including: a) "lumpy" items purchased sporadically; b) items that serve as inputs into production, or investments; c) items with low elasticity with respect to total expenditure; d) items acquired for other households.The next four bullets summarize the criterion for not including these items in a consumption aggregate for the purpose of welfare analysis. a) The "lumpy" and infrequent acquisitions Consumer durable purchases are typically large expenditures that occur very infrequently. A classic example is the purchaseof a car or motorcycle. A particularhousehold is likely to purchasea car only once every number of years. With a 12 month recall period, there will be a certain subset of households in the data who do indeed report purchasing a car. They will report spending a considerable sum of money for this item. Other households in the dataset will, in fact, own a car but will have purchased it in some preceding period, and will thus report zero expenditure in a car. Attributing a consumption value of zero to households that own but did not purchase a car in the specific recall period, will understate their welfare because they will in fact be consuming the services of a car. Attributingthe purchase value of the car to those households in the data that happenedto purchasea car duringthe referenceperiod will overstate their welfare because they will not be consuming all of the services provided by a car in this one-year reference period. The car's services will be consumed over a period of several years. The attributes of a consumer durable imply that it is unappealing to simply add expenditures over the reference period directly to the consumption aggregate.Where possible a flow of consumption from consumer durables can be addedto the consumption aggregate, imputed from the available information on ownership, age and replacement value of consumer durables. Deaton and Zaidi (2002) provide a good discussion of the available methods. In POF, although there i s a section on the inventory of durable stocks for householdsowning goods ina longer period span, the questionnaire does not include information on value (either original purchase value or current replacement value), so it is not possible to calculate the flow of services from the durables. b) Items that serve as inputs into production, or investments One key concern throughout the process was to not include expenditures in inputs for production, or investments, as consumption. If one includes expenditures on inputs into household production, and the income from household production is in turn devoted (at least in part) to consumption expenditure, then double counting occurs, and the consumption aggregate is overstating the actual welfare levels achieved by the household. In most circumstances, the distinction between productive inputs and consumption is rather obvious. For example, it i s clear that fertilizer expenditures should not be reflected in the consumption aggregatefor farming households c) Items with low elasticity with respect to total expenditure In some cases, it is difficult to determine the effect on welfare of the expenditure in items like health products and services with their effect on welfare. The analysis of whether to include health expenditures deserves an assessment of the elasticity of health expenses with respectto total expenditure. For instance, it i s complex to measure the extent to which health expenditures could increase welfare, since it i s not possible to measure the loss of welfare from illness and the increase in welfare from its alleviation. Including only the expenditure is incorrect, though excluding health expenditures altogether means that one may miss the difference between two people, both of whom are sick, but only one of which pays for treatment. Moreover, there are other considerations related to whether the health expenditures may also be discretionary and welfare enhancing, but it i s difficult to discriminate "necessary" from "unnecessary" expenditures. Therefore, Deaton and Zaidi (2002) recommend analyzing the elasticity of the expenditure in 66 APPENDIX-1 health items with respect to total expenditure. The higher the elasticity, the stronger the case for inclusion. We analyze the elasticity of health and education expenditures in POF when explaining the components of the consumption aggregate. d) Items acquiredfor other households Goods acquired for g f t s to other households should be excluded from the consumption aggregate, since its inclusion would involve double-counting if, as one would expect, the transfers show up in the consumption of other households.Therefore, it i s recommended to include only the goods acquired as a gift from others, which increasethe well-being of that household, but not the expenses made in that household for increasing consumption of other households. Foodconsumption The food component of the consumption aggregate comprises the value of expenditures and acquisitions of food items for consumption both inside and outside the home. Aggregating across all items, over the whole week, yields a measure of household weekly food acquisition. Multiplying this by the number of weeks in a month or in a year yields a measure of