Community Participation, Targeting Performance, and Administrative Capture in Community-Based Targeting: Experimental Evidence from Ethiopia

The interventions in this study occur at two levels: the community leaders and the community members (potential beneficiary households). We exogenously vary the nature of the transfer (hypothetical versus real) as well as the type of targeting process (without the participation of community members versus with participation of community members). This design allows us to study whether participatory targeting and hence inclusion of potential beneficiaries in targeting processes make CBT more effective in terms of identifying the poor and reducing administrative capture. For this purpose, we design alternative versions of CBT processes, which are described below.
Intervention: ranking and allocation of real versus hypothetical cash transfers by community leaders
We task community leaders to rank about 20 randomly selected households in their respective community from the most to the least needy, based on their need for potential social assistance. We provide community leaders with either a hypothetical or real lump-sum of cash transfers to allocate among the ranked households according to their perceived need for social assistance. The assignment of communities into the hypothetical and incentivized treatment arms is discussed in the experimental design.
Before the ranking and allocation of transfers, we administer a series of questions that aim to elicit community leaders’ objective functions in CBT. Specifically, we elicit community leaders’ priorities in the allocation of constrained budgets across households facing different constraints and challenges.
Each intervention will be implemented during structured field sessions facilitated by trained enumerators in collaboration with local authorities. Targeting committees, composed of six local leaders, will conduct the CBT exercises following standardized instructions provided by the research team. All sessions will be closely monitored to ensure adherence to the experimental protocol and consistency across communities.
The intervention tests whether community participation and real incentives in CBT can improve the accuracy of beneficiary identification. The experimental design isolates the role of incentives (real versus hypothetical transfers) and participation (with versus without public review of targeting decisions) to identify how these factors influence targeting performance.

The interventions in this study occur at two levels: the community leaders and the community members (potential beneficiary households). We exogenously vary the nature of the transfer (hypothetical versus real) as well as the type of targeting process (without the participation of community members versus with participation of community members). This design allows us to study whether participatory targeting and hence inclusion of potential beneficiaries in targeting processes make CBT more effective in terms of identifying the poor and reducing administrative capture. For this purpose, we design alternative versions of CBT processes, which are described below.
Intervention: ranking and allocation of real versus hypothetical cash transfers by community leaders
We task community leaders to rank about 20 randomly selected households in their respective community from the most to the least needy, based on their need for potential social assistance. We provide community leaders with either a hypothetical or real lump-sum of cash transfers to allocate among the ranked households according to their perceived need for social assistance. The assignment of communities into the hypothetical and incentivized treatment arms is discussed in the experimental design.
Before the ranking and allocation of transfers, we administer a series of questions that aim to elicit community leaders’ objective functions in CBT. Specifically, we elicit community leaders’ priorities in the allocation of constrained budgets across households facing different constraints and challenges.
Each intervention will be implemented during structured field sessions facilitated by trained enumerators in collaboration with local authorities. Targeting committees, composed of six local leaders, will conduct the CBT exercises following standardized instructions provided by the research team. All sessions will be closely monitored to ensure adherence to the experimental protocol and consistency across communities.
The intervention tests whether community participation and real incentives in CBT can improve the accuracy of beneficiary identification. The experimental design isolates the role of incentives (real versus hypothetical transfers) and participation (with versus without public review of targeting decisions) to identify how these factors influence targeting performance.

2025-11-13

2025-12-31

1. Households’ access to CBT transfers (whether the household receives a transfer or not)
2. Amount of transfer received by households.
3. Targeting error (inclusion and exclusion errors)
4. Distribution of transfers or inequality in the distribution of transfers
5. Administrative capture: the share of the budget devoted to running costs or administrative costs by community leaders.
6. Decision-making process and aggregation of individual preferences into collective decisions.

