Peru - Did the Ministry of the Presidency Reach the Poor in 1995?

Confidential 17817 - PE Peru Did the Ministry of the Presidency Reach the Poor in 1995? REVISEDOO CONFIDENTIAL July 20, 1996 t~~'j- U LAI l AtJ iiurt INFORIMI10N SERVICES 15-200 Country Operations Division 1 Country Department III Latin America and Caribbean Region The World Bank FILE COPY Peru Did the Ministry of the Presidency Reach the Poor in 1995? Contents 1. Introduction and Summary 1 2. Background: The Ministry in 1995 2 3. Results: Were the Poor Reached in 1995 3 4. Explaining Outcomes: Urban Tilt, Restrictions and Weak Targeting Mechanisms 10 5. A Proposal to Improve Geographic Targeting: Combining Consumption and Basic Needs Information 16 6. Summary and Suggestions 19 References 21 Annex 1: List of Programs and 1995 Budget 22 Annex 2: Geographical Bias Analysis 23 Annex 3 Comparing the FONCODES and an Income-Based Map 25 This report was prepared by a team comprised of Jesko Hentschel (Task Manager), Javier Poggi and Norbert Schady who visited Peru in June 1996. Kathrin Plangemann joined the team after the mission. Dr. Rosa Flores, Dr. Adrian Fajardo and Mrs. Zulema Rosano Estrada Vega from the Ministery of the Presidency worked closely with the team during the ussion in Peru. Peru Did the Ministry of the Presidency Reach the Poor in 1995 Peru Did the Ministry of the Presidency Reach the Poor in 1995? T. Introduction and Summary By announcing its intention to reduce extreme poverty by fifty percent over the next four years, the Government of Peru has set itself a laudable but very ambitious goal. More than four million Peruvians live in extreme poverty today, not able to finance a basic nutritional basket even if they spent everything they earned on food. Another seven million people live in conditions of poverty. While about seventy percent of the population reside in urban areas today, more than half of the poor and extremely poor live in rural and often remote areas.i With scarce public resources and a formidable task in poverty reduction ahead, it becomes crucial for the Government to select and execute programs which lead to substantial and sustainable poverty alleviation while ensuring that the programs reach the target population and show as little leakage as possible to the non-poor. This report examines whether and how the Ministry of the Presidency (MoP) -- the single largest implementation agency of social and basic infrastructure programs in Peru -- reached the poor with the services it offered in 1995. The Ministry -- with its broad sectoral mandate -- executes thirteen different programs which are associated either with the vice- ministries of infrastructure, social or regional development. The report aims to inform and evaluate the targeting effort of the different programs, drawing on detailed geographical expenditure information and a profile of beneficiaries of the largest programs. The report assesses both the geographical distribution of expenditures at the district level and the record of programs to reach poor households within the districts. With respect to geographical targeting, we measure the distribution of expenditures at the district level against the distribution of the extreme poor. Hence, our implicit yardstick was that programs should spend their resources in proportion to the number of extreme poor people in Peru. Obvioulsy, budget allocations are driven not only by the distribution of poverty in the country -- costs of reaching beneficiaries in different areas, local development potential and special support to previously inaccessible areas all play a role. We have abstracted from these considerations here. Conclusions about program performance reached in the report pertain only to reaching the poor' in 1995 and should not be interpreted as generalized program evaluations. A broader analysis of the type and quality of benefits provided by individual programs, the efficiency of their administration and their poverty reduction impact is necessary to complete such broader program evaluations. We did not aim for such a broad assessment here. Three key results derive from our analysis. First, the geographical distribution of the Ministry's expenditures in 1995 was clearly tilted against the poorest and the most rural areas. Second, an examination of the type of households which benefit from the services the Ministry offers, reveals a high degree of leakage of program resources to the non-poor. Evaluating the five largest social and basic infrastructure programs together in 1995, only 17 percent of the program benefits reach the extreme poor, 48 percent went to the poor; hence 35 percent leaked to the non-poor. Third, although about 65 percent of MoP resources were targetable, i.e. they could be channeled to specific districts and households, only a small portion was actually 1 According to estimates by the MoP based on the 1993 Census I Peru Did the Ministry of the Presidency Reach the Poor in 1995 targeted. All three of these findings are linked to the same underlying factors: the widespread lack of coherent targeting and monitoring mechanisms, restrictions on the way programs operated which limited their capacity to reach the poor, and the urban nature of many of these programs. Four suggestions follow from our analysis. First, the Ministry should evaluate the type and mix of services it offers with respect to its overall goal to halve extreme poverty by the year 2000. It goes beyond the scope of our analysis, however, to suggest a different nux or administrative decentralization -- this depends on the poverty reduction impact of alternative programs, the cost of supplying services to different groups and areas in the country, as well as on the program's targeting potential. These issues would need to be assessed in a separate study. Second, the Ministry should coordinate and monitor the geographical distribution of program resources closely. Third, the Ministry should help strengthen the capacity of affiliated programs to go beyond geographical targeting. The distribution of resources to the district level is only the first targeting step; thereafter the difficult task arises to ensure that the poor within the district are reached. Finally, the Ministry should ensure that programs monitor their performance to be able to make changes in program design. This Report is structured as follows. We start out with a short overview of MoP expenditures and programs in 1995 and assess the share of these expenditures which was indeed 'targetable', i.e. available for the Ministry in its poverty reduction campaign. Section 3 assesses how well expenditures of individual programs were geographically distributed in comparison to district-level theoretical asignaciones we derived on the basis of the FONCODES poverty indicator. This section also evaluates whether the expenditures of the five largest social and basic infrastructure programs indeed reached poor and extremely poor households. Section 4 explains the targeting outcomes of the previous section. In section 5 we present a proposal of a general geographical prioritization strategy the Ministry could use for all of its programs, going beyond the use of the FONCODES poverty indicator, which we show to have a technical problem. Section 6 concludes. 2. Background: The Ministry in 1995 The Ministry of the Presidency is the single most important implementation agency of social and basic infrastructure programs in Peru. Including the off-budget housing programs Ute-FONAVI, Banco de Materiales (BanMat) and ENACE which fall under the jurisdiction of the Ministry, its total expenditures amounted to 3.4 billion soles in 1995, representing 20% of overall central government expenditures. The Ministry executes thirteen social and basic infrastructure programs2 (see Annex 1), affiliated with the vice-ministries of social, regional and infrastructure development. 2 Additionally, the Ministry supports two regulatory agencies and cross-finances regional sanitation companies 2 Peru Did the Ministry of the Presidency Reach the Poor in 1995 About 65% of MoP's total expenditures Table 1: Targetable Expenditures of MoP were targetable in 1995, i.e. they could be (billion soles), 1995 channeled as benefits to the poor and extremely poor at the district level -- the Total Expenditures 3.4 smallest political unit which was used for - large hydro projects 0.8 geographical targeting in 1995 (Table 1). - other expenditures 0.4 Apart from social and basic infrastructure targetable benefits 2.2 programs, MoP also financed large Source: Estumates from MoP information. 