Re-targeting Humanitarian Assistance: Experimental Evidence from Lebanon

Last registered on May 21, 2024

Pre-Trial

Trial Information

General Information

Title
Re-targeting Humanitarian Assistance: Experimental Evidence from Lebanon
RCT ID
AEARCTR-0011831
Initial registration date
May 17, 2024

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
May 21, 2024, 11:33 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Locations

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

Request Information

Primary Investigator

Affiliation
World Food Programme

Other Primary Investigator(s)

PI Affiliation
ISDC
PI Affiliation
University of Milan-Bicocca

Additional Trial Information

Status
In development
Start date
2024-01-01
End date
2025-07-01
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
Humanitarian organizations operate within stringent budget constraints and regularly need to prioritize and re-prioritize their assistance, depending on needs and funding. In this study, we partner with the largest humanitarian organization in the world - the World Food Programme (WFP) - to investigate alternative approaches to prioritize assistance in the context of a scaling down of humanitarian operations. The study takes place in Lebanon, where WFP is planning to scale down its food assistance program from about 58,000 beneficiary households to about 25,000, due to resource constraints.

We will compare four different approaches to prioritize households during this downscaling: three data-driven approaches will rely on Proxy Mean Testing (PMT) to select households according to different predicted outcomes (expenditure, food security, or a combination of the two), while the fourth targeting approach will rely on criteria identified through a survey conducted among local experts. The study will investigate how the different approaches lead to different profiles of beneficiaries and how these differences impact the effectiveness of assistance. The experimental design will moreover allow us to causally assess the impact of the (dis)continuation of food assistance itself, through the definition of comparable groups of beneficiaries and non-beneficiaries across the four different approaches (more specifically, we will assess the impact on the “marginal” households, i.e. eligible under one targeting method but not another).

We will study dynamics over time through two rounds of follow-up data collection, and we will also take advantage of the staggered fading out of assistance (in two phases), to study how households might prepare for the end of assistance and which coping mechanisms they might adopt.
External Link(s)

Registration Citation

Citation
Baliki, Ghassan, Felipe Alexander Dunsch and Andrea Guariso. 2024. "Re-targeting Humanitarian Assistance: Experimental Evidence from Lebanon." AEA RCT Registry. May 21. https://doi.org/10.1257/rct.11831-1.0
Sponsors & Partners

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

Request Information
Experimental Details

Interventions

Intervention(s)
The experimental variation will stem from the random assignments of households into four different prioritization approaches, which include some of the most common approaches currently adopted by WFP as well as other humanitarian organizations whenever targeting their assistance. The four approaches are based on Proxy Mean Testing (PMT) or categorical targeting and the details are as follows:

1) PMT targeting consumption-based poverty (T1): beneficiaries are identified through a PMT model, designed to identify households with lower consumption per capita (this is the status-quo targeting approach within WFP Lebanon);

2) PMT targeting food (in)security (T2): beneficiaries are identified through a PMT model, designed to identify households with lower food security;

3) PMT using multidimensional vulnerability index (T3): beneficiaries are identified through a PMT model, designed to identify households with a combination of lower food security and lower consumption per capita (more specifically, the index relies on CARI or Consolidated Approach for Reporting Indicators of Food Security);

4) Categorical selection (T4): beneficiaries are identified following a list of priority dimensions identified through an expert survey administered to local experts and practitioners.

Within each one of the four study arms, households will be ranked and the most vulnerable households will be prioritised to keep receiving assistance during the next assistance cycle, while the remaining households will stop receiving assistance. The threshold is determined by resource constraints.
Intervention Start Date
2024-02-01
Intervention End Date
2025-01-31

Primary Outcomes

Primary Outcomes (end points)
Food Security: Food Consumption Score; Coping Strategy Index
Consumption & Expenditure: Household expenditure per capita for essential food and non-food items
Employment: number of hours worked
Well-being: Stress; Psychological well-being
Primary Outcomes (explanation)

Secondary Outcomes

Secondary Outcomes (end points)
Economic and Financial: income; savings; loans; transfers
Shocks & coping: coping strategies
Social cohesion: Trust; social networks; sharing practices
Satisfaction with the targeting process: formal complaints; self-reported satisfaction
Secondary Outcomes (explanation)