1. Households’ access to CBT transfers — whether the household receives a transfer or not — is a binary outcome capturing whether a household is selected to receive a transfer through the CBT process. The share of households deemed “eligible” and hence receiving a transfer from the community leaders is one of the key outcomes we use to measure the breadth and coverage of the targeting exercise and associated community-based cash transfer. Community leaders are tasked to identify “eligible” households, both in treatment arms involving actual cash transfer and in those involving hypothetical ranking without actual transfer.
2. Amount of transfer received by households measures the intensive margin of participation in the CBT program. This outcome complements the first by capturing the amount of transfer allocated to each household by community leaders. The amount received reflects trade-offs faced by community leaders between maximizing the number of beneficiaries covered versus the amount received per household.
3. Targeting error captures whether cash transfers reach poor households. Based on the list of households deemed “eligible” and their household consumption expenditure, we will compute measures of overall targeting errors as well as inclusion and exclusion errors. Inclusion error is defined as the probability of receiving a transfer conditional on consumption above the poverty line, while exclusion error is defined as the probability of not receiving a transfer conditional on consumption below the poverty line.
4. Administrative capture is measured by the share of the budget used to cover administrative costs claimed by community leaders. Community leaders are offered the option of retaining up to 10 percent of the budget allocated to cover “administrative costs”. This is in addition to a show-up fee of 350 Ethiopian Birr (about 2.5 USD during the survey) that will be paid to each community leader as a token of appreciation. This decision involves two stages: first, each community leader individually decides how much they would like to keep as an “administrative cost”, with the following options: 0 percent, 2 percent, 4 percent, 6 percent, 8 percent, and 10 percent. Second, the committee collectively decides on a negotiated amount to retain for administrative costs. We will use both the individual- and group-level shares, as well as indicator variable capturing whether the committee agrees to take the maximum amount allowed or not.
5. Distribution of transfers or inequality in transfer distribution will be measured using the Gini coefficient of transfer amount among community members. This measure allows us to identify whether specific CBT design features make community leaders more or less inequality averse in allocating transfers.
6. Decision-making process and aggregation of individual preferences into collective decisions will be analyzed using both individual and group decisions regarding the share of the budget withheld as “administrative cost”. This enables to explicitly test whether collective decision-making follows specific aggregation rules. We will examine whether different CBT designs affect negotiation behavior and whether group deliberation generates more pro-social outcomes than individual decision-making. Specifically, following Ambrus et al. (2015), we will analyze how individual preferences are aggregated into collective decisions by characterizing the relationship between individual proposals and group-level decisions (retention rates). This analysis will uncover the type of aggregation rule that best explains collective decisions and whether aggregation processes vary across incentive regimes (hypothetical versus incentivized targeting) and targeting formats (standard versus participatory CBT).

This section describes the experimental design of the CBT. The target group for this experiment is community leaders drawn from 181 communities. Thus, the key actors and targets of our interventions are the six community leaders recruited from each village. As described earlier, the CBT experiment brings together six community leaders who assume various roles and responsibilities within each village. To mimic actual targeting practices in Ethiopia, we define the six committee members to include: (i) a Kebele leader or member of the kebele leadership, (ii) an elder man/woman, (iii) a religious leader, (iv) a women’s representative, (v) a teacher or development agent or extension worker, and (vi) a youth representative. These members reflects the typical composition of local targeting committees in social protection programs in Ethiopia, including the PSNP. The same community leaders also participated in a similar targeting experiment conducted two years ago (Abay et al., 2024).

The CBT intervention follows community-level clustered randomization, in which the 181 communities are randomly assigned into one of the three treatment arms. The random assignment is stratified by region and by previous treatment status, as many of the community leaders had participated in a prior CBT experiment (Abay et al., 2024). The treatment assignment varies along two dimensions: (i) whether the community receives an actual transfer or a hypothetical transfer (control group), and (ii) whether the community members (potential beneficiary households) participate or not in the targeting process. The treatment arms generated by these combinations are outlined below.

