Community Targeting for Poverty Reduction Lessons from Developing Countries

Community Targeting for Poverty Reduction: Lessons from Developing Countries Community Targeting for Poverty Reduction: Lessons from Developing Countries Abstract This paper analyzes the efficacy of the community-based targeting approach as a means of identifying the poor in anti-poverty programs. It examines the performance of 30 community-targeted programs in developing countries, both in terms of the technique used to identify beneficiaries as well as broader targeting “design” issues such as targeting criteria, monitoring, transparency, accountability, elite capture, and corruption. This paper is intended to be a timely contribution to the ongoing policy debate on poverty targeting in which community-based approaches are enjoying growing support. Community-targeted interventions have tremendous potential to benefit the poor; the technique is undoubtedly preferable to universal poverty programs whose benefits are thinly spread across the entire population. Moreover, robust, program-specific design protocols are seen as critical success-inducing factors; monitoring, transparency, and accountability have a strong positive correlation with targeting performance, while elite capture—defined as the ability of a handful of individuals to hijack the beneficiary selection or benefit transfer process—and corruption are negatively correlated. Further, community targeting is better attuned to communities where societal tensions and extreme disparity are not a preexisting concern and where there is no known tendency towards cultural exclusion based on criteria not linked to poverty levels. On the other hand, the technique is not suited to situations where poverty reduction impacts are strictly dependent on following stipulated criteria. Communities inevitably digress from the criteria, and any effort to check this tendency offsets the potential benefit from allowing them to use local knowledge. Community targeting is also not recommended for programs where aggregation of poverty data is a high priority, such as programs that seek to create national or regional poverty rankings. Introduction Poverty reduction is the focus of most developing-country governments today, and development experts continue to grapple with the most effective means of assisting the poor. There is an increasing consensus that economic growth, while necessary, is not sufficient (World Bank 1990, 1997; Coady 2004). Targeted anti-poverty interventions are now believed to be an essential element of any poverty-reduction strategy. Targeting entails use of policy instruments to identify the poor among a population (Weiss 2005, 1). Targeted initiatives strive to maximize “target­ ing efficiency”—the proportion of benefits that reach those identified as ­“targets”—a feature pivotal to the future of poverty reduction. Inefficiency on this count reduces the utility of anti-poverty programs and wastes ­resources by channeling funds to the non-poor. While a number of different methods are available to identify the poor, there is growing emphasis on community-based targeting techniques. Community targeting is a sub-set of community1 participation, a broader concept that applies to all cases where communities lead the decisionmaking process. Existing studies that concur with this view link active participation to successful program performance (Bresnyan, Jr., Bouquet, Russo [undated], 9). Indeed, a number of donor agencies now make community participation a prerequisite for approval of poverty-reduction projects (Conning and Kevane 2002, 375). Specifically with regard to targeting, a community-led arrangement amounts to decentralization of the targeting process such that local knowledge of the relative poverty status among community members is used to identify program beneficiaries. Despite the emphasis on community-based targeting, we know little about the efficacy of the technique (Coady, Grosh, and Hoddinott 2004a, 59). Most literature on the subject is based on case studies that only allow for tentative generalizations across contexts. This paper seeks to expand the current understanding on the subject by analyzing a large number of cross-country initiatives. To our knowledge, this is the first large-scale study 1. We define “community” as a non-government, non-NGO affiliated collective bounded by geographical proximity. focusing specifically on community targeting. The paper is grounded in the “institutional approach” (Besley 1997) whereby program performance is seen to be affected not only by the “targeting mechanism”, a term we employ narrowly to refer to the specific technique used to identify beneficiaries, but also by broader “design” issues, which encompass all implementationrelated concerns that strike at the heart of governance issues relevant to the stipulated mechanism. We seek to identify targeting mechanism- and design-related trends across community-targeted interventions in order to highlight the key components that lead to success or failure of targeting performance. In that sense, the paper moves away from the more frequently visited question of welfare impact of poverty alleviation programs. Instead, we narrowly focus on determining whether communities are efficient in identifying and transferring benefits to the poor. The next section highlights conceptual issues related to targeting, and pre­ sents a brief overview of the current literature on community-based initiatives. This is followed by a discussion of the methodology of our analysis. We then analyze the performance of the selected initiatives and outline some broad trends related to targeting mechanisms. The next section deals with targeting design, specifically analyzing community behavior in terms of implementing targeting criteria, monitoring, transparency, and accountability of programs, and the pitfalls common to community-targeted initiatives. We end by summarizing the trends and outlining the key strengths and weaknesses of community targeting that should be internalized to extract maximum utility from future poverty-reduction programs. Targeting and Community-Based Initiatives: Conceptual Issues The theoretical debate about benefiting the poor has traditionally been dictated by two schools of thought. One cohort believes that the best possible way to execute any poverty reduction intervention is by “identifying” those in need and then directing the entire resource pool towards them. The opposing view prefers universal programs that are extended to all, irrespective of the level of poverty, but offer benefits that in theory end up benefiting the poor disproportionately (Besley and Kanbur 1991, 69; Weiss 2005, 3–6). Over time, universal programs have faded away in favor of targeted interventions (World Bank 1990, 1997; Besley and Kanbur 1991, 69–70). The need for targeting has become conventional wisdom and the success of anti-poverty programs is commonly judged by the accuracy of the targeting approach. On paper, the case for targeting is simple. It allows for maximum benefits to be directed towards the poor while minimizing expenditure on the nonpoor. In essence, it helps make programs more efficient (Grosh 1994, 8). However, this outcome is not easily achieved (World Bank 1990, 4). For one, targeting can entail substantial administrative and inforThe need for targeting has become mational costs; these are often conventional wisdom and the success inversely related to the accuracy of anti-poverty programs is commonly of the targeting methodology judged by the accuracy of the targeting (Atkinson 1995, 26, 50–60; approach. Besley and Kanbur 1991, 75–80; Sumarto and Suryahadi 2001, 4). Moreover, political economy considerations of balancing the needs of the poor and the vested interests of the middle and upper socio-economic strata can undermine the theoretical advantage of resource efficiency in targeted interventions (Besley and Kanbur 1991, 80–82; Besley 1997; Coady 2004). A failure to account for these concerns inevitably leads to one of the two types of errors associated with targeting. These include the error of undercoverage or exclusion (type I) where some of the poor are not reached, and the error of leakage or inclusion (type II) in which the non-poor— those outside the target group—receive direct benefits from the program ­(Weisbrod 1970; Hoddinott 1999). Programs that manage to keep type I and type II errors to a minimum while maintaining low targeting costs are considered efficient.2 To ensure maximum gains from targeted programs, a number of targeting ­approaches are frequently employed (World Bank 2000, 85; Grosh 1994, 2. Since our focus is targeting efficiency, the analysis in this paper naturally lends itself to a discussion of type II errors. Type I errors are most relevant when welfare effects of programs are under scrutiny, an aspect which remains beyond the purview of our discussion. 33–35; Weiss 2005, 7–8). These include activity (broad) targeting, direct (means) targeting, characteristic targeting, geographical targeting, and selftargeting.3 Community-based targeting, as defined in this paper, falls under the rubric of characteristic targeting. It is unique in that it relies on local knowledge about the status of households in the community. Local definitions of poverty are more contextualized and can capture community-specific traits that centralized proxies often miss (Conning and Kevane 2002, 378). Indeed, current research on community involvement in selection of beneficiaries suggests a positive correlation with improved targeting and successful implementation of development interventions (Subbarao, et al. 1997, 87). Proponents argue that community-based targeting is much less likely to encounter misinformation. Communities often maintain high social capital and societal accountability structures, which entail interactions that reduce costs for coordination and collaboration (Spagnolo 1999). Public voice and accountability are believed to be naturally attuned to community-based approaches. This also implies that design elements such as monitoring and accountability entail low costs in the community-targeted technique (Mansuri and Rao 2004, 10). Furthermore, even the overall costs of community targeting are believed to be low. In the most comprehensive large-N analysis of targeting programs featuring different techniques, Coady, Grosh, and Hoddinott (2004a, 61) conclude that community-based targeting is relatively inexpensive. Just as in the case of targeting, community participation does not guarantee gains that it theoretically should each time it is employed. Alderman (2001) argues that better community information can only bear dividends if communities find the incentives to reveal and base decisions on that information. The single biggest impediment in community participation, one that occupies the bulk of the discussion on the method’s efficacy, is elite capture. The concern emanates from the heterogeneous nature of communities, as the more influential individuals within communities—the elite—often have a natural informational advantage and thus can exercise greater leverage