Peru - Poverty Comparisions

DRAFT CONFIDENTIAL Report No.: 18459 PE Peru: Poverty Comparisons October 5, 1998 Country Department 6 Latin America and the Caribbean Region The World Bank FILE COPY PERu: PoVERy COmARSONs - A POLiCYNOTE 2 Part I: Report List of Content 1. Introduction and Overview 3 2. Prolog: Poverty Rates as Policy Goals? 8 3. Poverty, Inequality and Social Development, 1994-1997 10 1. Basic Developments 10 2. Regional Developments 14 3. Rising Inequalty and Its Components 18 4. Poverty -- Changing Faces? 21 1. Poverty Comparisons: How Did Groups in Society Fare? 23 2 Poverty Comparisons: Key Factors of Welfare Changes 29 5. Growth andEmployment 37 6. Social Expen ditures -- What and for Whom? 42 1. Social Expenditures in 1996 42 2. Poverty Impacts ofDirect Transfers 46 7. Institutions -- From Individual Sector Strategies to a Consistent 48 and Broad-BasedAnti-Poverty Focus References 50 Annex 1: Panel Study ofHouseholds 53 Annex 2: Methodology 55 This report is a product of the Country Department 6, Latin America and the Caribbean Region. Jesko Hentschel led the team that prepared the report.. The team included Alberto Chong (panel data analysis), Edgard Rodnguez (inequality) and Valeera Dorabawila (consumption definition, profile and simulations). It also draws on background papers by Rafael Cortez (health and nutrition), Lucia Fort (indigenous people and gender) and Jaime Saavedra (labor market and education). Juan Diaz provided valuable support The director of the country department 6 is Isabel Guerrero and the lead economist is Ernesto May. L overview Peru remains huge at a constant rate of about 45 percent of urban employment, even higher in rural areas. The positive Poverty and Social Develop2ent, social welfare trends are also due to 1994-1997 substantial Government efforts: from 1994 to 1997, more than half a million This report evaluates social households received water, electricity and progress in Peru since 1994. It carries sanitation connections; the public health mainly good news but also reports several sector attended more than one million worrying trends. The good news is that additional ambulatory patients per month; social welfare improved over the last years and 200,000 more children are in school -- and this is true when looked at from a in 1997 compared to 1994. variety of different angles. The poverty rate, the percentage of the population not But there are also some worrying able to finance a basic basket of goods, trends to report -- and most of them are has declined by several percentage points closely knit together. Economic growth and now stands at 49 percent -- roughly and government programs have not been 12 million Peruvians are therefore spread equally and have not benefited considered as poor. Severe consumption everybody. First, regional disparities have poverty -- an extremely austere measure -- grown with some regions showing has also declined, from about 19 to 15 enormous progress, especially Lima, and percent. This does, nevertheless, leave other regions falling relatively behind, three and half million Peruvians in this immediate danger of hunger and especially the rural areas in the highlands. meiateon daine oh hu and In the rural Sierra, overall poverty remains deprivation. In line with consumption stagnant while its severnty has declined. poverty rates, school attendance has risen Of the total reduction in poverty, almost slightly, literacy rates increased from 87 to 80 prent sed fm oions 90 ercntandthepoulaionisheathir. 80 percent stemmed from two regions 90 percent and the population is healthier. alone: Lima and the urban Sierra. In Most important among the latter, the rate inentoa opriosPr-ean of malnutrition for children below the age oneroah conis ith remely of five has further declined. About h va of reiona ncom 600,000 children younger than five, or one in every four, are malnourished in 1997. Second, with regional disparities increasing we also find, for the first time duthese iovee are without in ten years, inequality in society on the doubt due to the favorable overall rise -- a small but robust increase which economic environment with per capita can be observed when using several real growth rates from 1994 to 1997 at measurement methods and when looking about 3.5 percent. This growth, quite at the distribution of income or wealth contrary to public belief, did create jobs. alike. This has to be of concern to We estimate that about 1.3 million policymakers as evidence now exists that additional jobs were created in the more unequal societies tend to be more economy, absorbing both a population violent societies. Economic progress also increase and a higher participation rate in depends on equality with more unequal the labor force. Many of these new jobs societies showing a worse growth record. are informal jobs so workers are without And, clearly, inequality and poverty are formal contracts, pension insurance or also directly linked: for any given national health insurance, however. Informality in income, the more unequal the society, the PERu. PovERTYCO4fEARKgON$-- A PpLicyNOrE4 higher the poverty rate. We find two the survey data tells a sad story about driving factors behind these inequality child labor as more and more youngsters increases: first, the better educated between the ages of 6 and 14 work to Peruvians profited much more from the supplement family. In addition to current upswing than the less educated; children, many young adolescents are not and second, regional development varied faring well in Peruvian society today. strongly. While the government has made Youth unemployment is very high, 18 an effort to reach out more to the percent for females and 14 percent for marginal rural population, this effort has males in Lima (1996), and they show a only partially translated into measurable rising trend. benefits. Of the large achievements in education, health and infrastructure, about This report does not aim to describe 70 percent have been directed to cities. the situation of the poor in Peruvian society -- or sketch a 'poverty profile'. Third, Peru's development in the past Many other studies have done this. years has been inclusive for many but Rather, we are interested in assessing what exclusive for others. While we find determines whether families get ahead or gender differences narrowing and fall behind over time. This is of special vulnerable groups such as migrants or the relevance to policy makers. For example, landless sharing the benefits of a static view might tell us that informal development, certain groups appear to fall employment is a strong correlate of further behind or remain highly at risk of poverty. But a view over time will be able deprivation. The first one is clearly the to answer the question whether the depth indigenous population. Their social or of poverty increases if a household is political integration is still far from predominantly linked to the informal achieved. And we now find that even market or not. economically the native population has fallen further behind: while in 1994 an What helps Households Advance? indigenous family was about forty percent more likely to be poor than a non-native A number of factors have influenced family, in 1997 they were almost fifty household welfare over time, in both percent more likely to be poor. positive and negative ways alike. Using a Additionally, observing hundreds of the positive and negative ways. Using a panel same families from 1994 to 1997, the of identical households, we studied what indigenous families have clearly done characteristics were linked to the growth worse, even if we control for their lower rate of household per capita consumption educational training, lower access to from 1994 to 1997. First, surprisingly, services and lower land or housing households were more likely to advance if ownership compared to the non-native their income stemmed from the informal population. sector than from the formal sector. This is true in urban areas as well as in informal The social situation of children off-farm employment in rural areas. remains bleak. The youngest in Peruvian Second, household size matters. Larger society continue to have far higher families have done worse than smaller poverty and severe poverty rates than any ones. Third, access to services and credit other age group. And although poverty is not only of immediate help to rates decreased, the drop was slight and households but helps them advance faster much less than for other groups. Also, in ways not just directly connected to PERu: PovERTY ComPARisoNs - A PoLicyNoTE5 services access. Also, we find strong Employment creation going hand in hand support that bundling of such services with agricultural growth was not strong. matters: providing two services jointly has This could be due to a 'productivity a more positive effect than the sum of backlog' stemming from the recession at providing each one separately. the beginning of the 1990s. Finally, we present some findings A number of simulations show how about the incidence and impact of urban important growth remains for poverty violence on the families of the poor. reduction in Peru. However, these same While we cannot link insecurity directly to simulations also show that the type of welfare developments, violence is one of growth and its regional distribution will the main preoccupations of the urban matter enormously -- the more growth is poor. Incidence rates for various types of based in agriculture, construction and violence differ by poverty group, the poor commerce, and the more its impacts filter being about twice as much exposed to through to the rural high- and lowlands, physical aggression than the better off in the higher poverty reduction will be. society. Consequently, their feeling of insecurity is higher. Social Expenditure Prospects for Poverty Reduction -- The distribution of social and anti- Growth andEmployment Links poverty expenditures is disappointing. Examining the incidence of 8.4 billion One of the biggest concerns in the soles (about 40 percent of the total public Peruvian public debate on poverty is budget in 1996), it is mildly tilted towards whether growth has created employment the better-off in Peruvian society, i.e. the and whether this has lead to poverty poorest obtain less of these expenditures reduction. We find that, yes, growth over than their population share. In large part the past years has indeed created this is due to the anti-poor distribution of employment; about 1.3 million more higher education and hospital people have been in remunerated expenditures. employment in 1997 compared to 1994. The majority of jobs were created in the Several specialized government informal sector but they were not programs reach few of the poor and direct necessarily low paying jobs. A worrying public transfers play a significantly smaller trend is that urban productivity does not role than private transfers. The nutrition seem to pick up and that, consequently, program PRONAA and the social fund real wages are flat at best. FONCODES have the highest coverage and lowest leakage rates but the housing On face value, Peru's growth path credit programs as well as the over the past years was 'pro-poor' because infrastructure programs of COOPOP, the sectors where workers and their FONAVI and INFES reached only few of dependents are most likely to be poor the poor. Food aid has the largest (construction, commerce and agriculture) positive effect in rural Peru where the grew fastest. This appears to have helped severe poverty rate would have been three the poor in construction and commerce. percent higher had these programs not In agriculture, poverty reduction was existed in 1997. However, private slower than could have been hoped for. transfers generally play a significantly more important role than public transfers Today, the multitude of social policy in the country. programs operate independently, try to reach their beneficiaries with different means and lack stringent evaluation. From Individual Sector Strategies Expenditures of many of these programs, to a Consistent and Broad BasedAnti- although well intentioned, do not reach Poverty Focus the poorest in society and are often isolated in nature. Many different poverty This report does not aim to provide maps and targeting mechanisms are detailed recommendations as how poverty currently employed which would need to can be eradicated in Peru. Rather, we be harmonized. We find, however, that wanted to present a quick feedback about poverty is reduced most if interventions social developments and poverty after the are integrated, that is providing them new Living Standard Measurement data jointly and in a coordinated way. In Peru, from Cuanto S.A. became available in conflicting decrees empowering the June 1998, combining these with a policy- Ministry of the Presidency and the Social relevant analysis of growth patterns and Coordination Council CIAS currently the distribution of social expenditures. exist -- de facto, neither of the institutions Forthcoming World Bank reports on the has true power or manpower. CIAS, indigenous population, and the education although having made progress in and health sector will contain detailed coordinating the 'Gasto Social Basico' has policy proposals. Also, while global not functioned for almost one year now. strategies to reduce poverty are necessary and important, they do carry the risk of Second, and closely linked to the oversimplifying a very complex and above, pro-poor policy formulation needs difficult task. In Peru, with about half of to be accompanied by thorough and good the population in poverty, poverty evaluation. This goes beyond the need for eradication will take a long time and targeting and prioritization. It includes require coordinated efforts from all parts for policy makers to be able to assess of society -- the public, private and whether certain interventions did indeed voluntary sectors -- and the international help or not. And it also implies that community alike. policy-makers and technicians are able to assess how changes in program nature and This report does also not recommend how changes in expenditures are distributed the creation of new programs nor does it and what effect they have. make a statement about the appropriate size and mix of programs. In broad lines, Third, central coordination promises we find that the Peruvian anti-poverty to be effective if it goes hand in hand with programs with their mix of emergency decentralized execution, involving other help, nutritional focus and infrastructure partners in the fight against poverty. emphasize the right areas. However, we Examples from other Latin American believe that a much bigger impact could countries show that private-voluntary- be achieved with available funds: public partnerships in poverty reduction at the local level can be extremely successful. First, a central and powerful social One reason for why such partnerships are policy council needs to be established that successful is that each organization brings designs poverty reduction strategies in a its comparative advantage to the table: technical and non-political fashion. central government finance and PERu.: PovERTY COmPARIsoNs-- A PoLicyNoTE7 organization; municipal government local - Acknowledgements knowledge; and non-govemmental organizations often a good and direct We would like to acknowledge understanding of the problems of the close cooperation, fruitful discussions and poor. For this latter point we have some the generous sharing of data information 'hard' evidence: In 1996, NGO- with both Cuanto S.A. and the Peruvian administered programs have a significantly Statistical Institute (INE1). Much of the better targeting record than most of the material and statistics presented in this public programs and match the good report are based on four different targeting results of FONCODES or household surveys. Two of them are PRONAA. Living Standard Measurement Surveys by the independent Instituto Cuanto (Encuesta Nacional de Hogares Sobre Outline Niveles de Vida, ENNIV 1994 and 1997), one is a national household survey by the This report is structured as follows. National Statistical Institute INEI Section 2 contains a 'pre-cautionary (ENAHO 1996) and one a violence warning'. It is a short recap of what survey (ENHOVI 1997) by the same household surveys can and what they can institute. not do, including an assessment of how reliable poverty statistics are. Section 3 then turns to look at national and regional indicators of well-being between 1994 and 1997, including a closer look at why inequality rose between the two years. Section 4 presents our findings as to what groups in society benefited and which ones did not benefit from the general rise in living standards in Peru. Further, we examine which main factors can be made responsible for such welfare changes. Section 5 turns to examine what the prospects for poverty reduction are given different growth rates of the economy, different assumptions about inequality and -- most importantly -- different types of growth patterns. Section 6 takes a look at the distribution of social expenditures in 1996, i.e. which groups were reached and which weren't reached by the large social programs of the government. Section 7 describes key institutional ingredients for developing successful poverty reduction strategies, comparing it to other countries in Latin America. Section 8 concludes. 2. Poverty Rates as Polcy Goals? Much of the current political debate in Peru concerns whether poverty rates have fallen or risen over the recent years. Partly, the prominence of poverty rates in the public debates is due to the Government having set itself clear goals of poverty reduction. Partly, it is also due to a heated public debate about social conditions. Before launching into poverty measurement, profiles and correlates in the next section, we briefly want to argue in this section that, yes, setting poverty reduction goals is important and laudable. However, estimates of poverty rates are much more fragile than often thought. Further, lifting people out of poverty has a much broader meaning that raising them above a 'poverty line'. Poverty rate estimates are based on the analysis of household surveys which are an important and indispensable tool for poverty analyses and hence crucial for policy formulation. For example, the geographical distribution of poverty and severe poverty is a very important tool for expenditure targeting. Household surveys are indispensable to analyze the distribution and coverage of public programs -- which groups in society obtain what from the public portemonaie. Similarly, they can serve to track enrollment rates, illness patterns, service access and literacy rates at relatively moderate cost since the surveys are much less costly than conducting population censuses. And the determinants of poverty and well-being can be analyzed as well as the factors influencing malnutrition or child mortality rates. Studies covering these and many other topics have been used for policy advice in Peru for many years. But poverty analyses based on such surveys is by its very nature not an exact science. The choice of poverty lines and the determination of consumption and income depends on a vast amount of assumptions which, even if changed only slightly, can produce quite different 'poverty rates'. It is therefore of, no contradiction at all that different surveys will produce different poverty rates, as is the case in Peru right now. The Statistical Institute INEI and Instituto Cuanto have, for example, quite different questionnaires trying to capture food expenditures of households. With different questions for different types of products or product categories, it is only consequential that estimates of consumption and poverty will differ. This does not say that one is more right than the other. Whether poverty is 45, 49 or 53 percent should, in the end, not be at the center of attention. However, whether and to what degree poverty increased or decreased, using a consistent and common methodology is of importance to policy makers. Some of these aspects are further explored in Annex II. An example might also show that the very Graph 1: Intervals for Severe Poverty fact that poverty calculations are based on a Rate Rates sample of households, hence a subset of the 20.7 upperbound Peruvian population, carries implications. Samples are designed so that they reproduce the whole population but they can never be as exact 1... as information that covers everybody in the 16.8 .. country. Hence, they carry a 'margin of error'. Consequently, calculated poverty rates also have a margin of error. Graph 1 shows what this margin 12.8 of error means in the case of Peru. We use the 1994 1997 PERU PO.eRTY 7ComPARisoNs --A P IY No TE 9 calculated severe poverty rate as an example.' The two columns indicate the range in which we are very confident that the 'true' severe poverty rate lies. As can be seen, while we estimate that severe poverty has dropped substantially from 18.8 to 14.8 percent between 1994 and 1997, drawing such confidence intervals around them is quite revealing. If, indeed, we were at the lower end of the shown interval in 1994 and at the upper end in 1997, the rates for the two years might actually be very similar. And, on the contrary, if the true 1994 rate is rather 20 percent in 1994 and the 'true' value for 1998 is at the lower end, around 13 percent, the drop in severe poverty can actually be bigger than we estimate it here. Further, while a policy focus on lifting people 'over the poverty line' is laudable, poverty has obviously a much broader meaning. Income or consumption are only means of attaining better lives, and not ends in themselves. Lower malnutrition in children, better health of the population, longer lives, lower maternal and infant mortality rates, higher literacy, less hunger, more safety, less discrimination in work and social life, and more active participation in political and social affairs of communities and the country characterize better and 'less poor' societies. And several of these other dimensions might not be only linked to consumption poverty: crime and violence might effect large parts of the population, discrimination in the job market can exist against certain groups in society, or children might be malnourished although they grow up in rather affluent households. In conclusion, the answer to the question whether poverty rates serve as policy goals is clearly yes. This commits policy to be serving the most marginalized groups in society. However, policy makers need to be aware of the often quite fragile nature of such poverty estimates. And finally, reducing poverty rates is only a means to improving the lives of people so that they are healthier, and have longer, better lives. 1 The severe poverty rate presented here is not strictly comparable with the 'extreme' poverty rate generally reported due to the specifics of our consumption definition. See Annex II of this report. 2 The ranges indicate that we can be 95 percent confident that the true severe poverty rates lie within the upper and lower bounds of the interval. PERu:- PovERTY CompARIsoNs -- A PoLicyNorE 10Q 3. Poverty, Inequality and Social Development, .1994-1997 This section provides basic statistics on poverty and social development in Peru since 1994. It takes a closer regional look at where this progress was faster and where it was slower and also assesses who in society profited most from public investments in five areas: water, sanitation, electricity, education and health. Finally, we explore the rise in inequality and trace it to its underlying causes. 