Can Community-based Targeting Effectively Select Poorer Beneficiaries for a Large-scale Program? Insights from the LASED Project

CONTEXT With 34% of the population at the national poverty line in 20081, the Royal Government of Cambodia established Social Land Concessions (SLCs) based on Sub-Decree 19 (No. 19 ANK/BK/ March 19, 2003) to distribute state-owned land to the landless and land-poor. The Land Allocation for Social and Economic Development Project (LASED) was organized to pilot commune-based SLCs, allocating land and improving communal infrastructure. LASED is an ongoing project, organized into several phases. The first phase ran from 2008 to 2013. Over 11,000 households applied for SLCs during this phase, and over 4,000 land plots were distributed across five provinces: Kampong Chhnang, Kampong Speu, Kampong Thom, Tboung Khmum, and Kratie. LASED used community-based targeting (CBT) to identify beneficiary selection criteria and the final recipient households. EAST ASIA AND PACIFIC GENDER INNOVATION LAB

• Close targeting around poorer beneficiaries can be achieved across a wide project area and an implementation process spanning years.
• An iterative community-based selection process held over several smaller meetings may improve selection outcomes.

WHAT DID WE DO?
The LASED team invited EAPGIL to evaluate the project's effectiveness in improving the welfare of poor households. We aimed to use regression discontinuity (RD) analysis for the evaluation. This method compares households immediately on either side of the program eligibility cut-off score. As long as eligible and ineligible households are nearly identical before the program, differences seen later can be attributed to the program. Consequently, for LASED, if households close to cut-off score were highly similar in poverty characteristics before land was allocated, later differences in poverty could be attributed to the SLC grants.
We accordingly assessed the baseline poverty status of households a few points above and below the SLC cutoff scores to determine whether RD analysis could be used to evaluate LASED.

WHAT DID WE FIND?
Exploration of the data showed that RD analysis would not be feasible. The households on either side of the cut-off scores were markedly different in poverty status. However, though RD analysis could not be completed, the data revealed a few positive findings regarding program targeting.

LASED targeting was strongly pro-poor
Those above the cut-off score were significantly poorer than those below the cut-off score, across a range of characteristics. Land recipients had no, or significantly smaller plots of land, fewer cattle and other livestock, and were less likely to own a house. They were more likely to be widowed, to work as laborers, and to be registered 2 poor 3 . For example, while only slightly more than 10% of non-recipient households were registered poor, Residents had time to review the documents, submit complaints and corrections.
• Iterative process: 2-3 review rounds • 122 days reserved for community participation STEP 1 HHs complete applications for land. Applications capture demographic info.

STEP 2
Applications posted publicly, open to revision.

STEP 3
Applications ranked according to community criteria.

STEP 4
Priority recipients publicly posted their applications avalable publicly for review.

STEP 6
Final pool of potential recipients selected.

STEP 7
Ranking cut-off score determined.

STEP 8
HHS submit appeals.

STEP 9
HHs above cut-off score enter land plot lottery.

STEP 5
Communities review the information and submit appeals and corrections.

2-3 ROUNDS OF PUBLICLY REVIEWING AND REFINING THE APPLICATIONS AND RANKINGS
122-day minimum mandated for application submission, review, and selection process over half of recipient households were. Non-recipient households had nearly twice the amount of residential and agricultural land as recipient households. Notably, significant differences between recipients and nonrecipients remain for a range of band width around the cut-off score 4 .

Richer households were progressively siphoned away as the project proceeded
The iterative process of community review and correction of applications successfully pushed poorer households forward in the processes and removed richer households from consideration. All interested households could submit applications. During the initial review of applications, richer households were culled from the pool (Step 2 in Figure 1). Poorer households progressed to the ranking stage. This can be seen by comparing characteristics between households that progressed to the ranking stage and households that were dropped without being ranked. Those that were ranked are poorer than those that were dropped.
LASED selection continued to tighten around poorer households during the ranking phase. The poorer among ranked households predominately entered lucky draws to receive land, while the richer were again siphoned away.

Pro-Poor targeting is seen across provinces
Across provinces, land recipients had poorer characteristics than those who did not receive land. For example, though the amount of difference between recipient and non-recipient households varied by province, recipients were consistently less likely to own a house. Similarly, recipient households were more likely to be registered poor (IDPoor) -overall and by strata of poverty levels. The spread in pro-poor targeting across project areas indicates that LASED provided a systematic, rather than incidental, identification of poorer households to receive social welfare benefits.

WHAT CONTRIBUTED TO LASED'S SUCCESSFUL TARGETING?
Community Based Targeting (CBT) programs often experience tension between communal acceptability 2 For most areas in Cambodia, the Ministry of Planning maintains a list, updated every three years, of the registered poverty status ("IDPoor") of families.
3 Marital status and occupation differences were calculated via chi-square testing. All other comparisons were calculated through t-tests of differences in group means. 4 Detailed analysis of feasibility of RD design available from authors upon request. and targeting effectiveness 5 . LASED seemed to bridge this, with widespread community acceptance and participation, while also effectively identifying poorer households to receive program benefits. We reviewed the LASED's design to explore reasons for its success and saw that several project design factors aligned with CBT strategies recommended by the literature. The following factors were taken into account: Community definition of criteria and ranking: Rather than receiving definitions of eligible beneficiaries from external donors or other parties, local communities could adjust selection criteria and determine their own target populations. Distribution of project benefits therefore was more likely to resonate with the participants' values, contributing to acceptability.
Concurrent community-based and proxy means targeting methodologies: Projects often implement an initial CBT component to identify a pool of potential beneficiaries, and then use a proxy means targeting (PMT) survey to validate the CBT results. The PMT component may come from a source external to the community. When discrepancies are found between CBT and PMT results, the PMT results are frequently used to "correct" the CBT results, negating the contribution of the CBT method and potentially causing tensions in the community. LASED integrated the CBT and PMT components, and rooted both in community decision-making, unifying participatory methods with access to both the unique local knowledge base and an objective selection rubric.
Transparent selection process with public posting of potential and ultimate beneficiaries: All stages of LASED's selection process were open to the public's review and criticism, reducing chances of elite capture, and increasing program acceptability and use of local knowledge. Part of project transparency included diversifying the decision-making body. Project materials were open to all, and not just project and community leadership. Included among these community members, women also contributed to project monitoring and effective targeting. Others have identified including women in project committees as a way to improve targeting effectiveness 6 .
Iterative process extending over weeks: CBT is prone to exhausting participants. Restricting the CBT component to an hours-long meeting to review and rank potential beneficiaries, as is often done, often induces participant fatigue 7 . Testing the order in which potential beneficiaries are ranked has shown a higher likelihood of beneficiary inclusion and exclusion errors among those ranked last. With the overall process spanning several weeks, and the community members having several days to review project materials during each iteration, LASED may have mitigated participant fatigue.
Local decision-making by homogenous village and commune groups: Successful CBT programs are typically conducted among homogenous groups with shared values and local knowledge banks 8 . Although the overall project was large-scale, encompassing many households, decisions were orchestrated at the local level, promoting community consensus and acceptance.