over the terms of participation in an initiative (Bardhan 2000). Critics of community participation see this as common wisdom; they do not take 3. For a brief explanation of each targeting approach, see Yusuf (2007, 1). presence of social capital as a given. They also do not believe that community participation can instill a sense of ownership and undermine disparate societal power structures (Bardhan and Mookherjee 2000; Spagnolo 1999). For these critics, social cohesion is considered a necessary prerequisite for community mobilization and efficient targeting returns; its absence implies that the marginalized may not be able to defy local power disparities to demand genuine participation. Mansuri and Rao (2004, 20) argue that cases exhibiting a net negative benefit from collective action will likely see averseness to it. Conceivably, this is most likely in cases where community inequality levels are high. Indeed, a number of studies have found inequality to be positively related to elite capture (La Ferrara 1999; Bardhan and Mookherjee 2000). Related to the above are political economy concerns. Political support for programs is often negatively correlated with targeting effectiveness (Sen 1995). Essentially, political economy compulsions necessitate a certain level of tolerance for poorer targeting designs that allow influence of the political and social elite as a means of ensuring sustainability of the programs. The propensity to accept this “targeting efficiency–sustainability trade-off ” implies that the efficacy of targeting design elements built in as buffers against institutional perversions could be compromised. Methodology This paper presents a meta-analysis of existing assessments of communitytargeted programs. In order to select the sample, we conducted an extensive literature review to identify all anti-poverty programs with a community component designed to select individual or household level beneficiaries. We therefore excluded programs where communities were involved in selecting development projects or community groups. Moreover, to be included, the program had to have allowed communities to use their subjective knowledge about relative poverty levels among community members. Finally, each program had to explicitly aim at identifying the poor (or some subset of the poor) through the targeting exercise. Our literature review was limited to middle- and low-income countries as defined by the World Bank (2008). We exclude high-income countries as their greater resource and institutional capacity often introduces systematic differences in performance vis-à-vis the developing world. However, we have not restricted our analysis to any particular type of program. The final sample includes both cash and in-kind transfer initiatives. Within cash transfer programs, we have gone against the norm to include programs that offer smallscale credit as long as they explicitly ask communities to identify beneficiaries based on poverty status and not on capacity to repay loans. One major determinant of the final sample was availability of requisite information. To be useful for the analysis, we required information on the targeting performance of programs such that we could categorize them in terms of targeting efficiency. Although previous large-scale studies on related issues have attempted to rank programs in terms of targeting performance (Coady, Grosh, and Hoddinott 2004a, 25–26; Grosh 1994, 18, 24–32), the nature of the available data in our case would have induced a serious concern about comparability across programs if we had attempted to quantify and rank program performance. Therefore, instead of creating an ordinal performance scale, we have broadly categorized programs as “progressive,” “mildly progressive,” or “regressive” based on their relative performance. Given issues of comparability, we refrain from comparing intra-category performance, instead conducting the analysis across categories; the difference in the average performance across categories is large enough that we can safely rule out any non-random error induced by this methodology. For programs that had sufficient information about the proportion of beneficiaries across socio-economic levels, we used a methodology derived from Coady, Grosh, and Hoddinott (2004a, 25–26) and Grosh (1994, 18, 24–32). We benchmarked program performance against a hypothetical universal program that is “neutral” in its targeting such that each member of the population receives an equal share of the benefit.4 Performance is measured in terms of either (i) the share of benefits to a particular segment of the population representing the poor5 versus the percentage of the total 4. This is conventional for evaluating targeted anti-poverty interventions. Both studies from which we derive our methodology use the same benchmark. 5. One methodological limitation we face stems from the multiple notions of poverty (in terms of who qualifies as “poor”) entertained across the reviewed literature. Existing evaluative studies do not provide performance data against an identical benchmark. While the majority treats the lowest income quintile population the segment constitutes; or (ii) the share of poor beneficiaries in a program versus the percentage of the total population the poor represent.6 “Progressive” programs are ones where the share of benefits provided to the poor or the proportion of poor beneficiaries is more than twice the percentage of the total population the poor represent (this amounts to a ratio of greater than 2.0). Initiatives where this ratio was between 1.0 and 2.0 are considered “mildly progressive.” Finally, programs where the benefits to or share of the poor were less than their overall proportion in the population (ratio of less than 1.0) were deemed to be “regressive.” 7 For interventions where available performance information was only qualitative in nature, we subjectively determined the relative success, or lack thereof, of these programs based on how they were evaluated in the studies we examined. Programs for which assessments pointed to successful targeting in terms of having transferred an overwhelming majority of the benefits to the poor or where all reviews remained highly optimistic were in the progressive category. To be considered mildly progressive, there needed to be some indication that targeting was better than the “neutral” benchmark. However, assessments could have been mixed; they could have pointed to considerable leakages to the non-poor in absolute terms. Finally, programs where assessments were decidedly negative and suggested that the bulk of the benefits went to non-target groups were considered regressive. as the cut-off point to examine program performance, some use the bottom two quintiles, the bottom half, or country-specific poverty lines. Poorly defined poverty lines induce another source of targeting inefficiency. Notwithstanding, as others before us (Coady, Grosh, and Hoddinott 2004a; Grosh 1994), we were forced to use the cut-offs provided by the analyzed studies in each case and consider them comparable across the board. Again, this is permissible given that we only categorize programs broadly without conducting intra-category comparisons. However, the limitation does imply that we treat the “poor” as homogenous; we are unable to differentiate between specific segments of the poor. 6. The difference between these two measures is an indicator of welfare implications of differential (as opposed to uniform) transfers (Coady and Skoufias 2004). Theoretically, these two are comparable if one considers the per capita benefits transferred across beneficiaries to be equal (Grosh 1994, 18). However, this is an assumption that may not hold across cases. 7. To illustrate, say a program transfers 60 percent of the program benefits to the lowest quintile of the population, these being the “poor,” the performance ratio would be 60/20=3. On the other hand, if only 10 percent of the program beneficiaries are from the lowest quintile, the ratio would be 10/20=0.5. The methodology would remain the same when calculating the ratio for programs that list the proportion of beneficiaries (instead of benefits) in a particular segment of the population. For instance, if 60 percent of the beneficiaries are from the lowest quintile, the ratio would still be 60/20=3. Overall, we identified 42 programs that fit our criteria. Twelve of these were left out of the final analysis due to lack of requisite information. Apart from analyzing the targeting mechanisms, each program is evaluated for implementation protocols depending on the extent of information available. Given that the studies examined were of varying quality and focused on different aspects of the programs—sometimes even analyses of the same variables are not comparable across cases due to definitional issues—not all initiatives have been analyzed for the entire breadth of targeting mechanism and design issues. Table 1 lists the programs included and the factors against which each has been analyzed. Targeting Mechanism Community-based targeting has been employed in programs of various types, in combination with different targeting techniques, and in diverse geographical settings. Two models for community involvement exist; 14 of the 27 programs for which we have the requisite information limit community roles strictly to identification of beneficiaries, while 13 involve them in transferring benefits as well. In order to target individuals, all but three programs used selection committees with strong community representation. The only other method employed—as seen in the Bangladesh Rural Advancement Committee—CFPR/TUP (BRAC), Pakistan’s National Rural Support Program (NRSP), and South Africa’s Tshomisano—is the Poverty Wealth Ranking (PWR), a more formal technique that determines relative poverty through communityspecified indicators (Robb 1999; Communities are seldom used as Gibbons, Simanowitz, and sole targeting agents in anti-poverty Nkuna 1999).8 programs. Twenty-eight of the 30 Communities are seldom used analyzed programs also employed geoas sole targeting agents in antigraphic, self-targeted, means test, and poverty programs. Twenty-eight proxy-means test techniques. of the 30 analyzed programs 8. PWR also relies on community groups to identify the poor. However, unlike committees formed in other programs, the PWR technique stipulates a formal guideline on constituting the community groups and the method through which they are to identify the poor. Communities are monitored throughout and are not allowed any leeway in terms of digressing from the stipulated protocol. Table 1: Available Information for Selected Programs Country Program Years Quantitative Information on Performance Adherence to Program Criteria Monitoring Transparency Accountability Elite Capture Discretion High Corruption Albania Ndihme Ekonomika 1994-95      X   Bangladesh Bangladesh Rural Advancement Committee—CFPR/TUP 2002 onward    X     Bangladesh Food for Education 1993         Bangladesh Rural Maintenance Program 1983 X  X  X X X  Bangladesh Vulnerable Group Development 1975 onward X        Brazil North East