3.1. Basic Developments Indicators Used. We aim to look at social progress in Peru from a number of different angles. First, we use 'outcome' indicators of the development of society -- the rate of children below the age of five being malnourished and the literacy rate of the population. Second are poverty indicators. Throughout this report 'poverty' is defined as a state in which the effected population has per capita expenditures less than needed to purchase a very basic basket of food and non-food goods. The derivation of the cost of this basic bundle, the poverty line, is somewhat different than generally applied in Peru. Especially, for comparison reasons between the two household survey years 1994 and 1997, we had to make a number of important but tedious adjustments that are explained in detail in Annex II of this report.4 More important than the percentage of the population below this imaginary line is bow far the poor are away from it. For this we use the 'poverty gap' as a measure which expresses how many resources are needed to bring all individuals to the poverty line. This is expressed as a proportion of the poverty line itself. We also use an additional, much more austere poverty line to find out who in the Peruvian population is at a much more immediate risk of acute hunger and deprivation. Third, we compute inequality measures for consumption, wealth and income. The common indicator used here is the Gini coefficient; a measure which varies between 0 (totally equal society) to 1 (completely unequal society). Fourth, we record rates of child labor, defined as children age 6 to 14 year working more than 15 per hours per week. Starting from the position that child labor, especially at such young ages, is detrimental to both health and learning possibilities, we hope to find low and declining values. Both rates are derived from the LSMS surveys (ENNIV 1994 and 1997) which were carried out by the Instituto Cuanto in Lima. The malnutrition calculations are for stunting (height for age), defining a child as malnourished if it is more than two standard deviations below the age-adjusted international norm. This is in line with other studies of malnutrition in Peru. One of the crucial assumptions in these studies is that age-specific norms of height and weight are homogenous in the country. That is, they do not vary by location or ethnic group. Some of these adjustments change the consumption aggregate supphed by Cuanto S.A. for both survey years. As explained in Annex II, our major deviation from Cuanto's methodology includes: (a) leaving the total bundle of good entering into the calculation of the poverty line constant (while in Peru generally only the food basket is left unchanged), (b) excluding rent from the consumption aggregate as the survey question in this section changed significantly, (c) interpreting the benefit transfer from social programs differently, and (d) using a different regional price deflation method. PERu: PovERTY CompARIsoNs --A PoLICYNoTE i Fifth, we are interested in total school enrollments and the number of ambulatory care visits attended by the total public health network. Obviously, these are only 'inputs' as by these shear numbers we cannot tell the quality of education or why more people sought care in public health facilities -- access and/or quality might have improved or people might also be more in need of care because their health deteriorated. Finally, we record the percentage of the population with access to sanitation, water and electricity and the percentage of the population living in homes with mud floors. While we will look at the dispersion of some of these indicators by groups of people or regions below, there are some aspects of social development that we do not assess here. These pertain, for example, whether the population and different groups within it have secure property rights and whether they have the possibility to enforce these property rights, i.e. an equal and fair access to the justice system. Similarly, social progress of society will also be a function of the degree to which families and communities participate in local and regional decision making and how society integrates differences in cultural values and beliefs. Developments 1994 to 1997. Looked at from many different angles, Peru has made progress in many areas in the past three years. Table 1 contains the different indicators outlined above. Malnutrition rates decreased substantially, in both urban and rural areas. We estimate that about one out of four children below the age of five is now stunted, i.e. its weight for age is below a minimum acceptable level. However, this leaves still more than 600,000 children malnourished which will lower their learning ability and make them much more vulnerable to illness -- now and later on in their lives as well. Literacy rates of the Peruvian population above six years of age increased and now reach 90 percent. Poverty decreased in the past several years and this is a robust result, quite independent of the poverty line chosen.5 An especially positive development is the reduction of the severe poverty rate from almost 19 to now about 15 percent. In absolute numbers, this means that 600,000 Peruvians have managed to find livelihoods that helped them out of extreme consumption deprivation. However, 3.5 million people remain in such immediate danger. The proportion of the population in poverty, not able to finance a basic basket of food and non-food goods, decreased as well but remains very high. Half of the population or 12 million people are poor in 1997. And as we have seen when looking at severe poverty rates, the reduction in poverty is not only limited to the part of the population having consumption expenditures 'near' the poverty line. If this were the case, the poverty gap indicator (a measure of the depth of poverty in relation to the poverty line) would not show significant declines. As can be observed in the table, though, this is not the case and the 'gap' is reduced quite substantially, especially in rural areas. Notwithstanding the decline in poverty, important inequality measures show a (modest) rise. Table 1 reports that the distribution of consumption, especially in rural Peru, is less skewed. However, both the distribution of income and wealth -- and these are primary measures of inequality because they include current and accumulated savings of the population -- appear to have become more unequal over the past years. We varied the poverty line over a wide range and found that poverty rates decreased quite homogeneously. PERU: q.PO-vERTY.COMPARISONS --A.PqLjcY NOTE ........................................12 P ..p .P g .. . . . . .. . . . . .. . . ... .. .. ..I... ... .. .. ... .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Table 1. BasicIndicators.:y. National brban -Rural 1994 1997 1994, 1997 1994'.' 1997 . mnutrWtionlte % 0o.o ' 12 4,1 37.5 iteracyrate'(%) 87,6 90.2 92.3 943 ' 7 [mmioilife:expiectancy &s)1 , 8.o fimb,in faitmort4 100 births)] 42.0'' 2. poverty'ste,@%) 53.5 49. " 6.1 4 670 647 poverty gap (%} f8.9 16.0 14.4 "211 8 271 235 'severe orty rate (%) . 18.8 14-8 12,9, 9'3 29.5 245 ti Gin 360 348 351 345 ', `349 324 icome inequalRt Gsni .469 t484 4 A 500 inequality, Gitsi .695 .726 672 705 .06 .678 4" child iborj.2) 7.8: 1118 39 629 225 335 5, chool enirothnent (00) 4,88 5,080 2,960 3,030 1,920 250 c k 1,760 '2990' 1,250 2,160 5 (30 4. electicity connections (%1) 68.8 717 93.7 ',,,,,97.4 '232 303 aitaiconnection (%) 48.2, '58.6 :7 84- 2 16 wadter, public' sketwork ) 65.0 72.8 84.9 8 , 28.8 .' 43..1 homs w/mudfloors(%) 41.0 43.3' 20.4 2.2 77.9 79.6 1 Life expectancy rate and infant mortality rate is for 1996. 2 child labor percent of all children, 6 to 14 years of age, working more than 15 hours per week Sourc: Staff estimates based on ENNIV. Life expectancy and infant mortality rate from World Bank (1998b). With respect to child labor we have to record a negative trend. In 1997, more children were working (both in absolute numbers and percentages) than several years earlier. On the other hand, the public education and health sectors have attended more students and delivered more ambulatory care in 1997 than in 1994, although the rise of student enrollment rates in public primary and public secondary schools is very small. One cautionary remark is necessary here: The two household surveys on which we base our assessments were carried out in different months of the year, the 1994 survey in June/July and the 1997 survey in October/November. While this should not affect estimates of school enrollments since both periods are term times, health and illness patterns might be different. It would therefore be difficult to attribute the rise in ambulatory care treatments in health centers and hospitals to a larger and better public health service alone. And we estimate this increase to be very large: we record a rise of 1,2 million ambulatory care services. PERu: PovERryCOMPARISONS-A PoLiCYNOTE 13 The large public investments in basic infrastructure over the past couple of years through programs like FONCODES or FONAVI have increased connection rates significantly. About 700,000 households each have obtained sanitation, electricity and public water. Electricity connections in urban areas are now almost complete; urban sanitation is at 85 percent (which still leaves 2.4 million urban residents without adequate hygiene facilities) and public water is obtained by 89 percent (with 1.7 million residents without water). Inroads in rural electrification and rural water have been made but gaps remain very large. Rural sanitation, although showing success as well, is still very scarce. While public investments in infrastructure have increased living standards of many families, the housing stock itself does not seem to have improved. Actually, the percentage of dwellings with mud or earth floors have increased from 1994 to 1997, signaling that a substantial number of families constructing new dwellings -- in urban and rural areas alike -- have not the resources to install a more hygienic and stable floor. Comparisons. Some international comparisons with other Latin cw.ps2 LanAmence InfamM olay Rates, 19 American countries show that, despite the major improvements recorded se-M - above, Peru is still 'catching-up' with _OOn_a respect to several of the recorded * indicators. Graph 2 shows the infant e mortality rate in 1996, setting it in . relation to GNP per capita as recorded z - by the World Bank's Atlas method. __. 0 Although Peru reduced infant mortality 0 20 30 40 so 60 70 so from 54 (1990) to 42 death per 1000 .mW.WMm6any. live births in 1996, it is still lagging behind the regional achievements given its income level. For example, Graph 3: Latin America: Access to safe water, 1993197 both Colombia and Costa Rica have considerably lower infant mortality 10000 rates (Colombia 25, Costa Rica 12) 8000 aArgentn while they are roughly in the same income bracket than Peru. The same 6000 a picture -- although not that j L oo Pa pronounced -- can be seen for life z " ,o expectancy. Here, it is estimated that 2000 -tolomb11 Losta itS Peru has a roughly four year lower life 0 , . expectancy than its income level would so 60 70 so 90 100 suggest. Such a gap does not exist for Source World Bank (1998b)PercentofPopulaonwithaccess adult literacy where Peru performs according to expectations. Similarly, access to basic services is still low despite the big successes achieved. Graph 3 shows Latin American countries and their per capita income level, now with respect to the 6 See icks and Peeters (1998). PERU .PovERTY CompARISONs--A PoLicyNoTE 14 percentage of the population having access to safe water. While the relationship is not as visual as in the case of the infant mortality rate, we can nevertheless also detect that Peru with 72 percent water connection rates fares worse than Colombia or Mexico. Costa Rica -- as in many instances -- is the 'positive' outlier. Poverty rates are very difficult to compare internationally. As outlined in section 2 above, poverty estimates are based on household surveys and, across countries, these differ considerably in sample design and content. Many only record incomes and not consumption expenditures which we use here for poverty measurement. Poverty lines are generally determined in national contexts, with varying goods baskets. Real prices, even after converting them using exchange rates into one common currency, also show wide variation so that purchasing power parities have to be used to obtain comparability. While efforts to make poverty statistics internationally comparable are worthwhile, they have to be handled with 7 extreme care. 3.2 Regional Developments Changes in Poverty and Malnutrition Rates. While poverty and malnutrition indicators decreased in all of Peru's very diverse regions, rates of change have been quite different. Table 2 reports the perrentage change of the rates between 1994 and 1997, i.e. it normalizes for the level of poverty and malnutrition. A four percent decrease in absolute rates translates into a smaller percentage change in rural than urban Peru because prevalence of poverty and malnutrition is much higher in rural areas. Table 2 reveals that looking at these percentage changes, Peru made substantial inroads in reducing severe poverty (by 22 percent) and malnutrition (by 21 percent). However, the table also shows that a stark urban-rural difference in these advances exists. Overall, urban Peru was the driving force behind much of the gains for all three indicators. Two regions have performed better than the national average: the urban Sierra and Lima. On the other hand, the rural Sierra -- where almost two-thirds of the total rural population live -- has consistently performed worse than the national average. The one region where the indicators show an unusual trend is the rural Coast. While both poverty and severe pdverty reduction rates are below the national average, malnutrition declines very strongly. In its World Development Indicators the World Bank used to include Peru in such international compansons but, after reexamining the basic data, it has now stopped pubhshing estimated rates for the above reasons. PE,Ru: PovERTY ComPARIsoNs - A PoLicyNoTE 1 ........war................... c.ER y g gy US.NS gM O .................................................................................. Table 2.Peicentage ChAnges in RegionalPoverty and Maiuttion, 1994-1997 Povery Severe Poverty, Maindon RuralCosta ""' -4 -43-4 Urban Sierra 24 -42 RutalSierra 2 6 Urbanjungle 0 -23 Rura1Jng- -23 2 LTRBANPERU -13 28 30 RURALPERU 7" ,TOTAtcOaNTRY -8' -22,, 2 Soura :StafEstir'itatie"5baised on"ENNIV (199 1997) In accordance with these poverty and Graph 4: Percent of Total Por and Mlnourished in Runl malnutrition reduction A 1994-1997 patterns, the developments 1994 1997 over the past couple of 1994 199 years have led to a higher _19_7 A concentration of . deprivation in rural compared to urban areas of Peru. Graph 4 shows the distribution of the poor, extreme poor and malnourished children - below the age of five in poor severe poor rmffonshed 1994 and 1997. Almost 50 percent of the poor, 60 Source Staff Estimates based on ENNIV, 1994 and 1997 percent of the severe poor and 70 percent of the malnourished children lived in rural Peru in 1997. This happens against a continuing urbanization of the country that has led to more than two-thirds of the population living in large and small cities. PERU: -POVERTY COMPARISONS --A..PoLIcYN6.O ............I............................ 16 Dispersion in International Perspective. In line with the above, Gaph5:Realpercapitaconumptiongrowthates, income per capita and consumption per by rson 19419 capita differences between the diverse regions of the country has increased and Usbx.Coct Peru has an internationally high rate of internal regional dispersion. Per capita consumption in the urban Sierra and RuWuim Lima increased by more than 15 percent - while growth in the rural Sierra was negligible and even negative in the urban Law Selva and urban Coast (Graph 5). By uibusiffm _ international standards, the existing -5 0 5 10 is 20 regional dispersion in Peru is very high Source: Staffestur-ates based nENNIV 1994, 1997. as Table 3 shows.8 Peru has much higher regional inequalities than Colobia,Chie, Baziandeve MeioTable 3: Regional Dispersion Indicators, Ladn Colombia, Chile, Brazil and even MexicoAmrcnoutisva uyey and is only topped by regional dispersion in Argentina. Year Dispersion Distribution of new social and infrastructure services. As Table 1 Argentina 1995 .736 Brazil 1994 .424 showed, basic and social service access Chile 1994 .470 rates increased strongly between 1994 Colombia 1989 .358 and 1997. More than half a million Mexico 1993 .502 households have been newly connected Peni 1997 .561 to water, electricity and sanitation. Source: Fallon (1998). Peru estimate from ENNIV (1997). Ambulatory health visits have increased by about 60 percent. Who was reached Table 4: Distribution ofNewAccess to Basic and Social by these additional investments made in Services, 1994-1997 (percentage) the past years? This is an important question for public policy making given Urban Rural especially the increasing regional dispersion in Peru. Water 57 43 (100) Electricity 72 28 (100) Sanitation 78 22 (100) Ambulatory Health 74 26 (100) Education, enroll 33 67 (100) Source- Staff estimates based on ENNIV 1994, 1997. The dispersion indicator used here is the unweighted coefficient of vanation (i.e. the standard deviation across regions divided by the national mean). See Fallon (1998). PERu..PoVERTY.COMPARISONS -.A PoLcYNOTE ......................................................................1 Quite contrary to belief about the working of many of Peru's public programs9, we find that by far the largest portion of new beneficiaries live in urban areas. As Table 4 shows, except for the (modest) increase in absolute enrollment rates, the beneficiary population of all other basic services and health services has been in the urban areas. Indeed, the distribution of beneficiaries is quite close to the overall population distribution in Peru which would suggest that expenditure distributions for these programs have largely been driven by population densities rather than poverty criteria. Gains in new access to basic and social services were quite evenly distributed Table 5: Distribution ofNewAccess to Basic and n tSocial Sevices 1994-97, by population quindle in the population and do not show a significant pro-poor bias. We report in water electricity sanitation health Table 5 the new access by national population quintiles, with quintile 1 comprising the poorest 20 percent of the 1 (poorest) 20 18 18 16 Peruvian population and quintile 5 the 3 21 18 20 18 richest ones. The population quintiles are 4 18 20 18 26 formed on the basis of per capita 5 (richest) 15- 18 19 19 consumption expenditures. According to ------ the results, programs seem to be more (100) (100) (100) (100) successful in reaching the poor in the Source: Staff estimates based on ENNIV 1994, 1997. second quintile rather than the severe poor in quintile 1."o In part, this is driven by the observed regional distribution of the additional expenditures. Obviously, these statistics on supply of services will translate into welfare gains for new beneficiaries only if the quality of the services is adequate. In an urban qualitative investigation conducted in 20 centre poblados by the Ministry of the Presidency last year, communities ranked the quality of water and sanitation services as their primary problem -- while they had access, the services were not reliable." Similarly, especially in the social sectors, the bottleneck in many localities might not be the si!ply of services but rather the demand of beneficiaries. The demand for health care, for example, will be influenced by the costs of transport, waiting time and medicine but also by the compatibility of the offered services with traditional health beliefs. See Oxford Analytica (1998). 10 One important caveat has to be made here Comparing quintiles between 1994 and 1997 in this way would assume that there is no mobility between population groups As later pointed out, this is not the case. However, mobility is often restricted to movement to the closest quintile (e.g. from the bottom to the second quintile or from the nchest to the fourth quintile). The overall assessment that infrastructure connections were distabuted quite evenly across the population is therefore likely to hold. See Mrnisteno de la Presidencia (1998). lp,qER P ETy COMPAI"SONS -- A POLJCYNOTE 18 3.3. Rising Inequalty and its Components After having detected that social and economic progress was not evenly distributed in Peru over the past years, we now want to briefly return to the finding that, for the first time in a decade, inequality has risen in Peru. The section examines consumption, income and wealth changes by quintile and then looks at the sources of inequality increases. Consumption and Wealth Growth Rates. Consumption, Graph 6& Gmwth in Income, Consun and Weath (avcrage Growh Rtes. Conumpton,annual growth m per capita real tetrms), Peru 19974 income and wealth changes reveal a ': quite different distribution of gains in prosperity. Graph 6 shows that per . . capita consumption changes were 1) PCptaWea highest for the poorest and riches quintile alike over the past years. While true also for income changes, 2W- gains were clearly tilted toward the richest quintile that is reflected in the 2 3 5 rise in income inequality as reported above. 1 Changes in wealth of these same population groups show a wide variation which might in part be due to changes in reporting behavior of -e C Q-H (p".4b.d) wealth - however, according to these S-- StffEdei-hbsed-ENNIVjau4aad197 estimates the richest quintile again records the highest per capita growth rates of wealth. Wealth inequality increases. These results are confirmed by a different datasource: examining income inequality changes between 1995 and 1997, the Peruvian Statistical Institute (INEI) finds the Gini coefficient increasing as well -- with an even larger percentage change in inequality than we find here. According to preliminary estimates based on the large Encuestas Nacional de Hogares, the Gini coefficient increased by 10 percentage points between 1995 and 1997.14 12 See Annex II for a definition of income. In accordance with the definition of consumption, the income variable also does not include the rental value of the house because the question asking households for their hypothetical rental value changed between the 1994 and 1997 questionnaire. 13 The wealth variable includes the value of consumer durable goods, the value of owned houses (self- assessment), the value of property and equipment. The wealth changes reported in the graph correspond to population quintiles defined using household per capita expenditures. 