'Other hydroelectric investment projects for energy expenditures' include overhead of the Miustry, production and irrigation through the expenditures of the regulatory agencies and gross estunates of administrative costs (see Annex 1). National Development Institute INADE. Although benefits (and costs) of these projects also accrued to the population living close to the project areas, a large part of the benefits were realized in much larger regions or even at the national level. While MoP can review the desirability and scale of these large hydro projects, their very nature makes them non-targetable. The Ministry's targetable expenditures were heavily tilted in favor of the water & sewerage sector, education and housing in 1995. Services in these three sectors alone accounted for 70 percent of total MoP targetable resources (Graph 1). As we will Graph 1: Sectoral Distribution of MoP show later, housing expenditures through Expenditures,1995 the credit programs of ENACE and BanMat reached the poor rarely and the extremely Ety( urai ev. (5) Nutnc(3) poor not at all. At the other end of the m4m 7% 9% spectrum, nutrition, rural development and Educabon(6) health received a comparatively small weight 24% in the poverty programs executed by MoP. Programs affiliated with MoP worked on Water& Health(4) average in 4 to 5 different sectors; no clear se Housmg(2) I% assignment between programs and sectors 26% 19% existed. As shown in Graph 1, six programs Source. MP Numbers m brackets refer to progrars act3ve m each sector alone provided water & sewerage infrastructure and another six provided educational infrastructure. 3. Were the Poor Reached in 1995? Targeting of social and basic infrastructure programs involves two distinct steps. First, a geographical poverty map is used to concentrate resources in the poorest areas of the country. Resource assignments based on poverty maps, which provide a 'poverty index' per geographical area (i.e. districts in the case of Peru), are generally a function of the number of poor (or a related indicator) within each area. Second, programs then have to ensure that resources within the designated districts actually reach the poor. Even the poorest areas in 3 An alternative and possibly preferable indicator is the poverty gap per capita which is the amount of money needed to bring an 'average' citizen up to the poverty line. 3 Peru Did the Ministry of the Presidency Reach the Poor in 1995 countries will have a share of non-poor population, so that programs need to develop mechanisms to identify the poor. To reach the poor, both a good geographic Targeting Efficiency Indicators distribution of resources and a mechanism to identify the poor within the geographic areas are * Geographical Targetine Efficiency necessary; performing well on only one front will How well are program resources imply that not all the poor are reached. distributedgeographialy(at the Consequently, in our analysis below, we use two theoretical asignacrones derved from indicators when evaluating program performance. the FONCODES poverty indicator? The Geographical Targeting Indicator measures how * Household Targeting Efficiency- Do closely the actual distribution of program resources programs reach the poor? How closly te acualmuch of program resources 'leak' to matched the district-level theoretical asignaciones. the non-poor? To derive these theoretical asignaciones we used the FONCODES poverty index, which combines eight socio-economic indicators, and weighted it by the population in each district. Therefore, the theoretical asignacion roughly corresponds to the number of poor people in each district. The Household Targeting Indzcator assesses whether programs actually reached poor families. Hence, the household indicator measures the incidence of program benefits -- what share went to the poorest part of the population and what share went to the richest. Evaluated together, the indicators are a measure of two aspects of MoP's programs: were resources distributed in a geographically 'just' manner, did they reach the pockets of poverty in the country?; and: did programs actually reach the poor and extremely poor? While we can expect some close relation between the two indicators, this need not be the case. For example, a secondary education infrastructure program trying to reach the poor population could manage its budget allocation in a way that districts obtain funds in perfect concordance with the theoretical aszgnaciones. But it might build the secondary schools only in 'high demand' areas within the districts in areas which would tend to be rather wealthy. The household targeting indicator is then likely to show that the program does not reach the poor - - although its resources are distributed perfectly geographically. In this case leakage of program resources to the non-poor is high since they were not part of the population group the education program tried to reach. Similarly, a program could have a completely biased geographical distribution -- e.g. concentrating all its resources in Lima -- but on the other hand do a very good job in identifying the poor families within the city. Geographical Targeting Efficiency. Graphs 2.1 to 2.6 compare program expenditures to the theoretical asignaciones.4 On the horizontal axis, we have ranked districts by poverty index in twenty brackets, each bracket representing 5 percent of the total population. Due to the ranking of districts by poverty level, at the very left end of the axis are the five percent of the Peruvian population living in the poorest districts. Since the distribution of resources should be linked to the number of poor people living in every district, the share of expenditures going to the brackets at the left should be much higher than expenditures going 4 We exclude in this analysis CORDECALLAO and CORDELIMA as these programs only work in Callao and Lima Comparing them against a national poverty map would hence be misleading PRONAP is also excluded as its investments only indirectly benefit the population through its cooperation with sanitation companies 4 Peru Did the Ministry of the Presidency Reach the Poor in 1995 to the brackets on the right -- with the population per bracket equal but the poverty index lower, less poor people live i these richer districts. The vertical axis gives us information on what percentage of program expenditures are spent in each population bracket. Comparing the theoretical asignaciones to actual program expenditures visuahzes well programs targeted their expenditures geographically: The closer the theoretical asignaciones and the actual expenditure lines, the better the programs were doing Box 1 summarizes the results of the Box 1: Geographical Targeting geographical targeting indicator by program. Indicator, by Program, 1995 FONCODES achieved the best and INABIF, ENACE and Banco de Materiales the worst geographical Best targeting outcome according to our definition of the FONCODES theoretical distribution. The FONCODES line in Graph 2.4 follows the theoretical asignaciones very COOPOP, PRONAA closely, almost continuously decreasing from left to INFES, INADE, PRASBA right, which implies that FONCODES did indeed Ute-FONAVI spend its money in rough proportion to where the poor are. INABIF (2.5), Banco de Materiales (2.4) and ENACE (2.1), on the other hand, concentrated Banco Matenales their resources in geographical areas which are relatively rich (to the right of the spectrum). These ENACE programs heavily influenced the overall assessment of the geographical targeting effort of MoP in 1995: Taking all ten programs together (Graph 2.6) shows INABIF that the poorest areas of Peru were clearly Worst underserviced while the richer areas obtained a share of resources much higher than they should have. 5 We derive the Geographical Targeting Indicator in the following way: first, districts in Peru are ranked by the FONCODES poverty indicator. Second, we define 20 brackets each containing 5% of the population in Peru. Third, the hypothetical share of resources for each of the brackets is equal to the product of the poverty index times the population per district, divided by the sum of all these products across districts. Finally, the Geographic Targeting Indicator is the sum of the squared deviations of the hypothetical minus the actual share per bracket. Hence, the lower the indicator, the better the geographical targeting outcome. The numbers are as follows FONCODES 0 003, COOPOP 0.02, PRONAA 0 021, INADE 0 029, INFES 0 029,; PRASBA 0.029, Ute-FONAVI 0.031, BanMat 0.06, ENACE 0.078, INABIF 0.142. 