Rural Poverty Alleviation Program 1995       X          X   X  Chile School Feeding Program 1980s-90s Ethiopia Emergency Food Aid 1980s-90s   X India Indira Awas Yojana 1985-2001 X   X   X  India Integrated Rural Development Program 1976-1999      X X  India National Old Age Pension Scheme 1995 onward X  X  X    India Sampoorna Grameen Rozgar Yojana 2001 onward         Indonesia Inpres Desa Tertingga 1994-96 X   X X X X X Indonesia JPS Scholarship Program 1998         Indonesia Operasi Pasar KhususRaskin 2002 onward     X   X Indonesia Pemberdayaan Daerah Dalam Mengatasi Dampak Krisis Ekonomi 1999 X   X X X X X Table 1: Available Information for Selected Programs (continued) Country Program Years Quantitative Information on Performance Adherence to Program Criteria Monitoring Transparency Accountability Elite Capture Discretion High Corruption Malawi Targeted Inputs Program 2001-02        X Nepal Churia Food for Work 1993-95 X        Pakistan National Rural Support Program 1991 onward         Pakistan Zakat 1980 onward         South Africa Public Works Program– 101 Projects 1995-97 X  X X X    South Africa Tshomisano Program 1996 onward    X     Thailand Social Investment Fund– Menu 5 19992000 X   X     Uzbekistan Social Assistance Program (Child Allowance) 1994        X Uzbekistan Social Assistance Program (Low Income) 1994        X Vietnam Hunger Eradication and Poverty Reduction (Credit) 1998 onward         Vietnam Hunger Eradication and Poverty Reduction Low (Certificates) 1998 onward         Vietnam Hunger Eradication and Poverty Reduction (Health Insurance) 1998 onward         Vietnam Hunger Eradication and Poverty Reduction (Education) 1998 onward         Zimbabwe Child Supplementary Feeding Program 1992-93/ 1995-96 X    X    =information available; X =information not available also employed geographic, self-targeted, means test, and proxy-means test techniques. Geographic targeting is by far the most frequently used complementary technique; it was found in 16 of the 30 programs we examined. In each case, the geographic tool is used in the first tier to select the areas where the program will operate. Communities feature in the subsequent tiers; they are responsible for intra-community selection. Self-selection also features prominently, with 12 of the 30 programs using this technique in the first tier before employing community knowledge to select from among the applicants. There are only two programs—Pakistan’s Zakat and Malawi’s Targeted Inputs Program (TIP)—that base targeting solely on community knowledge. Both programs include the entire population in a defined geographical spread as part of the pool from which beneficiaries are to be selected, thus making the need for a geographical component redundant. Zakat covered Pakistan’s entire population and TIP was spread throughout rural Malawi. While the combination of targeting techniques makes it virtually impossible to isolate the value of the community component completely, this methodological complication does not confound our analysis since none The targeting performance of of the 30 analyzed programs satu­community-based anti-poverty prorated the poor population. All grams shows that simply relying on programs suffered from underlocal community knowledge does not coverage, thus implying that perguarantee success. fect efficiency at the community tier should theoretically have led to 100 percent targeting towards the poor irrespective of the performance at the first tier. Table 2 provides a summary of each of the analyzed programs including their relative performance. The targeting performance of community-based anti-poverty programs shows that simply relying on local community knowledge does not guarantee success. Of the 30 programs, 10 cases across seven countries were progressive in their targeting, 16 initiatives across nine countries were mildly progressive, and four programs across three countries were regressive. Further, the correlation between performance and other explanatory factors that could plausibly impact targeting efficiency—we examined the particular targeting technique used in a program, the type of program employed, and regional and country specific traits—seems weak; none of these factors seems to affect community targeting in a systematic manner. By itself, the technique used does not explain the targeting performance of an intervention. Programs with geographic and community targeting, self and community targeting, and a combination of geographic, self, and community techniques all show substantial variation in performance. Each of these is represented in more than one performance category. Programs with a combination of means-testing and community selection are the only exceptions as all four incidences produce progressive outcomes. However, this result needs to be treated with caution as each of these was a sub-component of the same overarching program: Vietnam’s Hunger Eradication and Poverty Reduction (HEPR). The type of programs, their regional spread, and country characteristics turn out to be equally weak proxies for targeting performance. Both cash and inkind transfer programs and credit and non-credit programs vary considerably in performance, and are found across the entire breadth of the performance spectrum. In terms of regional spread, programs in Africa, South Asia, and Southeast Asia have highly varied performance. (No meaningful analysis of Latin America, Central Asia, or Europe was possible given the small number of programs included from these regions.) Programs in South Asian and Southeast Asian countries have been progressive, mildly progressive, and regressive; African countries feature in the mildly progressive and regressive categories. We find no support for previously established evidence that richer countries are better placed in terms of targeting outcomes than their poorer counterparts (Coady, Grosh, and Hoddinott 2004a, 36). Vietnam, Bangladesh, and Nepal, all “low-income” countries, managed progressive programs. On the other hand, two of the four regressive programs belong to Indonesia, a “lower middle” income country (World Bank 2008). Substantial variation is even seen in programs within particular countries. Bangladesh, India, and Indonesia all have programs in more than one performance category. Further, while Vietnam shows relatively low performance variance, this is a function of all the interventions being part of the HEPR umbrella program. Table 2: Summary of Targeting for the Selected Programs Country Program Type of Benefit Target Population Targeting Mechanism Proportion of Benefits for Poor / Poor in Benefits Progressive Programs Albania Ndihme Ekonomika Cash—Supplemental income Urban unemployed and rural poor Self-selection, Proxy-means test, Community 53% benefits accrue to bottom quintile 1 Bangladesh Rural Maintenance Program Cash—Payment for employment Rural destitute women Self-selection, Community N/A Chile School Feeding Program In-kind—Free food Children of poorest in public schools Proxy-means test, Community 53% benefits accrue to bottom quintile 2 India National Old Age Pension Scheme Cash—Pension scheme Age > 65 with no alternate income Self-selection, Community N/A Nepal Churia Food for Work In-kind—Food for employment Poor in Churia catchment area Self-selection, Community N/A Thailand Social Investment Fund—Menu 5 Cash and In-kind—Multiple cash and in-kind transfers for emergency assistance Poor, especially those unemployed or marginalized and impacted by the financial crisis Geographic, Community N/A Vietnam Hunger Eradication and Poverty Reduction (Credit) Cash—Small-scale credit Poorest in the county Self-selection, Means-test, Community 60.2% beneficiaries are in the bottom quintile 3 Vietnam Hunger Eradication and Poverty Reduction Low (Certificates) In-kind—Poor household certificates allowing free medical treatment Poorest in the county Means-test, Community 58.5% beneficiaries are in the bottom quintile 3 Vietnam Hunger Eradication and Poverty Reduction (Health Insurance) In-kind—Health insurance cards allowing free medical treatment Poorest in the county Means-test, Community 57.8% beneficiaries are in the bottom quintile 3 Vietnam Hunger Eradication and Poverty Reduction (Education) In-kind—Exempted or reduced school fees Poor with school-going children Means-test, Community 56.9% beneficiaries are in the bottom quintile 3 Geographical, Poverty Wealth Ranking, Proxymeans test Over 95% accrue to rural poor [53% of population constitutes rural poor] 4 Mildly Progressive Programs Bangladesh Bangladesh Rural Advancement Committee—CFPR/ TUP Cash and In-kind—Micro-credit, healthcare services Poorest in NGO’s Rural Program l areas Table 2: Summary of Targeting for the Selected Programs (continued) Country Program Type of Benefit Target Population Targeting Mechanism Proportion of Benefits for Poor / Poor in Benefits Bangladesh Food for Education In-kind—Free food Poor households with at least one child aged 6–10 years Geographic, Community 60% accrue to the poorest 50% 5 Bangladesh Vulnerable Group Development Cash and In-kind—Food, training, and cash for income-generation assistance Poorest rural women Geographic, Community N/A Brazil North East Rural Poverty Alleviation Program Cash—Infrastructure development, employment, and small-scale credit Rural poor in poorest villages of the North East Geographic, Self-selection, Community 93% benefits accrue to rural poor [64% of population constitutes rural poor] 6 Ethiopia Emergency Food Aid In-kind—Free food Poor in rural areas Geographic, Community Probability of receiving food is 42% at 25th percentile and 35% at 75th percentile 7 India Indira Awas Yojana In-kind—Free rural housing units Rural poor Geographic, Community N/A India Integrated Rural Development Program Cash—Small-scale credit and subsidy Poor in the country Self-selection, Community 81% beneficiaries are poor [44% of population is poor] * 8 Indonesia JPS Scholarship Program Cash—Educational scholarships Households with school-going children in lowest two categories of BKKBN national prosperity list Geographic, Community 28.8% benefits accrue to bottom quintile 9 Indonesia Operasi Pasar KhususRaskin In-kind—Subsidized food Households in lowest two categories of BKKBN national prosperity list Geographic, Community 26.4% benefits accrue to bottom quintile 9 Malawi Targeted Inputs Program In-kind—Free maize seeds Poorest rural farmers Community 78.8% beneficiaries are food insecure (proxy for poverty) [73% of population is food insecure] 10 Pakistan National Rural Support Program Cash—Multiple benefit schemes Poorest in program area of the NGO Geographic, Poverty Wealth Ranking 40% beneficiaries are poor [28.1% of population is poor] 11 Pakistan Zakat Cash—Supplemental income Poorest Muslims in the country Community 21.6% benefits accrue to bottom quintile 12 South Africa Public Works Program–101 Projects Cash for employment Poorest in Western Cape province Geographic, Self-selection, Community N/A Country Program Type of Benefit Target Population Targeting Mechanism Proportion of Benefits for Poor / Poor in Benefits South Africa Tshomisano Program Cash—Micro-credit