14 Communication with Departamento de Estadistica, INE. PERU: .PO.VERTY_COMPARIsONs --A Po4fCY NOTE ........................................19 Sources of inequality increases. What are Table : Expected Changes in Income the contributing factors behind this increase Inequality by Income Source, 1997 in wealth and income inequality? We first (percent of Gini change) look at the impact the different components of income. What happens if income of one of the components of income were to increase? does this lower or increase inequality? Table 6 self-employment income -4.9 reports the result on total inequality from wages 0.6 tranfers2.2 increasing income of each component by one transfers 2.1 percent: income from self-employment property income (comprising both micro-enterprise owners, Source: Rodriguez (1998) professionals and entrepreneurs) reduces inequality strongly (by almost 5 percent), while all other categories -- wage income, transfers and income from property -- tend to increase inequality. The rise in inequality in Peru over the past three years can be largely attributed to a declining share of self-employment income and rising shares of transfer and property income. Repeating the exercise for wealth, the Table 7: Expected Changes in Wealth driving forces behind wealth inequality are Inequafity by Wealth Source, 1997 housing and urban property (Table 7). Wealth (percent of Gini change) is defined here as the total value of housing, durable goods (resell value), urban property, Wealth source Expected Change agricultural property and enterprises. Increasing each of the components by one percent -- while housing 1.9 leaving all the others constant -- would increase durable goods -1.5 the Gini coefficient in the case of housing and urban property 1.3 urban property while decreasing it if wealth agncultural property -1.6 from durable consumer goods or agricultural enterprises 0 property would increase. Hence, it appears that Source: Rodriguez (1998) land distribution and the allocation of durable goods among households is more equal than the overall wealth distribution in the country. Turning to the factors associated with changing inequality, two variables are associated with the observed rises in income inequality: education and regional income differences. Education increases inequality significantly over time. This reflects a common finding in many analyses over the past couple of years, i.e. that incomes of the highly educated are rising by more than the incomes of the less educated. Average incomes of families with higher education rose by 63 percent between 1994 and 1997, while average incomes of those with primary education or less rose by just 5 percent. Similarly -- and as already outlined above -- the widening in regional differences has contributed to a rise in inequality as well. The above, rather technical, discussion has the purpose of showing what type of policies would help to improve income and wealth inequality in Peru. First, education and training of the now less-educated and their children will help equality. Second, a balanced regional pattern of growth will do the same. Lastly, policies that help income earning possibilities from self-employment -- such as in micro-enterprises -- would have the same effect. P R :P V R YC M A SO S-APoLicyNoTE ---------------------------------------20 International Comparisons. While we cautioned against direct comparisons of poverty rates across countries, a stronger case can, however, be made to compare distribution statistics across countries because these do not depend on the fixing of some real baseline as is the case with an absolute poverty line. Table 8 shows that Latin American income inequality is the highest is the world, even higher than for Sub-Saharan Africa. While Peru's inequality, measured by the Gini coefficient, was above the Latin American average in the 1970s and 1980s, this appears to have changed in the 1990s. There is quite general agreement that income inequality in Peru has decreased from the mid-1980s to the mid-1990s.1 Table 8. Inequality by Region (Gini coefficients, multiplied by 100) 1960s 1970s 1980s 1990s Eastern Europe 25.1 24.6 25.0 28.9 OECD and high income 35.0 34.8 33.2 33.8 East Asia/Pacific 37.4 39.9 38.7 38.1 South Asia 36.2 34.0 35.0 31.9 Middle East/North Africa 41.4 41.9 40.5 38.0 Sub-Saharan Africa 49.9 48.2 43.5 47.0 Latin America 53.2 49.1 49.8 49.3 Peru n/a 55.0 51.8 47.2 Sources: IMF (1998, p.2). Peru's rates for 1980s (=1985) from Saavedra and Diaz (1998), for 1990s (=average 1994 and 1997) from Rodriguez (1998). However, in the period Graph 7: Latin America: Income Inequality (Iatest available year) from 1994 to 1997 income inequality -- for the first time since BuaidI 1985-86 is on the rise and with a G=atern Gini coefficient of 48.5 very close casnk11 to the Latin American average. Ci _-1 Graph 7 shows the latest ______ -- - estimates of income inequality in 1an Latin American countries. Peru is considerably more equal than nu . . . . Brazil, Columbia and also Mexico but more unequal than Costa V-e -1 Rica, Venezuela and Bolivia. 1M 0 10 20 30 40 50 60 70 Gtni coefficient Source 1World Bank (1998) 15 See Saavedra and Diaz (1998) and Escobal et al (1998). PERU:POVETy COMPA.RISONvS --A PoLicYNoTE 2 4. Poverty -- Changing Faces? This section takes a look at how different groups in society have fared over the past years and how -- if at all -- the main factors linked to poverty have changed. Whenever possible, we will also make references to some of the other welfare indictors looked at in the previous section; however, data limitations will restrict us to largely concentrate on consumption poverty. We will abstain from presenting a full poverty profile in this section -- much has been written and is by now well-understood about the main causes and correlates of poverty in Peru. To briefly summarize, nevertheless, compared to better-off groups, the poor continue to live in worse and smaller dwellings, generally with adobe walls and earth floors. They have less access to markets, especially in rural areas and their possessions are limited; almost all of the urban poor have access to radios, most to TVs but only one in five families has a telephone. With few exceptions, there are no telephones at all in rural Peru. Electricity connections in the cities are almost universal but only about a fifth of the rural poor can rely on it for lighting. Poor city dwellers keep having to cook with kerosene; the rural poor -- if they can -- rely on wood. Although needed, about 40 percent of poor, ill people can't afford to see a doctor or nurse. With health access, especially in rural areas, still sparse, transportation, waiting and medicine costs are simply too high and only ten percent of the poor are covered by some form of health insurance. Finally, education still is one of the main driving forces behind welfare differences in Peru -- looking at the 'average' years of education, poor household heads in urban areas have about 6.5 years of formal training and non-poor heads 9 years. Educational attainment in rural Peru is even lower and so is the gap between groups: for poor household heads 4.5 years (less than needed for finishing primary school) and for non-poor ones 6 years or barely into secondary school. Instead of describing static differences in living conditions, we are interested here how powery iskS have changed over time. That is, by belonging to a specific group in society, e.g. an age group or ethnic group, how high was the risk of being poor in 1994? And how did this risk change over time; is it increasing or decreasing? Such an analysis will help us to identify relative trends, e.g., has the indigenous population caught up relative to the developments for the non-indigenous groups? Or has the gap between them increased and their degree of economic exclusion widened? The 'selection' of these risks and if they are really important independently follows an analytical work of examining hundreds of identical households in 1994 and 1997 alike (see Box 1 and Annex 1). This allows us to determine the most important factors linked to households having expanded their consumption or having had to reduce it. 16 See, for example, Moncada and Webb (1996) PER: PO.VERTY OMPARISONS ..Al C'......._N.. . ......................................... 22 Box 1: hat'Induences Success ofFilure? Results froni Studying Identical Households Over Time Examining how nine hundred identical households have fared between 1994 and 1997 sheds' light on the determinants of which households did better, and which,ones did worse over the time period; As IAnnex 1 describes in more detail, we studied what the determining factors of per capita tonsumption growth were over the period. The results are: ,a. Female-beaded households had hgherpermtia gath rates ban makheaded'houstholds. Controlling for allother variables like education experience, household composition, initial consumption), we fnd strong evidericeithat female-headed'.household had mor positive welfaire dages than male-headed households; b. Migranifami4efamed better than non-mgrantfamiies;, c Nade Angua.ge peakeryt//clean) behind. One of the strongest result we find is that language, and with it indigeneity, matters a lot. Even when we control for other variables that are correlated with language such as geographic location, do nativeilanguage speaking households fall further behiid spanish-speeking households; T.hftutiry is c kajot importane. and d nd i~meamr,,g roturns ro sif We ind evidence f increasing retums to the number of,services households command Hence a household with four services (telephone, ater, electricity, sanitatio) obtains Tiot than diuble th re'turn thin, households with two, services-' Loeking at services.by type, electricity is: the:mos important service linked to household welfare improvements in rural areis while a telephone appears to be the most important service in urban areas. .e. Aa s to credit andsangs increase per capita growth rates f. Household jiZe and dependemy rates are both determining factor of nefa4re changes. 'Perhaps the most important result,'we find that household size and dependency ratio significantly influence welfare developments. Results suggest 'that (a) larger,households fare worse than smaller ones; (b) tis relationship is,not linear; the larger'the,household the less negative is the effect; and (c) the dependency ratio (number of non-income 'eamers to income.eamers):has an independent negative influence on household per capita consumption growth g., Better educadon and more experience means fasteradvance. The higher the education of the household head in 1994, the larger the growth in per capita expenditures. This reflects the common opinion in Peru that the better educated and more experienced have benefited more than the, less-educated from the recent economic upswing. h. Households with home-based husinesiesfare better. Households that stated that they used at least one room in their house for business purposes -- both urban and rural --'have managed,to achieve a signifIcantly higher growth of welfare than households, which did not have this possibility Again, 'this result holds when controlling for all other factors that influence consumption growth. i. If at -a4 household heads employed in the iitformal sector did better. We find some, although not completely robust, evidence that household headed by an informally employed person did better than formally employed ones over the past years. This would confirm that the informal sector is not necessarily a dead-end street but for many the road out of poverty. PERu:- PovERTY CompARisoNs -- A PoLicyNoTE 23 FER....v .o.m s.. cou.... .P I. OA'S foLICENOTE........... ...............................................................................3. 4.1. Poverty Compadsons -- How Did Groups in Society Do? This section takes a closer look at a number of different groups in society which have, in the past, been identified as particularly poor. These include the native speaking population - - and we use language here as a corollary of ethnicity --, children of different ages, adolescents, women-headed households (and rural widow-headed households), migrants and the landless. Our main aims is to see whether members of these groups are particularly at risk of being poor and whether group characteristics are important in the changes of this risk. While we will briefly touch on other dimensions of well-being, such as political and social integration, we will limit ourselves mainly to this 'material' view of poverty below. Table 9 reports Table 9. Poverty Risks of Selecied Grups - More orLikely to poverty risks for each group be,Poor? relative to the other (percent members of society. A positive entry means that National Percentof Total the population belonging to 9 '1997 Por, 1997 this group is more likely to be poor than the rest of the nativelanguageispeakers +24.0 +2.0 [209 population; a negative entry children,`0-`5'years +26.0 +27.0 8]] stands for the reverse. Hence, these are not absolute children, 6-14,yeams +24.5 +25.5 [155] poverty rates but relative youth, 15-17 years. + 5.5 + 8.6 [6.7] ones compared to all other ruralhouseholds w/o land 43.4 '4.0 6.2] groups. The last column of ruralwidow-headed households 5.0 -15.0 [251 the table shows what share of the total poor population female headed,Imsehlds 3.5 4.5 10.7] in Peru belongs to this migrants 46,0 -18.0 [ group. The latter is important for policy makers as the poverty risk might be 1 child labor percent of all children, 6 to 14 years of age, working extremely high for a group more than 15 hours per week inte io b u Sou.. rce Staff estimates based on ENNIV in the population but they might represent only a tiny fraction of the total poor population. Native-Speaking Population. The native speaking population is, of all the groups looked at here, the one with the highest relative poverty risk and we find this relative risk increasing over the past years -- which implies that the native speaking population is falling further behind the Spanish speaking population. We use native speaking population as a proxy for 'ethnicity' here as indigeneity cannot be 'defined' -- indigenous languages, traditional clothing, heritage, and observed traditions and beliefs can, but need not be, part of the life of the indigenous people. Accordingly, estimates of the indigenous population vary widely PERu:- PovERTY ComPARIsoNs-- A PoLrcyNoTE 24 P U. POVERTY COM/g g aNS ............... 1. PO...CY. ...O ..E .... ... ......... . ................................... ........ . 2.4. between 10 percent of the population and 40 percent.'7 However, with language being an integral part of indigenous culture, we employ it here as a 'proxy' for indigeneity. We find that the native-speaking population was 24 percent more likely to be poor than the spanish- speeking population in 1994 and 29 percent more likely to be poor in 1997 -- in other words, they are falling behind. Native language was also one of the most robust and important factors when we examined several hundred of identical households between 1994-1997 (see Box 1) -- controlling for everything else (e.g. education, geographic location, experience, household size), the native-language speaking families had significantly lower consumption growth rates than the Spanish-speaking population. Integration of the indigenous population has long been recognized as one of the most important challenges in the fight against poverty and deprivation in Peru. Education levels of adults are low and even illiteracy levels are still substantial (21 percent of the rural native speaking population above 6 years old are illiterate). School attendance of indigenous children is significantly behind the national averages and children from indigenous families are more than twice as likely to be malnourished than . Table 10: Distribution ofNewAccess to Rural children from a non-indigenous background. Basic and Social Services, by Language, Other factors like education or experience 1994-1997 (percentage) being ejual, native language speakers earn less income. Partly due to the majority of the Native Non-Native native-language speaking population living in New access Speakers Speakers rural areas, their access to electricity, sanitation, water and health services is lower Water 22 78 - (100) than the for Spanish-speaking population. Electricity 47 53 (100) Sanitation 60 40 (100) Examining the distribution of new ambulatory, hospital 23 77 (100) access to basic and social services from 1994 ambulatory, center 48 52 (100) to 1997 in rural areas, success in reaching the memo: native speaking population was mixed. Only Distribution of severe poor 60 40 (100) new sanitation investments went in its Distribution of rural poor 48 52 (100) majority to native speakers. All other, and Rural Population 42 58 especially water and ambulatory hospital care, Source: Staff estimates based on ENNIV 1994, reached mainly the non-indigenous 1997. population. These figures appear especially forceful compared to the composition of the extreme poor: almost 60 percent of the extreme poor were native speakers in 1997. 17 We classify all those households as 'native-speaking' for which the household head reported that her or his mother tongue is Quechua, Aymara, Campa or another native language. See MacIsaac and Patrnos (1995) or Davis and Patnonos (1997) 18 MacIsaac (1995). See also Lopez (1998, p 20-21) who finds that, controlling for other factors, native background reduces income of farmers and non-farm workers by 44 percent. he reports, though, that this difference disappears after controlling for village effects which could be a proxy for geographic isolation. PERU POVERTY-------COMPARISONS------------A--------------NOTE 7 The above suggests that the Government of Peru needs to take a careful look at how it tries to reach one of the most deprived groups in the county, its indigenous population. With sixty percent of the severe poor in rural areas native speakers and the massive public investments over the past couple of years having only limited success in reaching them, one possibility would be to try to bring in groups Tablell: ChildLabor, 1994 and 1997 as partners in the poverty reduction effort (percent of children ages 6 to 14) that have long worked in and with isolated, poor communities. The Ministry of the morr than 15 hours mor than 10 hours Presidency in its 'Lucha contra la Pobreza' Ou"tik 1994 1997 1994 1997 has attempted such a partnership approach, 1 11.9 16.7 17.6 24.2 especially in rural areas, and now works 2 7.7 13.1 12.1 21.1 closely with a number of respected and 3 7.4 10.1 9.4 16.3 knowledgeable non-governmental 4 6.4 9.2 8.9 14.3 organizations to facilitate, plan and 5 2.0 6.1 4.3 8.5 implement community-based projects. While an evaluation of this approach is still total 375 616 549 940 outstanding, it does represent an innovative and very promising approach to channel Source: Staff estimates based on ENNIV 1994, much needed help to the severe poor 1997. indigenous population. Children. In the age distribution, Table 12: Malnutrition Rates by Region and children are the poorest group in Peruvian Gender,1994andl997 society.19 The below fourteen-year-olds had (percent of children below age 5) a 25 percent higher risk of being poor than the rest of the population and this relative exatile 1994 1997 risk slightly increased over the past three years. We also find a rather worrying trend Urb. Coast 12.0 10.0 when looking at child labor rates (Table 11): Rur. Coast 32.0 19.0 child labor rates doubled for children Urb. Sierra 27.6 15.9 between 6 and 14. Child labor is much higher Rur. Sierra 47.7 45.8 in the poorer segments of society. Almost Urb Jungle 28.8 22.1 one out of four severely poor children ural jungle 44.7 36.1 worked more than 10 hours in 1997. Both of Total girls 31.5 24.8 these trends point to children remaining a Total boys 28.7 22.9 particularly vulnerable group in society.20 Total country 30.0 23.8 malnutrition Source: Staff Estimates based on ENNIV 1994 On the positive side, mantiinad 1997. rates have declined strongly, both by region 19 Household surveys contain very limited information about the intra-household distabution of resources. Hence if a households is 'poor' then all of its members are assumed poor although this need not be true. Similarly, depending on the intra-household resources distabution, 'non-poor' households might have poor members as well. The discussion about child poverty should therefore more precisely be framed as 'children living in poor families' 20 See also Rodnguez and Abler (1998) He finds the surpnsing result that child labor in Peru tends to pro- cyclical, i.e. falling with an economic downturn and increasing with an economic boom. PERU .POVER!Ty.COPAi.RISONs-A PoLJ'cyNOTE ......................26 and gender. Table 12 shows that malnutrition in the rural Sierra remains very high; almost every second child is malnourished. A qualitative and quantitative inquiry by Caritas in selected Sierra communities (1997) confirmed the high prevalence of malnutrition. Females remain to be slightly more at risk of being malnourished than boys are but the gender gap appears to be closing slowly. Preliminary results from examining the determinants of malnutrition rates confirm earlier analyses in Peru: the poverty level (household consumption), and educational status of the mother. Access to basic services like clean water, sanitation and light correspond with decreasing malnutrition levels. Coverage rates of the nutritional programs are, by international standards, very ofReal Program Bene Dits, 1997 big. In 1997, about 60 percent of poor households received some nutritional benefit monetary from one of the many nutritional programs. poup OMWe . benw6t The biggest of these programs are the 'Glass of Milk Program' (Ministry of Finance, malnutr.& poor 66.3 38.0 working through municipalities), the School malnour. & non-poor 43.3 22.3 Breakfast Program (administered by non-malnuri & poor 47.0 15.9 FONCODES) and the local soup kitchens (comedores populares, largely financed by PRONAA). Table 13 breaks the households (100) in the population in four different groups: by poverty and malnutrition level (in which Source: Staff Estimates based on ENNIV 1997. households with at least one malnourished , child would fall). Based on the ENNIV 1997, Table 14: it appears that coverage rates show relatively Distribution ofmalnourished children and good targeting with the households being both distribution ofnutrition transfers, 1997 poor and having at least one malnourished infant showing a coverage rate of 66 percent. &stributon of the dstibution of The non-poor households with well-fed mabiourished program exp. infants show a much lower coverage rate, of 23.7 percent. However, the 'progressiveness' LUna 8.9 31.6 of this transfer is considerably smaller if we Urb. Coast 6.9 8.8 calculate the net monetary transfer equivalent Rur. Coast 5 1 9.6 (taking into account frequency and quantity of Urb. Sierra 7.7 5.3 -.Rur. Sierra 51.3 31.9 the transfer). Minimum leakage of the ur. Junrle 5.1 4.4 Urb. jungle 5.1 4.4 nutritional programs taken together (which go Rural Jungle 15.0 8.4 neither to the poor nor the households with malnourished children) is about 24 percent. (100) (100) Looking at the distribution of benefits on a . - Source. Staff Estimates based on ENNIV 1997. regional basis, the rural Sierra receives and 1997. considerably less transfers in 1997 than it 'should' given that about 51 percent of all malnourished children live there (Table 14). Lima receives a much larger share than it would were expenditures distributed purely on a geographical basis in accordance with the prevalence of malnutrition. As can be seen from the coverage rates, Peru's nutrition programs are very large. According to the Budget Commission, spending on the nutritional programs of PRONAA, PERu: PovERTY CompARIoNs -- A PoLicy NoTE 2 Vaso de Leche, of the Ministry of Education and FONCODES alone increased from 190 $US million in 1994 to about $250 million in 1997. But many more programs, administered centrally and by municipalities, exist, for example PACFO with the Ministry of Health which targets the 5 poorest departments in the central Sierra. One of the biggest problems in the sector is that the Ministry of Women and Human Development is formally responsible for diagnosing the nutrition situation and establishing nutrition policies, program objectives, implementation norms and standards. But the Ministry has only very limited authority over many policies and programs. Many nutritional programs have no nutritional objectives per se but are rather geared towards income generation or school attendance. Youth. The third age group we find at an increasing relative risk of poverty are young table 15. adoesens.While they were about 5 percent Unemployment Rates by gender, 1992-19%6, adolescents. WMetropolitan Lima more likely to be poor than all other age groups in 1994, they were 8.5 percent more likely to be 1993 1994 1995 1996 poor in 1997 (Table 9). For two more reasons are young adolescents a group requiring male . .14-18 yrs 19.4 12.1 14.9 18.2 particular concern, especially in urban areas: first, it is the only age group for which total male 10.0 9.0 7.1 7.2 unemployment rates remain very high -- for all female other age groups open unemployment dropped 14-18 yrs 21.0 11.8 11.4 13.9 after the recovery started at the beginning of the total female 12.4 12.0 8.7 8.5 1 990s . As shown in Table 15, unemployment Source: Encuesta de Hogares, Ministry of Labor, rates increased substantially for both female and and Saavedra (1998). male adolescents between 1995 and 1996 in Lima. Second, the young are also increasingly involved in violent acts. A recent survey from the statistical Institute INEI shows that young groups are made responsible for 90 percent of acts of vandalism and about 25 percent of acts of physical violence. Landless. Anthropological studies geneallypoin to he rralGraph 8: Secondary School Enrollment, by Age, 1997 generally point to the rural landless population as being 0.8 a particularly poor group. According to the household 06 surveys, however, while the female landless rural households 04 were slightly more likely to male be poor than the rural 02 households owning land in 1994 were, this relative risk a has disappeared in the past 1 0 15 20 several years (Table 9). age Sociological studies point Source Statf Estimates based on ENNIV 1997 out, though, that the impact of landlessness might not show up in official data as PERu:- PovERTY COMPARiSONS--A PoLicy NoTE ........... ............................28 households lead by the elderly or widows often do not have the physical strength to work their own land after the previous head of the household migrated or died. These would be de facto landless households (although not titled as such) as they do not derive income from farming their land. We tested this assertion by computing the relative risk of rural widow-headed households to be poor (Table 9). It appears that these households, astonishingly, fare better in relative terms to other rural households and that there relative risk of poverty decreased over the years. Gender. Viewed from many different angles, gender gaps in Peru seem to be closing. First, households with female heads -- given its limitation as an indicator -- have fared better than male headed households over the last years. Their relative risk of poverty is considerably lower and continues to decline. Similarly, when examining which group of the same households did well over the three years, we found that female headship -- while controlling for all other factors -- had a strong positive influence on per capita expenditures (Box 1). Second, labor market discrimination (i.e. women earning less than men although they have the same education, experience and age) have disappeared in the urban formal and informal sector. The only labor market in which women continue to be disadvantaged is rural non- farm employment.23 Third, malnutrition rates are not equal yet between boys and girls but the gap is slowly narrowing as well. Finally, regarding education, of most concern in the past was secondary school enrollment of girls. Graph 8 shows the percentage of secondary school attendance by age group of the population and also confirms that, at least in terms of school attendance, there is almost no difference between boys and girls any more. Participation rates in the labor force do remain considerably lower for women than for men. Increases, however, are faster for women. Briefly looking at the Table 16. role of women in high levels of Women in High Levels ofPower and Decision Making power and decision making, seats held in administrators & prof. & tech. women account for about 11 parhament managers workers percent of seats in parliament, (o women) (o women) (%/ women) 20 percent of administrators and 40 percent of professional Bolivia 6.4 16.8 41.9 and technical workers. Brazil 6.7 17.3 57.2 Although still far behind their Colombia 9.8 27.2 41.8 men, women's role in politics Costa Rica 15.8 21.1 44.9 and in management is about Mexico 13.9 20.0 43.6 Peru 10.8 20.0 41.1 comparable to its neighbors Uruguay 6.9 25.3 62.6 (Table 16). Peru does lag Venezuela 6.3 17.6 55.2 behind other Latin American countries regarding the role of women as profession technical Source: United Nations (1997a). workers, though. Also -- and as 21 See, for example, Luerssen (1994). 