5 Peru Did the Ministry of the Presidency Reach the Poor in 1995 Graphs 2.1. to 2.6.: Geographical Distribution of Program Resources, 1995 1 ENACE and INADE 22 Ute-FONAVI and INFES MO0% 6 p200%%WP 150% - 1 150% I0k. 11 I- 50% 50I 10 .. ..... '10% 1Parntdsatncts Ibpuldon Rdvtdu9tncts 1 P . Ophm ap) - EDJA - -NACE ......OpinvI (Nap) - -LTIEAVI - EB 21 COOPOP and PRASBA 24. IONCODES and BanMat 200% %'.200% 150% 150% 100% *.100% 1 Poomt distncts IbpULd n Ri dtsdtncts 1 Pocrstdstcts IpuLan Rid tdutncs . Oprru(Na) - CoPC - -PRASBA ...0.Otl (Map) - -RF MS -BANMAT 220% N5. PRONAA and INABIF 26 Total 10 Pagranu 2%ep 200O%' 2C150 150% 150% 100% 100 .* 50% **... 50%M 00% 00% 1 Part dstncts pbpullm Rid tdastncts 1 Pacd.stct ibpulat Ridarcdutncts . OPtnr (Map) - -PRONAA -NAE[F . OpbnW (Map) -Total 10 Pranw Source: Own calculations based on MoP expenditure data. The 'theoretical map' distribution of funds is derived from the FONCODES poverty indicator; actual distributions are mapped for each program. The Y-axis presents percent of program expenditures; the x-axis are population brackets compnsing 5% of the population each. The first bracket is hence the 5% of the Peruvian population living in districts with the highest poverty indicator, the bracket on the far right contains the 5% of the Peruvian population living in districts with the lowest poverty indicator 6 Peru Did the Ministry of the Presidency Reach the Poor in 1995 Overall, the Ministries resource distribution in 1995 showed a pronounced tilt discrimination agamst the most rural and poorest areas in Peru. That can be shown by applymg statistical analysis to the geographical distribution of MoP's expenditure a the district level in 1995. As shortly described in Annex 2, we tested what factors drove the distribution of the Ministries' expenditures aggregating expenditures of ten programs.6 The results show that geographical budget allocation was largely driven by population by district and to a much lower extent by the poverty level. A significant tilt against the most rural and poorest areas was present. For example, our results show that -- on average - for every 100 soles spent per an extreme poor person in the most rural areas, 195 soles were spent per extreme poor person in the rest of Peru. Household Targeting Indicator. Rather than limiting ourselves to asking whether programs reached the poorest areas in the country, we now assess if they reached poor households. As mentioned above, a program concentrating its resources in rather wealthy districts can nevertheless do a good job in identifying the poor within this district and minimize the participation of non-poor in the services it offers. We can conduct this household level analysis for five programs for which we have information from the Encuesta Nacional de Hogares Sobre Niveles de Vida y Pobreza which INEI conducted in 1995. The survey asked households whether they participated or benefited from FONCODES projects (separate questions for education infrastructure and nutrition/basic infrastructure), whether INFES had constructed or improved the local school which children of the household attended, whether the household had received credits from Banco de Materiales or ENACE, and finally whether the household had received food aid sponsored by PRONAA. Graphs 3.1 to 3.6 show how much of the money spent in these six programs goes to the poor and how much leaks to the non-poor. The horizontal axis m the graphs contain the Peruvian population, distributed in quintiles where the first quintile contains the poorest 20 percent or the extreme poor.7 We can classify the next two quintiles (2 and 3) as the 'poor' while the richest two quintiles contain the non-poor population.8 6 These programs are FONCODES, COOPOP,PRONAA,Ute-Funavi, INFES, PRASBA,INABIF, BANCO DE MATERIALES, ENACE and INADE. 7 Households are ranked according per capita consumption as computed by INEI. 8 Most estimates of poverty in Peru correspond broadly with the above classification. E.g, World Bank (1995) estimates the extreme poor population at 20% and the poor plus extreme poor at 50% 7 Peru Did the Ministry of the Presidency Reach the Poor in 1995 Graphs 3.1. to 3.6.: Household Distribution of Program Resources, 1995 3.1. FONCODES - various projects* 3.2. FONCODES - educational infra. 05 ~ exdudeducization 03 02 02 01 0.1O 0 0 1 2 3 4 5 1 2 3 4 5 Poorest Population Richest Poorest Population Richest % expend. 3.3. PRONAA % expend. 3.4. INFES 0 4 0.3/ 0 3 0. 00 2 01 0 0Ui 1 2 3 4 5 1 2 3 4 5 Poorest Population Richest Poorest Population Richest % expend. 3.5. Banco de Materiales % expend. 3.6. ENACE 0 4 03 01 01 2 3 4 5 1 2 3 4 5 oorest Population Richest Poorest Population Richest Source Own calculations based on INEI, Encuesta Nacional de Hogares Sobre Niveles de Vida y Pobreza, 1995 We calculate the incidence of program expenditures as a function of participating persons. Benefits per person are hence assumed not to vary among beneficiaries 8 Peru Did the Ministry of the Presidency Reach the Poor in 1995 FONCODES and PRONAA did the best job Box 2: Household Targeting Indicator9, in reaching the poor while ENACE and Banco de by Program, 1995 Materiales showed the highest amount of leakage to the non-poor. While avoiding any leakage to the Best non-poor is non-practical and costly, we could expect that programs at least favor the poor population and therefore show a declining share of FONCODES (various) resources going to richer quintiles. But this was only PRONAA the case for FONCODES (various projects in Graph FONCODES (education) 3.1; educational infrastructure in Graph 3.2.), PRONAA (Graph 3.3.) and to a certain degree for INFES INFES, although it had problems reaching the extreme poor in the first quintile. Banco de Banco Matenales Materiales and ENACE, on the other hand, gave more housing credits to the richest two quintiles in Peru rather than to their own proclaimed 'target N group' -- the population in quintiles 2 and 3. As we Worst will show later on, both FONCODES and PRONAA used various mechanisms to ensure that beneficiaries within selected districts were indeed needy of assistance, thereby minimizing leakage of program resources to the more wealthy. Box 2 summarizes our evaluation of the household targeting results for the five programs. Coverage Rates. Coverage rates of programs are generally low. Table 2 contains the coverage rates, i.e. the percentage of the population benefiting from the programs, per quintile and area. It shows that even progarams which we found to be relatively well targeted leave ample room for improvements as they need to get to a larger number of the extreme poor. As can be seen, the educational infrastructure programs of FONCODES and INFES as well as PRONAA reached the most people but also showed relatively low coverage numbers in 1995. ENACE's impact measured by the number of people it reaches with its operation, was negligible. Summary. This section examined the geographical and household targeting record of MoP programs. As for geographical targeting, we examined ten programs which represented about 90 percent of targetable expenditures of the Ministry in 1995. We found a significant variation in program performance with FONCODES and PRONAA showing the best indicators and INABIF, ENACE and BanMat the worst. But even for the relatively well performing programs substantial progress can be made as programs within districts show considerable leakage and low coverage rates. MoPs expenditures were largely driven by the population per district and to a much lesser extent by poverty. A pronounced tilt against the most rural and poorest districts is apparent. With respect to the household targeting result, we found that of five large social and infrastructure programs (amounting to 60 percent of targetable expenditures in 1995), more than one third leaked to the richest forty percent of the population -- resources that were badly needed for the extreme poor to fulfill MoP's mission in poverty reduction. 9 For the Household Targeting Indicator we simply calculate the percentage of expenditures per program going to the wealthiest two expenditure quintiles, i e. leakage to the non-poor 9 Peru Did the Ministry of the Presidency Reach the Poor in 1995 Table 2: Estimated Coverage Rates of Five Programs, 1995 Program Population Quintiles Total Estimated 1 2 3 4 5 Beneficiaries FONCODES* 3.5 1.8 1.6 1.2 0.6 394,000 FONCODES -- education 10.7 16.2 12.2 11.7 5.5 (2) INFES 4.8 10.2 10.2 8.7 5.8 (2) PRONAA 12.0 12.6 6.6 5.6 2.8 1,824,000 Banco Materiales 0.7 2.1 2.6 3.9 2.5 537,000 ENACE 0.0 0.5 0.7 0.1 0.7 155,000 Source: Calculations based on INEI, Encuesta Nacional de Hogares Sobre Niveles de Vida, 1995. 