monthly or annual food "expenditures". While it may not be strictly the case that all food acquired in a given week i s consumedthat week, the general assumption i s that at the monthly or annual level, total food expenditures indicate the value of total food consumed by the household. This procedure providedthe first component of the overall householdconsumption aggregate. There were 1636 households (3.4% of the total) with no reports on food consumption. A possible explanation to these missing reports could be that the 7 days period may be short to capture the food consumption of families that might not have purchasedany food items duringthe week in which the survey was carried out, since it i s expected that many families make their food purchases in a monthly or quarterly basis. To the extent that this problem occurs only with respect to food consumption, one might hope that for those households with significant nonfood expenditures, their overall rankingin the welfare distribution may not be affected too badly by this problem. As a result this issue may be of less concern when trying to identify the rich (inan analysis of inequality, for example). However, amongst those with low incomes, for whom food expenditures are typically particularly important, the presence of noise in the food consumption data i s likely to lead to an over-estimate of overall poverty and to make less sharp the distinction between the "poor" and the "non-poor" interms of household and individual characteristics. Unfortunately, there is no way of knowing whether these households in fact did not spend anything in food due to difficulties, or whether they happened to have zero expenditures simply because of the short recall period in the survey for food expenditures (7 days). Alternatively, we checked on the sensitivity of the poverty and inequality rates to the imputing of the food expenditures for these missing reports in the calculation of the consumption aggregate. We predicted the food expenditure of the households with missing reports based on a model for food expenditure as a function of a set of households' housing and demographic characteristics and area of residence. The parameters estimated by the model allowed the imputation of food expenditures equal to zero.53The food expenditure was imputed for all 1636 households, except for 97 households with per capita income below the political indigence line of R$50, who were expected to have no reports on food consumption because of difficulties, rather than because of the short recall period. We tested the sensitivity of poverty and inequality measures to this imputation in the final consumption aggregate, and the measures showed almost no sensitivity to the imputationinfood expenditure. 53We also triedthis procedurewith Propensity Score Matchingand foundvery similar results for the imputation. 67 APPENDIX- 1 Housing This component of the consumption aggregate comprises rents; basic services; small scale home renovations; furniture and household items; appliances and appliance repairs; and cleaning materials. For those households that are renting their home, the rent payments were included as a measure of the consumption of services that derive from housing. Households owning their dwellings do not pay rent, but are clearly consuming housing services, therefore we included the implicit rent from own housing reported by household owners in POF. As respondentsare likely to be well informed about the value of their home and the kind of rent they would have to pay for a home with similar quality and location attributes, this estimated responsei s generally found to be quite satisfactory. The expenditures on basic services (water, sewage, etc.) were included in the consumption aggregate. They represent a large share of total expenditure for some households. Deaton and Zaidi (2002) generally recommend against the incorporation of expenditures on publicly provided services in the consumption aggregate. This i s because finding the proper set of prices with which to value these goods is difficult. Including expenditures on networked water and sanitation, for example, while not being able to properly take account of the fact that some households are not connected to a water network at all, that some households do not receive bills although they are connected, and that some households receive only sporadic supply of water and supplement their publicly provided water with purchases from private vendors, could introduce important biases in rankings of households. If there i s any reason to think that expenditures on networked water, electricity and gas is only weakly linked to the welfare that i s associated with the actual consumption of those services the general recommendation would be to exclude these expenditures from the consumption aggregate. Other services, such as internet access, telephone expenditures andtelevision subscriptions, are more straightforwardly addedto the consumption aggregate. The expenditures with home renovations in the consumption aggregate include the more frequent expenditures on housing maintenance, such as: upkeep, gardening and home repairs; which were collected as expenditures within the 90 days reference