22 Saavedra and Chong (1997). 23 Lopez (1998). PERu: PovERTY ComPARisoNs -- A PoLrcyNoTE 29 ?........ .m ........ lso.... s........r..Y.. N.... T.. .......................................................................... . _ _ 29 explored below in more detail -- a new 'face' of urban poverty, social violence, affects especially women. Migrants. Based on the relative risk of being poor, migrant families appear to be integrating well into their new environment. Such families were at 16 percent lower risk of being poor in 1994 and this relative risk dropped to 18 percent in 1997 (Table 9). While most rural to urban migrants state that they have moved place of residence for income and employment reasons, their educational level tends to be higher than for non-migrant families, explaining their relatively good economic integration in cities they migrated to.24 Also, migration is one of the strongest integrating and assimilating factors of the rural and often indigenous countryside with the generally mestizo cities because migrants maintain their rural links and community networks.25 Internally Displaced. One group often considered as a very high risk of deprivation is the internally displaced people who had to leave their rural residence for reasons of political violence. The United Nations estimates that still today, years after the marked decline of the incidence of political violence, about one half million Peruvians are internally displaced. These are especially rural-rural migrants who have had to leave their house, land and family networks and have not (yet) returned to their original place of residence. To help families resettle to their homes, the Peruvian Government created the Proyecto de Apoyo al Reboplamineto y Desarrollo de Zonas de Emergencia. in 1993. Until the year 2000, the program plans to have helped a total of one million internally displaced persons. 4.2. Poverty Comparisons -- Some Key Factors of Welfare Changes We now turn from a focus on groups to a recap of factors that underlie poverty changes. Obviously, these two categories are often intrinsically interlinked. The native speaking population, for example, is at a higher relative risk of poverty than the Spanish- speaking population, partly because their education attainments are lower, their access to markets is scarcer and their dependencies ratios are higher. The factors analyzed below, however, are significant independently -- or better in addition -- to looking at different groups in society. Again, their selection is mainly based on examining hundreds of identical households over time and relating the change in their welfare level (consumption per capita) growth to household and individual member characteristics. Table 17 reports the again relatim risks of poverty for such factors. As above, the interpretation of the data is relative to the rest of the population -- e.g. people living in households with a dependency ratio bigger than 4 were twenty-two percent more likely to be poor in 1994 than the population in households with a lower dependency ratio. Houses Used for Business Purposes. In both 1994 and 1997, Peruvians living in households that can use at least part of their house for some income-activity are almost thirty 24 See Escobal et al. (1998) and White (1995). 25 Altamirano (1988). 26 Unted Nations (1997b). PERU POER TY.m CMAIgSONS rn A POLJCY NOTEq ---------------------------------- percent less likely to be poor 'Table 17. Poverty Risks ofSelected Factors than the population not - More or Less L ikel to be Poor? having such an incomepouainerntgs possibility. And this was one of the very robust and National Percent of Total, strong factors, which we 1994; 1997 ofT oor,1997 found helped families advance over the last years houieholds using'house ,29.0 ) 1 9 (Annex 1). Such businesses forbusinesspuroses can obviously be formal rural housel6I8s wfatleaid -23.0 -2.0 [2 (e.g., a formal store in the ne;snembi Uiff house, professionals ouoe.. working from home) but in or partner of head iworldng the large majority of cases households w/o water and F537: +48.4 5.7) such businesses are sanitation informal. Renting out a room, performing home- households w/o electricity +63.0 +69.5 [3761 based piecework, selling households whead,less +73.0 +73.0 161.3 merchandise out of the thansecondary edusato house or small houAseholdsof7pesons and +71.0 4990 [3171 manufacturing on a sub- larger contract basis would fall :__ under this category. 1 spouse working is defined in remunerated work in the last seven days before the survey was conducted. To the degree that S : Staff estimates based on ENNIV (1997). legal home ownership is related to the home-use for business purposes, than the current drive to providing titles for urban dwelling in Peru can indeed be of major importance for the poor. Home ownership in Peru is very large with about three out of four families living in their own houses. This, interestingly, does not vary much with the poverty level of families. However, many of the poor, especially in Lima, who invaded their land in the big migration flow at the end of the 1980s and beginning of this decade, do not posses legal title to their property. If, as found in several qualitative and quantitative studies,27 legal home ownership encourages home investment and this raises the likelihood of using the home as an income-generating asset, the current urban property drive can have very positive poverty effects. The finding that such informal businesses can be important for helping families advance is in line with a rather positive, or at least neutral, view of the informal sector in Peru. Informality remains huge, accounting for almost half of urban employment. However, for the self-employed opening micro-enterprises, informal sector employment in Peru is largely a choice. People are not 'forced' into informality by distortionary policies or labor market practices. Those policies have been dismantled to a very large extent at the beginning of the 1990s. The main motive to remain in the informal sector in self-employment is 27 See Moser (1996) and Persaud (1992) PER u: Po vERTY CompPisoNs - A PoLicyNoTE 31 entrepreneurial skill.28 For informal wage earners the picture is different. Here a segmentation still exists -- i.e., given their education, experience and other individual characteristics, wage earners in the informal sector are earning less than they would in the formal sector. Such workers tend to be young, single and non-heads of households. Hence, it is suggested that they lack experience and are in a 'waiting position' to find formal sector employment. Confirming this view of the informal sector in Peru, households whose head was employed in the informal sector had a btgber per capita growth rate of consumption than household heads working in the formal sector even when we control for other factors. Services. Households without basic services like water, electricity, sanitation and telephones are at a much higher relative risk to be poor than households who command these services. And, as shown in Table 17, for some services like electricity, this relative risk has improved over the past years. Obviously, we will find a mutual dependency between poverty level and services access: poorer households tend to live in poorer neighborhoods so that the likelihood of them having access is low. Simply stating that access is lower does not imply that there exists a causal link. However, in the panel analysis of identical households, we found very strong evidence that households that had access to basic services in 1994 had a significantly higher growth rate of per capita consumption than households that did not have such access. Many reasons for this can be found, such as a positive impact on health through clean water supply and sanitation services or the importance for home enterprises of electricity and phone connections. Furthermore, the 'bundling' of services is very important for households. The additional, positive impact of one new service increases with the total number of services available. Based on the analysis of the household panel, we show in Graph 9 that adding a fourth service has about seven times a higher additional impact than linking a second service to households. The logic behind this is the joint Bundling Services: Increasing Retusu (increase in per capita provision of services is growth rate, 1994-i"7) important to realize welfare effects: for example, clean 0 s water access will improve a household's well being more if it comes logether with sanitation. Health risks might decline with 00 water only access but the real 01:13 benefit might only be . A j materialized if the services are second service third service fourth service provided together. The same Source. Staff Estimates based on ENNIV (1994 and 1997). logic holds for all services considered here (sanitation, electricity, water, telephone). The positive impact of such bundled, integrated interventions has recently also been observed relating to other projects in Peru. In a recent study, electrification and sanitation services were found to increase the returns to education significantly in rural and urban Peru 28 See Saavedra and Chong (1997) and Yamada (1996). PERu: PovERTY ComPARisoNs - A PoLicyNoTE 3 alike -- as children can read and study longer at night, they profit more from the provided schooling. Sanitation is likely to lower illness and malnutrition and have a positive effect on learning possibility. Also, better rural roads and rural transport were shown to have a very positive impact on the returns to rural education. Education. Educational attainment Table 18: remains not only one of the central determining Urban Peru: Educational Preria, factors of poverty levels but it is also one of the 1991 and 1996 (percent) main driving forces explaining who advances rapidly or falls behind in the Peruvian society. 1991 1996 change In Table 17, we see that in 1994 Peruvians living in an household whose head had less than pry/ 40.0 33.0 -7.0 secondary education were seventy percent more no education likely to be poor than the rest of the population. This huge relative risk stayed constant till 1997 secondary/ 7.0 17.0 10.0 but the panel study showed that the higher the primary initial education of the household head, the non-university 13.0 25.0 12.0 higher was per capita consumption growth: the higher/secondary better educated have benefited proportionately more than the less educated in recent years. university/non- 47.0 70.0 33.0 This is in line with many findings in Peru over university higher the past years that not only the returns to Source: Saavedra (1998). education have increased across the board since the beginning of the 1990s but also they have increased more for those with more education. Table 18 shows education premia (i.e. the difference in income levels according to educational attainment) for urban Peru in 1991 and 1996: educational premia have increased strongest for university educated Peruvians. Hence it was one of the driving forces behind the increase in inequality over the past years. On a side note, Peru's returns to education are not below those observed in other Latin America countries. And together with Colombia, Chile, Argentina and Costa Rica, education returns have increased during the structural reforms. Current concern in the education sector rests with enrollment in secondary school, quality and financing. Both primary and secondary school enrollment Graph 10* Secondary school enrollment, rural Peru, rates have increased steadily since the beginning of the decade and gaps in enrollment rates by poverty 07 level have disappeared for school beginners. Differences in school 05 ---7 top 20% attendance remains for secondary X school attendance, especially in rural 04 areas, as Graph 10 shows: for the E 3T poorest group (bottom quintile of 0.2 - bottom 40% the population), enrollment rates are 0.1 ,* still considerably lower than for 0 ./... . . . children from well-off families. 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Quality indicators show that much Source Staff Estimates based age on ENNIV, 1997 PERu:- PovERTY ComPARisoNs- -A PoLicy NoTE 33 remains to be improved in the sector: a large Table 19: difference between net enrollment and gross Peru: Drop-Out Rates in Secondary enrollment rates indicates that many children are School,1994andl997(percent) in primary and secondary school longer than they should be. And drop out rates in secondary quintile Urban Rural school have increased between 1994 and 1997. 1994 1997 1994 1997 Table 19 shows this increase and also the continuing, very strong link between drop-out 1 11.0 22.0 14.0 27.0 rates and poverty levels. 2 8.0 18.0 15.0 22.0 3 6.0 8.0 13.0 9.0 Although having increased by 30 percent 4 4.0 8.0 15.0 11.0 in real terms from 1994 to 1997, education 5 3.0 4.0 15.0 4.0 expenditures are still low in historical perspective Source: Staff estimates based on ENNIV and highly tilted towards the more well-off 1994 and 1997. regions in the country. It is estimated that educational per capita expenditures were roughly 20 percent below their 1970 level in 1997.30 And these per capita expenditures are significantly higher in better-off departments: Graph 11 plots per capita student expenditures by department against the FONCODES poverty index. the poorer the department, the lower educational per student expenditures. Such a spending pattern perpetuates regional and general inequality in Peru. Graph11: Poverty Index and Current Per Student Public Expenditures in Basic Education, by Departamento - 1994 260 240 220 200 180 160 140 120 100 09 14 19 24 29 34 39 44 49 Poverty Index From Saavedra, Melzi and Miranda (1997) Expenditures coms from GRADE's educational database, the poverty index from FONCODES 29 We calculate the drop out rates by looking at a constant age cohort, the 16 to 19-year-olds in both years (1994 and 1997). Since Graph 10 shows that a large percentage of an age cohort still attends secondary school in these ages, the calculated rates are likely to underestimate the true drop-out rates. 30 Saavedra (1998) PERu:- PovERTy CompARIsoNs- A PoicyNoTE4 Dependency ratio and household size. The more dependents live in a household, the less well did the household do. Additionally, we find that household size is negatively linked to welfare developments over time. There has been considerable debate whether the use of 'per capita' consumption or income is a good indicator of household welfare since larger households tend to use more goods in common and might have a quite sizeable benefit from this. We find, however, that independent of such assumptions about 'economies of scale' in household consumption, larger households have fared worse in Peru over the past years. Off-farm employment. Off-farm employment, although still very thin, has the potential to be a road out of poverty in rural areas. We find that the likelihood of being poor is about 24 percent lower for rural families that have at least one member in off-farm activities in 1994 and 1997 (Table 17). Also, in the panel study of households, off-farm employment influenced consumption growth of households strongly and positively. The market, though, which is comprised of trade, small-scale manufacturing or transport remains relatively unimportant in comparison to agricultural employment: in 1994, only about 13 percent of total workday equivalents were devoted to off-farm labor market activities in rural Peru. This is considerably lower than in other countries in the region.? The strong link we found between off-farm activities and welfare improvements of households points to a potential role for these markets in the fight against poverty. Studies from other countries have shown that, besides general economic performance, rural off-farm markets depend heavily on infrastructure, especially rural roads. Urban Violence. Urban violence is one of the biggest preoccupations of the urban poor. Although acts of political violence have been contained since the beginning of this decade, public concern about criminal and social forms of violence -- including robberies, armed attacks and sexual assaults -- has risen sharply in recent times as mirrored by the importance they receive in the media. How important violence, its different forms and its impacts are for the life of the urban population shows the outcome of a ranking exercise. The Ministry of the Presidency had asked 20 poor neighborhoods in the Lima district of Ate32 (with a total of almost 40,000 people) to conduct neighborhood meetings in which main problems of the population would be prioritized, their main causes identified and possible solutions suggested. Each community could specify as many problems as they deemed important. Table 20 records the results if all 20 case studies are pulled together. 31 See Saavedra (1998) and Lanjouw (1998). 32 Ate is a big district in Uma's eastern cone. It has almost 300.000 inhabitants and according to the FONCODES poverty map (denved from the Census), about half of its population has no connection to water or sewerage, 30 percent are without (legal) electricity connections. The malnutrition rate of minors below the age of 6ve, extrapolated using census data, is about 30 percent. About half of the houses were in the 'vivienda precana' category. School assistance rates are very high (95/6) and analfabetism 5.1 percent. These indicators, aggregated into one index used by FONCODES, rank ATE as the tenth poorest district within bma (out of 41). PERu: PovERTY CompARIsoNs --A PoLIcy NoTE 35 Violence was one of the main Table 20: Urban Problems: Ranking Exercise of Twenty problems identified by these twenty Communities in Are, Lima, December1997 urban communities. Fourteen communities specified assaults, Rank# Problem Problem in # of robberies and domestic violence as one of their largest problems. After 1 water & sewerage: not working 17 water and sewerage problems, child 2 child malnutrition 15 malnutrition, and street conditions 3 street condition bad/accidents 15 (with associated high accident rates) , 4 violence (assaults, domestic) 14 ilenc ranked hig h inthe houses: unfinished 14 violence ranked fourth in the environmental pollution and communities' self-assessment of related contagious diseases 14 problems". Further, youth gangs 7 no employment 12 were generally named separately, 8 distance to basic education 9 reflecting that in addition to their 9 no local community aula 8 10 youth gangs 8 often criminal activities, the 11 bad basic health care 7 formation of youth gangs worries 12 insecure land ownership 6 parents and shows that the social 13 no recreational space 5 fabric of communities is threatened -- 14 vaso de leche program not well equipped 3 be it for unemployment or other 15 drugs, alcohol 3 reasons. Source: Ministerio de la Presidencia, Plan de Accion Social, Ate (1997) The communities also discussed the causes of violence and how these could be tackled. Communities have a clear view about causes and possible solutions. As to youth violence, causes are perceived to be economic (youth unemployment) and at the same time lacking supervision and family values which do not provide the youth with a strong 'corset' of accepted behaviors and values. Proposed solutions are practical and restrained to what the community itself can do (possibly with some outside help): building sports facilities to lure the young of the streets, organize parents' training classes, and conducting vocational training courses. With respect to domestic violence, all 'asambleas generales' agreed that the major reason for family violence is a deterioration of respect among family members. Unanimously, the four asambleas who prioritized domestic violence suggested that educational classes be held which would have to be attended by the male household heads -- such classes would discuss basic family values and the rights of women and children. Lastly, the asambleas agreed in large part what causes a high incidence of robberies and assaults: absent (or infrequent and irregular) police controls and neighborhood security committees. Strengthening such security measures is consequently proposed. As an additional datasource, the Peruvian Statistical Institute (INEI) conducted a survey in 1997 that assesses certain acts of violence in Lima. Due to the sensitive nature of the topic, the survey included only certain types of violence and excluded domestic violence in the family. Using simple predication models, we combined this violence survey with the Encuesta 33 These rankings were obtained through general community meetings, specialized focus groups (e g by gender) could have resulted in different results but they were not available. PERu: PovERTY CompARIsoNs- A PomIcyNoTE 36 Nacional de Hogares, also from INEI (1996) and were able to impute consumption levels for each household which allowed us to look at the incidence of violence by consumption decile.34 The overall rate of violence 40 Graph 12: Incidence of Violence by 16 is very high and certain types of Consumption Level, percentey violence levels are linked to poverty 2 .5-o level. Overall, more than one third < a i / of the population in Lima were victim or witnessed a violent act in " 1997. As Graph 12 shows, W3 robberies (in the street, house or of 2 L20-- the car) were more prevalent -- as 2 could be expected -- for richer Lima residents but still significant for the C..siif.o. Dec - poor. About 23 percent of the loet fou de s werentim or Source: Staff Estimates based on ENHOV (1997) lowest four deciles were victims or witnessed such violence. The and ENNAHO (1996), INEI. incidence of physical aggression was significantly lower than that for robbery but concentrated among the poor: the risk of being exposed to physical aggression in the poorest decile was about double the risk than that in the richest decile. The INEI survey reveals that in about half of the cases of physical violence the perpetrator was know to the victim. Finally, it is interesting to note that about 90 percent of all violent acts were not reported to the police, in one quarter of all cases for lack of trust in the authorities. Mirroring the ranking exercise of the neighborhood communities in Ate, 25 Graph 13: Security Feeling in less than ten percent of the poor felt safe eve rnt 1 )Consumption in their neighborhood. Graph 13 depicts 20 the percentage of the population in Lima who states that they feel secure in their neighborhood. The security feeling is a io clearly linked to poverty levels with about a four times as many Limenios in the richest 5. decile feeling secure than in the poorest decile. This extremely high insecurity 2 3 4 5 6 7 8 9 10 Total feeling of the extreme poor limits their Source: Staff Estimates based on ENHOV (1997) mobility and with that both their social and ENNAHO (1996), INEI. interactions and earnings possibilities. 