1 FONCODES activities here include nutrition programs, rural electnfication, water and sewerage projects and literacy campaigns. 2 The 1995 LSMS did not record the number of children benefiting from educational infrastructure programs. We therefore assumed that all household members benefit from these programs. Reported coverage rates will be good approximations of the proportion of children attended as long as the ratio of benefiting children to other household members is relatively constant across quintiles. If, as we know from demographic data, the ratio of children to other household members decreases from the first to the fifth quintile, we will be under-estimating coverage rates in the poorest quintiles and over-estimating them in the richest quintiles 4. Explaning Outcomes: Urban Tilt, Restrictions and Weak Targeting Mechanisms The findings of section 3 are all linked to three underlying factors: (a) the often urban nature of the programs; (b) restrictions attached to many programs which limit their capacity to reach the poor; and (c) little use of targeting and monitoring mechanisms. This section presents our program evaluations with respect to nature, restrictions, and targeting & monitoring mechanisms in a condensed form. It also includes evaluations of the targeting efficiency discussed in the previous section. For a more detailed program descriptions, the interested reader is referred to an available background document.1o The remainder of this section discusses in a general way the conclusions reached above. 10 Peru - Targeting Mechanisms and Evaluation Systems of the Ministry of the Presidency, mimeo, World Bank, Washington DC, July 1996 10 Peru Did the Ministry of the Presidency Reach the Poor in 1995 (i) Urban Tilt MoPs operations had an urban tilt." Many programs offered their services only in urban Table 3: Rural versus Urban Program areas. Of the thirteen programs, six were Expenditures, 1995 exclusively geared towards the urban population. These programs together spent 51% of MoP Program Naturel number percent of MoP resources in 1995. Five other programs had expenditure operations both in urban and rural areas but spent only urban 6 51 over 70% of their resources in urban areas with the only rural 2 5 exception of FONCODES (50 percent) and urban and rural 5 44 COOPOP (62 percent). The only programs that spent exclusively in rural areas, PRASBA and percent expenditures urban areas -- 75% INADE (rural part), were relatively small. Overall, percent expenditures rural areas - 25% 75% of MoP resources went to urban areas while 1 Classification:: Urban. Ute-Fonavi, Enace, BanMat, about 55% of the extreme poor live in rural areas.12 Cordelima, Cordecallao, Inabif Rural: Inade (only As shown in Graph 4, there is a clear relationship part), Prasba; Both. FONCODES, Infes, Pronaa, between the degree of rurality of departments and Pronap,Coopop. the expenditure per extreme poor person.13 (ii) Restrictions Eight of the nine programs included in Table 4 had restrictions attached to their operations which limited their flexibility to reach the poor. BanMat, ENACE and Ute- FONAVI show the most serious restrictions. These three programs alone accounted for close to 50% of the targetable resources of MoP. Since these are credit operations, programs require a minimun income for beneficiaries to access funds which excludes the extreme poor. In addition, these programs have other legal and technical restrictions such as requiring beneficiaries to be FONAVI members and legal owners of land. These are likely to reinforce the tilt against the extreme poor. Although a better targeting and monitoring methodology can help these programs to reach more of the poor, substantial changes will not be obtained if the nature of the programs is maintained. As can be observed in Table 4, BanMat is a good 11 Due to migration and urbanization, only about 30 percent of the population lives in rural Peru today. However, according to the FONCODES poverty indicator, about half of the Ministry's resources should be directed towards rural areas. The Census reports on the percentage of the population per district that lives in rural areas. We calculated the share of the resources to be directed towards the rural poor and extreme poor by assuming that poverty (measured by the FONCODES indicator) is equally distributed between urban and rural areas within the districts - an assumption that will bias rural poverty estimates downward as long as rural poverty is indeed higher than urban poverty Indeed, according to consumption data from INEI 75 percent of the extreme poor and 55 percent of the poor (including the extreme poor) live in rural areas. 12 Closely linked to this finding of underserving rural areas is the tendency to underspend in areas with a high indigenous population. According to the FONCODES map, 21 percent of MoP's resources should go to such areas - most of them rural - but only 13 percent of expenditures reached these geographical regions with a high indigenous population 13 The reported data in the graph are derived by diiding the actual expenditures by the amount of extreme poor persons per department. Obviously - and as shown above - only a fraction of the expenditures actually goes to the extremely poor 11 Peru Did the Minustry of the Presidency Reach the Poor in 1995 example of a program which had a coherent targeting mechanism but nevertheless was not able to achieve a good targeting outcome.14 2000- Graph 4: Expenditures per Extreme Poor, 1995 (soles) Rurality 1000 - 500- (iii) Lack of Coherent Targeting & Monitoring Mechanisms. Table 4 includes evaluations ('rankings') for the targeting & monitoring practices of nine programs. These rankings are arbitrary to a certain degree but reported here to summarize how we evaluated the targeting effort. Box 3 describes what constitutes good targeting and monitoring practices, hence against what yardstick we have evaluated programs. We limit ourselves here to summarize the targeting and monitoring performance of programs affiliated with MoP: in 1995, only two programs (FONCODES and PRONAA) used a poverty map to determine geographic budget allocations. Even these programs did not monitor actual disbursements against the allocations on a continous basis but only once a year. The relatively poor geographic targeting outcome discussed above is a reflection of the limited use of poverty maps; on-site visits (BanMat, FONCODES, PRASBA) and means-tests (INABIF) are applied more frequently but their use is not universal. Programs which received FONAVI funds (Ute-FONAVI, ENACE, BanMat) conducted means-tests to assess the repayment capacity (and not the poverty level) of the intended beneficiaries; few programs encouraged self-targeting. The ones that did were PRONAA, locating distribution centers in unattractive areas and distributing lower-quality food, and FONCODES paying lower than market wages; 14 Examples from many other Latin American countries show that publicly supported housing credits mostly end up in the hands of the upper-middle income classes For this reason many countries now switch to community-based upgrading projects which comprise the legalization of lands and technical assistance for house and community upgrading. See Persaud (1992). 