Leveraging Social Connections: Using decentralized targeting to deliver cash transfers in Monrovia, Liberia
Last registered on March 15, 2019

Pre-Trial

Trial Information
General Information
Title
Leveraging Social Connections: Using decentralized targeting to deliver cash transfers in Monrovia, Liberia
RCT ID
AEARCTR-0001164
Initial registration date
July 18, 2018
Last updated
March 15, 2019 2:15 PM EDT
Location(s)

This section is unavailable to the public. Use the button below to request access to this information.

Request Information
Primary Investigator
Affiliation
University of California, Berkeley
Other Primary Investigator(s)
PI Affiliation
University of California, Berkeley
PI Affiliation
University of California, Berkeley
PI Affiliation
Northwestern University
Additional Trial Information
Status
On going
Start date
2018-02-01
End date
2019-09-15
Secondary IDs
Abstract
Targeting beneficiaries for social programs in urban areas are increasingly important as urban populations grow and poverty or emergency relief programs become more common in densely populated settings. However, the current targeting strategies and tools may not be best suited for these dynamic urban environments. For example, the tools for targeting social programs often rely on methods developed in rural settings. These rural programs often leverage pre-existing social and political institutions to target beneficiaries. We theorize that the effectiveness of these structures may break down in shifting, urban environments. Another popular beneficiary targeting tool is the proxy means test. While proxy means tests are promoted as a quick option for assessing program eligibility, it requires regular updates to calibrate the means testing and does not translate well outside of welfare-based eligibility.

This study aims to provide evidence and recommendations for identifying beneficiary households/individuals for social programs in urban areas. In parallel to the traditional approach to targeting, we will use a decentralized targeting mechanism that aims to gather information held by socially knowledgeable members of an urban community. The principal social program that we will implement is a one-time unconditional cash transfer to households. For the cash transfer, we intend to verify whether or not this decentralized targeting of the unconditional cash transfer is effective in reaching poor households and households that have experienced an economic or health shock. We propose a decentralized mechanism for reaching program beneficiaries. The decentralized targeting mechanism, if found effective, could be used more widely to apply local community knowledge and identify key social nodes to improve program beneficiary selection.

We set out to examine how a decentralized targeting method differs from beneficiary selection in a more traditional regime. To do so, we conducted a study that tested multiple channels of identifying poor and vulnerable households in urban neighborhoods of Monrovia, Liberia.

The primary research questions we set out to address were:
• Do certain members of an urban community have better access to information about households that are most likely to benefit from a social program?
• Are certain members of an urban community better positioned to share information about a social program?
• Depending on how they are selected, do beneficiaries of a cash transfer differ in their consumption and investment patterns?

Through this study, we hope to contribute evidence about targeting of diverse social programs through an experiment with residents of Monrovia, Liberia. We will solicit the advice of individual residents within the community to identify beneficiaries for two separate programs. We sought to provide evidence to verify whether leveraging social connections lead to differential beneficiary selection than other methods. The project investigated how local, urban social networks can be resourced to reach households within a densely populated community with diverse social protection programs.
External Link(s)
Registration Citation
Citation
Beaman, Lori et al. 2019. "Leveraging Social Connections: Using decentralized targeting to deliver cash transfers in Monrovia, Liberia." AEA RCT Registry. March 15. https://www.socialscienceregistry.org/trials/1164/history/43461
Sponsors & Partners

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information
Experimental Details
Interventions
Intervention(s)
We will canvas community blocks (the smallest increment of formal community division in Monrovia) in selected communities and conduct the household census with heads of households or their partners. Subjects will be invited to participate in the research voluntarily. Written consent will be recorded. This household census will include a form of the proxy means test, questions about recent health and economic shocks, questions about their inter-household social interactions/who the respondent pays attention to in the community, and questions about their perceptions of other community members’ relevant knowledge.

After the household census, some community members will be invited to complete a one-on-one interview with an IPA staff member. Invited community members will be drawn from 15% of households interviewed in the household census. We refer to the individuals invited for one-on-one meetings as “Targeting Assistants” (TAs) as they will provide input to help determine how the unconditional cash grant will be distributed among households living within the community.

We refer to the one-on-one interview as a “targeting assistant survey” as the interview aims to elicit the beliefs and preferences of the Targeting Assistants. Subjects will be invited to participate in the research voluntarily. Written consent will be recorded. During the targeting assistant survey, TAs will be asked to express their belief about whether or not other households in the community are among the poorest 20% of households in the community.

Targeting Assistants will not know about the cash grant before targeting assistant survey. We will then randomize way in which the targeting assistant survey is introduced. One of the following two paragraphs will be read to the targeting assistant.

"The goal of this program is to reach the poorest households in this community. We would like to request your assistance in deciding how to reach the poorest households in this community."

OR

"The goal of this program is to reach the households affected by a major loss of wealth (fire, flood, theft) in this community. We would like to request your help in deciding how to reach households affected by a major loss of wealth (fire, flood, theft) in this community."