Experimental Design

Experimental Design
The study builds on the ongoing effort of WFP Lebanon to prioritize households during the scaling down of its current food assistance program. WFP Lebanon is currently providing food assistance to about 75,000 Lebanese households, which include more than 5% of the population. Assistance is provided in the form of food parcels, whose weight and value depend on household size but correspond to roughly 10Kg or 10USD per person per month. Due to limited resources available, WFP is planning to significantly scale down its program. In 2023 WFP administered a rich census survey to about 58,000 households currently receiving food assistance. The census will be used to identify the 25,000 most vulnerable households among current beneficiaries, who should keep receiving assistance in the next assistance cycle, while assistance will end for the other households. Such prioritization exercise is common among WFP and, more in general, humanitarian organization operations, although different approaches have been used for identifying vulnerable households. In order to shed light on the trade-offs across different approaches, WFP will randomly assign the 58,000 households into four different groups of equal size (about 14,500 households each). Within each group, a different prioritization approach will be used to identify the most vulnerable households. The 6,250 households that are identified as most vulnerable within each group will be those that WFP will keep assisting over the next cycle, reaching in this way the stated target of 25,000 beneficiary households.

The four prioritization methods cover some of the most common approaches currently adopted by WFP as well as other humanitarian organizations whenever targeting their assistance, which are based on Proxy Mean Testing (PMT) or categorical targeting. More specifically, the four different approaches are the following:
1) PMT targeting consumption-based poverty (T1): beneficiaries are identified through a PMT model, designed to identify households with lower consumption per capita (this is the status-quo targeting approach within WFP Lebanon);
2) PMT targeting food (in)security (T2): beneficiaries are identified through a PMT model, designed to identify households with lower food security;
3) PMT using multidimensional vulnerability index (T3): beneficiaries are identified through a PMT model, designed to identify households with a combination of lower food security and lower consumption per capita (more specifically, the index relies on CARI or Consolidated Approach for Reporting Indicators of Food Security);
4) Categorical selection (T4): beneficiaries are identified following a list of priority dimensions identified through an expert survey administered to local experts and practitioners.

In practice, the PMT models used in the first three approaches were developed and trained using the Lebanon Vulnerability Assessment Panel (LVAP) dataset, which was also collected in 2023. The models are then applied to the census collected in 2023 by WFP. Households will then be randomly allocated to the four targeting approaches (stratifying by eligibility to ensure balance along that dimension) to define four equally sized groups of 14,500 households. Within each group, households will be ranked based on the associated targeting approach and the 6,250 households deemed most in need of assistance within each group (for a total of 25,000) will be selected as the priority households that WFP will keep assisting in the new cycle, while the remaining 33,000 households will instead stop receiving assistance.

The current practice during prioritization is to inform households a month in advance whether they will be included in the new assistance cycle or not. However, a random subset of about 900 households will be selected among those that will stop receiving assistance in the next cycle, to keep receiving assistance for a period of 7 (rather than 1) months after being informed that assistance will end (i.e. they will have an extended warning window).

The study sample will build on this design and include a representative sample of 6,400 households, selected stratifying by targeting arm, eligibility, and regular or extended warning window.

The objectives of the study are threefold:
First, we will compare outcomes across the four targeting arms to learn which method leads to overall better food security, consumption, and psychological well-being.
Second, we will study the impact of fading out food assistance in terms of food security, consumption, and coping strategies, by comparing outcomes between people assigned to receive assistance and people excluded from assistance while controlling for the initial eligibility status of the households (this will give us the estimated impact on the “marginal” households, i.e. eligible under one targeting method but not another).
Third, we will study whether and how the “extended warning window” leads households to take actions aimed at smoothing consumption over time, by taking advantage of the random variation in the number of months for which assistance will continue after notification.
We plan to explore heterogeneity across the following dimensions: starting level of food security and poverty; household size; gender of the household head
Experimental Design Details
Not available
Randomization Method
The randomization will be done in office by a computer using RStudio. The randomization will use a seed for replicability.
Randomization Unit
The population of 56,000 households will be randomized (at the household level) into the 4 different targeting approaches. The study sample of 6,400 households will be randomly drawn from that population, stratifying by targeting arm, eligibility, and regular or extended warning window
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
0
Sample size: planned number of observations
6,400 households
Sample size (or number of clusters) by treatment arms
The study sample will include 1,600 households from each one of the 4 study arms (across the population, there will be 14,000 households allocated to each one of the 4 targeting methods).
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
The proposed sample of 1,600 households classes per study arm (i.e. 6,400 in total) will enable us to detect at the 5% significance level with 80% power a difference across targeting methods on our primary outcome of interest equal to 0.1 standard deviations or larger (assuming 10% attrition). The same design will also enable us to detect at the 5% significance level with 80% power a difference between assistance recipients and non-recipients equal to 0.08 standard deviations or larger, and a difference between long-warning window households and other households equal to 0.14 standard deviations or larger.
IRB

Institutional Review Boards (IRBs)

IRB Name
Humboldt-Universität zu Berlin Ethics Committee
IRB Approval Date
2023-12-07
IRB Approval Number
2023-08-W