Poorest 30% in program area of NGO Geographic, Poverty Wealth Ranking 99% beneficiaries are poor [67.3% of population is poor] 13 Uzbekistan Social Assistance Program (Child Allowance) Cash—Child allowance Poor with under-18-year-old school-going children Geographic, Self-selection, Community 28.4% benefits accrue to the bottom quintile 14 Uzbekistan Social Assistance Program (Low Income) Cash—Low-income benefits Poor who cannot supplement income Geographic, Self-selection, Community 22.6% benefits accrue to the bottom quintile 14 Regressive Programs India Sampoorna Grameen Rozgar Yojana Cash and In-kind—Food and cash for employment Poor in disaster-hit districts Self-selection, Community 25% accrue to the poor [26 % of population is poor] 15 Indonesia Inpres Desa Tertingga Cash—Small-scale revolving credit Poor in ‘neglected villages’ Geographic, Self-selection, Community N/A Indonesia Pemberdayaan Daerah Dalam Mengatasi Dampak Krisis Ekonomi Cash—Employment and small-scale credit Unemployed poor Geographic, Community N/A Zimbabwe Child Supplementary Feeding Program In-kind—Free food Poor and malnourished children in droughtstricken rural areas Geographic, Community N/A * Estimate is an average of figures reported in two different studies. [1] Alderman (2002); [2] Grosh (1994); [3] World Bank (2003); [4] Chavan and Ramakumar (2002), United Nations “Millennium Indicators” (2007); [5] Coady, Grosh, and Hoddinott (2004b); [6] Quinn (1998); [7] Jayne, et al. (2000); [8] Srivastava (2004), Swaminathan (1990), MAKER (2002); [9] Perdana and Maxwell (2004), Pritchett, Sumarto, and Suryahadi (2002); [10] Chinsinga, et al. (2002), Levy (2003); [11] Khan (undated), Government of Pakistan (2006a); [12] Heltberg (undated); [13] “Small Enterprise Foundation” (undated), Punt, et al. (2005); [14] Mehra and Rashid (2002); [15] Srivastava (2004). Table 3 highlights the spread of programs across the performance spectrum by targeting technique, program type, and geographic location. TARgeTINg DeSIgN This analysis points to the low explanatory power of the targeting technique, program type, and geographical location of initiatives in terms of explaining targeting performance. Targeting design elements focused on implementational effectiveness constitute the other set of variables that could impact targeting performance, and thus may have a more important role in explaining the performance of community-targeted interventions (Coady, Grosh, and Hoddinott 2004a, 33). In this section, we examine elements of targeting design—namely monitoring, transparency, and accountability— and assess their impact on commonly feared institutional perversions, such as elite capture and corruption. One key question is whether the variability in targeting efficiency is caused by the behavior of selection committees. Do communities choose to ignore targeting objectives in some cases? Does this result in poorer performance? Moreover, what role do monitoring, accountability, and transparency protocols play in outcomes? Do programs that exhibit good governance manage to extract better behavior from community groups in charge of selection? And how are these related to the incidence of elite capture and corruption? By answering these questions, we hope to establish whether targeting design elements have high explanatory power in terms of overall targeting performance. We begin by highlighting the relationship among various design elements and between these and overall program performance. Next, we contextualize the analysis by undertaking a detailed qualitative discussion of each of the design elements. Before proceeding further, it is important to put targeting performance in perspective. Thus far, we have posited progressive programs as successful, mildly progressive programs as less so, and regressive ones as unsuccessful. While this is true relative to the benchmark of a “neutral” universal program, it does not imply that the successful programs are free from institutional anomalies; even successful programs could have substantially high leakages. In fact, each of the progressive programs evaluated allows Table 3: Spread of Programs across the Performance Spectrum Total Incidence Progressive Mildly Progressive Regressive By targeting technique Geographic and community* 9 1 6 2 Self-targeted and community 5 3 1 1 Geographic, self-targeted, and community 5 0 4 1 Means-test and community (with or without selfselection) 4 4 0 0 Cash transfer 23 5 14 4 Non-credit 15 4 9 2 Credit 8 1 5 2 15 6 7 2 By program type ** In-kind transfer Food/Farm input 9 2 5 2 Social services 6 4 2 0 38 11 21 6 South Asia 11 3 7 1 Total Regional distribution South East Asia 9 5 2 2 Africa 5 0 4 1 Total 25 8 13 4 Bangladesh 4 1 3 0 India 4 1 2 1 Indonesia 4 0 2 2 Vietnam 4 4 0 0 By Country * “Community” does not include programs that conducted a PWR exercise. PWR was treated separately and is excluded from the table given its low incidence among the analyzed programs. ** Total is greater than the number of analyzed programs as some programs offered both cash and inkind benefits, and in some cases, the same program offered credit and non-credit benefits. populations in the top four quintiles to receive between 40 and 50 percent of the benefits. The leakages are staggeringly high for programs in the lower performance categories. For instance, India’s Integrated Rural Development Program (IRDP), a mildly progressive program, included 15 to 26 percent of recipients above the poverty line; in some regions the proportion of ineligible beneficiaries was 40 to 50 percent (Chavan and Ramakumar 2002, 957). Similarly, for Bangladesh’s Food for Education (FFE) program, also a mildly progressive initiative, 35 percent of the beneficiaries were from the richest two quintiles (Ahmed and Ninno 2002, 23). In essence, even though programs may seem successful against our benchmark, they can still have substantial type II errors. Notwithstanding, if design issues are positively related to program performance, we should expect better performing programs to have fewer perversions in terms of implementation. Linking Elements of Design to Targeting Efficiency A strong trend between functional9 design elements and positive performance is easily discernable. All 10 progressive programs managed a robust programspecific or overarching monitoring protocol. On the other hand, less Further, transparency, monitoring, and than half of the mildly progressive accountability are perfectly correlated. programs and none of the regressive Each program that instituted some programs exhibited effective monimechanism for monitoring also had toring. The pattern is the same in properly functional transparency and terms of transparency. Two-thirds of accountability protocols. the evaluated programs instituted a properly functioning mechanism to apprise communities of the program and its benefits and/or employed transparent disbursement criteria. None of the remaining, non-transparent programs were progressive. Further, transparency, monitoring, and accountability are perfectly correlated. Each program that instituted some mechanism for monitoring also had properly functional transparency and accountability protocols. 9. In this section, we are concerned with the functionality, not mere stipulation of design protocols. Therefore, we consider design elements to be “absent” in programs where they were not implemented effectively in practice, irrespective of whether they were stipulated on paper. The link between these governance-enhancing protocols and institutional perversions is also evident. Monitoring, transparency, and accountability are inversely related to elite capture. In absolute terms, the incidence of elite capture—defined as the ability of a handful of individuals to hijack the beneficiary selection or benefit transfer process—is relatively low; it features in seven of the 25 programs evaluated on this count. None of these is progressive; only Bangladesh’s Vulnerable Group Development (VGD) and FFE initiatives suffered from elite capture despite instituting monitoring, transparency, and accountability protocols successfully. On the other hand, failure to employ functional design protocols leads to elite capture in every case. That said, the propensity of the elite to hijack the program is not synonymous with discretion. At least 11 of the 25 evaluated programs allowed some level of discretion in selection to an individual—these were mostly local officials, village chiefs, or heads of the selection committees—but this only translated into elite capture in three instances. Moreover, discretion is not a necessary precondition for elite capture; at least three programs were hijacked by the elite despite absence of any provision to allow discretion. Incidentally, elite capture and corruption are perfectly correlated, and like elite capture, occurrence of corruption is inversely related to design protocols. Out of the 22 programs evaluated for corruption, six reported corruption as a major concern but only two—Bangladesh’s FFE and VGD programs—possessed functional monitoring, transparency, and accountability protocols. Table 4 highlights the presence, or lack thereof, of various design elements across the analyzed programs. In essence, while none of the governance-strengthening protocols seems to be a sufficient condition for successful performance, the general pattern suggests a strong negative correlation between initiatives with positive design attributes and institutional perversions. Moreover, the fact that monitoring, transparency, and accountability are strongly correlated with each other suggests that, for the most part, programs either manage to institute robust, overarching targeting designs or fail to do so completely; the latter raises the likelihood of both elite capture and corruption and, in turn, relatively poorer program performance. Figure 1 captures the link between monitoring, transparency, and accountability and elite capture and corruption. Table 4: Targeting Design Across Programs Country Program Monitoring Transparency Albania Ndihme Ekonomika Present Present Bangladesh Rural Maintenance Program * Chile School Feeding Program India Accountability Elite Capture Discretion Corruption Present * Present Absent Present * * * Absent Present Present Present Absent Present Absent National Old Age Pension Scheme * Present * Absent Absent Absent Nepal Churia Food for Work Present Present Present Absent Present Absent Thailand Social Investment Fund—Menu 5 Present * Present Absent Absent Absent Vietnam Hunger Eradication and Poverty Reduction (Credit) Present Present 1 Present 1 Absent Present Absent Vietnam Hunger Eradication and Poverty Reduction Low (Certificates) Present Present 1 Present 1 Absent Absent Absent Vietnam Hunger Eradication and Poverty Reduction (Health Insurance) Present Present 1 Present 1 Absent Absent Absent Vietnam Hunger Eradication and Poverty Reduction (Education) Present Present 1 Present 1 Absent Absent Absent Bangladesh Bangladesh Rural Advancement Committee—CFPR/ TUP Present * Present Absent Absent Absent Bangladesh Food for Education Present Present Present Present Absent Present