34 We imputed household consumption using the following procedure. First, we selected all variables which were both in the violence survey and the Encuesta Nacional de Hogares (1996). Second, we derived simple econometric models in which household consumption is a function of occupational status of household members, household size, access to services and education attainment of household members. Third, we used these models to impute household consumption. Finally, we derive consumption deciles by ranking households accordig to their per capita consumption level and using expansion factors supplied in the survey as weights. PERu:- PovERTY CompARIsoNs - A PoLIcyNoTE 37 5. Growth and Employment One of the biggest concerns in the Peruvian public debate on poverty is whether growth has created employment and whether this has lead to poverty reduction. This section examines this question. It finds that, yes, growth over the past years has indeed created employment, about 1.3 million more people have been in remunerated employment in 1997 compared to 1994. Many of these new jobs are informal jobs so workers. Worrying trends, which can explain at least part of the negative public sentiment, are that productivity does not seem to pick up and that, consequently, real wages are flat at best. The major impact of growth on poverty reduction has thus been through employment creation and not through real wage increases. The section starts with general labor market trends and then analyzes the link between growth, poverty and employment. We then present a number of different simulations about future poverty reduction possibilities, taking into account both possible varying regional and sectoral growth patterns. Labor Market Trends. In the years since Table 21. Labor Force Participation 1994, 1.3 million new jobs were created in Peru. Rates, 1994 and 1997 People finding jobs were to a very small percentage male female the unemployed but many more were newcomers 1994 1997 1994 1997 on the labor market. The participation rate in Peru, already on the rise since the beginning of the urban 75.6 79.9 45.2 53.1 decade, has again strongly increased. Table 21 rural 91.4 91.3 64.9 72.9 shows that participation rates for men increased by total 80.7 83.0 51.2 59.0 2.3 percent and for women by almost 7 percent between 1994 and 1997. Source: Staff Estimates based on ENNIV, 1994-1997. New jobs were mainly created in the informal sector and the urban sector of the Table 22 RemuneratedJob Creation, economy. Using a 'legalistic' definition of by Formality, 1994-1997 ('000) formality, the increase in formal sector f informal TOTAL employment was slightly less than one half million while informal employment grew by more than 800,000. (Table 22). However, 'informality' urban 430 585 1015 generally refers to the urban sector only since rural rural 45 235 280 employment -- small-scale agriculture -- is by its TOTAL 475 820 1295 very nature informal. But even for the urban sector alone , the majority of jobs were created in informal Source: Staff Estimates based on ENNIV, employment and that share of the informal sector 1994-1997. increased slightly over the past years. As pointed out above informality should not be equated with 'bad' jobs as they can offer for many of the poor a route out of poverty. The legalistic definition defines the formal labor market as comprised of all wage-earners or the self- employed who pay taxes, are insured with the IPSS, have a signed contract, have rights to vacation or belong to a union. See Saavedra and Chong (1997). PERu:- PovERTY ComPARISONs --A PoLicyNoTE 38 P y. Po.......s .C.... ar.....o ............Pa cr ..g.............................................................. ..... ...._ _ _3. Growth Pattern, Poverty Reduction and Sectoral Employment Growth. Employment growth was closely linked to poverty reduction. Table 23 looks at severe poverty rates and employment growth for the different sectors in the Peruvian economy.36 As can be detected, the three sectors with the highest employment growth rates (construction, trade and commerce, and services) are also the three sectors, which achieved the highest percentage decrease in poverty. Much of this employment growth provided families with a second source of income. Similarly, agriculture and mining/manufacturing had the lowest employment growth rates and also showed the lowest percentage reduction in the severe poverty rate. Sectoral growth rates and employment creation are connected. As Table 23 reports (column 4), the 'push' sectors in Peru over the last years were agriculture, construction and trade. However, these real growth rates will capture only output of formal enterprises and would not necessarily account for many informal economic activity. Nevertheless, the two can be thought of to be closely linked - if formal sector growth is high in a specific sector, supporting or parallel informal enterprises should also realize an upswing. On face value, Peru's growth pattern was 'pro-poor' over the time period considering that it was driven by the sectors in which severe poverty rates were highest (columns 1 and 2). And in construction and trade real growth translated into employment growth and poverty reduction. Table 23: Sectoral Poverty Reducdon and Growth Rates, 1994-1997 Sectors Severe Poverry Rate perc. change Employment Real Growth Distribution of 1994 1997 1994-1997 Growth (formal,94-97) Severe Poor, 1997 (1) (2) (3=2/1) (4) (5) (6) Agriculture & 31.8 26.4 -17.0 10.3 23.4 (30.4) forestry Construction 25.2 17.4 -31.0 63.9 33.8 (7.2) Transport and 11.8 10.2 -13.0 18.0 n.a. (7.8) Communications Trade and Commerce 13.8 8 6 -37.5 43.9 22.8 (18.4) Mining, Petrol & 9.2 8.4 - 8.5 7.9 13.7 (12.7) Manufacturing Services 11.9 8.8 -26.0 21.6 8.4 (23.5) TOTAL COUNTRY 18.8 14.8 -21.0 19.0 ???? 100 Source: Staff Estimates based on ENNIV 1997. AD households have been assigned a 'pnmary' sector, e.g. the sector of the main income earner. Real growth rate from Central Bank of Peru (1998). 36 To link sectors of the economy and poverty, we 'assign' a household to a sector based on the primary occupation of the household head. This leaves only about 80 households unaccounted (non-active household heads) which we neglect in the table and the calculations. PERu: PovERTY CompARisoNs- A PoLA'cyNoTE 39 But one of the key factors explaining inequality and poverty developments over the past years in Peru is that the impressive agricultural growth rates did not translate fully into employment creation. Real growth rates of the sector are estimated at 23 percent over the 1994-1997 period (column 4 in Table 23); after construction the best performing sector. Growth has especially been strong in non-traditional exports. Agricultural productivity was seriously depressed at the beginning of the 1990s so that one can expect growth to be generated in large part by the existing work force working longer hours. This would be porewosi elainger for. Threivulow bGraph 14: Reduction in Severe Poverty Rate for Varying one possible explanation for the relatively low Growth Rates, percentage change, five years growth elasticity of employment generation of agricultural growth. This explains to a large extent the growing regional inequality in Peru 4. as well as slower social progress in rural Peru . can be explained. Growth and Poverty Reduction: Simulations. A number of simulations show . -.5. how important continued growth is for .,, poverty reduction but also how unequal, anti- r p c g r poor growth can lower or eradicate the source: Staff estimates based on ENNIV (1997) potential benefit from economic expansion. Table 24: Simulation ofSevere Poverty Growth will remain the backbone of any Reduction: DifferentAssumption About successful poverty reduction strategy in Peru. Inequality Using a simple simulation in which we (real gro nib rate of3 percent forfive year) distribute the gains from growth completely equally in society,37 Graph 14 shows that Peru Simulation Severe Poverty could reduce severe poverty by a further 25 Reduction (percent) percent in the coming five years if it were to achieve a real per capita growth rate of 3%. Inequality Constant -23.0 Higher growth rates would mean faster severe poverty reduction -- a seven per cent real per Inequality Increase' 0 capita growth rate would halve severe poverty reduction in five years. However, as we have Inequality Decrease2 -62.0 seen over the last three years, inequality cannot Source. Staff Estimates based on ENNIV 1997. be assumed constant per se. If the trend of 1 an increase in inequality implies that the increasing inequality would continue and the nchest 20 percent of the population in crease society would become considerably more their share of total consumption from 43 to unequal, growth might not translate into 50 percent. poverty reduction at all : if the richest 20 2 a decrease in inequality means here that the e opoorest 40 percent of the population increase percent of the population increase their their share of total consumption from 20 to consumption by 10 percent while the economy 25 percent. This assumes that incomes (and consumption) are related to the overall growth rate in the economy through productivity (and wage) increases or through the creation of new employment for secondary work or additional income earners. would grow at 3 percent, the rich would simply reap all the benefits from growth. Severe poverty would not fall. In the reverse case, if inequality falls and the poorest 40 percent of the population would increase their share of total consumption from 20 to 25 percent, severe poverty could be reduced by over sixty percent. The pattern of growth matters Table 25: Simulation ofSevere Poverty Reduction: for poverty reduction. Again, these Different Sectoral Growth Rates simulations are highly stylized as they (realgrowth rate of3percent for five years) assume that sectoral growth translates directly into growth of household Simulation Sectors Severe Poverty consumption via additional employment Reduction and real wage changes (and we saw above that for the agricultural sector this relationship did not hold from 1994 to growth of high- agriculture, -49.2 1997). Further, these simulations poverty sectors construction assume no 'feed-back' effects -- for growth of medium- mining, petroleum, -27.7 example, that growth in export sectors poverty sectors manufacturing, trade would lead to technological spillovers as transport, commun. generally found. However, the . . growth of low- services -24.3 simulations do provide an interesting poverty sectors comparison of how different patterns of growth matter for the poor. Table 25 Source: Staff Estimates based on ENNIV 1997. The recaps the findings. We find that if simulations assume that the growth rate of high growth would be 'pro-poor', that is growth sectors as 6 percent with all other sectors concentrated in' agriculture and growing equally at the residual growth rate. construction, severe poverty would be reduced by more than half in five years Table 26 Simulation ofSevere Poverty Reduction: with an overall annual per capita growth Different Regional Growth Rates of 3 percent. On the other hand, if (realgrowth rate of3percent for five years) growth were to be concentrated in services, the impact on severe poverty Simulation Severe Poverty would only twenty-five percent. Reduction Closely linked to this strong La22.3 impact of sectoral growth path is the impact of different regional patterns of Other urban areas -26.4 development. Severe poverty reduction would be strongest if the rural sectors Rural areas -47.0 (with all their activities from on-farm agricultural and off-farm activity alike) Source: Staff Estimates based on ENNIV 1997. The Peruvian simulations assume that the growth rate of high were to carry Peconomic growth region is 6 percent with all other regions progress in the coming years. Table 26 growing equally at residual calculated so that overall shows the results for this calculation. real growth rate is 3%. The severe poverty decline would be similar to the one observed with agricultural & construction growth: almost fifty percent. If Lima would continue its role of the engine of growth it has played in the past, severe poverty reduction could be much smaller, at around 22 percent. P R :P V R YC M A IO S-A POLICY'NOTE .......................................41 The task ahead: raising productivity and real incomes. One of the major tasks in the coming years lies with raising productivity and with it real incomes in the economy. Till now, employment generation has not been accompanied by real income increases. For Lima, Graphs 15 and 16 shows real income developments in both the formal and informal sector since 1986. As can be clearly seen, the hyperinflation contracted real wage income enormously in 1990 and the recovery period of the economy went hand in hand with increasing real wages. However, with very few exceptions, real incomes have remained flat since 1991 and some even show declining trends. This trend is matched very closely by overall productivity developments in the country.38 In terms of levels, real wages remain -- independent of sector and type of occupation repressed below their level twelve years ago. The Peruvian experience of real income and productivity developments is not atypical for countries having undergone structural reforms. In Brazil, wages increased only slightly since the liberalization in 1991. In Chile, real wages were basically stagnant for about ten years after the recession of 1982 and the market oriented reforms. Nevertheless, the more the government can support the private sector to raise productivity, the quicker will real income changes filter through the economy: supporting public infrastructure investment, training support and education of the next generation of wage-earners can all contribute to this. Graph iS M*tropolitain Umas Monthly real income of informal workerl', 1986-1996 90000 / :1 1 ""0renepse 700 00 /Dm- Sel kmpoye 0.0 . Unpsid flimfly worker 400 00 30000 20000 /---- -- 1oooo ... --.. . . 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 Year Graph 16 Metropolitan Lima Monthly real income of formal workers, 1986-1996 140. 1200 Y .00 .. ... ... o 600 A.- ---- -. 400 /-Tot1 Se11onployed p.ooe onels& .. ch... ssn *0-. d wo - a 6.49 olkc fiam, 200 --- 5sIsisewoe n 5099ke-e fim *. .aedwoer-nl 00+ wokahm 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 Year Socll Ences.. de Hog... d. MTPS 1986 1996 I Tr.d.t-on.a defin.on 38 Saavedra (1998). PERu.- PovERTy ComPARisoNs -- A PoLrcyNoTE ................................42 6. Social Expenditures -- What and for Whom? To complement the analysis of poverty comparisons, we have one major question still to answer: what was the role of public programs in poverty reduction over the past years? Did they help to reduce poverty? Or were they in the end of little value to the recipients? This section will present available material to this end -- but we will fall short of establishing a clear link between public programs and poverty reduction. To shed light on this topic, we would need detailed information on households before and after an intervention, e.g. the well-being of a household before and after it receives nutritional aid through the 'Glass of Milk' program. And we would need a 'control group', i.e. families that are similar in their characteristics but did not benefit from the 'Vaso de Leche'. We had hoped that the panel data, containing information about identical households over time, would serve us to this extent but -- unfortunately, we were not able to clearly establish which specific programs had what kind of effect on household welfare, partly because the sample size was severely limited. The only clear result we obtained and reported earlier regarded the provision of public services in water, sanitation and electricity: these raised families welfare and, in addition, had a 'bundling' effect - - three services having more than three times the effect of one service. However, since literally dozens of programs finance this type of infrastructure we could not distinguish which program was successful and which ones not. What we can do in this section, though, are two things: to take an 'aggregate' view of social expenditures and anti-poverty programs and assess whom they benefited and how many people they reached -- without judging whether they were 'successful' or not. For this analysis we used the Encuesta Nacional de Hogares (1996) from the Statistical Institue of Peru (INE1), a survey considerably larger (20,000 households) than that of the Instituto Cuanto. Second, we can analyze what the short-term impact on poverty direct transfer programs have, i.e. those that provide direct nutritional support, employment, or income transfers. 6.1. Social Expenditures in 1996 -- Who was Reached and How Many? Before presenting our results on the distribution of aggregate social expenditures, one cautious remark needs to be made: figures presented here relate to the 'average' incidence of program expenditures, i.e. 'what percentage reached which group in the population'. Basing policy decisions on such an average incidence might be misleading as the distribution of marginal expenditures might be very different. Or in other words, a program might benefit largely the non-poor in society at a given point in time. However, the additional budget might go directly to the poor. A policy decision on the average performance might not be wise. Such a gap between the distribution of average and additional expenditures is likely to be high for programs that have a large part of their current budget linked to past investments (e.g. education or health programs). The difference will not be that pronounced for programs that finance short-maturation projects and then move on to different sites such as common under the Social Investment Fund. Aggregate Distribution. We examine expenditures in the education and health sectors, in housing and infrastructure programs, and in a number of specialized anti-poverty 394 programs. Together, these programs accounted for a total expenditure of 7.6 billion soles or roughly 40 Analysis (1996), bilon sales percent of the total public budget. Of those about 55 percent were in the education sector, 25 percent in health, twelve percent in housing and basic Education 4.1 infrastructure programs and 8 percent in the anti- Health 1.9 ro s Housing & Infrastructure 1.0 poverty pgram. Anti-Poverty Programs 0.6 Looking at first at the general distribution of Total 7.6 expenditures, the first observation is that, in 1996, they went largely to the urban areas. We estimate that about Source: Comision de Presupuesto del 70 percent of the examined expenditure went to urban Congreso, INEI, Saavedra (1998) Peru and this is likely to be an underestimate as we assume -- as is common in incidence analyses -- that the per capita benefit of beneficianies is Table 28: Rural/Urban Distribution ofExpenditures, equal in the country. It is all too well 1996 known, however, that per beneficiary expenditures in health and education sector share of exp. are much lower in rural areas and we received by rural also showed evidence for this in residents Graph 11 above (distribution of education expenditures by Basic education education 47 departments). Table 28 reports that secondary education education 19 only one program or line of activity university education education 6 bai elhhealth 50 spent the majority of its expenditure basic health health 16 hospital carehelh1 in rural areas in 1996: the social fund Foncodes various 68 FONCODES which had formulated Pronaa nutrition 44 an explicit strategy to target the rural Inabif children 2 poor at the beginning of the year. It Fonavi elec & water 20 is likely that FONCODES increased Enace housing 10 Banco Materiales housing 11 its total budget allocation to rural Infes education 16 areas in 1997 even further given its targeting strategy. It is therefore a TOTAL 30 clear exception to the largely urban- based other social programs. memo: share of poor in rural (1997) 47 share of severe poor in rural (1997) 58 Source. Staff estimates based on ENAHO (1996). The anti-poverty programs include FONCODES, PRONAA, COOPOP, INABIF and INFES The housing and infrastructure programs include ENACE, Banco-Matenales and Ute-FONAVI PERU PO E T CO P RS N - PoLicyNoTE ........... ............................44 Aggregate Social Expenditures Distribution. Table 29: Aggregate Distibution of Examining the total budget of 7.6 billion soles Socia Expenditures considered here, expenditures were mildly tilted b towards the better off in society. Only about 17 qhta oftotalexpend.ra percent of expenditures went to the poorest twenty percent in the Peruvian society. It appears as if 1 (poorest) 16.6 expenditure distribution was largely driven by the 2 18.6 population distribution rather than by poverty level 3 21.2 as was already observed above when we looked at the 4 22.4 distribution of total expenditures by urban and rural 5 (wealthiest) 21.1 area -- there, as well as by poverty group, we cannot Source: Staff Estimates based on ENAHO (on such an aggregate level) detect the targeting of (1996). expenditures to the weakest in society. Given that the budget we look at here is the social and anti- poverty budget, this is a disappointing result. Education and Health Expenditure Distribution. As Graph 17: Education Expenditure by Population observed in many countries, Quintile, 1996 Peruvian expenditures in basic education and basic health were 100 progressive in 1996 while higher 80 Primary .. level spending went in its majority 6. to the better of in society. Graphs 17 and 18 present education and 40 . c , health Lorenz curves for 1996 with 20 *Universit the horizontal axis representing the 0 population distribution and the 0 1 2 3 4 5 vertical axis the distribution of population quintile Source: ENNAHO (1996), INEI. expenditures. As can be seen primary education expenditures wre considraly expogres Graph 18 : Health Expenditure by Population were considerably more progressive Quinie, 1996 than secondary and higher education expenditures. For the latter, about half of total 100 expenditures supported the richest 80 Primary Care '. / twenty percent in society. Similarly, 60 hospital care (both ambulatory and . for stationary) was considerably 40 1 ol- more regressive than primary health 20 Hospitals care -- thirty percent of resources . a went to the top population quintile. 0 0 1 2 3 4 5 population quintile Source: ENNAHO (1996), INEI. PERu: PovERTy compARIsoNs --ARP(?mCYNOZTE ...................................... 