Half of the targeting assistants will hear the first framing (poverty) and the other half will hear the second framing (shocks).

The variation is important to understand whether or not framing the motivation changes the way in which targeting assistants assess the welfare of other members of the community.

In addition to questions to assess the TAs familiarity with other members of the community, TAs will be asked the following two questions about the relative welfare of 30 households within their community:

Question 1:
Please imagine a 5-step ladder. [Show picture of ladder] On the bottom, the first step stand the poorest 20% of households in {community}. On the highest step, the fifth stands the richest 20% of households in {community}.
Where do you think {name}’s household stands when you think of how poor or wealthy {name}’s household is compared to others in {community}? (Ask the respondent to provide a number 1-5)

In place of the {name} placeholder, we will ask the targeting assistant about specific people that live in their neighborhood. The names will be drawn from the household census. As such, using a fictitious name, the question as asked will be asked as: Where do you think {Joesph Tubman}’s household stands when you think of how poor or wealthy {Joseph Tubman}’s household is compared to others in {Banjor}?

Question 2:
If we asked other people in your community, what percentage would say that {name}’s household is on the first step of the ladder, that is in the poorest 20% of households in {community}?

As with Question 1, we will insert names of members of the neighborhood. For example, using the fictitious case above, If we asked other people in your community, what percentage would say that {Joseph Tubman}’s household is on the first step of the ladder, that is in the poorest 20% of households in {community}?

We ask you to report this probability in percentages, ranging from 0% to 100%. For example, if you are completely sure that others in your community block would say that {name}’s household is on the first step of the ladder, you should report a probability of 100%. If you are not sure whether others in your community block would agree that {name}’s household is among the poorest 20% of households, you should report a probability between 0% and 100%.

The 30 households that each TA is asked about will be randomly drawn from the total population of households in the TA's community. This will provide 5 data points from targeting assistant surveys for every household in the community.

All households within the study area will be eligible for a cash grant drawing, independent of their participation in the study. Beneficiaries of the cash grant drawing will receive an expected value of USD$80. We anticipate a maximum of 280 households receiving a cash grant. Participation in the study is not required to be eligible for the cash grant drawing. Using the rankings and data collected from the targeting assistant surveys and community leader meetings, one-time cash grants will be provided to selected beneficiaries. This selection will include the households identified in the community targeting exercise in addition to households most likely to be considered poor by TAs. Effectively, if the TAs identify a household as poor, the likelihood of a household receiving a cash grant is increased.

Heads of selected households will be invited by phone to collect their cash grant at a designated central location near the community. Cash grant beneficiaries will be asked to verify their identity (name, age, gender, phone number). Cash grants will only be distributed to verified members of the urban neighborhoods included in this study. There is only one cash grant per household.

Responses from the Targeting Assistant Surveys will be aggregated to give each household in the community is given a score (i.e. number of Targeting Assistants that classify a household as poor). We refer to the score as the poverty-likelihood score. The households most commonly identified as poor are more likely to receive a cash grant. The poverty-likelihood score from the targeting assistant surveys effectively serves as a probability weight for the likelihood of receiving the cash grant. To give all households a non-zero probability of receiving a cash grant, we will also randomly select (weighted by the poverty-likelihood score) households from each community to receive the cash grant. The random assignment is intended to provide exogenous variation in the amount of cash received by a household, thus allowing us to evaluate the causal effect of additional cash on household behaviors. We anticipate that the households selected to receive the cash grant will be 20% of households within each community.
Intervention Start Date
2018-07-19
Intervention End Date
2018-09-01
Primary Outcomes
Primary Outcomes (end points)
Consumption; Business assets
Primary Outcomes (explanation)
Secondary Outcomes
Secondary Outcomes (end points)
Inter-household transfers
Secondary Outcomes (explanation)
Experimental Design
Experimental Design
Using the rankings and data collected from the select community member surveys, one-time cash transfers will be provided to beneficiaries. This selection of beneficiaries is informed by three factors. First, a group of 80 households receives a cash transfer if nominated by a leader within their community block. Second, a group of 120 households receives a cash transfer if nominated by a non-leader within their community block. Third, a group of 80 households will be randomly selected from the poorest households within the community block.
Experimental Design Details
Not available
Randomization Method
Randomization performed by computer using matched pairs of households.
Randomization Unit
Household
Was the treatment clustered?
No
Experiment Characteristics
Sample size: planned number of clusters
Treatment is not clustered
Sample size: planned number of observations
560 households
Sample size (or number of clusters) by treatment arms
280 control households, 280 cash transfer households
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
0.5 SD
IRB
INSTITUTIONAL REVIEW BOARDS (IRBs)
IRB Name
University of Liberia PIRE
IRB Approval Date
2018-01-08
IRB Approval Number
17-12-082
IRB Name
Innovations for Poverty Action
IRB Approval Date
2017-12-14
IRB Approval Number
13958
IRB Name
University of California Berkeley
IRB Approval Date
2018-04-20
IRB Approval Number
2018-01-10649