Bangladesh Vulnerable Group Development Present Present Present Present Absent Present Brazil North East Rural Poverty Alleviation Program Present Present Present Absent Absent * Ethiopia Emergency Food Aid * Absent * Present Absent * India Indira Awas Yojana Absent * Absent Present * Present India Integrated Rural Development Program Absent Absent Absent * * Present Indonesia JPS Scholarship Program Present Present Present Absent Absent Absent Indonesia Operasi Pasar Khusus-Raskin Present Present * Absent Present * Malawi Targeted Inputs Program Absent Absent Absent Present Present * Progressive Programs Mildly Progressive Programs Country Program Monitoring Transparency Accountability Elite Capture Discretion Corruption Pakistan National Rural Support Program Present Present Present Absent Absent Absent Pakistan Zakat Absent Absent Absent Present Present Present South Africa Public Works Program—101 Projects * * * Absent Present Absent South Africa Tshomisano Program Present * Present Absent Absent Absent Uzbekistan Social Assistance Program (Child Allowance) Present Absent 2 Absent 2 Absent Present * Uzbekistan Social Assistance Program (Low Income) Present Absent 2 Absent 2 Absent Present * India Sampoorna Grameen Rozgar Yojana Absent Absent Absent Present Present Present Indonesia Inpres Desa Tertingga Absent * * * * * Indonesia Pemberdayaan Daerah Dalam Mengatasi Dampak Krisis Ekonomi Absent * * * * * Zimbabwe Child Supplementary Feeding Program Absent Absent * Absent Absent Absent Regressive Programs * Information not available 1 Transparency and Accountability are not program specific; rather, they are built into the societal norm. 2 The Mahallas that are responsible for targeting are institutionalized setups but do not have oversight or transparency. Yet, Mahallah heads are well-respected and are under a social obligation to act in the community’s interest. Figure 1: Design Elements and Institutional Perversions Monitoring Transparency Accountability Minimize elite capture Minimize corruption Contextualizing Design Issues Having underscored the importance of strong targeting design, we now move on to highlight how the design-related protocols were instituted across programs (also see Appendix 1). This is essential since none of the design elements are uniform in their application and it is entirely plausible that some forms are more suited to particular contexts. A nuanced view is necessary if we are to identify targeting designs that are most likely to succeed if replicated. Official Program criteria and community Flexibility. A consistent trait in community-targeted programs is the propensity of the communities asked to select beneficiaries to diverge from the stipulated program criteria. The tendency was seen in programs across the performance spectrum; there seems to be no correlation between performance and adherence to the program criteria. Virtually all analyzed programs exhibited “delegation”—not “devolution”—in that none handed over total responsibility for all aspects of targeting to communities (Conning and Kevane 2002, 381). Instead, communities were given criteria on which to base the selection process. Flexibility was permissible—its extent varied considerably—but only within the bounds of the criteria. We find that communities consistently mold, if not completely ignore, stipulated targeting indicators to identify the beneficiaries. In as many as 19 of the 26 programs that stipulated criteria, communities either disregarded it entirely or followed it only loosely (see Table 5). Table 5: Flexibility Beyond Stipulated Criteria Flexibility Present Absent Progressive Programs 8 2 Mildly Progressive Programs 7 9 Regressive Programs 4 0 The following discussion focuses on two specific aspects of community behavior regarding: (i) c riteria stipulating greater weight to specific indicators of poverty as the basis for selection; and (ii) c riteria for the total number of beneficiaries (or the total benefit allocation). Conventional wisdom suggests that the tendency of communities to move away from stipulated criteria is a negative attribute. The presupposition is that the criteria laid out by program conceivers are based on calculations that seek to optimize the outcomes in terms of achieving their broader poverty reduction goals. However, in the analyzed cases, flexibility in the use of program criteria does not necessarily produce poorer results. Instead, we find support for the view that centralized impressions of poverty are unable to capture the differences in profiles of the poor across communities. In Vietnam, local community discretion produces improved outcomes over income-based means tests stipulated by the government (Nguyen and Rama 2007, 18). Similarly, in Bangladesh’s FFE program, while the community selected 21.3 percent households that did not meet the official eligibility criteria, 57 percent of these had incomes lower than beneficiaries who satisfied the criteria (Ahmed and Ninno 2002, 25). On the other hand, Clay, Molla, and Habtewold (1999, 401–02) partly blame poor targeting performance in Ethiopia’s Emergency Food Aid (EFA) program on the official selection criteria; the recommended indicators—households with aged or female heads—turned out to be weak proxies for poverty in some program areas. The potential for improved targeting notwithstanding, community flexibility in selection of indicators makes it impossible to aggregate poverty counts across communities. Community-based poverty evaluations are known to suffer from aggregation problems, an attribute that makes the task of developing overarching anti-poverty strategies using community-targeting difficult. In fact, this is the reason that program initiators stipulate proxies for selection in the first place (Yusuf 2007; Simanowitz, 1999). However, the stipulation seems to have little effect. Therefore, the best case scenario— this is applicable to programs where flexibility in using stipulated criteria enhances targeting performance—entails a “targeting efficiency–crosscommunity comparability trade-off.” On the other hand, initiatives where flexibility is not correlated with improved targeting performance present an even more unattractive prospect of having to forego aggregation without gaining on the targeting count. The second aspect where communities tend to ignore stipulated criteria deals with the level of benefit transfers. A number of programs—again spanning the breadth of the performance spectrum—that tasked communities to transfer benefits or determine the amount of per capita benefit saw them frequently spread transfers across a higher-than-stipulated number of beneficiaries. In some cases, the poverty reduction rationale of programs was completely ignored. In Indonesia’s Operasi Pasar Khusus (OPK)-Raskin, for example, some villages adopted a first-come, first-serve approach regardless of the beneficiaries’ poverty level (Perdana and Maxwell 2004, 27). Others divided the villages’ quotas equally among all members (Coady, Grosh, and Hoddinott 2004a, 61). In South Africa’s Public Works Program (PWP) and Bangladesh’s Rural Maintenance Program (RMP), the communities were viewed as egalitarian, leading to a process that selected beneficiaries through a lottery among all applicants (Adato and Haddad 2002, 25; Robinson 2006, 7). Similarly, in Zimbabwe, pressure to distribute food among all children led to the per capita benefit dropping from the stipulated 10 kilograms per person per month to a mere five kilograms (Kaseke, Dhemba, and Gumbo 1997, 28). The desire for a wider-than-permissible transfer of program benefits strikes at the heart of the social capital debate, given that this is often a consequence of concern for intra-community harmony. In OPK-Raskin, communities implementing the first-come, first-serve approach were under threat of communal conflict (Perdana and Maxwell 2004, 28). In Malawi, another program that saw thin dispersion in some villages, areas that The desire for a wider-than-permisrefrained from this practice saw sible transfer of program benefits targeting-related violence. Morestrikes at the heart of the social capital over, those who were left out of debate, given that this is often a consethe program refused to assist in quence of concern for intra-community other community development harmony. works (Chinsinga, et al. 2002, 50–51). In the same vein, South Africa’s trade unions opposed community participation in the targeting of PWP, citing a correlation between community involvement and conflict (Adato and Haddad 2002, 28–29). The thinly spread benefits in Zimbabwe were also a result of the social norm, which opposed selective distribution; Chitekwe (undated) found conflict among beneficiary and non-beneficiary neighbors in the program area. The trade-off between positive social capital and poverty reduction should be evident. Community-based targeting seems to confront a moral hazard problem whereby what communities consider fair and based on true deservingness actually constitutes mis-targeting from the perspective of those conceiving the programs (Conning and Kevane 2002, 381). Presuming that the stipulated criteria are optimal in terms of achieving poverty reduction targets, the leakage or reduced per capita transfer to the poor caused by thin distribution is counterproductive: it is likely to have adverse incentive effects on labor supply and savings of beneficiaries, and in turn on efficiency and growth objectives (Ravallion 2003, 3). This runs counter to the view of proponents of community participation who see social capital as complementing overarching anti-poverty goals. There is yet another troubling tendency related to the “social capital– poverty reduction trade-off ” in community-based programs: the propensity of selection committees to deliberately ignore certain segments of the community based on social dynamics unrelated to poverty. In Uzbekistan for instance, the Mahallas10 are attuned to Islamic ethics and are less popular with non-Central Asian, Slavic civilizations. Their outlook biases targeting practices (Coudouel, Marnie, and Micklewright 1998, 4, 24). Similarly, in Vietnam decisions by village chiefs tend to leave out unregistered migrants, people with socially unacceptable behavior, and those not considered to be working hard enough (Nguyen and Rama 2007, 9–10; World Bank 2003, 83). The South African PWP witnessed strong resistance among recipient communities on the question of selecting women for employment in the construction sector, a program stipulation that went against the established norm (Adato and Haddad 2002, 29). Monitoring of Programs. The evaluated programs used a wide variety of monitoring approaches. An overwhelming majority of public programs retain the monitoring role for government officials; for non-governmental initiatives, program staff tend to monitor programs. Out of the 22 programs analyzed on this count, only seven left communities in charge of monitoring without employing any additional layer of monitors. However, five of these—the four sub-components of the Vietnamese HEPR and the Thai Social Investment Fund (SIF)—are set in contexts where a strong, overarching monitoring and accountability mechanism exists courtesy of societal norms (World Bank 2003, 125; Satterthwaite 2004).11 The other two programs—the Indonesian Inpres Desa Tertingga (IDT) and Pemberdayaan Daerah Dalam Mengatasi Dampak Krisis Ekonomi (PDM-DKE)—lacked any such norms; we found both to have regressive outcomes. 