period. The POF survey also collected expenditures on renovations over the 12 months reference period. Inthis case, the renovations are less frequent and lumpy since include reconstruction and reforms. Therefore, these last expenditures were not included as part of the consumption aggregate.As commented in point a) above, this type of occasional and high expenditures can introduce a wedge between the welfare levels of households which incurred in this type of expenditure in the reference period and the households who spent on them in a previous period. The same consideration was taken for deciding on the inclusion of durable items as furniture, appliances (fridge, televisions) and repair of these appliances. We scrutinized each item in order to decide if the purchase of durable goods was to be considered occasional and lumpy expenditure. We included only the items which were to be considered more frequent and less lumpy. Healthand Education Ifone were to include expenditure on healththen one should alsotake into consideration the implicit loss of welfare due to illness (something that i s very difficult to do). However, some items related to prevention and care can be consideredas more discretionary and welfare enhancing (and thus reasonably included in the consumption aggregate). The decision to include or exclude these expenditures, according to Deaton and Zaidi (2002), must be based on the analysis of the income elasticity of the health expenditures. These authors have shown that in developing countries, this elasticity is relatively low (varying between 0.74 and 0.86), which does not justify including health related expenditures inthe consumption aggregate. We also computed the elasticities for education, for which there are also concerns about its inclusion (i.e. the inclusioncan introduce a wedge in welfare level betweenhouseholdswithout children going to school 68 APPENDIX-1 and households with children in school age). As explained inpoint c), the higher the elasticity, the stronger the case for inclusion inthe consumption aggregate. The results presentedin tables 1, 2 and 3 compare the elasticity of health and education with respect to the total expenditures and to family income for Brazil using the POF. As table 1 indicates, the elasticity of education expenditures is larger than the elasticity of healthexpenditures, whichjustifies the inclusion of all of the education expenditures. The elasticity of health i s 0.97, which i s lower than the elasticity of education expenditures, but greaterthan the elasticity found inthe countries analyzed by Deaton and Zaidi (2002). The elasticity of the health and education expenditures was estimated by income deciles. It i s observed that the elasticities are always higher for education expenditures than they are for health expenditures. In the case of health, the elasticity is higher for deciles four and six. For the bottom deciles, this elasticity i s lower. Based on the results for the elasticities, which were not very low for the health expenditures and were greater than one for the education expenditures, the followingprocedures were adopted: We included expenditures in health and dental insurance plans, because they provide insurance, which can be related to a higher level of welfare and there i s no indication of decrease in welfare from illness in insurance plans. Moreover, these expenses represent an important part of the expenditures incurred by the families. Other type of health expenditures, such as the purchase of pharmaceutical products and analysis were excluded, since in this case it is not possible to capture the welfare loss from the diseases they are supposedto alleviate. The expenditures ineducation were included, since the expenditures in private school fees can be directly related to a higherlevel of welfare of householdspaying for educational services. Although education can also be considered an investment instead of consumption, the inclusionof education expenditures inthe consumption aggregate i s unlikely to lead to double counting as the returns from this particular investment will probably not be reflected incurrent consumption levels. Current practice typically treats education as a consumption item, but it is obviously a matter ofjudgment. Table 1.1Elasticitiesofhealthandeducationexpenditures Variable Elasticity Standard deviation t P Value Health Income 0,81 0,0136 59,64 <.0001 Health Expenditure ** 0,97 0,0100 69,80 <.0001 Education Income 1,13 0,0200 54,88 <.0001 Education Expenditure ** 1,30 0,0200 62,59 <.0001 Source: 2002-03 POF. Note: The sample design of the survey was considered for the calculation. Table 1.2 Elasticitiesof healthexpendituresby decilesof incomedistribution Income decile Elasticity Standard deviation t PValue Observations 1 0,037 0,061 0,60 0,548 2762 2 0,567 0,222 2,56 0,011 3116 3 0,550 0,276 2900 0,046 3421 4 1,589 0,324 4,91 0,OOO 3655 5 0,572 0,316 1,81 0,071 3782 6 1,214 0,289 4,20 0,OOO 3968 7 0,953 0,248 3,85 0,000 4122 8 0,921 0,190 4,86 0,000 4338 9 0,964 0,123 7,86 0,oOO 4502 10 0,655 0,033 19,64 0,OOO 4633 Source: 2002-03 POF. Note: The sample designof the survey was consideredfor the calculation. 