(1) Control: Hypothetical targeting and ranking for distributing a 20,000 Birr transfer (C): Community leaders in this group are instructed to assume a hypothetical budget of 20,000 Birr, which needs to be distributed to those deemed “needy”. Before allocating transfers, community leaders are first asked to rank 20 sample households from the most to the least needy, based on their perceived need for social assistance. They are then asked to allocate this notional budget among the 20 households included in our sample according to their ranking. During this exercise, leaders are required to strictly adhere to pre-defined criteria provided by the research team. These criteria are carefully selected to mirror the targeting criteria used in actual social assistance programs in Ethiopia, such as the PSNP (Gilligan et al., 2009; Berhane et al., 2024). Community leaders are instructed to prioritize households that: (i) had difficulty satisfying their food needs; (ii) own no or little asset (e.g., livestock, land); (iii) have limited income-generating activities or capacity; (iv) have lost productive assets due to shocks (e.g., conflict, drought); and (v) have lost family members recently. These targeting criteria closely follow those used by the national safety net program in Ethiopia, the PSNP (e.g., Gilligan et al., 2009; Hoddinott et al., 2012; Abay et al., 2022).
(2) Incentivized targeting involving actual cash transfer (T1): Another group of communities is randomly assigned to a treatment arm that receives real transfer funds with a total budget of 20,000 Birr (about 140 USD). In this treatment arm, community leaders are required to rank households based on the same five pre-determined targeting criteria in the control group and allocate real cash transfers to those deemed “needy".
(3) Incentivized and participatory targeting involving community members (T2): This group of communities follows similar rules as those in T1, but the targeting process incorporates public engagement by involving potential beneficiary households. Community members (potential beneficiary households) will actively participate in in reviewing and approving the proposed beneficiary list. All decision made by the committee will be publicly disclosed and reviewed by at least two-third of the potential beneficiary households (community members) for final approval and implementation. This includes disclosing the following information to the community members (potential beneficiary households): (i) Total budget allocated, 20,000 Ethiopian Birr, (ii) Amount (and share) of transfer/budget they withhold for administrative costs by community leaders; (iii) A list of potential beneficiaries identified by community leaders, as well as the amount allocated to each beneficiary, and (iv) justification of community leaders’ decision to community members.
Community leaders are asked to rank households from the most needy (1st) to the least needy (last) and propose how funds should be allocated to those households they deem needy, but final allocation requires support by at least half of the community members (households). Most importantly, community members will vote on whether to include or exclude each proposed beneficiary. If a simple majority supports or rejects a decision, the community members' vote overrides the committee’s proposal. This treatment arm allows us to examine whether public disclosure and participation of potential beneficiaries improve targeting accuracy and reduce elite capture.

The randomization is conducted at the village (EA) level using the list of villages and communities from our previous surveys and experiments. The initial selection of villages into our sample considers their accessibility for a survey. The selected and accessible villages are then randomly assigned to one of the three groups, with stratification by region and by previous treatment assignment in a related community-based experiment. A reserve list of villages has been prepared in case some sites become inaccessible due to conflict and other unforeseen factors.

Village or community level

Yes

181 communities and about 3,000 households

181 communities and about 3,000 households

C (Control: hypothetical targeting): 61 villages
T1 (Incentivized targeting): 66 villages
T2 (Incentivized and participatory targeting): 54 villages

We compute the number of clusters needed for the primary outcomes described above, assuming a fixed number of households and community leaders in each cluster (village). Because of the nature of the outcomes that we examine in this study, the analyses will be conducted at three levels: household, community leader, and community level. In the baseline sample, there were an average of 20 households in each village, and we anticipate being able to trace about 85 percent of them (about 17 households per village). Our power calculations aim to achieve the standard and widely adopted 80 percent power at a significance level of 5 percent. Given that we have several hypotheses and primary outcomes, we computed the number of clusters and associated sample size needed for each outcome separately, and then selected the maximum number of clusters needed to detect plausible impacts across these outcomes. To evaluate the impact of participatory targeting on overall targeting outcomes, we compute statistical power and sample size needed to detect a reasonable impact on households’ access to CBT transfers (whether the household receives a transfer), the amount of transfer received by households, as well as the targeting error and share of budget devoted as “administrative cost”. For each outcome, we assemble means and standard deviations from previous studies in similar settings. We heavily rely on Abay et al. (2024) for extracting the mean and standard deviations outcomes. We complement these with estimates from additional studies (e.g., Brown et al., 2018). As shown in Table 1, to detect an 8-percentage point difference in households’ access to the cash transfers, we need about 39 clusters in each arm. Similarly, detecting an 8 percent difference in the amount of transfer going to beneficiary households requires 59 communities in the control group, and three times that for all arms. Based on these calculations, we allocate about 33 percent of the 181 communities into the hypothetical arm with no actual transfer and divide the remaining communities into two groups. We stratify the random assignment of communities across regions and previous exposure to similar targeting exercises.