Innovations in Causal Inference: Quantile Mixtures, Empirical Bayes Aggregation, and Synthetic Controls

Session Submission Type: Panel

Abstract

This session brings together three studies concerned with precision, inference, and interpretability in policy evaluation studies. The first paper, by Luis Alvarez (University of São Paulo), develops sampling theory for nonparametric quantile mixture models, demonstrates the applicability of the theory to statistical inference in synthetic control studies, and applies the method to evaluate the effects of the Brumadinho Dam failure (an industrial-environmental disaster in Brazil) on local wage distributions. The second paper, by Coady Wing (Indiana University Bloomington), Alex Hollingsworth (Indiana University Bloomington), and Jacob Goldin (University of Chicago), introduces empirical Bayes shrinkage estimators designed to aggregate group-time ATT parameters from staggered adoption difference-in-differences designs, balancing treatment effect heterogeneity with sampling error. The third paper, authored by Felipe Lozano-Rojas, Amanda Abraham, and David Bradford (all University of Georgia), and Sumedha Gupta (Indiana University Indianapolis), uses synthetic control methods to estimate the effects of medical cannabis laws on opioid and non-opioid pain medication use across 22 U.S. states. They find reductions in opioid prescribing and evidence of substitution toward NSAIDs, with important heterogeneity across patient demographics and health conditions.

Together, these papers showcase innovative estimation strategies—combining nonparametric sieves, Bayesian shrinkage, and synthetic counterfactuals—to improve both the statistical power and substantive insights of causal impact analyses. The discussion will focus on practical implementation, inferential trade-offs, and guidance for selecting the most appropriate technique given specific research designs.

Policy Area

• Methods and Tools of Analysis

Chair

• Jacob Goldin, University of Chicago

Organizer

• Jennifer A Mangano, Indiana University

Individual Presentations

• The Effect of Medical Cannabis Laws on the Use of Pain Medications Among Commercially Insured Patients - Presenting Author: FELIPE ANDRES LOZANO-ROJAS, University of Georgia

• Precision and Heterogeneity in Staggered Adoption Designs: An Empirical Bayes Approach - Presenting Author: Coady Wing, Indiana University

• Quantile Mixture Models: Estimation and Inference - Presenting Author: Luis A. F. Alvarez, University of São Paulo

• Direct And Indirect Treatment Effects With Time-Varying Covariates - Presenting Author: Kyle Butts, University of Arkansas, Fayetteville