10. Mahallas are traditional, community-based social institutions and have existed in Uzbekistan for centuries. They allow for community participation and remain central to daily life, especially in the rural areas. 11. In Vietnam, a grass roots democracy initiative is in place. The role of the village chiefs remains central to the initiative. In the same vein, Thailand has a dedicated functionary, the Community Organizations Development Institute, set up in order to enhance community networking through Community Organization Networks. A number of programs institute multi-tiered monitoring. While multitiered monitoring was not limited to successful programs,12 studies assessed such frameworks to be beneficial for targeting performance whenever they functioned efficiently. Arguably the most robust monitoring structure featured is Albania’s Ndhime Ekonomika, where triangulation and redundancy were intrinsic to the protocol: at least two independent government functionaries were asked to monitor the program in addition to the community and monitoring was dynamic, mandating regular updates of beneficiary lists (Kolpeja 2005a). Other programs that combined community and government roles in a multi-tiered monitoring framework include the Chilean School Feeding Program and the Bangladeshi FFE (Kain, Uauy, and Taibo 2002, 600–601; Ahmed and Ninno 2002, 9–11). The Brazilian North East Rural Poverty Alleviation Program (RPAP) left monitoring primarily to the state authorities but retained an active role for the donor agency, the World Bank, which evaluated monthly progress reports generated by the government (Quinn 1998). The Indonesian JPS scholarship program used another approach that positively affected targeting; a donor-funded initiative employed independent monitors through the Central Independent Monitoring Unit (CIMU), which collected data regularly to analyze program performance (Baines 2005, 39–41, 105). The above notwithstanding, as Baines (2005, 105) points out, independent monitoring is often criticized for being an expensive exercise and thus not feasible for non-donor funded programs. All three NGO-run programs—the South African Tshomisano, the Bangladeshi BRAC, and Pakistani NRSP—included extensive monitoring of the identification processes (Simanowitz 1999; Gibbons, Simanowitz, and Nkuna 1999, 39–62). However, this was partly a function of the PWR technique, which has a built-in monitoring mechanism for beneficiary selection, requiring program staff to monitor the entire process. The most elaborate protocol was followed by the Tshomisano Program where triangulation of results was ensured by using three to four reference groups to 12. An example where the system did not work is Zimbabwe’s Child Supplementary Feeding Program. In this case, government-appointed District Nutrition Coordinators were deployed in insufficient numbers and were thus unable to perform their task efficiently. Another serious constraint that made monitoring virtually impossible was the absence of data on the number of beneficiaries and amount of transfers (Kaseke, Dhemba, and Gumbo 1997, 37). rank each household separately before tallying the results (Simanowitz 1999; Simanowitz 2000). Perhaps most important from the point of view of program designers is the fact that, irrespective of the monitoring technique and program performance, monitoring across programs was restricted to overseeing anomalies that surfaced despite the community’s opposition to them. There Monitoring across programs was is no evidence in any program restricted to overseeing anomalies that of monitors challenging aspects surfaced despite the community’s opon which community consensus position to them. There is no evidence existed. Therefore, the use of in any program of monitors challenging flexibility beyond the stipulated aspects on which community consenprogram criteria continued irsus existed. respective of the strength of the monitoring protocol. In fact, in extreme cases such as Vietnam’s, communities were even allowed to ignore the condition of conducting a means test altogether in favor of local knowledge (Nguyen and Rama 2007, 9). Incidence of flexibility despite monitoring hints at two possibilities. First, specifically in terms of criteria for selection, authorities may have realized the value of deferring to local knowledge given that centralized poverty estimates do not always capture context-specific poverty profiles. Second, the political economy of targeted programs also seems to have weighed in. Anti-poverty programs are often sold by politicians as a major achievement on their part. Therefore, it is only natural for them to shy away from creating resentment within the community; this would amount to a reversal of any good will they have generated by portraying the program as an outcome of their efforts. Given that implementing program criteria, especially in terms of per capita disbursement, could lead to intra-community tensions, it would be politically damaging for elected representatives to force adherence. Transparency. Transparency in targeted programs is primarily required at three steps: (i) while apprising communities of the program and benefits; (ii) when ensuring that those in charge of selection understand the identification criteria; and (iii) during disbursement of benefits. Public education campaigns about a program’s features were the most popular component of information dissemination. Indonesia’s JPS scholarship program undertook a major media campaign to raise awareness among communities (Baines 2005, 39–40). An effort to publicize the OPK-Raskin scheme was also undertaken (Sumarto and Suryahadi 2001, 18). In B ­ razil’s RPAP, communities were apprised of the means to benefit from the programs through information campaigns (Quinn 1998). In Bangladesh’s VGD, while no information dissemination protocol was mandated, the BRAC NGO staff, which was involved with VGD communities, apprised locals about the program and their entitlements (Hossain 2007). Programs in Vietnam suggest that established mechanisms that favor transparency may reduce the need for program-specific protocols to the same ­effect. For instance, “village chiefs” in Vietnam already internalize information dissemination as a societal norm (World Bank 2003, 83, 113–14). Therefore, despite the fact that no Vietnamese program instituted a program-specific transparency mechanism, the established community-level protocols seem to have buffered against elite capture and corruption. Next, programs applying PWR conducted a detailed briefing of the communities to inform them of the program, the PWR technique, and potential program benefits (Simanowitz 1999; Matin and Halder 2004, 6–9; Yusuf 2007). Again, such transparency is built into PWR’s formal methodology. That said, BRAC’s experience throws up a dilemma in terms of providing ex ante information about program benefits. Communities tended to find the PWR exercise cumbersome, and most only joined reluctantly. In Bangladesh, a study conducted by BRAC (2004) points to high community disinterest despite repeated requests to participate in PWR. This suggests that for programs employing PWR, it may be prudent to carefully balance between questions of social science research and project implementation ethics, thereby withholding information that may lead communities to become disinterested in the exercise (Hoddinott 1999, 20). In terms of apprising selection committees about program objectives and selection criteria, the data exhibits a strong link between transparency and complexity of selection design. Complexity can be a result of complicated documents that are to be evaluated according to set criteria, as was the case with self-targeted programs like the IRDP and National Old Age Pension Scheme (NOAPS) in India, Ndhime Ekonomika in Albania, and Zakat in Pakistan (Srivastava 2004, 25, 30; Alderman 2001; Government of Training or briefing communities Pakistan 2006b). Alternatively, becomes important as there are often programs could include multiple no pre-established communication targeting tiers that community channels between program conceivers selectors need to be informed and communities. about. There could also be specific groups of marginalized within the poor to whom programs’ conceivers wish to allocate a disproportionately high level of benefit. Training or briefing communities becomes important as there are often no pre-established communication channels between program conceivers and communities. Moreover, unless advised, communities have little knowledge of what is expected of them (Coady, Grosh, and Hoddinott 2004a, 61). Overall, 12 out of the 30 programs can be characterized as having complex selection requirements or program criteria. Five of these, the progressive NOAPS, the mildly progressive Zakat and TIP, and the regressive PDMDKE and CSFP, lacked a training component. Three others, while missing program-specific performance, relied on pre-existing capacity among communities courtesy of institutionalized structures regularly tasked to perform such functions. This was the case in Vietnam, Thailand, and Uzbekistan.13 Indonesia’s JPS scholarship program had the most elaborate training and socialization component: it used independent monitors and master trainers to assist community groups (Baines 2005, 80–81). Chile combined the two traits, whereby school teachers conducted census exercises periodically even before the School Feeding Program but were also trained specifically to identify the health problems that would make children eligible for program 13. The village chiefs in Vietnam are mandated to maintain up-to-date knowledge about community members and are called upon to take the lead in all community-led initiatives. The Thai Community Organizations are also part of all programs that involve community participation. By the same token, the Uzbek Mahallas are instituted in the country’s governance structure by law and remain central to community issues. All three benefit from institutional memory in terms of conducting tasks such as targeting and program implementation. benefits (Kain, Uauy, and Taibo 2002, 602). To the contrary, the three cases where lack of training seems to have affected performance adversely are Indonesia’s PDM-DKE, Malawi’s TIP, and Zimbabwe’s Child Supplementary Feeding Program (CSFP). Malawi’s case was extreme in that even the selection committees—the Village Task Forces (VTFs)—were not formed in one fourth of the villages included in a survey conducted by Chinsinga, et al. (2002, 32). In Zimbabwe, lack of proper training resulted in poor data records and incorrect preparation of food packets (Kaseke, Dhemba, and Gumbo 1997, 37; Walker and Ncube 1997, 14; Munro 2002, 248). Finally, in terms of disbursement, programs that featured transparency