Research
Abstract:
This paper explores the extent to which pollution externalities and market-based effects from industrial mining affect local agricultural output in Sub-Saharan Africa. I combine mine geolocations, topographical data and satellite-based measures of pollution, yields and weather, to identify areas around mines that are disproportionately exposed to pollution but not market-specific effects. Leveraging the staggered openings of mines across Sub-Saharan Africa, I find that air and water pollution externalities from mines account for almost 50% of the overall reduction in yields caused by industrial mining. Finally, I use both standard heterogeneity analysis as well as machine learning methods to document that pollution externalities are larger for mines in countries with poor governance and regulatory environments, as well as for mines located in areas with low initial levels of pollution.
Rainfall index insurance can enable farm households to manage production risk, but demand in developing countries remains low at market prices, in part because the insurance trigger may not correlate well with individual farm losses. Area–yield crop insurance, which links payouts to average yield in a geographic zone, attempts to increase demand by more accurately targeting insurance payouts to production shortfalls. However, shifting from an exogenous weather-based to an endogenous yield-based index introduces concerns of asymmetric information, which can lead to market failures that constrain supply from providers. These features are inversely related: larger insurance zones inhibit index manipulation, but average yield is less informative about any individual plot. We quantify this tradeoff for maize in Ghana using a spatial yield model calibrated to match observed production. Insurers must demarcate zones of no more than 5,000 farmers for area–yield insurance to outperform weather insurance. The framework presented in this paper allows assessment of the relationship between index performance and asymmetric information in new crop insurance products.
Recent advances in earth observation and machine learning have opened new frontiers in impact evaluation that appear well-suited for agricultural settings. We apply these promising methods in the context of Ethiopia’s Direct Seed Marketing (DSM) program, which rolled out after 2011 and aims to enhance farmer access to improved seed varieties. Our satellite-based impact assessment focuses on maize productivity as a summary outcome. Satellite-based yield predictions enable a high-resolution, landscape-level analysis of DSM impacts using a difference-in-difference identification strategy, but yield prediction errors introduce new sources of potential bias in subsequent causal inference. We test for this prediction error bias and compare our DSM impact estimates to those that use farmer-reported and crop cut yield measures. We find evidence of small positive but insignificant effects of the DSM on maize yield, explore how errors in predicted yields introduce bias in causal estimation, and discuss implications for the selection of prediction models.
Cellphone data has proven successful in predicting socioeconomic status. However, data limitations have hindered the study of the capacity of these models to predict welfare over time. By relying on a panel sample with information collected two years apart, we test for decay in the ability of an algorithm to predict wealth levels. We find evidence of model decay, with the predictive capacity of a model trained on the first panel wave being 15-30% lower than a model trained on the second wave survey and contemporary cellphone data. We link the lower performance to re-ranking of households across the wealth distribution, changes in the distribution of cellphone features over time and the rise of internet-based communication apps. Finally, we explain how the COVID-19 pandemic serves as a mechanism through which these effects could occur
This study examines coverage, costs, cost-effectiveness, and cost burdens of a hybrid vitamin A supplementation (VAS) event in Burkina Faso. Data were collected from randomly chosen populations within two health districts. Post-event coverage surveys measured impact; spatially scaled primary data provided estimates of costs. Costs of caregiver participation were measured. We include data provided by and on all the national, regional, district, and local actors involved in the design and implementation of the VAS event.Campaign-based Vitamin A supplementation (VAS) programs persist but are expensive and reliant on international assistance, and hence are unsustainable. The study revealed differences in coverage, costs, and cost-effectiveness within and across districts, signaling potential efficiency gains from tailored approaches in Burkina Faso.
Link to AEA RCT Registry pre-analysis plan
Link to AEA RCT Registry pre-analysis plan