12 Table 4: Targeting and Monitoring in MoP Programs, 1995-1996 Program Intended Targeting Outcome, Targeting Mechanism-19953 Targeting Mechanism- Targeting Restnctions Monitoring & Evaluation-19955 Monitonng & Evaluation-1996 target groups2 1995 1996 PRONAA Rural and Geographical Ranking 2 Rankinp 2 Ranking - International food Ranking 3 Ranking 2 (nutrition) urban, - see Box I and Graph - Poverty map used to - Use of Lucha contre IA donations often have - Limited to some programs - Survey of comedores populares extremely 35 allocate about 80% of pobreza poverty unap strings attached - No exitcriteria for comedores - No regular ex-post beneficiary poor and poor budget to regional offices, - Construction of intra- populares assessment Households Ranking 2 no clear mechanism for district poverty maps - No regular ex-post beneficiary - No regular ex-post assessment of - leakage (to qutntiles 4 allocation of these resources with INE[ assessment changes in nutritional status and 5) 21% within catchment area, - self-targeting - No regular ex-post assessment - see Box 3 and Graph Asignaciones sometimes - use of mirco-surveys to of changes in nutritional status 43 disregarded determine location of - Self-targeting at comedores comedor populares - Some priority to rural areas and vulnerable populations - Attempt to purchase food from low-income farmers INABIF Mainly urban. Geographical Ranking 4 Ran 3 Rankine3- None Ranking 4 Ranking 4 (family extremely - see Box 1 and Graph - Self-selection, verified by - No changes envisioned - No ex-post evaluation - No changes envisioned welfare) poor and poor 3 5 social worker, but without clear evaluation criteria Households Ranking na - No use of poverty map Banco de Urban, poor Geographical Ranking 4 Ranking 2 Ranking 2 Beneficiaries must be Ranking Ranking Mateniales - see Box I and Graph - Geographic targeting based - No changes envisioned FONAVI members, - Correspondence between - No changes envisioned (housing) 34 on "housing deficit", able to repay the loan, asignaciones and disbursements proportion of population in and legal owners of monitored on a quarterly basis Households Ranking 4 urban areas, and previous the land and home in - No regular ex-post beneficiary - leakage (to quintiles 4 Banco de Materiales loans question assessment and 5) 54% - On-site visit - see Box 3 and Graph - Self-targeting (cap on 45 maximum size of loan) ENACE Urban, poor Geographical Ranking 4 Ranking 4 Ranking: 4 Beneficiaries must be Ranking: 4 Ranking- 4 (housing) - see Box I and Graph - Criteria for pnorituation - No changes envisioned FONAVI members, - No ex-post monitoring and - No changes envisioned 3 1 mainly technical and able to repay the loan, evaluation financial and legal owners of Households Ranking 4 the land and home in - leakage (to quintiles 4 question and S) 51% - see Box 3 and Graph 36 UTE- Mainly urban, Ranking 4 Ranking 4 - Beneficiaries must be Ranking 4 Ranking: 4 FONAVI poor Geographical Ranking 3 - Criteria for prioritization - No changes envisioned FONAVI members - No ex-post evaluation - No(hanges envisioned (electrification - see Box and Graph3 2 mainly technical and and able to repay the , water & financial loan sanitation) Households Ranking na - 80 % of households in the community in question must be willing to take out the loan - Approval by water or power company in the area in question Table 4 (cont.): Targeting and Montoring in MoP Programs, 1995-1996 Program Intended Targeting Outcome, Targeting Miechanism-1995 I Argeting Mechanism- Targeting Restrictions lontonng & Evaluation-1995 Monstonng & Evaluation-19% target groups 1995 1996 COOPOP Rural and Geographical Rankin9 2 Rankin3 Rankin - All individual projects Rankini 4 Ranking 4 (multi-sector Urban, - see Box 1 and Graph Ad-hoc priontzation. - No changes pre-approved by the - No ex-post evaluation - No changes envisioned microprojects) extremely 33 trying to give preference to envisioned MEF poor and rural and under-served poor Households Ranking na areas FONCODES Urban and Geographical Ranking I Rankins 2 Rankin- I - Approval by Ministry Ranking 2 Ranking I (multi-sector rural, - see Box 1 and Graph - 80% of budget allocated - District-level poverty of Educatzon, Health, - Yearly ex-post beneficiary - Quarterly monitoring of correlation microprojects) extremely 34 according to province-level map Transportation, or other assessments between asignaciones and poor and poverty map - New survey at the inshtution which has - No regular monitoring of expenditures poor Households Ranking 2 - On-site visit level of "centros signed a legal correlahon between asignaciones - Monitoring of access by indigenous - leakage (to quintdes 4 - self-targebing poblados" agreement (convenio) and expenditures at the groups and 5) 24% - Demand-stimulation with FONCODES departmental level - ex post beneficiary assessments - see Box 3 and Graphs in under-served areas 4 1 and 4 2 - tow wages used to self-target investments - Computerized prioritizaton system, which includes information on poverty level and lack of basic services INFES Mainly Geographical Ranking 3 Rankin. 3 Ranking 2 Approval by Ministry Ranking 4 (education) urban, poor - see Box I and Graph - Prioritization according to - Use of the of Education - No ex-post evaluation - No changes envisioned 3 2 the size of the student FONCODES poverty body, physical condition indicator Households Ranking 3 and age of the school - leakage (to quintiles 4 buildings and 5) 35% - No use of poverty maps or - see Box 3 and Graph means testing 44 PRASBA Mainly ruraL Geograhical nking 3 Rankin 3 Ranking3- Projects must be Ranking 4 Ranking 4 (water & extremely - see Box I and Graph - First priority given to - No changes proposed directly by - No ex-post evaluation - No changes envisioned sanitation) poor 33 projects identified directly envisioned President's Office by President Households Ranking na - Cross-check with FONCODES poverty indicator - on-site field visit S IK,nefaary nuisixrs are estunate froin the programus theniselves and need not reflet t true Ix-nefistry nunilwrs 2 Por the purposes of this paper, the two lowest population decales are defined as "extremely poor" and the next three deales are defined db "poor" Tiss generally (orre.Iuids withs other esttmates of t prevalence of extreme poverty and poverty in Peru 3 Targeting mechanism rankings are based on a number of criteria, Mduding (i) use of a poverty map to assign resources on a geographical basis, (it) use of on-site visits or other forms of means-testing, (ii program design which explicitly encourages self-targetrig. 4 Monitoring and evaluation rankings are based on a number of criteria, including (i) regular monitoring of the correspondence between asignaciones and actual expenditures, (u) beneficiary assessments other forms of external program evaluation; (in) updating nformation to reflect program impact and changes in the beneficiary population Peru Did the Ministry of the Presidency Reach the Poor in 1995 monitoring and evaluation was generally insufficient: programs did not regularly monitor the correspondence between asignaciones and actual expenditures; only one program (FONCODES) conducted regular ex-post beneficiary assessments; ongoing programs (INABIF, PRONAA) have no clear exit criteria; only one program (Banco Materiales) updates information to take into account program impact; attempts to monitor the impact of interventions using relevant indicators -- such as nutritional status -- were limited; and there were no exphcit criteria to evaluate access to programs by vulnerable populations, such as indigenous peoples and women. Box 3: Good Targeting & Monitoring: Concepts 'Targeting mechanism' describes the process through which a program channels its resources to an intended beneficiary group, trying to maximize the number of recipients in the group whule mninumnng 'leakage' to people outside the intended group. Successful programs often combine three different instruments to achieve effective targeting: (a) a geographical poverty map is used to assign resources to as-small-as-possible areas; (b) built-in incentives encourage the poor to participate in programs while they deter the non-poor. Examples here include offering work at wages below the imphat market rate or distributing lower quahty foods in marginal areas; and (c) means-tests, social worker evaluations or on-site visits are used to ensure that the poor within the designated areas are indeed reached; often decentrahzed administration of programs helps to use these individual assessment mechanisms. Obviously, the nature of different programs often dictates to what degree these instruments can be combined to aclueve a good targeting outcome. Efficient targeting is closely linked to the ability of programs to change and fine-tune the program design, i.e. a well-designed monitoring and evaluation system. Such a system is composed of (a) periodic and independent evaluations of the socio-economic profile of program beneficiaries through surveys; (b) monitoring of actual and planned expenditures by geographic region; and (c) evaluating and updating the indicators used to identify the target population. As an overall judgment, we find that few of MoP's programs made use of targeting and monitoring systems to reach the poor. But these same few that invested in targeting design showed better results as they spent their moneys in a geographically more just fashion and deliver it to poor households within designated geographical areas. Both FONCODES and PRONAA are good examples here and both programs have undertaken further refinements in 1996 (Box 4). 