Pricing People into the Market: Combining Field Experiments and ML to Target Aid
Friday, April 17, 2026 · 11:30 AM – 12:00 PM · Auditorium 3
In many contexts, the objective is not to minimize a statistical loss function, but to improve social welfare. In Burkina Faso, we combined field experiments and machine learning to estimate demand for household sanitation products. This required predicting how households with different observable characteristics would respond to different levels of financial assistance. Using those estimates, we then deployed a “smart subsidy” pricing structure that sought to maximize the number of households who switched from manually emptying their latrines to a safer technology, and evaluate the market design using a randomized controlled trial.
The treatment led to an increase in the market share of a socially beneficial good of 12.7 percentage points in the group likely to purchase the inferior product, 5.3 percentage points overall. We estimate and use a model of household behavior to compare alternative market designs, information structures, and market sustainability. We use counterfactual simulation techniques to compare our market design to others, like posted prices or auctions, and show how targeted pricing improves on common market structures.
This work combines causal inference, market design, and machine learning to show that we can design better systems that stretch limited subsidy resources further, and make people happier, healthier, and more productive.
About the Speaker
Terence Johnson
UVA-SDS, Assistant Professor of Data Science
Terence Johnson is an Assistant Professor of Data Science at the University of Virginia. He is an intervention scientist working on problems at the intersection of economics and data science, particularly using machine learning to design markets and evaluate the impact of interventions. He is the analytics curriculum designer for the School of Data Science, where he teaches probability and machine learning. His recent work focuses on increasing agricultural productivity in Lesotho using generative AI, the design of just-in-time auctions in Senegal for sanitation services, targeting aid to poor households with machine learning in Burkina Faso, and an audit of the Virginia Pretrial Risk Assessment Instrument tool. His scholarly work is forthcoming or has appeared in the Review of Economic Statistics, Journal of Development Economics, American Economic Journal: Applied, Journal of Economic Theory, Economic Theory, and World Bank Economic Review. Johnson received his doctorate in economics from the University of Maryland, and holds a B.S. in economics, mathematics, and political science from Syracuse University.