Agricultural Economics Seminar Series – Dr. Tor Tolhurst

About

Dr. Tol Tolhurst, Michigan State Unviersity Department of Agricultural, Food, and Resource Economics, will present "Probability Bounds for Inference with a Single Treated Cluster," during the Agricultural Economics Seminar Series.

Dr. Tor Tolhurst is an applied microeconomist who studies rare, extreme events in food and agricultural systems and how they affect risk, resilience, and policy. He holds a Ph.D. from UC Davis, an M.Sc. from the University of Guelph, and a B.Comm. from the University of British Columbia.

Abstract:

Many quasi-experimental studies have a single treated cluster: a state passes a law, a hurricane strikes a city, or a school district adopts a new curriculum. A single treated cluster biases cluster-robust standard errors severely downward, even with many control clusters and many observations per cluster. I propose a non-asymptotic, distribution-free approach to single cluster inference based on concentration inequalities. The test compares the treated estimate to a reference distribution of placebo estimates. Size control requires the treated and placebo estimates are weakly exchangeable under the null, a condition familiar from permutation and bootstrap tests. Closed-form critical values come from a finite-sample Chebyshev inequality, for which I derive a one-sided version. I apply the test to asset bubbles and consider where it may or may not offer advantages over current methods, particularly under cluster heterogeneity.