15 Peru Did the Ministry of the Presidency Reach the Poor in 1995 Box 4: Good Practice - Targeting and Monitoring of FONCODES and PRONAA in 1996 While FONCODES and PRONAA already had quite elaborate targeting and monitoring schemes in 1995, both are currently working on further improvments. The new systems are 'good practice' and can be examples for other programs to refine their targeting and monitorng efforts: In 1996, FONCODES and PRONAA have combined geographical targeting with means-tests and incentives for self-selection of the poor. FONCODES 1996' asignaci6n combines mformation about the number of people livmg in populated centres which have between 40 and 300 households with a district-level poverty index from the poverty map used in previous years. Small corrections to the asignacion were then made to reflect spending capacity of regional offices, flows of funds from directed donations and the special needs of border regions. A computerized system helps to prioritize projects according to poverty level, lack of services and absorptive capacity of communities. Demand stimulation is used in areas of extreme poverty where communities cannot propose projects on their own. Individual assessment mechanisms such as mandatory on-site visits are used to verify informally the poverty level of the community in question, providing valuable additional information for refined targeting. Finally, self-targeting by low wages (below the market rate) is a mechanism employed to ensure that only the poor have the incentive to take the employment offered by projects. In 1996, PRONAA adopted the MoP or Lucha contra la Pobreza poverty map which is selectively augmented by intra-district poverty maps and on-site visits (up to now only in Lima) to determine locations of future popular kitchens in heterogeneous areas. By their very nature, the popular kitchens contain an important self-targeting element. Both programs also employ a thorough monitoring system. Expenditures are regularly measured against asignaciones. FONCODES uses beneficiary assessments to keep track of whom its projects reach; PRONAA now relies on surveys of the popular kitchens to do the same. These surveys will help both programs to improve the selection of indicators and project design to reach their intended target group. Costs of installing good targeting and monitoring systems are quite low when compared to the benefits of directing a higher share of program resources to the poor. Staff of FONCODES estimates their targeting costs (i.e. the resource cost of developing poverty maps, conducting on-site visits and other individual assessment methods) as 5 to 10 percent of operating costs with PRONAA gives a somewhat higher estimate -- classification of costs as purely due to 'targeting' is arbitrary. International comparisons have shown that, depending on the type of instruments used, targeting costs range from 0.5 to 1.5 percent of total project costs or about 5 to 15 percent of operating costs. At the same time the effectiveness of programs to reach the poor is clearly linked to these investments in targetmg and monitoring systems.15 5. A Proposal to Improve Geographical Targeting: Combining Consumption and Basic Needs Information While the last section emphasized that different instruments need to be combined to achieve an efficient targeting outcome, the first step for many programs is to prioritize and 15 Grosh (1994) 16 Peru Did the Ministry of the Presidency Reach the Poor in 1995 pre-allocate their expenditure on a geographic basis. Many MoP programs currently do not conduct such a geographical allocation. In this section, therefore, we shortly discuss the MoP and FONCODES poverty maps and the characteristics which may reduce or affect its accuracy as a targeting mechanism. If maintained, these baseline maps would have to be reformulated. We make an alternative proposal of how a generalized geographical map could be developed and how each individual program could operationalize it. One word of caution is in place here. While geographic poverty maps are an essential part to guide the distribution of resources of implementing agencies, there are also other factors which play a role in deciding final budget allocations. These are, for example, the costs of providing the population with a specific service. Often these costs vary with the remoteness of the area and might become prohibitive.16 Similarly, the regional development potential, local economy aspects or serving previously inaccesible areas also have to be taken into consideration. The MoP and FONCODES Poverty Maps. These maps combine eight socio-economic indicators, seven of which are drawn from the 1993 IX Population and IV Housing Census, including illiteracy rates, school attendance rate, a housing quality index (roof material), crowding index, and access to basic services such as potable water, sanitation and electricity at the district level. To standardize the indicators (which are all defined as 'population share with deficiency'), each one is divided by the value in the 'best' district, hence by the minimum value. The seven indicators are then added and they account for half of the poverty index. Information on the rate of chronic malnutrition from a 1994 census of the height of school- aged children accounts for the other half of the poverty index, standardized in the same way. Problems Associated with Table 5: The MoP and FONCODES Poverty Maps - the FONCODES Map. There are Intended and Actual Weights three problems associated with the poverty index which is the Variable mtended actual basis of the maps. First, the weight (%) weight method of deriving the composite % children malnourished 50.00 15.3 indicator assigned arbitrary % pop. dliterate 7.14 3.4 weights other than those intended to % hholds in crowded houses 7.14 3.0 its components. Originally, % children not attending school 7.14 2.2 nutrition was meant to carry a % hholds without water 7.14 8.8 weight of 50% and each of the % hholds without sewerage 7.14 7.4 % hholds without electricity 7.14 21.6 othe cometa wveity o. I% hholds with madequate roof 7.14 38.3 7.1% in the total poverty index. But after dividing each variable by its minimum value, the range of values per indicator was different -- e.g. the 'housing indicator' ranges from 1 (best) to 524 (worst) and the nutrition indicator varies from 1 (best) to 46 (worst). Since the standardized indicators were then added, the implicit weights are a direct function of the ranges -- or better variances -- of the different components. Table 5 shows what was intended and what the actual outcome was: the 'inadequate roof' and the 16 For example we found that the cost per beneficiary household in Ute-FONAVI projects (electricity and water) increases in a statistically significant way with the degree of rurality, i e the percentage of rural population per district. 17 Peru Did the Ministry of the Presidency Reach the Poor in 1995 electricity indicator together carry a weight of 60 percent in the poverty index. Both education variables -- which m cross-country studies clearly emerge as the closest links to poverty -- carry a combined weight of only 6.4 percent. If the Ministry decides to continue using this indicator, these weights would have to be adjusted.17 The second problem with these maps concerns the link between basic needs and poverty. The maps contain measures of unmet basic needs with the highest (implicit) weights attached to basic services and living conditions. But access to such services as electricity, water and sewerage as well as housing materials vary considerably across the very diverse regions in Peru. Giving fixed weights to these indicators implies that, for example, a household without a water connection in a rural village endures the same hardship as an urban household in a poor area in Lima, although the rural family might have access to river or well water and the costs of supplymg different types of households with the sewerage are very different. Further, while lacking access to water might be a good indicator to distinguish poor from non-poor households in urban Peru, it is likely to be a very bad indicator in rural Peru, where almost none of such connections exist -- independent of the welfare level of the families. Finally, the maps are difficult to update. All of the indicators which make up the map