45 Coverage and Targeting Rates. While education and health programs, by their very nature are universal programs, the other programs and projects looked at here intend to reach specific groups in society. FONCODES, for example, aimed at reaching the extreme poor in the rural areas in 1996 or PRONAA wanted to reach poor families with malnourished children. To assess how well these programs did in 1996, we want to measure their success using two indicators. The first is the coverage rate. This is simply the percentage of the poor population reached. The second is the targeting performance, i.e. the percentage of total expenditures that actually went to the intended beneficiaries and did not 'escape' to better of groups. It should be noted here, though, that several of the programs considered here, especially the housing and basic infrastructure programs did not have the articulated aim to reach the poor in society. Nevertheless, we want to hold them against the above measures. We find that FONCODES and Graph 19: Coverage and Targeting in Peru, 1996 PRONAA have the best record in reaching their 60 beneficiaries and targeting E 5o A their expenditures. Graph PonFonay an 19 shows both of these I A Foncodes .30 A indicators together. Against A COOPOP the horizontal axis we plot BanM&t the coverage rate, i.e. the 10 - NACE percentage of the poorest 0 forty percent reached. 0 2 4 6 8 10 12 14 16 Against the horizontal axis coverage race of bottom 40/ we measure the concentration of expenditures in the same two quintiles -- our targeting indicator. A program that reaches a substantial portion of the population (i.e. a high coverage rate) and at the same time manages to concentrate its resources in the poorest two quintiles would get an entry in the upper right corner. On the other hand, programs that reach few of the poor and concentrate only a small percentage of their total resources on them would end up in the lower left-hand corner. As can be seen in the Graph, coverage rates for most programs were relatively small -- mostly below 5 percent. And concentration shares were also relatively low; programs spent mostly less than forty percent in the bottom two quintiles. The social fund FONCODES and PRONAA are exceptions here -- both of them show a progressive distribution of expenditures and a relatively large coverage rate. Reasons. Many studies have analyzed the quite meager coverage and targeting performance of the social programs described in this section. These are (a) little use of (existing) poverty maps (except for FONCODES and PRONAA); (b) an urban tilt in expenditure distribution which means that many of the extreme poor in rural areas cannot be reached; and (c) criteria for program access which exclude many of the poor (especially in the housing credit programs ENACE and Banmat). For both nutrition programs and programs of selected ministries, like the Ministry of the Presidency, it has been repeatedly observed that many programs overlap in functions, are not centrally coordinated and are subject to discrete PERu. PovERTY ComPARIsoNs -- A Po...........................O.........E... 46 expenditure decisions.40 Targeting maps, if used, have been found useful for expenditure allocation. However, the most widely used map of FONCODES was found to contain a serious error. The map is based on the construction of a poverty index which is itself an aggregate of several indicators. As a consequence of the aggregation procedure, the indicator that was supposed to have the highest weight in the poverty index, malnutrition, was assigned a much lower importance -- fifteen instead of fifty percent. The World Bank (1996) found that forty percent of the poverty index is actually determined by the roof type of households. Similarly, recent analytical work also questions the poverty map now used by the Ministry of the Presidency in the 'Lucha Contre la Pobreza'. 41 Because this map uses the number ofthe poor per district as one of the key variables determining expenditure allocation instead of the poverty rate or poverty depth, it heavily tilts resources to larger districts in urban areas. In fact, the per capita expenditure distribution under this scheme is found to be worse than if resources were allocated on a pure population basis. 6.2. Poverty Impact of Direct Transfers Different from basic and social services programs, direct transfer programs are geared to help poor families in the short run. Public transfer programs in Peru are mainly the large and many nutritional programs but they also consist of the employment programs, e.g. of FONCODES. Further, we can also count pension payments from the National Security Institute as a public transfer. How important are these public transfers for the families? And how important are these public transfers compared to private transfers from friends, neighbors and family members? 40 See, for example, Homedes (1996) and World Bank (1998a) 41 See Schady (1998a). PE,Ru:.PovERTY.COMPARIsoNs -.A PO4.XKN6OTE ........................................4.7 Table 30. Impact ofPrivate and Public Transfers on Poverty and Severe Poverty, 1997 How much higher would poverty have been? poverty rate severe poverty rate urban rural urban rural food aid 1.2 1.7 0.9 3.0 other public 0.3 0.3 0.1 0.5 transfers public pension 0.1 0.1 0.2 0.1 private pension 4.6 1.3 3.5 1.3 or employer benefit private transfer, 4.3 2.7 3.5 3.5 national private transfer, 1.1 0.1 0.8 0.1 international alpubHc combined 1.4 2.4 1.2 3.6 all private combined 9.6 3.9 8.2 5.1 Source: Staff estimates based on ENNIV 1997. We analyze the effect of transfers by estimating the impact they have on poverty rates. In reality, transfers might not increase household welfare in the short run for all families. For example, if food donations replace the purchase of food, family income increases which might be saved, used to repay credits or spent in a way which is not beneficial to the poorest in the family. However, this exercise assumes that all transfers received have in their full amount increased households' consumption. Since we want to compare public with private transfers, we make the same assumption for family aid, private pensions and remittances from home or abroad. Public transfers here do not include education and health expenditures and basic and productive service investments. We find that total public transfers have 'hypothetically' a much lower impact on poverty than private transfers. Table 30 examines the impact on poverty and extreme poverty of a number of different transfers. By far the most important transfer for the poor and extreme poor alike are not public transfers but private national transfers followed by private pensions and employer benefits. Food aid is the most important public transfer and has a quite significant impact on severe poverty in the rural areas but is less significant in urban areas. PERu: PovERTY ComPARisoNs -- A PoLicyNoTE 48 7. Institutions -- From Individual Sectoral Strategies to a Consistent and Broad-Based Anti-Poverty Focus This report does not provide detailed recommendations as how Peru can make further and effective inroads to fight poverty. Such specific policy recommendations have been and will be made by specialized studies. The mainstay of this report has been to take a more global, aggregate view of social progress over the past years and much of the evidence presented here was concerned with the distribution of public investment and social programs. This section, therefore, takes a look at social policy formulation in Peru at a more 'macro' level. Current Social Policy Formulation. Today, the multitude of social policy programs operate largely independently, try to reach their beneficiaries with different means and lack stringent evaluation. Expenditures of many of these programs, although well-intentioned, do not reach the poorest in society and are often isolated in nature. The Ministry of the Presidency alone has six programs in the education sector -- outside and in addition to all those of the Ministry of Education. Nutrition programs are plenty and administered by the Ministries of Finance (Vaso de Leche), Women and Human Development (PRONAA), Ministries of Health (Basic Health project, PACFO), Education, to the Ministry of the Presidency (FONCODES). While a social policy coordination council exists, it has long had no mandate, resources or staff. The Inter-Ministerial Council on Social Affairs (CIAS) has the responsibility to ensure smooth inter-ministerial information flow and guide social policy development. But inter- ministerial meetings have been suspended for almost one year now. This leaves a void in social policy making with each government entity on the one hand side 'freer' to develop its own policies and on the other hand more susceptible to political pressure. Conflicting Decrees. The need for the empowerment of one central social policy unit becomes quite apparent if we look at two important, recently passed presidential decrees. The first decree (012-97-PCM from April 1, 1997) makes the Presidency of the Council of Ministers, and with it the social policy coordination council (CIAS), responsible to coordinate and improve the better allocation of social expenditures between agencies and ministries. The second decree (030-97-PCM from June 20 1997) officially adopted the targeting and coordination strategy of the 'Lucha Contra La Pobreza'. This strategy was developed and is managed by the Ministry of the Presidency. The decree called for the 'widespread application' of the strategy by the whole public sector, making -- de facto -- the Ministry of the Presidency responsible for social policy coordination. Recommendations. We believe that one of the most pressing needs in the fight against poverty in Peru is institutional reform. Then, a much bigger impact could be achieved with available funds. First, a central and powerful social policy council needs to be established which would design poverty reduction strategies in a technical and non-political fashion. As pointed out in section four, the biggest poverty reduction effort is established if interventions are integrated, that is providiig two services jointl has a more positive efect than the sum ofproviding each one separately. Experiences from other countries in Latin America show that such central coordination is possible (Box 2). PR:PVRYCMASNS-APoLICyxNOTE--------------------------------------- 49 Box 2: Muli-Sectoral Social Policy Formuladon in BrazU: The Comunidade Soldaia The Comunidade Solidana (CS) constitutes a direct link between the government and society to identify and address cross-sectoral social problems from outside the sphere of the Federal Government but linked to its policymaking and program mechanisms. The Consultative Council of CS is comprised of 11 ministries of state, the executive secretariat of the CS and 21 representatives from civil society with the aim of mobilizing social efforts, implementing innovative experiences at local levels, and identifying social priorities. The Executive Secretariat of the CS is linked to the Civil Affairs Office of the President of the Republic, with representatives from sectoral ministries, provinces and municipalities, and civil society. Consultations between local organizations, and the municipal and federal government lead to the formulation of a 'Basic Agends' which constitutes an action plan for local anti-poverty programs. Funds provided to the Basic Agenda social programs are in the form of periodic transfers from ministries and federal bodies to state and municipal governments. They have grown from R$980 million in 1995 to R$2.5 billion in 1997, and are predicted to increase even more to RS2.9 billion in 1998. Most funds are designated for the poorest regions in Brazil across social sectors. The total number of municipalities participating has also increased significantly, beginning in 1995 with 302 and in 1997 totaling 1,368. Second, and closely linked to the above, pro-poor policies requires good targeting and thorough, good evaluation. Many different poverty maps and targeting mechanisms are currently used in Peru and they can be harmonized. However, program planning and monitoring goes beyond the need for targeting and prioritization. It includes for policy makers to be able to assess whether a certain interventions did indeed help or not. And it also implies that policy-makers and technicians are able to assess how changes in program nature and how changes in expenditures are distributed and what effect they have. As pointed out in this report, a large part of new electricity, sanitation and water connections went to urban areas in the last three years. The same is true when we look at additional school enrollments or public health visits. One of the central roles of the social policy council has to rest with building an integrated, transparent information and evaluation system. Third, central coordination can go hand in hand with Gph 20- Targeting of Social Programs: (percent of expenditure decentralized execution that per quntile includes other partners in the fight against poverty. Examples from many Latin American countries show that private-voluntary-public HNGOs partnerships in poverty BGenera Public reduction at the local level can be extremely successful. One 1 2 3 4 5 reason for why such u. reason fo why such Source ENNAHO (1997) qitl partnerships are successful is that each organization brings its comparative advantage to the 42 Fiszbein and Lowden (1998) have collected a large number of examples how government, business and civic partnerships have worked successfully for poverty reduction in Latin America. table: central government finance and organization; municipal government administrative and local knowledge; and non-governmental (and community-based) organizations often a good and direct understanding of (and link to) the problems of the poor. For this latter point the Encuesta Nacional de Hogares from INEI (1996) shows NGO-administered programs have a significantly better targeting record than most of the public programs and match the targeting results of FONCODES or PRONAA administered programs (Graph 20). References Altamirano (1988), Cultura Andina y Pobreza Urbana, Pontifica Universidad Catolica del Peru, Lima. Caritas (1997), La Extrema Pobreza en el Area Rural en Peru, Lima. Central Bank of Peru (1998), Boletin Mensual, Lima. Cuanto (1997), Mil Quinientas familias Dos Ailos Despues: La Pobreza en el Peru, 1994-1996, iUma. Davis, Shelton and Harry Patroninos (1996), Investing in Latin America's indigenous peoples: The human and social capital dimensions, Seminar on Indigenous Peoples Production and Trade, Copenhagen, Denmark, 15-17 January 1996. Deininger, Klaus and Lyn Squire (1996), A New Data Set Measuring Income Inequality, World Bank Economic Review 10, pp. 565-591. Encuesta sobre Niveles de Vida (1994, 1997), Instituto Cuanto S.A., Lima. Encuesta Nacional de Hogares (1996), INEI, lima. Encuesta de Hogares Sobre Victimizacion (1997), INEI, lima. Escobal, Javier, Jaime Saavedra and Maximo Torero (1998), Los Activos de los Pobres en el Peru, processed, Lima, GRADE. Fallon, Peter (1998), Dispersion of Sub-National Regional Income per Capita, KMS site Sub-National Economic Policy, World Bank. Ferreira, Francisco and Juhe Litchfield (1998). Fiszbein, Ariel and Pamela Lowden (1998), Working Together for a Change: Government, Business and Civic Partnerships for Poverty Reduction in LAC, Economic Development Institute, World Bank. Francke, Pedro (1996). Francke, Pedro (1997), Realmente Ha Aumentado la Pobreza en los Ultimos Dos Aiios , mimeo, Banco Central de Reserva, Lima. Glewwe, Paul and Gillete Hall (1995), Who is most vulnerable to macroeconomic shocks? Hypotheses tests using panel data from Peru, Living Standard Measurement Study 117, World Bank. Hentschel, Jesko and Peter Lanjouw (1996), Constructing an Indicator of Consumption for the Analysis of Poverty, Living Standard Measurement Discussion Paper Senes xxx, World Bank, Washington D.C. PERU: POVERTY COMPARiN ISON-A PoLIC YNOTE 51 Hicks, Norman and Pia Peeters (1998), 'Social Indicators and Per Capita Income in Latin America', processed, World Bank. Homedes, Nuria (1996), Nutrition Note, processed, World Bank. IMF (1998), Peru - Selected Issues, Western Hemisphere Department, Washington. Lanjouw,Jean Olson (1997), Behind the line, United Nations. Lanjouw, Peter (1998), Ecuador's Rural Nonfarm Sector as a Route out of Poverty, Policy Research Working Paper Series 1904, World Bank. Lanjouw, Peter, Giovanna Prennushi and Salman Zaidi (1996), Building Blocks of a Consumption-Based Analysis of Poverty in Nepal, mimeo, Development Economics Research Group, World Bank. Lanjouw, Peter and Martin Ravallion (1995), econ, Economic Journal, pp. Lopez, Ramon and Cada della Magiiora (1998), Rural Poverty in Perii: Stylized Facts and Analytics for Policy, processed, University of Maryland, College Park. Luerssen, Susan (1993), Illness and household reproduction in a highly monetized rural economy. a case from the southern Peruvian highlands, journal of Anthropological Research 49, pp.255-281. MacIsaac, Donna and Jesko Hentschel (1996), Urban Poverty, in Ecuador Poverty Report, World Bank, pp..... Macisaac, Donna and Harry Patnos (1995), Labour market discrimination against indigenous people in Peru, Journal of Development Studies. Ministerio de la Presidencia (1997), Plan de Accion Local en Ate, Lima. Moncada, Gilberto and Richard Webb (1996), Como Estamos? Analisis de la Encuesta de Niveles de Vida, Insituto Cuanto, Lima. Moncada, Gilberto (1996), Perfil de la Pobreza en Peru, 1994, in: Webb, Richard and Gilberto Moncada, Como Estamos?, Instituto Cuanto, Lima, pp. 97-135. Moser, Caroline (1998), Urban POverty: How do Households Adjust0, in World Bank (1996), Ecuador Poverty Report. Oxford Analytica (1998), Peru: Poverty Programs, The Fujimon Government's Strategy to Tackle Rural Poverty, July 30. Persaud, Thakoor (1992), Housing Delivery Systems and the Urban Poor, A Companson Among Six Latin Countries, Latin America and the Caribbean Regional Studies Program 23, Washington D.C. Ravallion, Martin (1994), Poverty Companons. Rodriguez, Edgard (1998), Toward a More Equal Income Distribution> The Case of Peru 1994-97, background report for this study, processed, World Bank. Rodriguez, Jose and David Abler (1998), Asistencia a la escuela y participatcion en el mercado laboral de los menores en el Peru entre 1985 y 1994, processed, Pennsylvania State University. Saavedra, Jaime (1998), What Do We Know About Poverty and Income Distribution in Peru with Emphasis on Its Links with Education and the Labor Market, background report for this study, GRADE, Lima, Peru. PERu. PovERTY ComPARisoNs - A PoLicyNoTE .52 Saavedra, Jaime and Alberto Chong (1998), Structural Reform, Institutions, and Earnings: Evidence from the Formal and Informal Sectors in Urban Peru, Journal of Development Studies, forthcoming. Saavedra, Jaime and Juan Jose Diaz (1997), El Rol del Capital Humano en la Distnbucion del Ingreso. Mimeo, Lima, GRADE. Saavedra, Jaime, R. Melzi and A. Miranda (1997), Financiamiento de la Educacion, Documento de trabajo 24, Lima, GRADe. Schady, Norbert (1998a), Picking the Poor. Indicators of Geographical Targeting in Peru, Woodrow Wilson School of Government, Princeton University, processed, Princeton. United Nations (1997a), Humand Development Report, New York. United Nations (1997b), Informe Sobre el Desarrollo Humano del Peru, Lima. White, Michael J., Lorenzo Moreno, Shenyang Gua (1995),The interrelation of fertihty and geographic mobility in Peru: a hazards model analysis International Migration Review 29, pp. 492-514. World Bank (1998a), Did the Ministry of the Presidency Reach the Poor in 1995, processed, World Bank. World Bank (1998b), World Development Indicators, Washington D.C. Yamada, Gustavo (1996), Urban informal employment and self-employment in developing countries: Theory and Evidence, Economic Development and Cultural Change 44, 289-314. PERu:- PovERTy CompARisoNs --A PoLicy NoTE 5 Annex 1: Panel Study ofHouseholds This annex shortly recaps the results of studying the panel sub-sample of the Living Standard Measurement Surveys (ENNIV 1994 and 1997) to examine what happened to several hundred identicalhouseholds which were interviewed in both rounds of the ENNIV. Our basic model consisted of expressing the growth of per capita household expenditures as a function of a large number of exogenous variables, such as acess to basic services, initial education and consumption level etc. For this purpose we looked at three different, increasingly restrictive panel subsamples. First, the total panel of 891 households. Second, all households which had the same head in the two years (690 households). Third, all households with the same head, household size and household composition. The latter group comes as close to a controlled experiment as possible. Per capita growth in household consumption is not influenced by additions or attrition from the household. Our first result confirms other panel analyses in Peru: mobility is very high, i.e. a large percentage of households changes its relative position by more than one welfare decile over the three years (see also Glewwe and Hall 1995, Escobal et al 1998). In absolute terms, about 55 percent of all households recorded a per capita consumption growth of more than 10 percent, about 15 percent recorded more or less the same consumption level and 30 percent had a consumption per capita level in 1997 more than 10 percent below their level in 1994. Robust results from the regressions, which hold for a variety of different specifications and application to the different sub-samples, are: a. Female-headed households caught up. Controlling for all other influences (e.g. education, initial consumption, household size, dependency ratio etc.) female-headed households fared better than male ones, increasing per capita consumption per capita growth by .12 percent. b. Migrant families farev well. Although very small (.04 percent), the fact that a household had migrated before 1994 increased growth of per capita consumption significantly. c. Native language speakers fell clearly behind. One of the strongest result we find is that language, and with it indigeneity, matters a lot. Even when we control for other variables that are correlated with language such as geographic location, do native language speaking households fall behind spanish-speeking households (0.12 percent). d. hrastmicture is of major mportance and we find increasing returns to services. We find evidence of increasing returns to the number of services households command. Holding all other influences constant, if a household had access to one of four services (phone, electricity, water, sanitation) in 1994, this increased the per capita growth rate by 0.08 percent. If the household had access to two services per capita growth was, on average, .10 percent higher (the marginal return on the second service is .02 percent). However, marginal returns increase: the third service has a marginal return of 0.12 percent and the fourth service even 0.15 percent. Looking at services by type, electricity is the most important service linked to household welfare improvements in rural areas and a telephone in urban areas. PERU POVERTY CompARISONs -A PoLicyNOTE ................................ 