Treatment Choice in Heterogeneous Populations Using Experiments without Covariate Data (Invited Paper)

TL;DR

This paper explores optimal treatment assignment in heterogeneous populations when covariate data for experimental subjects is unavailable, proposing methods to improve decision-making based on population-level information.

Contribution

It introduces a nonparametric approach to treatment choice that accounts for population heterogeneity without requiring individual covariate data during experiments.

Findings

Provides a framework for treatment assignment without covariate data

Shows how to approximate optimal rules using population-level information

Addresses practical limitations in experimental data collection

Abstract

I examine the problem of treatment choice when a planner observes (i) covariates that describe each member of a population of interest and (ii) the outcomes of an experiment in which subjects randomly drawn from this population are randomly assigned to treatment groups within which all subjects receive the same treatment. Covariate data for the subjects of the experiment are not available. The optimal treatment rule is to divide the population into subpopulations whose members share the same covariate value, and then to choose for each subpopulation a treatment that maximizes its mean outcome. However the planner cannot implement this rule. I draw on my work on nonparametric analysis of treatment response to address the planner's problem.