are based on population and height census information. While currently accurate, this type of information will be outdated very soon. Updating will be difficult since census data collections are not conducted regularly. An Alternative: A Consumption-based Poverty Map. Household consumption is generally believed to be a better indicator of poverty than an Unmet Basic Needs indicator -- for two reasons. First, household consumption is a direct measure of how much family members consume of food and non-food items. Minimum nutritional requirements and a minimum consumption basket can be directly compared with a household's consumption level, determining whether families are 'poor' or 'extreme poor'. Second, a well-defined measure of consumption encompasses the satisfaction of basic needs. Evaluating the consumption of water, electricity and sewerage services at their value to the consumer implies that they can be included in the measurable basket of goods of the consumption measure. Together with INEI, the Ministry can develop such a geographical consumption based poverty map. This map would result from crossing information based on the recently completed Encuesta de Hogares de Niveles de Vida y de Pobreza (1995) with information from the Census (1993). Starting from a comprehensive and analytically sound definition of household consumption, econometric models which examine the relationship between consumption and a variety of exogenous variables could be developed based on the Encuesta. These models would be used to impute household consumption using Census information and to derive a generalized geographic poverty map which goes beyond the basic services and nutrition information. Further, maps could be updated regularly as long as the household surveys are conducted frequently. 17 The adjustment to ensure that the weights of the variables are equal can take two forms first, each indicator can be ranked ranging from 1 (best district) to 1790 (worst district) Ranks can then be simply be added or given weights as desired. Second, indicators can be normalized so that their mean is zero and their standard deviation equal to one Aggregations and weights can then proceed as desired 18 Peru Did the Ministry of the Presidency Reach the Poor in 1995 INEI has made great advances to develop such a generalized map. INEI has produced an 'extreme poverty map' using income as the baseline variable which is applied from the Encuesta de Hogares to the Census. The map includes the percentage of extremely poor population per district and could be easily extended to mirror other poverty indicators such as the poverty gap. Although consumption is undoubtedly a better indicator to determine poverty levels, the existing INE map is well suited to test the accuracy of the basic indicator maps used by the MoP and FONCODES. As Annex 2 shows, significant variations between the two maps exist, which reinforces the notion that the Ministry should consider developing a generalized consumption-based poverty map. Pnoritization: Combining a Consumption-Based Poverty Map with Basic Needs Information. Combining a consumption-based poverty map with basic needs information could provide individual programs with a straightforward prioritization methodology. The joint use of both information goes to the very heart of defining a target group for a service offered. For example, what should the target group for FONCODES water projects be? The poor population, the population currently without access to water, or the poor population currently without access to water? Combining the two maps would allow FONCODES to target the latter -- the poor population without access to water. Currently, budget allocations are based on the an Unmet Basic Needs indicator, independent of whether the population actually has water access or not. Such a prioritization could be applied to almost all of the programs the Ministry offers - be it in nutrition, education, health, electricity or water & sewerage infrastructure. The following pages show how such a combination would work. The colored map shows poverty levels in the department of Arequipa according to the generalized map; the poorest districts are marked as red. Assume now that a program wants to determine how to allocate its water projects. In this case it would 'overlay' the generalized map with its own information on the percentage of the population without access to water. The transparency includes this latter information. Here the districts with the highest percentage without water are horizontally striped. Red areas with horizontal stripes now become the priority areas for investments as they are poor areas with a high percentage of the population without access to water. 6. Summary and Suggestons This paper examined whether the Ministry of the Presidency reached the poor and extremely poor with the large number of different programs it administers. We can summarize our results as follows: the geographical distribution of the Ministry's expenditures showed a clear preference for urban areas and a tilt against the most rural areas; - examining what type of households benefited from the services the Ministry offers, we found that a high degree of leakage of program resources to the non-poor occured in 1995. Evaluating the five largest social and basic infrastructure programs together, only 17 percent of the program benefits reached the extreme poor, 48 percent went to the poor; hence 35 percent leak to the non-poor; 19 DEPArrå~rT o DE ARQUIPA Poorest 2 É5 Richest i' ý4 5 3'\ 4z S,a Af" 6n /A("~ i ýi«it4k lø Peru Did the Ministry of the PresidenSy Reach the Poor in 1995 although about 65 percent of MoP resources were targetable m 1995, i.e. they could be channeled to specific districts and households, only a small portion was actually targeted. The following reasons expain the above findings. many programs offered their services only in urban areas. Of the thirteen programs, six were exclusively geared towards the urban population. For example, by their very nature, housing programs did not reach out to the rural poor; restrictions inhibited many programs from reaching the poor and especially the extremely poor. Among others, housing programs and Ute-FONAVI make it mandatory for beneficiaries to be FONAVI members, which generally requires work in the formal sector. Minimum income and legal requirements of the same programs exclude the extreme poor from obtaming credits; many programs lacked coherent targeting & monitorng mechanisms. In 1995, only two programs (FONCODES and PRONAA) used a geographic poverty map to determine budget allocations. On-site visits and means-tests were applied more frequently but their use was not universal. Few programs explicitly encouraged self- targeting of beneficiaries; all but one program lack the use of monitorng devices and hence were not able to assess whether they reached their intended target groups. Four suggestions follow from the above. First, the Ministry should evaluate the type and mix of services it offers with respect to its overall goal to halve extreme poverty by the year 2000. Water & sewerage, housing and education investments currently have a much higher importance than nutrition, health and rural development investments. By their very nature, most programs cater to the urban population, leaving the fifty percent poor who live in rural areas unreached. It goes beyond the scope of our analysis, however, to suggest a different mix of programs - this depends on the poverty reduction impact of alternative programs, the cost of supplying services to different groups and areas in the country, as well as on the program's targeting potential. These issues would need to be assessed in a separate study. Second, the Ministry should coordinate and monitor the geographical distribution of program resources. If the Ministry decides to continue using an Unmet Basic Needs map, this would have to be corrected for its technical problems. Alternatively, the Ministry could develop a consumption-based geographic poverty map together with INEI which could be used by individual programs to prioritize their investments. Third, the Ministry should help strengthen the capacity of affiliated programs to go beyond geographical targeting. The distribution of resources to the district level is only the first targeting step; thereafter the difficult task arises to ensure that the poor within the district are reached. On-site visits and other means tests offer possibilities which many programs have not yet fully exploited. A resource team within the Ministry could provide individual programs with the necessary technical support. Finally, the Ministry should ensure that programs monitor their performance to be able to make changes in program design. Monitoring though surveys and expenditure control will give programs the yardstick against which they can measure their cost and success. 