69 APPENDIX-1 Table 1.3 Elasticities of education expenditures by deciles of income distribution Income decile Elasticity Standarddeviation t PValue Observations 1" 0,027 0,076 0,36 0,718 2067 2" 0,830 0,278 2,98 0,003 2190 3" 0,730 0,357 2,05 0,041 2397 4" 1,018 0,454 2,24 0,025 2491 5" 1,053 0,424 2,49 0,013 2755 6" 0,907 0,395 2,29 0,022 2852 7" 1,688 0,355 4,75 0,000 3070 8" 1,567 0,289 5,43 0,000 3283 9" 1,382 0,190 7,29 0,oOO 3619 10" 0,835 0,053 15,79 0,000 3954 Source: 2002-03POF. Note: The sample designof the survey was consideredfor the calculation Transport services Expenses in transport services were included as part of the consumption aggregate. Although these expenditures are to be considered "regrettable necessities" for getting to the work place, in this case it was not possible to distinguish them from transportation expenses for other purposes. Clothing, Culture and leisure, Personal Services and Personal Hygiene and Care These components of the consumption aggregate comprise all types of expenditures in clothing, leisure (tickets to cinema, etc), personal services (haircuts, beauty, etc) and personal care; which were considered to increase welfare of the households without introducing biases in the comparability of households' welfare levels. Notwithstanding the fact that expenditures in clothing and shoes can be considered infrequent purchases, the value of these purchasesi s rather modest, so they were included inthe aggregate. Considerations for Other expenses The remaining components of the consumption aggregate comprise professional services (such as notaries, lawyers); expenditures in ceremonies, celebrations and anniversaries (that are collected for the 12 month reference period); and expenses related to taxes, contributions, bankingfees, among others. The procedure followed was to include all items except for occasional expenditures (such as occasional ceremonies). As with consumer durables these are often infrequent expenditures that can become very costly and ideally we would like to have some smoothed value rather than actual, total expenditure on the event. Following Deaton and Zaidi (2002), we excluded these items from the consumption aggregate. The sole exception was made with respect to birthday parties and wedding anniversaries - events that occur on an annual basis. For such items the 12 month reference period i s the appropriate one and one could thus justify including these items inthe consumption aggregate. Regarding taxes and contributions, following Deaton and Zaidi (2002) expenditures on levies are not part of consumption, but a deduction from income, and should not be included in the consumption aggregate. Therefore the consumption aggregatedoes not include this type of payments. The authors suggest including property taxes when there is evidence that they could be linkedto the provision of a specific service to the households. Inthis case there were no grounds to relate the property taxes (IPTUandITR)54to a better level of well-being. Incontrast, we did include payments that could be linkedto service provision, like insurance payments. Inparticular, we did not include taxes related to the acquisition of goods already excluded (e.g. 54IPTU: Impostosobre aPropriedadePrediale Territorial Urbana. ITR: IrnpostoTerritorial Rural. 70 APPENDIX-1 purchases of cars). Expenses related to financial transactions, regarding the paying off of debts were not includedas part of the aggregate. Considering the suggestions in Deaton and Zaidi (2002) for gifts and transfers, these expenditures were excluded from the aggregate. Includingthem would involve double counting if the transfers show up in the consumption of other households. Large expenditures that may be considered investments, such as the purchase of real estate, gold bars, etc. were excluded from the consumption aggregate.They can also introducebias inthe comparison with householdsalready owning these assets. Source: World Bank estimates using the 2002-03 POF. 71 APPENDIX-1 72 APPENDIX-2 APPENDIX 2: TESTINGTHE SENSITIVITYOFTHE FOODPOVERTYLINESWITH THE CBN METHOD Inthis appendix we summarizethe various tests conductedfor the sensitivity ofthe regionalfood poverty line to changes in a) the reference population (for quantities and prices) for estimating the food poverty line. we tried definitions different from the 20 to 40 percentiles of the per capita consumption distribution. b) In the estimation of the quantities and prices. apart from the mean quantities and prices. we also testedthe same procedure takingthe median. c) Changes in the composition of the poverty basket to allow for more variety within each component of the poverty basket. Table 2.1 summarizesall the procedures followed in estimating the regional food poverty lines. Each raw of this table corresponds to alternative reference population, while each column corresponds to different methodsused for estimating the quantities and unit values (prices) of the food items in the basic needs food basket (means vs. medians etc.) Table 2.1: Summary of all the tests usedto investigate the sensitivity of the CBNfood poverty line 20 to 40 percentiles Mean Median Mean Median 10to 50 percentiles Mean Median Mean Median I I BPLl=Half Minimum Wage=R$100adjustedby prior 0'7