Publications

Working Papers

Work in Progress

Micro-enterprise saturation: Critical mass or overcrowding?

Draft coming soon

Abstract: Should entrepreneurship support programs to extremely poor people be scaled by saturating a few villages, or maintaining low saturation and enrolling more villages? On the one hand, saturation may create a critical mass of entrepreneurs whose income generation and associated spending multipliers make each other viable. On the other hand, encouraging the start-up of many small firms in close proximity may be a recipe for overcrowding and artificially fierce competition in a small village economy. In a large randomized control trial in Malawi, I exogenously vary the saturation of an entrepreneurship support program to assess whether business performance increases or decreases in the share of treated neighbors. I also test a simple intervention to mediate overcrowding by encouraging entrepreneurs-to-be to coordinate which sectors they enter. One year after the start of the intervention, the number of successful businesses scales linearly with the number of supported entrepreneurs to be, but the overall earnings do not appear to grow as saturation increases. Instead, aggregate earnings are split among the larger number of supported entrepreneurs and I do not detect lower consumer prices. The coordination intervention is not affecting business choice.

The role of shocks in poverty

Draft coming soon

Abstract: In any narrative of persistent, extreme poverty, shocks loom large. Households living below the poverty line face an incessant onslaught of shocks to their health and livelihoods, their housing and assets, livestock and crops, infrastructure, or the social networks they are embedded in – each difficult to predict or prevent, each capable of wiping out the household’s progress toward lifting itself out of poverty. Our ability to formally model and study the role of shocks for turning poverty into a trap is inhibited by how challenging they are to recall and measure well in baseline-intervention-endline RCT frameworks. At the same time, Ultra-Poor Graduation programs (UPG) that support the diversification out of subsistence agriculture and into micro-entrepreneurship have emerged as the closest that development economics has come to a silver bullet for sustained ultra-poverty alleviation. The mechanisms behind their success are not well understood. I leverage the roll-out of a large UPG in Malawi to study the role of shocks in poverty and its alleviation. I conduct an RCT that generates unique high-frequency data of the shocks that households face. Random program assignment traces out the impact of (alleviated) poverty on the number, severity, and composition of shocks. An event-study analysis traces households' reactions to shocks. A structural model explores their interaction.

Seasons of change: Experimental evidence on the optimal timing for poverty alleviation

Data collection ongoing

Abstract: In a large randomized control trial, I allocate an identical, bundled poverty alleviation program to three cohorts to receive either around the agricultural planting, harvest, or lean season. This study will illuminate the trade-offs between humanitarian pressures to cover basic needs when resources are low, and opportunities for transformational big-push livelihood advances in times of plenty.

The economic impacts of the novel Malaria vaccine

Data-sharing agreement signed

Abstract: Malaria remains one of the most common causes of death for children under five, imposing severe health and productivity tolls on survivors. We combine data from a recent large-scale medical trial of the novel malaria vaccine in Malawi with the country's administrative and census data infrastructure. We assess the impacts of childhood malaria immunization on households' economic wellbeing in the short run. Further, we lay the groundwork for tracking the children in the treatment and control group of this trial through their childhood, adolescence, and adulthood.

Research Assistance

Public and Private Transit: Evidence from Lagos

Working paper

Included and excluded instruments in structural estimation

Abstract: We consider the choice of instrumental variables when a researcher’s structural model may be misspecified. We contrast included instruments, which have a direct causal effect on the outcome holding constant the endogenous variable of interest, with excluded instruments, which do not. We show conditions under which the researcher’s estimand maintains an interpretation in terms of causal effects of the endogenous variable under excluded instruments but not under included instruments. We apply our framework to estimation of a linear instrumental variables model, and of differentiated goods demand models under price endogeneity. We show that the distinction between included and excluded instruments is quantitatively important in simulations based on an application. We extend our results to a dynamic setting by studying estimation of production function parameters under input endogeneity.

Building Resilient Health Systems: Experimental Evidence from Sierra Leone and the 2014 Ebola Outbreak (2021, QJE)

Developing countries are characterized by high rates of mortality and morbidity. A potential contributing factor is the low utilization of health systems, stemming from the low perceived quality of care delivered by health personnel. This factor may be especially critical during crises, when individuals choose whether to cooperate with response efforts and frontline health personnel. We experimentally examine efforts aimed at improving health worker performance in the context of the 2014–15 West African Ebola crisis. Roughly two years before the outbreak in Sierra Leone, we randomly assigned two accountability interventions to government-run health clinics – one focused on community monitoring and the other gave status awards to clinic staff. We find that over the medium run, prior to the Ebola crisis, both interventions led to improvements in utilization of clinics and patient satisfaction with the health system. In addition, child health outcomes improved substantially in the catchment areas of community monitoring clinics. During the crisis, the interventions also led to higher reported Ebola cases, as well as lower mortality from Ebola – particularly in areas with community monitoring clinics.  We explore three potential mechanisms: the interventions (1) increased the likelihood that patients reported Ebola symptoms and sought care; (2) unintentionally increased Ebola incidence; or (3) improved surveillance efforts.  We find evidence consistent with the first: by building trust and confidence in health workers, and improving the perceived quality of care provided by clinics prior to the outbreak, the interventions encouraged patients to report and receive treatment. Our results suggest that accountability interventions not only have the power to improve health systems during normal times, but can additionally make health systems resilient to crises that may emerge over the longer run.

Main paper, Companion Paper (AEA P&P), Non-technical summary

Media Coverage: New York Times, IPA, CBS

Put to rest

Robust results in the presence of treatment interactions: The case of unconditional cash transfers

Abstract: Recent contributions to the analysis of randomized experiments have shown that cross-randomizing several treatments can lead to small cell sizes, which can result in distorted treatment effect estimates. In addition, it has been suggested that failure to include interaction terms in regressions may also bias results. In light of these concerns, we re-evaluate a large study of unconditional cash transfers in Kenya with three cross-randomizations (recipient gender, transfer magnitude and timing). To circumvent the failure of asymptotic results in small cells, we first re-estimate the treatment effects of the study using randomization inference,bootstrapping, and a jackknife method. The results are very similar to those obtained with conventional asymptotic methods. Second, we show that in the class of experiments which assign multiple versions of the same treatment, rather than several independent treatments, inclusion of interaction terms does not yield “pure” treatment effects, but rather treatment effects for particular subgroups,which are not generally estimands of interest. Finally, we quantify the possible bias introduced by slight imbalances in cell sizes generated by randomization, and suggest a simple re-weighting to correct for it.