4 e. Household siZe and dependeng rate are both determiining factors. Perhaps the most important result, household size and the dependency ratio significantly influence welfare developments. Results suggest that (a) larger households fare worse than smaller ones; (b) this relationship is not linear; the larger the household the less negative is the effect; and (c) the dependency ratio (number of non-income earners to income earners) has an independent negative influence on household per capita consumption growth. f. Better education and more experience means faster advance. The higher the education of the household head in 1994, the larger the growth in per capita expenditures. This mirrors the common option in Peru that the better educated and more experienced have proportionately more than the less educated in recent years. g. Access to credit in the initial year and finanial satings. Credit access in 1994 or new credit access between 1994 and 1997 had a positive and significant influence on consumption growth. Financial savings of households (in 1994) had the same effect. h. Households nith home-based businesses fare better. Households that stated that they used at least one room in their house for business purposes -- both urban and rural -- have managed to achieve a significantly higher growth of welfare than households which did not have this possibility. Again, this result hold when controlling for all other factors that influence consumption growth. PERu.: PovErRTry CommuPsoNs -- A PoLicy NoTE 55 Annex II: Methodology Al. Introducdon Peru is one of the 'pioneer countries' fielding comprehensive household surveys aimed at measuring poverty and well-being. The first Living Standard Measurement Survey in Peru was conducted in 1985 by the Statistical Institute and since then the Instituto Cuanto has conducted a host of other surveys. While each survey added or modified specific questions, which was partly a reflection of the specific interest of individual funding organizations, the core of the survey with its focus on housing conditions, education, health, migration, the labor market and agricultural activity, has stayed remarkably constant. The two most recent and largest surveys are those used in this study: the 1994 and 1997 Encueastas de Niveles de Vida. They employed a sample frame to achieve representability in the urban and rural areas of the three agro-climatic zones in the country (Costa, Sierra, Selva) plus Lima. Most users of micro data can tell picturesque stories of how working with individual and household survey data requires many -- often cumbersome -- steps of data cleaning and consistency checks before the actual empirical investigations can -begin. Especially when comparing different variables over time, one of our primary aims, caution is necessary. And this statement hold even more when the aim is to compare an 'artificially' created variable between surveys. For poverty analysis this 'creation' is central as we first have to derive a monetary welfare measure (consumption or income). Further, the monetary aggregate need to be deflated over both time and space. And then the (in)famous poverty lines need to be derived before the simplest of all comparisons -- calculating headcount rates -- can take place. Along the way, many assumptions have to be made. This annex contains a detailed description of how we used the two Cuanto LSMS surveys for poverty analysis. We first start with a short background section on poverty comparisons in general which stresses the importance of defining welfare in a consistent manner when conducting comparisons over time. Section three describes how we aggregated consumption expenditures, paying particular attention to their comparability across survey years. Income definitions are also included. Section four explains how we derived poverty lines for 1991, 1994 and 1997 and is concerned with the necessary price adjustments. Section five reports the results of several sensitivity analyses with respect to adult equivalency and economies of scale. A2. Background: Poverty Comparisons in Time One of our aims when analyzing the consecutive Living Standard Measurement Surveys from Cuanto is to compare poverty and welfare changes of the Peruvian population over time. At face value this does not seem to be very difficult. All of the surveys include household income and consumption which can be converted into per capita terms and then compared to certain absolute standards, the poverty lines. In reference to these lines, headcount rates, poverty gaps and poverty severity can be calculated and compared over time. But for a number of reasons, poverty comparisons using consecutive surveys are quite cumbersome and difficult. First of all, it has to be ensured that the sampling frame is the same PERU: POVERTY COPRisoNs -A PoLicy NOTE .......................................5.6 (from which factor expansions are computed) in the two years and that definitions that determine the sampling process are identical. For example, if stratification of the sample is conducted with respect to urban and rural areas, the latter need to be defined in the same way in consecutive years. The same holds for other stratified variables be they of political or socio- economic nature. For example, in addition to the nationally representative household surveys mentioned above, Cuanto also fielded one relatively small survey in 1996. This was a pure panel survey as all 1,491 households that were interviewed as part of the larger 1994 survey. While statements can be made about the comparison of poverty between 1994 and 1996 for the panel householdW, generalizations for the whole country cannot be made. Even if the selection of the panel had been completely random (all households forming part of the 1994 survey had the same chance of being selected for the panel), the 1996 would not have been nationally representative. Newly formed households after 1994 had a zero probability to be selected in the 1996 survey. Although the bias between two years might be small, its effect cannot be quantified. Second, poverty comparisons are based on a number of very stringent assumptions. The most common method to conduct poverty comparisons is to base all nominal income and consumption data from different surveys in one time period and one region, i.e. to deflate nominal variables in space and time and hence convert them into real values." These real values are then compared to a constant poverty line representing a minimum consumption basket. Generally, the basket itself is derived (at least the food basket) from actual consumption patterns of the poor so that it is 'appropriate' for the type of analysis being carried out. This basket of goods is supposed to present a certain welfare level that can be compared across households. One of the important assumptions underlying such welfare comparisons is that households have homothetic tastes, i.e. that the welfare households derive from consuming the basic bundle of commodities is identical. Although this is almost certainly unlikely, we could nevertheless think of the basic bundle of goods as a 'yardstick' against which we measure people's (relative) welfare -- we express the (relative) welfare of households as 'how many times can they consume a given bundle of goods'. Welfare comparisons over time leave, if at all possible, this yardstick (or basic basket of goods) constant. Hence, the composition of the food basket with all its components of fruits, vegetables, meats etc. is kept constant as well as the non-food components such as housing, clothing, services, the use of durable items and the like. However, as the assumption of homothetic tastes was a shortcut, so is this. Over time, relative prices between goods change. Households adjust to these relative price changes by choosing a different mix of commodities, generally increasing the consumption of relatively cheaper products and reducing intake of relatively more expensive ones. A basket of goods which was representative of the poor's consumption pattern in a base period hence does not need to be representative of the consumption pattern in a different time period. Both relative price change and modifications 43 Cuanto (1997) produced a study comparing poverty in the 1994-1996 panel and was careful not to generalize results to the whole of Peru. 44 See, for example, Ferreira and Litchfield (1998), .... PERu: PovErRTY ComPARisoNs -- A PoLtcy NoTE 57 in preterences can account for such shifting consumption patterns. Welfare comparions using a fixed basket in time hence are -- again -- only approximations.45 Third, and touched on above, careful price deflation is crucial. A number of different ways are used in the literature to derive spatial and time indices in order to make consumption expenditures comparable between households in different locations and of different survey years. But all of them require the availability of regionally distinct price indices for broad product groups. In many countries such information does not exist. In Peru, however, detailed price information exists for food products in all regions and for non-food categories in twenty-five urban centers. As will be shown later, we can use this information to adjust nominal consumption in all survey years. Fourth, if the basket of goods is kept constant over time, the definition of consumption needs to be the same as well. Poverty measurement can be seriously distorted if the definition of consumption changes over time. For example, consecutive surveys might add a question on expenditures and auto-consumption of a very specific food item that was not asked for before. Obviously, even if the 'true' food consumption of households is completely identical in the two survey years, it would appear on paper to be higher in the year in which the additional question was asked. If the basket of goods against which we compare consumption expenditures of households is fixed, however, adding additional consumption items will lead to an unequivocal reduction in poverty."' Caution also needs to be taken if the meaning or phrasing of questions in the consumption module change over time. Even if items for which households are asked to report expenditures or auto-consumption are identical, the phrasing of questions can have a profound impact on the level and structure of responses. An example from the Cuanto surveys 1994 and 1997 will be used to illustrate this point in section 4. Finally, poverty comparisons should ideally establish whether observed trends in statistics are robust or not. This would imply varying some of the underlying procedures consumption aggregation such as imputation procedures or testing the effect of implicit assumptions about economies of scale or adult equivalency scales. Further, the poverty line can also be varied over a wide range of different values to test whether the choice of poverty line changes the conclusions as to the direction of welfare changes. This study uses a fixed basket of total goods (food and non-food goods alike) to denve poverty lines and conduct poverty compansons This differs markedly from the tradition of poverty analysis in Peru which kept only the food basket constant over time and denved the non-food component endogenously. 46 Lanjouw and Lanlouw (1998) show that if consumption questions indeed changed over time so that the definition can simply not be held constant, a second possibility is to only keep the food basket constant and denve the non-food basket implicitly by calculating the Engel coefficient They show that this will give consistent estimates of poverty under a number of assumptions, including a homogenous relationship between food expenditures and total expenditures. Further, between the survey years no or little relative pace changes between food and non-food goods should occur. PER U: PO V-R TY COMPARSN- PoLiCY No TE .......................................58 A3. Defining Welfare A3.1. Consumption The consumption aggregate we are using is one designed to be comparable for the 1994 and 1997 survey years. Although the surveys show a very high degree of homogeneity, the questionnaire did change at several places: new products were added, product groups were changed, separate products in one year summed up in the next year, or questions were reformulated which gave them a different meaning.* We made a large number of small adjustments in almost all components of the consumption aggregate. Table I contains the description for different product cate ories and the consumption definition we used (computer programs are available on request). As can be seen, certain exclusions and inclusions of sub-components were made in the education, health, semi-durable, transfer and auto-consumption frpm business section. We could not include the depreciation stream from durable consumer goods because the 1994 survey did not include the age structure of the household durable consumer goods, which meant that 48 depreciation rates could not be calculated. Similarly, furniture purchases were not included. The food, rent and social program modules require some more elaborate explanations: Food Module. While the food module of the questionnaire appears to me almost completely identical between 1994 and 1997, Cuanto introduced one considerable modification. Specifically, Cuanto added one supplementary question to the food module in 1997 which reads "total autoconsumo y autosuministro". The consequence of including this complementary (and well-intentioned) question was that more than one third of all sample households and their interviewers (close to 1,300) chose to respond only to the aggregate question -- and thereby avoided detailed answers. Generally, it has been shown that detailed consumption questions have a clear advantage over aggregate questions as recollect quantities and expenditures better. Lanjouw (1997) reports that under-declaration can be significant in shorter'questionnaires, especially for lower income groups. 47 Send emad to jhentschel@worldbank.org The 1997 survey includes this age structure Current consumption from the stock of durable goods can then be estimated as the median age can be calculated for each type of good as the depreciation rate can be assumed half the life duration of the products. Given the age of individual products per household, one could individually compute the expected remaining lifetime of each product. PERu..Po.vE,RTY.ComPARisoNs -A..P0LjcYNOTE ........................................ Table 1: Defintion of the Consumption Aggrejate, 1994 and 1997 Item 1997 1994 Consumer durables excluded excluded Daily Non-Food, Other all purchase and auto-consumption Included included (Z3A, Z3B) except for expenditures on public telephones as not in earlier 1994 (z2=109) In-knd consumption from firm/busness Includes auto-consumption from firms Included (W19) In-Kind consumption from work primary and secondary work over last 7 Included as in 1997 (identical questions) days, primary and secondary work over last 12 months (adjusted for tume-penod worked) (MlIB,M12B,01lB,012B, R12B, R13B, TI2B, TI3B) Education direct expenditures in education section school uniforms excluded (as included in included (F1OA, Fl0B, Fl0C, FIOD, Fl0E) the services sections) Also, 1997 questionnaire asked for frequencies but 1994 survey did not. Hence, we used median frequencies from the 1997 survey used for evaluating payment of matniculation, books, transport in 1994. Also, a separate questions on expenditure for children under age 6 in 1994 surveys excluded because not in 1997 survey. Furniture Excluded (and depreciation rate cannot be Excluded calculated) Food module purchases and valued autoconsumption Purchases and valued autoconsumption (AE4, AES, AE6 and AE7) (identical apart from minor and negligible different grouping); however, an additional question which allowed households to only give ONE aggregate figure for total food purchases did cause comparability problems (see text). Health expenditures direct health expenditure in health section same as 1997: expenditures in health section included (HI1A, HI iB, HIS, H19), but included, in service section excluded. excluded in services section However, examining the share of health (AAl= 125,126,127) as recall penod expenditure in total expenditure, we found a different and not clear if these are number of outlier households which stated additional or the same products mentioned that they spent more of 50 percent of their in the health section total expenditure on health. This did not happen in 1997. Hence, we excluded these outliers (xx). House expenditures water (D9A), light (DI2A), heating and all included cooking fuel (DI4A), telephone (D18B and D20A), municipal fees (D22A) Inote- municipal arbitration excluded] Payments for House (repayment of credit) Excluded Excluded Rent Excluded (questions different) Excluded (questions different Semi-durables and Services Only purchase included as the 1994 only purchase included; questionnaire did not include auto- consumption Social Programs Only food aid included as 1994 survey did Food aid included (as captured in the food not ask for other social program transfers, section under 2l02=327 and ak02=327); excludes 'ahmento por trabalo' ghl=506 as excludes al0l =09 as Cuanto maintains that not in 1994 survey question al01=09 was imputed into a103=327 Transfer Expenditures ceremonies (ADI=02), direct taxes Ceremonies (in 1994 in services module), (AD1=03), social security (AD1=04), social security, membership fees, donations, membership fees (AD I=07), donations direct taxes (AD1=09) Insurance questions excluded as not in 1994 survey (ADI=06) E R COM RISONS POL N OT ...............................................................................0 While we do find a pattern in Wh e co inthod paere of Table 2. Detailed andAggregate Food Module responses when comparing the food share of Responses households answering the detailed questionnaire with the food share of quintile foodshare of hhs foodshare of hhs households that only give one aggregate food answering detailed answering aggre- expenditure value, we have opted for non- questions_gate_question adjusting this variable. As can be seen in the attached table, the foodshare of households 1 67.9 681 2 60.3 60.2 answering the detailed questions tend to be 3 56.2 534 significantly higher than the foodshare of 4 497 47.9 households only giving one, aggregate 392 estimate for the richest three quintiles. We Source: LSMS (1997), own calculations. opted, however, not to make adjustments for Population quintiles defined by total real two reasons: (a) in the lowest two quintiles evpend6irinue (which are of our primary concern in a poverty study), the mean difference is not very large; and (b) it is extremely difficult to make food adjustment as the foodshare per quintile shows enormous fluctuations across households (this is generally observed in household surveys, see Lanjouw and Lanjouw 1998). It is therefore questionable to lmit the foodshare -- which accounts for the bulk of ex%pendtures to one 'average' number across households as in rvaly fluctuations arr tery high. The same argument would holdfor making food inputations on the basis of re gaion mode/s. ent. Next to food, the actual (or imputed) rental value of the house tends to be the most important budget item for the poor and non-poor alike. We therefore were very eager to include this variable in our consumption aggregate as it adds an important welfare component of families: how much space households have, if the house is made of weather-resistant material, how close the shelter is from the nearest market, what transportation possibilities exist. All such factors enter into the determinants of the housing value, which we were keen to incorporate, albeit ensuring that we have consistency over the years. Both surveys (1994 and 1997) collected information on (a) actual rent paid; and (b) self- estimated imputed rent from the households that were owner-occupiers. Two problems arose here. First, we had to undertake some simple imputations of the value of housing. These had to be carried out for those households that neither provided actual nor estimated rental value of their housing. For this purpose, we used simple hedonic regressions in which we postulated the value of housing to be a function of the stock of assets of the household, regional dummies (capturing price variations), and both housing (material, size) and household (size) characteristics. Based on their housing, household and asset characteristics, we predicted housing expenditures for those households that had reported a value for actual or estimated rental value of their house -- these were 238 in 1994 and 13 in 1997. The second problem that arose was much more severe. The questionnaire had changed between the two survey years so that we needed to test whether we could indeed reported rental values of the house without compromising the comparability between the surveys. While both surveys had a question recording the actual rent paid by households, the 1994 survey had a follow-up question for owner-occupiers in which households were asked how much they would charge if they were to rent their house. In contrast, the 1997 questionnaire queried owner-occupiers how much they would be willing to pay had they to rent PERu.- Po vERTY CompPisoNs -- A PoLicyNo TE 61 their own house. Our initial hypothesis was that the change in the question would only have a marginal impact on structure and level of this variable. We conducted several tests to find out whether the rent sections were comparable. The first one involves the imputation regressions reported above. As Lanjouw et al (1996) have shown, such imputation regressions can be used to predict housing prices in different regions of the country while controlling for the quality and characteristics of housing. The idea is simple: using the median values for all exogenous variables (asset variables, housing & household characteristics, geographical dummy variables) and the estimated parameter values reported in Table 4, we derived the expected price of a 'standard house' with in all regions for 1994. Using the same median values of the exogenous variables (as we want to control for quality over time) but now employing parameter estimates for 1997, we could calculate rental values for 1997 as well (as this is when the change in the questionnaire took place). While the change in the housing price itself is interesting, we can now also add one more control variable as INEI, the Peruvian Statistical Institute, compiles a value for housing ('alquiler) for 25 cities in the country. Table 3 shows the results for the change between June 1994 to October 1997 (the months of the survey). According to the regional INEI price data, all urban regions experienced considerably slower rent inflation than (implicitely) recorded in the survey. Table 3: Rental Values in 1994 and 1997 Area Predicted Rental Predicted Rental Change in Predicted Change in Rent Value, 1994 Value, 1997 Value, 1994/97 (o) Index, INEI (%) Lima 1701 2593 52.4 34.2 Costa Urban 1184 1822 53.9 35.5 Costa Rural 221 445 101.4 .. Sierra Urban 731 1452 98.6 35.9 Sierra Rural 144 537 2729 . Selva Urban 764 1371 79.5 267 Selva Rural 200 372 86.0 Source: Own Estimates Based on Tables 3 and 4, INEI Regional Offices The second possibility to test the rent variable changed over time was to take a look at how the same households evaluate their rental value in 1994 and 1997. In line with its tradition, Cuanto included in its sample in 1997 about nine-hundred families that had already been interviewed in 1994. We divided this panel data-set into ten 'rent' deciles, i.e. the household reporting the lowest rent values are grouped into decile 1, the households with the highest self-declared rent values are included in decile 10. Using a transition matrix, we can then see how this ranking of households with respect to the rent variable changes. Since the households live in the same dwelling, we would assume that the ranking of households stays very stable if the different question does not have an impact on households' reporting patterns. However, we find that rankings change considerably: only about 25% of the panel sample can be found on the diagonal in the transition matrix which implies that they have not changed their decile ranking between 1994 and 1997. An additional 25% are off by one decile but PERU .PovERTY CompARISONs--A PoLicyNoTE 62 roughly half of the total sample can be found with changes of at least two decile rankings which implies significant ranking changes -- although these are identical households living in the same dwellings in 1994 and 1997. Finally, we took a look at the overall expenditure composition. Our hypothesis was that the change in the questionnaire would likely lead to a lower share in households' subjective evaluation of their rental value in 1997 compared to 1994 (since in 1997 households were asked what they would pay for their own house in rent). But this did not prove to be the case -- actually, the share of total expenditure on (actual and self-declared) rent increased marginally over the 1994-1997 period (from 14 to 16 percent). However, this average does mark quite large variations in answers between the years: actual and imputed rent was the variable with the highest fluctuation as a share of total expenditures -- rent patterns almost reversed in certain regions. Table 4 tabulates rent as a share of total expenditures in 1994 and 1997. Given these findings, we concluded that the changing question regarding the rent variable had a significant impact on responses and, especially, their structure in different expenditure groups. Since one of our main aims were welfare comparisons, we therefore chose to exclude the rental value from the consumption aggregation. The income definition also did not include the rental value. Food Donations. The questionnaire also changed considerably between 1994 and 1997 with respect to how food donations are treated. In 1994, these were not explicitly asked for but entered in two different questions: first, in the section on food consumption households entered the value of 'prepared food products'." Second, households were asked how much income or value in products they received from non-profit organizations (examples qiven in the questionnaire were 'Vaso de Leche', Club de Madre and CARITAS). o Cuanto holds that the value of food donations was actually included in the 'prepared food' question in the food module when the original data were processed. The questionnaire was considerably different in 1997. Cuanto added a whole section on access to social services in which households reported the value of food received by program and funding source. In addition, the food module continued to include the same question as had been asked in 1994, i.e. the value of prepared food products consumed by the household. This is in the food module secton in 1994, vaiables a102 and akO2 (rubrique 327). 