Estimating Heterogeneous Effect on Clustered Data Using Mixed-Effects Model

Junlong Zhou (New York University)

Abstract: Estimating the heterogeneous treatment effect is essential to assess the generality and mechanism of randomized experiments. In this paper, we propose an extension of the regression forest combining mixed-effects to analyze heterogeneous treatment effect using aggregated data sets. We show that including mixed-effects can improve estimation by accounting for the cluster-level heterogeneity, and help assess the population-level conditional average treatment effect. We demonstrate its performance using a Monte Carlo simulation and real-world application.

View Poster in a New Tab