either mandated public distribution or sought to reduce human contact in benefit transfer. Distribution through checks deposited directly in beneficiary bank accounts in lieu of cash seems to have reduced incidence of misappropriation in India’s NOAPS (Srivastava 2004, 30). Another mechanism adopted by programs such as the Albanian Ndhime Ekonomika and Indonesian JPS scholarship programs disbursed money through post offices—the postal service was regarded as relatively honest—rather than local program facilities (Kolpeja 2005; Baines 2005, 38). Apart from that, programs such as NOAPS transferred cash in public,14 while Nepal’s Churia and Bangladesh’s FFE mandated grain distribution in the presence of communities (Meagher, Upadhyaya, and Wilkinson 1999, 19–20; Ahmed and Nonno 2002, 11). Assessments of these programs do not report any concern about misappropriation during disbursement of benefits. Accountability. Of the 21 programs analyzed for accountability, 14 exhib- ited a functional protocol. Unlike monitoring and transparency, however, no accountability protocol was left to the community. Each program that featured a system of accountability required a party detached from beneficiary selection or benefit transfer to manage the protocol. Administrative ­accountability mechanisms were most frequently adopted. Albania’s Ndhime Ekonomika allowed any household or government-appointed program inspector to contest the selection decision of the committees (Kolpeja 2005b). Contestation was to be taken up by the regional offices in charge 14. NOAPS provisions for both options, i.e., distributing cash in public or transferring benefits directly to recipient bank accounts. of the program. Vietnam’s HEPR tasked a civil servant in charge of social issues in the Peoples’ Committee to review and act upon all contested cases (World Bank 2003, 114). BRAC and NRSP allowed direct complaints against PWR group members to the field staff or to the senior program staff in regional offices (Yusuf 2007). Indonesia’s JPS scholarship program used the dedicated CIMU. The Unit actively pursued cases of misappropriation; it seems to have acted as a deterrent as only 12 cases of petty corruption were brought to its notice in the first six years of the program—a significant achievement for a country that is otherwise ranked in the bottom half of global corruption indices (Baines 2005, 96, 115–118). Two programs relied heavily on public accountability. Nepal’s Churia, a progressive program, apart from conducting regular audits and displaying all information in the public domain, exposed cases of misappropriation to shame culprits in large community gatherings and forced them to cover the shortfall in benefits before the next tranche could be released to the communities (Meagher, Upadhyaya, and Wilkinson 1999, 18–20). Interestingly, while Nepal has traditionally suffered from chronic institutional perversions in anti-poverty programs and has poor accountability indicators,15 the Churia program successfully isolated the initiative from the broader context by developing a sense of community ownership and creating a system that made a good reputation an imperative for beneficiaries to continue taking advantage of the program (Meagher, Upadhyaya, and Wilkinson 1999, 1–9). The Thai SIF targeted accountability towards the community networks responsible for benefit allocation and disbursement. It relied on public announcements as a means to deter excesses. A website was also used to post developments, including any instances of misappropriation (Salim 2001). This mechanism succeeded because community networks are a permanent feature of the Thai civil society and thus they are very concerned about their reputation. 15. In the most elaborate and recent compilation of global governance indices, Kaufmann, Kraay, and Mastruzzi (2008) rank Nepal’s “Voice and Accountability” environment at 170th out of the 212 Elite Capture and Corruption: Elite capture is a concern that all community- centered initiatives have to grapple with. Keeping the elite out of the targeting process is unrealistic (Wade 1987). Often, the elite are the only cohort who can comprehend program objectives and bring community groups together to implement initiatives (Bardhan 2000). Elite capture can take place both at the behest of the community influentials as well as without their concurrence. Among the analyzed programs, the most non-transparent ones saw communities themselves being involved in manipulating the process. In some villages in Malawi’s TIP, the VTFs designated to select beneficiaries deliberately kept the larger community out of the information loop and changed the beneficiary lists after selection at the whims of the village heads. VTF members even self-selected themselves regularly (Chinsinga, et al. 2002, 33– 34, 46). Pakistan’s Zakat is another program where the Local Zakat Committees (LZCs) collude with social and political elite to misdirect benefits deliberately. According to Arif (2006, 37), 42 percent of the beneficiaries are recommended by local landlords, elite, and religious leaders, or are relatives of the members of the LZCs. Indonesia’s PDM-DKE also saw community selection committees restricting access to the decision-making process for the broader community (Perdana and Maxwell 2004, 23–24). One major factor in avoiding elite capture, especially in programs that accorded discretion, was the personalities of those in charge of selection. Perdana and Maxwell (2004, 28) argue that in Indonesia’s OPK-Raskin, targeting performance was best in villages where local personnel were honest and understood the ultimate objectives of the program. Similarly, in Pakistan’s NRSP, anecdotal evidence recorded by the author (Yusuf 2007) found the charismatic personality of the community organization leader to be positively related to the success of the initiative. In Vietnam, the village chiefs are well respected and under a social obligation to act in the community’s interest (World Bank 2003, 125–26).16 This is also the case with Uzbekistan’s Mahallas. Mahallah heads have considerable discretion but are not found to be colluding with the elite (Sievers 2002, 153; Coudouel, Marnie, and ­Micklewright 1998). 16. The village chiefs in Vietnam were not accorded any discretion except in HEPR’s “Credit” program. In the other three sub-components, decision making was highly democratic. The other set of programs that experienced elite capture were hijacked despite—not because of—communities. In India’s Sampoorna Grameen Rozgar Yojana (SGRY), all projects selected under the program were decided by the political elite in line with the preference of contractors. Elite simply forced Gram Sabhas17—these are also established institutions but did not manage to keep the elite at bay—to announce their selected projects and beneficiaries during the meetings to fulfill the formality. The elite even managed to have Gram Sabha-based selection abandoned in the second phase of the program. The social caste system contributed to mis-targeting, with the scheduled castes facing discrimination in beneficiary selection (Deshingkar, Johnson, and Farrington 2005, 580–85). India’s Indira Awas Yojana (IAY) also suffered from elite capture whereby local politicians used beneficiary selection and benefit distribution as a means of patronage, thus favoring their constituencies. Even then, beneficiaries had to bribe officials to receive their entitlements (Srivastava 2004, 28–29). In Bangladesh’s VGD program, Ninno (2001) found that the selection committee was involved in identification in only 69 percent of the sample cases. There was also evidence of political patronage. The local-level Union Parishad18 officials retained household VGD cards against the stipulated protocol in order to keep communities indebted to them for the transfer of benefits (Hossain 2007, 15). Women in VGD communities reinforced the patronage potential by lobbying aggressively with the politicians to encourage selection committees to include their names (Hossain 2007, 15). The analysis hints at a relationship between social disparity and elite capture. While we have been unable to formally test this relationship, social hierarchies do seem to play a role. Indian programs are notorious for exhibiting community behavior that suggests their helplessness in the face of excesses committed by the elite. A classic example was the SGRY where extensive corruption and nepotism led communities to seek assistance by making phone calls to the local police. However, the influentials simply had the community’s phone access cut off prior to any planned large-scale sale of rice (contractors paid employees in cash and sold the rice at a higher price in 17. Gram Sabhas are village-level community groups, widespread throughout India, designed to ensure community involvement in decisions that affect their development. 18. Union Parishad is an administrative unit roughly comprising 15 villages. the market) originally intended to reach the beneficiaries (Deshingkar, Johnson, and Farrington 2005, 586). Similarly, using inequality and profession of household heads as a proxy for societal disparity, Gallasso and R ­ avallion (2000, 23–24) show that targeting was poorer in villages with greater disparity in Bangladesh’s FFE program. The social disparity in FFE’s case may explain the program’s anomalous nature in terms of being plagued by elite capture and corruption despite exhibiting functional accountability and transparency protocols.19 The FFE seems to be an example of a case wherein the poor are more amenable to upholding social norms and not challenging established instiThe analysis hints at a relationship tutional structures for fear of between social disparity and elite reprisal from the elite (Coady, capture. While we have been unable Grosh, and Hoddinott 2004a, to formally test this relationship, social 60). Governance protocols, while hierarchies do seem to play a role. functional, may be useless in such cases. This also runs counter to the idea that community participation in community-driven initiatives increases their leverage and inclination to challenge existing power structures (Thomass-Slayter 1994). Quite to the contrary, inequality itself may become a deterrent for community participation. Moreover, viewing the situation from the elites’ perspective, the fear of a challenge to the status quo would tend to keep them on guard against any such possibility. This explains elite behavior in India’s SGRY. To be sure, our argument is not designed to be deterministic. Indeed, as Nepal’s Churia and India’s NOAPS programs show, even communities with high inequality levels can overcome this problem. In the Churia project, it was the incidence of elite capture prevalent in prior Nepalese programs that led donors to include strong buffers. Quite remarkably, by the end of the program, politicians, otherwise considered to be a threat to such initiatives, were largely positive about community ownership. At the same time, 19. The only other program to do so was Bangladesh’s VGD. Given that both were set in the same overarching context (their areas of operation