Projects
Recent working papers
Remitting against poverty: Eight-year evidence on digital remittances and rural poverty (with Jean Lee, Mahreen Mahmud, Saravana Ravindran, Abu Shonchoy, and Shashank Sreedharan). Upward mobility for poor, rural workers often involves migrating to cities. To enable safer and more efficient remittance-sending, a randomized intervention introduced mobile banking to 815 “ultra-poor” urban migrant-household pairs in rural northwest Bangladesh. One year later, the intervention, which cost $PPP 32 on average, had increased urban-to-rural remittances, decreased rural poverty, and improved conditions in the lean season. Eight years later in 2023, treated households had 33 percent more productive assets by value. The treatment and control groups were using mobile money to comparable extents by then, and treatment effects on rural household consumption, income, poverty, and financial outcomes are no longer detectable.
The Design and Impact of Cash Transfers: Experimental Evidence from Compton, California (with Sidhya Balakrishnan, Sewin Chan, Sara Constantino, and Johannes Haushofer). Randomly chosen low-income households in Compton, CA received unconditional cash transfers averaging roughly $500 monthly. Half received transfers twice monthly, half quarterly. Eighteen months later, twice-monthly transfers improved food security relative to quarterly transfers, but had no other differential effects on pre-specified main outcomes. Averaging across frequencies, monthly income (excluding transfers) was lower than controls by $333, and expenditures (excluding major durables) by $302, without changes in other primary outcomes, including overall labor supply. In line with this, we find suggestive evidence that households paid down debt and purchased durables. Transfers also affected part-time work, housing security, and violence.
Poverty at Higher Frequency (with Joshua Merfeld). Experiences of poverty often shift during the year. Rural households’ struggles intensify in lean seasons, for example, and urban communities regularly face short-term economic downturns. Official poverty statistics based on headcounts, however, are designed to eliminate evidence of variation during the year. We introduce the timecount as a complement to the standard headcount, providing a simple way to capture within-year instability. The timecount measures the average fraction of the year that households experience poverty. Weighted versions capture changes in the intensity of deprivation. We show, unexpectedly, that timecounts are simpler to approximate than headcounts in low- and middle-income countries when data are collected following expert guidelines. Moreover, without intention or recognition, timecounts have replaced headcounts as the de facto poverty rates in many countries. As a result, global poverty statistics today incorporate measures of poverty at higher frequency, registering seasonal poverty and the short-term deprivations of “nonpoor” households. In monthly longitudinal data from rural India, we show that the timecount is 28% larger than the headcount, captures a wider range of experiences of poverty, and is a better predictor of child health outcomes. We describe consequences for analysis, policy, and ethics.
World Development: What will it mean to end poverty?
Ideas for India: India’s Poverty Rate Does Not Measure What You Think It Does
Selecting Experimental Sites for External Validity (with Michael Gechter, Keisuke Hirano, Jean Lee, Mahreen Mahmud, Orville Mondal, Saravana Ravindran, and Abu Shonchoy). Policy decisions often depend on evidence generated elsewhere. We take a decision theoretic approach to choosing where to experiment to optimize external validity. We frame external validity through a policy lens, taking a Bayesian approach and developing a prior specification for the joint distribution of site-level treatment effects using a microeconometric structural model and allowing for other sources of heterogeneity. With data from South Asia, we show that, relative to basing policies on experiments in optimal sites, large efficiency losses result from instead using evidence from randomly-selected sites or, conversely, from sites with the largest expected treatment effects.