50 This refers to codes a101=09 an the 1994 questionnaire. PERu: PovEry CoMfPARISONS --A PolcyNOTE 63 Table 4: Expenditure Patterns and Consumption Defiition (to be changed) Inding Imputed Rn Eedekdfg Impted Rent Asmo QUnte Foodban Foodsbarr d(Foodsbora) d(Rentbar) Foodsbar Foodbarr d(Foods)ar) 1994 1997 1994/1997 1994/97 1994 1997 1994-1997 Lima 1 56.88 50.42 -6 46 65.85 59.88 -5.97 2 52.83 47.56 -5 27 64.39 58.25 -6.14 3 50.44 45.63 -481 61.39 56.89 -4.50 4 46.14 38.82 -7.32 5830 51.77 -6.53 5 3448 29.11 -5.37 50.15 45.84 -4.29 CoutA 1 56.81 50.41 -6.40 7.53 63.5 65.77 2.27 Urian 2 52.98 47.23 -5.75 5.29 62.81 60.69 -2.12 3 50.74 49.54 -1.20 1.68 61.08 61.56 0.48 4 48.05 45.87 -218 0.74 57.32 56.88 -0.44 5 39.44 41.09 165 0.99 51.23 5405 2.82 Costa 1 72.14 6861 -353 0.66 79.4 74.82 -4.58 Rum/ 2 68.62 68.41 -021 -1 05 73.99 71.53 -2.46 3 68.49 68.66 0.17 0 27 73 74.98 1.98 4 59.82 65.56 5.74 -1 03 66 2 69.37 3.17 5 5781 60.16 2.35 -1.35 63.34 64.06 0.72 Stemr 1 58.69 4848 -10.21 473 65.74 67.45 1.71 Urban 2 56.91 47.42 -949 4.87 63.39 64.39 1.00 3 48.49 46.09 -240 1.96 59.9 60.79 0.89 4 48.15 43.32 -4.83 2.73 54.74 56.33 1.59 5 412 40.62 -1.58 0.61 54.76 53.41 -1.35 sise 1 71.96 69.41 -2.55 2.68 77.68 78.23 0.55 Rwa/ 2 71.52 65.74 -5.78 5.96 77.34 78.13 0.79 3 69.54 65.42 -4.12 3.82 76.27 73.80 -2.47 4 70.02 64.35 -5.67 3.30 7447 74.04 -043 5 63.2 66.58 3.38 -0.31 6882 73.94 5.12 SOlm 1 62.97 52.32 -10.65 1178 70.58 68.21 -237 Utiwn 2 62.41 55.39 -702 7.28 6907 68.65 -0.42 3 56.67 53.28 -339 462 6419 6560 141 4 5342 51.85 -1 57 078 62.31 61.68 -0.63 5 47 09 4795 0 86 -094 57.95 57.72 -0.23 Selm 1 73.04 68.16 -488 0 81 82.36 76.94 -5.42 Ruml 2 73.36 68.25 -5 11 066 77.54 74.42 -3.12 3 68.19 67.66 -0.53 041 74.63 72.52 -2.11 4 68.17 67.76 -0.41 -0.31 7281 72.56 -0.25 5 69.19 65.43 -3.76 -0.33 7440 70.05 -4.35 Source: Staff Estnmates based on ENNIV (1994 and 1997). PE-Ru: PovE.RTY CompARIoNs --A PoLicyNoTE 64 Comparing the 1994 and 1997 surveys shows considerable changes in the values of food donation variables. In 1997 prices, total estimated food donations in 1994 were 180 million soles51 while the total estimated benefits of food programs was 1.3 billion soles in 1997.52 These are low estimates for 1994 but also very high estimates for 1997. We included food donations in a different form than Cuanto for 1997. First and most important, Cuanto multiplied 'daily' food donation receipts by 365 to obtain annual values. However, several of the donation programs do not work every day per week and certainly not all weeks per year. The most important one is the school breakfast program which operates from Monday to Friday and about 2/3 of the whole year. Also, comedores populares and club de madre food distribution points are generally operated from Mondays to Fridays, with exceptions also on Saturdays. Second, studying the data in depth it becomes clear that many responding households were confused as to whether they were supposed to give the (a) daily value of donations received; or (b) the value in the recall period. For example, many households responded that they obtained five times a 'glass of milk' for two children in a recall period of a week and they listed 10 soles as the value they received. Obviously, the value must refer to the LyAI of the five times two milk rations received rather than to a single glass (10 soles was about 4 dollars at the time of the survey). In its estimates, Cuanto interpreted these 10 soles as the daily value of one glass of milk which was received 365 times a year -- hence adding about US$1,600 to the expenditure of the particular households. In our approach, we calculated the median value of one glass of milk or school lunch per region and used this value to estimate the value of these items for the households. We thereby assumed that schools operate five times a week and 8 months a year. The glass of milk program was assumed to work 5 times a week during the whole year. Making these adjustments, our estimate of total food aid in 1997 dropped from 1.3 billion soles (Cuanto) to 800 billion soles - a figure much more in line with expenditure reports of the large nutrition programs. As shown in the main body of this study when discussing social expenditures, the in- or exclusion of food aid has a marked impact on calculated poverty rates, especially in the rural highlands. The severe poverty rate would have been three percent higher had we excluded the donations. Poverty calculations are quite sensitive to changes in definitions here and further in depth analyses from other researchers would be welcome.7 5i The survey codes in 1994 is a102=09. 52 The survey codes in 1997 are ab1=501-505, 507-508. For the measurement of welfare and poverty, a problem nevertheless remains since there seems to have been an underestimation of food donations in 1994. Hence, we are overestimating welfare improvements by including food aid. On the other hand, leaving food donations aside while in realty they have indeed increased, we are underestimanng welfare increases PERu: PovERTYCompARIsoNs- -A PoLtcyNoTE 65 A3.2. Income With one major exception, we employ the income definition of the Instituto Cuanto. Income definitions of the different components (self-employed income, wages, transfers and property income) were remarkably similar between 1994 and 1997 with one major exception: the above-mentioned question on owner-occupied housing (rent) which should be included as an income component. 54 In line with the income employed wages, our preferred income aggregate excluded rental income. A4. Poverty Lines & Price Deflation Food Basket and Value in 1994. From their inception, the Cuanto surveys have not collected quantity or price information in the consumption module. This has important implications for the derivation of poverty lines: it implies that the composition of the basic food basket cannot be derived from the survey itself and has to be obtained from an external source. We use the basic food basket from Cuanto (see Moncada and Webb 1996) as the starting point of our analysis. It is priced in 1994 and 1997 using detailed regional price indices supplied by the National Statistical Institute INEI. Non-Food Basket and Value in 1994, Poverty Line in 1994. Different from common practice in Peru, we keep the basket of non-food goods constant over time. This has been suggested several times by Francke (1996 and 1997) and goes back to the argument made earlier that ideally we want to fix a certain welfare level (associated with a fixed bundle of goods) over time. In international practice, this seems to be the preferred way of welfare comparions in time (Ferreira and Litchfield 1998, MacIsaac and Hentschel 1996, Ravallion 1994). We use 1994 as the reference year to derive the basic consumption bundle. We employ the three normative food baskets used in Peru and their value and then derive the upper bound poverty lines by comparing how much the population group that spends on food equal to the food poverty line spends in total. Weights for the non-food basket can be consequently derived (Table 5). Poverty Lines in 1997. The food and non-food basket from 1994 was then priced in 1997 using data from INEI regional offices. We computed price indices for non-food categories from price information provided by INEI for 26 cities in the Sierra, Costa, Selva and Lima. For each region, we calculated the average price index and used it to value the 1994 basket in 1997. Since INEI only reports urban prices, we assumed that the relative price rise (not its absolute level) between rural and urban areas in all regions is the same between the survey years. Again, programs are available upon request. Please send email tojHentschel@worldbank.org. The figures on inequahty presented in the main body of this study also do not take into account other non- cash income. We tested the robustness of inequality changes over tnie, though, and found that inclusion or exclusion of (other than rent) non-cash income did not change results PERu:- PovE-RTY CoMPARIsoNs-- A PoucyNomE 66 Table 5. Dervation ofPoverty Lines, 1994, 1997 Category Weights 1994 Value 1994 Inflation Value 1997 Oct-97/June-94 Lima food 0630 91104 1.30' 1182.38 clothing 0.049 7067 1.37 96.60 water, electncity 0086 12505 1.44 179.58 cleaning 0.027 39 12 1 38 53.78 health 0.075 108.33 1.56 16943 transport 0.075 10833 1.37 148.20 education 0.055 80.12 1 54 12339 other 0.004 11.44 1.31 14.99 [povertyline] 1454.10 1968.34 Urban Const food 0.610 789 13 1.31* 1032.75 clothing 0.028 35 95 1 28 45.83 water, electricity 0.111 14391 1.36 195.00 cleaning 0.031 39.97 1.23 49 28 health 0.052 66.96 1.42 94.75 transport 0.065 83 83 1 23 10278 education 0.061 79 55 1 45 115.66 other 0.043 5840 129 75.16 [poverty line] 1297. 70 1711.22 Rural Coas food 0.660 700.07 1.31 91747 clothing 0.068 7267 1.28 92.66 water, electricity 0.041 44.03 1.36 59.66 cleaning 0.027 29 18 1.23 3597 health 0.066 70.32 1 42 99 50 transport 0.064 67.86 1 23 83 20 education 0.032 33 88 1.45 49.26 other 0.043 50.69 1 29 65.24 (poverty line] 1068.70 1402.96 Urban Soerr food 0.610 66868 1.30* 866 45 clothing 0.049 53 58 1.29 6932 water, electricity 0.105 116.00 1.36 157.60 cleaning 0.029 31 82 1 23 39.24 health 0.046 50.60 1 42 71.96 transport 0.054 59.44 1.29 76.59 education 0.079 8673 1.47 12742 other 0.030 3795 1.31 4976 [poverty line] 1104.80 1458.34 Rural Sicrr food 0.760 58323 1.36* 791 48 clothing 0.063 4803 1 29 6214 water, electricity 0.029 22 56 1 36 3065 cleaning 0.034 26 17 1 23 32.26 health 0.037 28.24 1 42 40 16 transport 0.026 19.80 1 29 25 51 education 0.025 19 49 1 47 28.63 other 0.026 1979 1 31 2595 [poverty Inel 767.30 1036.78 PERu. POVERTYCO4fPARisoNS --A POlicY NoTE 67 Table 5. Denvadon ofPoverty Lines, 1994, 1997 (continued) Category Weights 1994 Value 1994 Inflation Value 1997 Oct-97/June-94 Urban Sely food 0.650 70199 1.31* 922.57 clothing 0.049 53.18 1.20 63.71 water, electricity 0.097 10410 1 27 131 79 cleaning 0.032 33 91 1.20 4076 health 0.051 54 79 1 42 77.80 transport 0.043 5533 1.26 69.83 education 0.043 45 75 1.34 61.22 other 0.028 26.45 1.20 31.60 [pore ykne] 1076.50 1399.27 Ruda Sel food 0.730 647.51 1 36* 880.23 clothing 0.064 57.37 1.20 6873 water, electricity 0.018 15.61 1.27 1977 cleaning 0.047 4102 1.20 5051 health 0.049 4407 1.42 62.58 transport 0.050 44 16 1.26 5573 education 0.018 15.61 1.34 20.89 other 0.025 25 83 1 20 3087 1poverryinef 89220 118931 Notes: Poverty lines in all years are derived using the poverty basket of the year 1994. Consumption is defined as outlined in the previous section (e g., total consumption e'-cludes rental value of the home), product group definitions follow the Instituto Cuanto. The food share in 1994 (by region) is determined by the decile of the population which spends on food products the value of the exogenously determined food basket (Moncada 1996) Expenditure shares in 1994 refer to this population group We calculated price changes for non-food categories between the different survey years using INEI city price indices by broad product group for 26 cities, calculating average indices by region. Since INEI only reports urban prices, we assumed that the price rises (not its absolute level) in rural areas is the same than in their urban counterparts Price index of the exogenously given food basket derived from dividing the nominal value of the basket in different years. The value of the food basket in all regions in 1994 and 1997 was calculated by the Instituto Cuanto Different from other food basket values which were taken from Instituto Cuanto, we computed the value of the food basket in 1991 for the rural Sierra and urban Coast the Instituto Cuanto had assumed that absolute prices in these regions are the same than in Lima (urban Coast) and urban Sierra (rural Sterra) See Moncad. (1996, p 133) Rather, we apphed the implicit food basekt price deflation derived for Lima to the urban Coast and for the urban Sierra to the rural Sierra This produces the same relative price difference in the value of the food basket between the regions in 1991 than in 1994 Price Deflation. Two price deflations were applied. First, price deflation in time as the surveys in 1994 and 1997 were conducted over several months. Such adjustment for inflation was carried out by Cuanto and is included in the basic database. Second, rather than working with seven different poverty lines, we adjusted all household consumption (and income) values to the price of Lima. For this exercise we used the computed poverty lines and price deflators, defining Lima as '1' and using the ratio between the Lima poverty line and each individual regional poverty line as a deflator for regional monetary values. This allowed us to compare welfare levels between households directly. Definition of Severe Poverty Line. The lower or severe poverty line used in the report is not strictly comparable to the 'extreme' poverty line used in most other poverty studies. This is for a simple reason. As outlined at length above, we had to exclude several certain PERU..PO.vUR7Y.COMPARISONS --A..... Po..........c.......NO.......E.......... important consumption components from our aggregate in order to achieve comparability between the 1994 and 1997 survey. Most important of these exclusions was the rent value. The definition of the extreme poverty line is generally the value of the food basket alone. Extreme poverty rates would then be the percentage of the population whose total expenditures is not enough to purchase such a basic food basket. However, if we were to apply this definition here, extreme poverty rates would be severely inflated since our total consumption aggregate is lower due to the exclusion of the rent. Therefore, we opted to apply an 'arbitrary' severe poverty line that nevertheless has the property to be perfectly comparable across time. We chose two-thirds of the upper poverty line (in Lima prices) as this rate. A.5. Sensitivity Analyses The sensitivity tests concern whether poverty estimates presented in the main body of the report are sensitive to household composition, size and the poverty line chosen. In the baseline estimates presented above, we compared the per capita poverty line to per capita consumption expenditures in the different years. Although this is common practice in poverty analyses, it is important to note that a number of very stark assumptions are necessary to conduct welfare comparisons with these assumptions. Equivalency Scales. The first test we want to conduct concerns adult equivalency scales. The food basket used in Peru was developed for a 'typical' family of two adults and three children. In this prototype family, it was assumed that different household members have different nutritional requirements. For five members, the food basket contains about 11,900 calories (and 320 proteins) for the Costa and Selva and 13,200 for the Sierra. In both cases the per capita requirements in the family are lower than 2700 calories which the World Health Organization classifies as the minimum caloric intake for an adult male.s' Hence, in the derivation of the basic food basket, children's food requirements were given a lower importance than adult requirements. Although the basic food basket in Peru does take different needs of different household members into account, our base line poverty measurement does not take account household composition. This stems from the (widely applied) 'shortcut' to derive one general per capita (food) poverty line and apply this to all households. For example, a one-person household is measured against this per capita food poverty line and declared poor if (s)he records consumption expenditures below the treshhold -- independent of the person's age or sex. Similarly, a ten-member household with nine children in it would also be measured against (ten times) per capita poverty line which was developed for a family of quite different characteristics. Hence, although derived from a normative concept that different people have different nutritional requirements, the way we measured poverty implicitly assigns everybody an 'adult equivalency' weight of one. The 'pure' way to measure poverty would assign each household in the dataset an individual poverty line that reflects the unique composition of the household. We tested to 56 The per capita requirements of these food baskets are bigh in inteniational compansons. The food poerly kne in the Costa and Selva corresponds to 2367 calones, in the Stena to 2648 calones In most other Latin Amencan countnes, the arage per capita food reqtarements are jet lower at between 2100 and 2200 calonej. PERu: PovERTY CoM4PARISONs --A PoLIcyNoTE- .69 what degree our poverty comparisons are dependent on the implicit choice of an adult equivalency scale of one. We did not derive a new 'exogenous' poverty line for an adult but we conducted the following experiment: first, we chose an equivalence scale which is very different from the one in the base scenario and quite often applied in other countries: 1 for adults, 0.5 for children between ages 5 and 14, and 0.3 for children below age 5.s Second, we then chose a poverty line which results in the same percentage of the Peruvian population being poor in 1994 than when we use no adjustment for adult equivalency scales. This provides the advantage that we can control for the absolute number of poor and can now assess the impact of the adjustment on the regional distribution of poverty and its chance. Explicit adjustment for adult Table6: AdultEquivalencyScalesandPovertyRea, Peru equivalency scales does not alter the 1994ndl997 distribution of the poor nor the ranking ranking change in change in change of poverty from 1994 to 1997 w/ ADS w/o AES poverty poverty 1997 1997 w/AES w/o AES very much. Table 6 includes the 94-97 94-97 results of the robustness test. The first two data columns show the ranking of National - the seven different regions with .Lima i 1 -9.3 -8.3 respect to the simple headcount rate -- Costa Urban 4 4 -1.2 -05 the ranking is not influenced by the Cost. Rural 5 5 -12.0 -4.3 Sierra Urban 2 2 -14.5 -13.2 introduction of adult equivalency Siera Rural 6 6 .61 -3.8 scales (AES). The third and fourth Seli Urban 3 3 -0.5 -0.1 data columns show the change in the Seia Rural 7 7 -7.7 -6.4 poverty rate between 1994 and 1997. For all regions changes go in the same direction, with the urban Sierra and Lima showing substantial gains in poverty reduction. However, the rural Coast shows a much stronger headcount reduction ratio with the adjustment for equivalence scales then without. Here, family structures have changed significantly in the last three years: the average household size in the poorer groups decreased, possibly due to outmigration to urban centers. Economies of Scale. The second robustness test of our result is concerned with economies of scale in consumption. Here, we want to test the assumption that larger households have a distinct advantage over smaller households as they can benefit from sharing commodities (like stoves, furniture, housing infrastructure) or purchase of products in bulk which might be cheaper. However, economies of scale in consumption would pertain to larger households independent of their age composition and is therefore quite distinct from the adult equivalency discussion above which derived from different 'needs' of different household members. There is no single agreed on method to estimate economies of scale in consumptions". However, to assess the importance of scale consumption, analysts often choose a value of 'theta' (the degree of economies of scale) of around 0.6. See Hentschel and LanIouw (1996) for a short discussion of adult equivalency scale ranges. See Lanjouw and Ravallion (1995). 59 This derives from the transformation of household expenditures (E) into per capita terms as E, = E/(n) where n is the household size and 0 is the scale parameter. With 6 equal to 1, no scale economies are assumed. The lower 0, the higher the scale effect PERu:- PovERTY CompARIsoNs-- A PoLicyNoTE 70 We conduct the following evaluation. In order to assess the Graph Al: Economies of Scale, Household Size and Poverty Risk, 1997 importance of the scale effect, we choose a poverty line that produces the same 1 national poverty rate as if we were to use 08 the unadjusted data. Having identified the subset of households which is poor and not-poor in both datasets, we calculate the 04 VAM iconwies . ,escale adjurment poverty risk per household size and 02* compare the scale-adjusted results to the non-adjusted results. These are portrayed a 1 2 3 4 5 6 7 8 9 10 11 in Graph A.. As can be seen, adjustment Household Size for economies of scale has, as expected, a 'flattening' impact on the poverty risk/household size curve. While it remains to be the case that larger households have a higher liklihood of being poor, the differnces in poverty rates between larger and smaller households becomes smaller. Conducting a similar analysis using the poverty gap as the welfare indicator, a very similar outcome results. We also find the relationship between the dependency ratio (i.e. the ratio of non-income earners to income-earners in the household) and household size to be robust with respect to scale economy assumption (Graph A2). Finally, we want to look at another Poverty Risk by age Group and Economies of variable which is intertwined with the above Scale, Peru, 1997 discussion: the age structure and poverty risk. 08 Since household survey data does not permit 0e wfth economs of scale to assess the intra-household distribution of resources, households in their entirety are 4 classified as 'poor' or 'non-poor'. Hence, if it 02 certain age groups (e.g., the elderly) are more likely to be in certain household structures 0 (e.g., larger households), the age profile of Age Groups poverty might also be influenced by scale adjustments. Graph A2 contains the outcomes of our estimations. We find that consistently in all permutations -- independent from the poverty indicators looked at and whether we make adjustments for equivalency and economies of scale or not -- children are the highest poverty risk group. Dominance Analyses. Dominance analyses allow us to study if the observed decline in the poverty and other poverty indicators are dependent on the chosen poverty line. This is accomplished by testing, or more specifically plotting, the development of poverty over classes of poverty measures. We compare the cumulative income distribution functions for 1994 and 1997. One distribution dominates another if the consumption distribution functions for that year lies above that of another year at all levels of consumption. If however, the distributions do not exhibit a clear dominance relationship and the functions cross, it is possib7le to derive a partial ordering over a range of incomes. If we find that first order dominance holds between two different years, this implies that the whole range of poverty measures in one year is higher than in another, i.e. the headcount index, the poverty gap and any measure of the severity of poverty. Graph 7 shows these graphs for Peru (to be added). CATALOGUERSIFILE MICROGRAPHICS DRAFT CONFIDENTIAL DRAFT CONFIDENTIAL Repot No.: 18459 PE Report No.: 18459 PE Type: ER Type: ER