Data and code for Heterogeneity in Effects of Poverty-alleviating Interventions
Domain:
socioeconomicRecord type:
datasetpaper
Creator:
GarWalOrk
Publisher:
ICP
Host:
In this paper, we explore heterogeneity in treatment effects with respect to pre-treatment recipient and household characteristics for each of our proposed interventions. Taken together, our results indicate that the treatment effects of the workshop, cash intervention and their combination vary little with respect to both pre-specified measures of heterogeneity and a broader class of pre-treatment variables. Once we adjust for multiple hypothesis testing, there is little-to-no evidence for heterogeneity. This is true under both our pre-specified median split strategy and modern machine-learning-based heterogeneity analyses.
The paper revisits a 415-village cluster-randomized field experiment involving 8,300 women living in poverty in rural Kenya. The trial had four treatment arms. The first was an unconditional cash transfer of 2,237 USD PPP. Our implementing partner provided transfers to 2,085 households, with women selected by households to receive the transfer in 91% of cases. It was delivered in three monthly tranches of 203, 1,017, and 1,017 USD PPP via the mobile money payments system M-Pesa. The target population was made up of poor households identified by a proxy-means test.
Presence of Common Scales: See Orkin et al. (2024. We use a village census to collect village-level variables for stratifying the village-levelrandomisation. We then use a household census to determine study eligibility and draw the baseline sample. Of the 41,322 households censused, 32,964 participated in the study, 4,677 refused to participate in research, and 3,681 were not at home. Of those who chose to participate, the group relevant for our study is those who meet our implementing partner's proxy means test for living in poverty. Households are classified as poor if they met any one of four simple criteria: (1) they had per capita housing space less than 62,000 cm squared; (2) they had a mud floor and no mobile telephone; (3) they had a mud floor and the household head was a widow; or (4) the household included an orphan.
Our study sample excludes some means-tested-poor households: those which do not contain an adult female (4.5%) and those which are polygamous (11%), due to difficulties associated with household definition. These households were still eligible for GiveDirectly transfers. From the remaining households, we sample roughly 20 households per village. If one of the 20 target households could not be found or refused to participate, the field officers included one household on the reserve list as a replacement. Sampling probabilities vary by village because we don’t sample exactly the same share of treatment-eligible households in each village.
Further details about the data are provided in Orkin, Kate, Robert Garlick, MahreenMahmud, Richard Sedlmayr, Johannes Haushofer, and Stefan Dercon. 2024. “Aspiringto a Better Future: Can a Simple Psychological Intervention Reduce Poverty?” National Bureau of Economic Research. Working Paper 31375. Smallest Geographic Unit: Village
The paper revisits a 415-village cluster-randomized field experiment involving 8,300 women living in poverty in rural Kenya. The trial had four treatment arms. The first was an unconditional cash transfer of 2,237 USD PPP. Our implementing partner provided transfers to 2,085 households, with women selected by households to receive the transfer in 91% of cases. It was delivered in three monthly tranches of 203, 1,017, and 1,017 USD PPP via the mobile money payments system M-Pesa. The target population was made up of poor households identified by a proxy-means test.
Presence of Common Scales: See Orkin et al. (2024. We use a village census to collect village-level variables for stratifying the village-levelrandomisation. We then use a household census to determine study eligibility and draw the baseline sample. Of the 41,322 households censused, 32,964 participated in the study, 4,677 refused to participate in research, and 3,681 were not at home. Of those who chose to participate, the group relevant for our study is those who meet our implementing partner's proxy means test for living in poverty. Households are classified as poor if they met any one of four simple criteria: (1) they had per capita housing space less than 62,000 cm squared; (2) they had a mud floor and no mobile telephone; (3) they had a mud floor and the household head was a widow; or (4) the household included an orphan.
Our study sample excludes some means-tested-poor households: those which do not contain an adult female (4.5%) and those which are polygamous (11%), due to difficulties associated with household definition. These households were still eligible for GiveDirectly transfers. From the remaining households, we sample roughly 20 households per village. If one of the 20 target households could not be found or refused to participate, the field officers included one household on the reserve list as a replacement. Sampling probabilities vary by village because we don’t sample exactly the same share of treatment-eligible households in each village.
Further details about the data are provided in Orkin, Kate, Robert Garlick, MahreenMahmud, Richard Sedlmayr, Johannes Haushofer, and Stefan Dercon. 2024. “Aspiringto a Better Future: Can a Simple Psychological Intervention Reduce Poverty?” National Bureau of Economic Research. Working Paper 31375. Smallest Geographic Unit: Village
Similar
Data and Code for: Testing for Urban Heterogeneity in Separability in TanzaniaData and Code for: A New Perspective on Spatial Heterogeneity in African DevelopmentPolicy, Ethics and Future Directions in Alleviating Poverty from SpaceAlleviating Poverty through Agricultural Projects: A Systematic Literature ReviewPrompting Towards Alleviating Code-Switched Data Scarcity in Under-Resourced Languages with GPT as a PivotData and code for study "Economywide effects of climate-smart agriculture in Ethiopia"