20 Peru Did the Ministry of the Presidency Reach the Poor in 1995 References Grosh, Margret (1994), Administering Targeted Social Programs in Latin America, World Bank Regional and Sectoral Studies, Washington DC Persaud, Thakoor (1992), Housing Delivery Systems and the Urban Poor: A Comparison Among Six Latin Countries, Latin America and the Caribbean Technical Department, Regional Studies Program 23, Washington DC World Bank (1995), Peru Policy Notes, Washington DC 21 Peru Did the Ministry of the Presidency Reach the Poor in 1995 Annex 1: List of Programs Affiliated with the Ministry of the Presidency and 1995 Budget (thousand soles)1,2 Ministry o the Presidency (overhead) 51,572 Vice-ministry social and other: COOPOP Oficina de Cooperación Popular 20,620* INABIF Instituto Nacional de Bienestar Famihar 35,285* INFES Instituto Nacional de Infraestructura Educativa y de Salud 363,000* PRONAA Programa Nacional de Asistencia Alimentaria 175,348* FONCODES Fondo Nacional de Compensación y Desarrollo Social 328,735* Vice-minLstery Infrastructure: Ute-FONAVI Unudad Tecnica -- Fondo Naaonal de Vivienda 632,641* SSS Superintendencia Nacional de Servicios de Saneamiento 15,515 SENASS Servicio de Agua Potable y Sanamiento -- Provincias 47,202* SEDAPAL Servicio de Agua Potable y Alcantarillado de Luna 27,900 ENACE Empresa Nacional de Edificaciones 279,105* BANMAT Banco de Materiales 149,458* SBN Superintendencia de Bienes Nacionales 5,010 PRONAP Programa Nacional de Agua Potable y Alcantarillado 44,589* PRASBA Programa de Apoyo al Saneamiento Básico 689* Vice-ministry Regional Development: PDZE Programa Desarrollo Zonas de Emergencia 1,235 INADE Inst. Nacional de Desarrollo 931,013* CORDECALLAO Corporación de Desarrollo de Callao 71,692* CORDELIMA Corporacion Departamental de Desarrollo de Lima 7,152* Total 3,187,761 1 lf marked with a (*) budget estimates stem from the respective programs 2 The discrepancy between the total estimate given here and the one given in Table 1 is due to the inclusion of adnunistrative costs in the latter. 22 Peru Did the Ministry of the Presidency Reach the Poor in 1995 Annex 2: Geographical Analysis A regression analysis shows that MoP's expenditures in 1995 were largely driven by the population per district and were significantly tilted against the poorest and most rural areas in Peru. We used the following model to explain the determinants of the MoP's expenditures: Expen, = A Pop, a Pov, P exp(xDRur,i + 4 Dpov,i + 8 DInd,i) + S where Expeni is the total MoP expenditure in district i; A is a constant term; Popi is the population in district i; Pov, is the FONCODES poverty index in district i; DRur,i is a dummy variable: 1 if the district is among the most rural districts home to 20 percent of the population; and 0 otherwise; Dr,, is a dummy variable: I if the district is among the poorest districts home to 20 percent of the population; and 0 otherwise; D ind,, is a dummy variable: 1 if the district is among the most indigenous districts home to 20 percent of the population; and 0 otherwise; 61 is the error term. If MoP's expenditures were to follow exactly the theoretical asignaciones, we would expect the parameters a and P to be equal to 1 and all other parameters equal to zero. Regression results for the log-regression are the following: Parameter value t-statistic A -7.32 -12.25 a 1.32 36.49 p 0.48 3.01 x -0.25 -2.24 -0.95 -7.18 8 0.28 3.03 R-squared 47.1 f-value 317.8 These results show that MoP's expenditure distribution was largely driven by the population (a, the population elasticity has a value significantly greater than one) and much less so by poverty (p, the poverty elasticity, is significantly smaller than one). Further, other things being equal, the most rural and the poorest districts obtained significantly smaller budget allocations than they should have (both X and * are significantly negative). For 23 Peru Did the Ministry of the Presidency Reach the Poor in 1995 example, for every one hundred soles spent in the poorest districts (amounting to 20 percent of the population), 195 soles were spent in the richer ones. Finally, we can note that there does not seem to appear an independent discrimination against areas with a high indigenous population. On the contrary, contrblling for the other exogenous variables, we found a significantly positive relationship between indigeneity and expenditures per district. This does not mean that indigenous districts did not obtain a smaller share of funds than they should have according to the theoretical asignaciones. However because rurality and indigeneity are closely linked what appeared to be the driving factor in the expenditure allocation was the tilt against rural areas rather than against indigenous ones. 24 Peru Did the Ministry of the Presidency Reach the Poor in 1995 Annex 3: Comparing the FONCODES and an Income-Based Poverty Map This Annex compares two geographic poverty maps and emphasizes that the Ministry should consider the development of a consumption-based poverty map. The first geographic map stems from FONCODES which is discussed in detail in section 4. Districts obtain a poverty index based on a weighted average of eight unmet basic needs indicators including access to basic services, education and nutrition variables. The second map was developed by INEI by crossing information from the Encuesta Nacional de Hogares sobre Niveles de Vida y Pobreza with Census information. Based on the household survey, INEI researchers developed models which explained household income as a function of a host of exogenous variables, all of which are also included in the Census. Employing the models to impute individual household income with the Census information allowed INEI to calculate poverty and extreme poverty rates per district in Peru (applying poverty lines). The two maps can be compared by developing a transition matrix which is presented in the table. Each cell in the matrix contains the percentage of the population in the decile on the left (the INEI extreme poverty map) which is included in the decile on top (the FONCODES map). If all cross-diagonal entry elements were one, the two maps would be perfectly identical. The more significant percentages we detect further away from the diagonal, the more do the maps differ. As the reader can observe, on both ends of the table correlations are high on or adjacent to the diagonal. However, a large number of variation between the maps can be observed in deciles 3 to 8 which without doubt include a large proportion of the poor population of Peru. Transition Matrix Comparing District Rankings Between FONCODES Poverty Map and Extreme Poverty Map from INEI FONCODES Map Decile 10 9 8 7 6 5 4 3 2 1 10 59.5 36.4 0.4 0 0 1.9 0.3 0.5 0.8 0.3 INEI 9 30.4 17.0 33.6 0.5 12.5 1.3 4.7 0 0 0 Extreme 8 6.4 4.3 22.3 19.3 42.5 1.6 2.7 0 0.8 0 Poverty 7 0.1 12.3 6.8 19.3 2.7 34.9 15.5 4.5 2.8 1.1 Map 6 0.8 22.4 16.9 15.9 10.9 6.6 8.0 7.0 7.7 3.4 5 0 7.5 15.7 17.2 8.9 9.4 21.5 7.0 6.9 6.1 4 0 0 6.2 13.5 17.1 20.0 11.9 10.1 8.8 12.5 3 0 0 0 7.3 5.7 11.6 17.7 23.5 16.2 18.1 2 1.8 0 0 0 3.3 8.7 10.2 25.9 29.7 20.5 1 1.2 0 0 0 0 4.3 8.9 21.4 26.6 37.5 Source own calculations. Districts are sorted by poverty level and each decile represents ten percent of the Peruvian population in the poorest districts according to the FONCODES map and according to the extreme poverty map from INEI. Poverty is highest in decile 25 Peru Did the Ministry of the Presidency Reach the Poor in 1995 The results demonstrate that the FONCODES map shows considerable variation with a generalized poverty map. Although the extreme poverty map from INEI falls short of the 'best' poverty map which could be developed combining Census and household survey information (which would be poverty gap per capita calculation based on a consumption based map), we can assume that the extreme poverty map from INEI is relatively closely related to such a consumption-based map. With such a high degree of variation detected between the FONCODES and the INEI map, the development of generalized consumption-based map is an important step to improve MoP's success in reaching the poor. 26