overlapped), the following argument made for FFE may also be applicable to VGD. (However, our data does not allow us to substantiate this empirically.) leakages of grain saw a fifty-fold reduction compared to traditional Nepalese Food for Work programs (Meagher, Upadhyaya, and Wilkinson 1999, 27–28, 35). Conclusion: Putting Community Targeting in Perspective The foregoing analysis has highlighted a number of concerns that must be taken into account when designing community-targeted interventions. Arguably, the single most important finding is that community-targeted interventions have tremendous potential to benefit the poor; the technique is undoubtedly preferable to universal programs that are available to the entire population. Out of the 30 analyzed programs, only four were regressive. However, the generally positive performance should not imply that community targeting necessarily complements anti-poverty goals of program conceivers. In fact, we find an inherent disconnect between community behavior and overarching program objectives. The informational advantage—this is the very factor that makes use of communities attractive—acts as a double-edged sword. In terms of beneficiary identification, better information automatically lends itself to a provision for flexibility. Flexibility in beneficiary identification can lead to greater efficiency, as was seen Arguably, the single most important in cases like Vietnam’s HEPR and finding is that community-targeted Bangladesh’s FFE. Even though interventions have tremendous potenthe result is not uniform, this tial to benefit the poor; the technique implies that communities are is undoubtedly preferable to universal indeed able to use accumulated programs that are available to the social capital to acquire and use entire population. the informational advantage. However, social capital and informational advantage are bound by a tautological relationship. While leading to an informational advantage, capital stock is also responsible for pushing communities to ignore program criteria in terms of benefits transfer. The latter inevitably clashes with poverty reduction goals. Consider that the very notion of social capital drives towards strong individual bonds that harness people’s ability for growth and development while improving the prospects for community betterment (Grootaert 1998). This automatically implies an overriding concern for community norms and interpersonal harmony. In terms of benefits transfer then, thin distribution across communities is only inevitable. This is especially true in cases where community tensions are feared to be a realistic possibility—that is, in communities with low social capital. This view also refutes the argument that hard budget constraints ensure positive targeting outcomes (Rai 2002). Such constraints actually raise the potential for tensions, given that following the transfer criteria would leave out many; thin distribution acts as a coping mechanism meant to offset the negative externality. Political realities make the social capital–poverty reduction trade-off virtually irreconcilable. For one, program conceivers, be they governments, donors, or NGOs, always remain wary of spurring societal tensions and are thus likely to accept diluted poverty reduction impacts as a fait accompli. Moreover, political economy concerns are certain to prompt local politicians to favor flexibility in disbursement since this portrays them in a favorable light. The flexibility-induced conundrum is not the only pitfall associated with community-targeted programs. Our analysis reinforces the perception that there is a substantial disconnect between the theoretical advantages of community participation and their empirical performance. Community members tasked to select beneficiaries remain susceptible to collusion as long as robust governance-enhancing protocols are missing. Moreover, elite capture is evident in a number of cases. Not only that, but as reflected by ­Bangladesh’s VGD program, communities at times even play into the hands of the political elite to achieve benefits. Social accountability and social capital seem to assist only in cases where strong community-based governance mechanisms exist as a societal norm. However, these do not come about simply by involving communities in initiatives; they seem to be institutionalized through long-established community-centric lifestyles. The above said, perhaps the most encouraging finding from the point of view of policy makers is that institutional perversions in communitytargeted programs can be addressed through program-specific design elements. The absence of any obvious link between targeting performance and country incomes, regional location, targeting techniques, or program types suggests high explanatory power for targeting design elements. While a robust targeting design is imperative, there is no evidence that community involvement makes implementation of a robust design easier and less costly or that communities are better able to oversee proper implementation. Indeed, community-led design mechanisms should not be seen as substitutes for administrative protocols. In fact, we find that successful design elements are seldom run solely by communities; rather, they are meant to check community behavior (apart from flexibility) as much as they are aimed at avoiding any anomalies that communities are unable to question given social and institutional realities. Multi-tiered monitoring, which involves communities at the lower tier and individuals or teams with no direct stake in the program at the second tier, seems preferable. Successful accountability protocols were also managed by arrangements that did not involve communities and ensured absence of any conflict of interest on the part of the executors. Further, transparency in program design and information dissemination among communities, as well as in selection and disbursement, is essential. However, we do not see simplicity of program design as an imperative. Targeted anti-poverty programs are often complex, and complexity is bound to increase if an elaborate targeting design is employed, as recommended here. Given that, pursuing simplicity in program design as an end in itself may in fact be counterproductive to enhancing targeting efficiency. A better option is to ensure comprehensive briefing/training of the community members involved in implementation of the targeting protocols. By implication, training of those tasked to brief communities and oversee composition of selection committees becomes vital. Moreover, in terms of disbursement, transparency can be achieved through public distribution, disbursement through honest intermediaries, or through direct transfer mechanisms that remove the need for intermediaries altogether. Finally, programs that are dynamic in their designs such that they undertake regular updates of beneficiary lists and design procedures seem to be better placed. The latter is important to preempt the “learning effect” whereby the non-poor and intermediaries may extract benefits once they learn how to manipulate the targeting process (Lipton and Gaag 1993, 35). While our analysis does not lend itself to any “one-size-fits-all” formulation, the broad contours of a context that is supportive for community-targeted initiatives is discernable. To begin with, conceivers of community-targeted interventions should consider lower-than-stipulated per capita benefit transfer as a fait accompli and provision for this tendency in their feasibility analysis. Moreover, where communities are used, poverty reduction impacts should not be strictly dependent on following stipulated criteria in terms of poverty proxies. Communities are certain to digress from the criteria and any effort to check this tendency would undermine the very rationale for employing community targeting, i.e., using local knowledge. Moreover, it would inevitably result in community and local political opposition and risk spurring societal discord. The cost-benefit analysis of the initiative may become unfavorable if the effects of these potential pitfalls are internalized. Similarly, community targeting is not recommended for programs where aggregation of poverty data is a high priority. Community targeting is better attuned to communities where societal tensions and extreme disparity are not a preexisting concern. Moreover, communities can be expected to uphold social norms and thus would exclude segments of the community that do not fit their framework. While this does not necessarily imply the need for homogeneity, it does suggest that communities bearing excessively disparate power structures or ones inclined towards cultural exclusion based on criteria other than poverty are less suited to community targeting. Finally, while overarching societal norms that lend themselves to strong governance matter, this is not a luxury most developing countries enjoy. Therefore, the focus ought to remain on instituting a robust, programspecific design protocol. Monitoring, transparency, and accountability tend to operate in tandem in programs for the most part. Therefore, program ­designers should think of design elements as an overarching whole and seek to integrate these functions rather than treating each one in isolation. ­Specifically for monitoring, synergies should be developed between the community and administrative protocols that are operating together. The dictum of “trust but verify” should remain central for those who conceive programs. Moreover, while we have not been able to analyze procedures through which selection committees are formulated, it is obvious that the person heading the selection process remains vital. Thus, even if the role of the elite is inevitable, program managers must retain some say in the choice of the “head.” Table 6 lists the ideal conditions for community-targeted programs. Table 6: Ideal Conditions for Community-Targeted Programs Context Preferred Status Importance of following stipulated criteria for beneficiary selection Low Importance of transferring the stipulated per capita benefit to beneficiaries Low Priority for aggregation of poverty data across multiple communities Low Preexisting social tensions within the community Low Level of societal disparity in power structures Not extreme Tendency towards cultural exclusion of marginalized groups Absent Robust, program-specific design protocol Present Societal norms towards accountability and transparency Present Program managers’ ability to choose “head” of the beneficiary selection process from within the community Present While the above analysis is still far from providing a comprehensive picture of community targeting, it does present a nuanced view on the strengths and pitfalls associated with the practice. Since this targeting technique continues to enjoy growing support as a means of identifying the poor in anti-poverty programs, development experts must play close attention to the factors that are most likely to enhance targeting efficiency; this is imperative if the global quest to increase the utility of poverty-reduction programs is to be realized in the coming years.