Contamination Bias in Linear Regressions

2021/06/09 by Paul Goldsmith-Pinkham, Peter Hull, Michal Kolesár · 1 voice · 8 citations
Mathematics · #Statistical Methods and Inference #Advanced Statistical Methods and Models #Advanced Causal Inference Techniques

paper · pdf · doi:10.1257/aer.20221116

Abstract

We study regressions with multiple treatments and a set of controls that is flexible enough to purge omitted variable bias. We show these regressions generally fail to estimate convex averages of heterogeneous treatment effects—instead, estimates of each treatment’s effect are contaminated by nonconvex averages of the effects of other treatments. We discuss three estimation approaches that avoid such contamination bias, including the targeting of easiest-to-estimate weighted average effects. A reanalysis of nine empirical applications finds economically and statistically meaningful contamination bias in observational studies; contamination bias in experimental studies is more limited due to smaller variability in propensity scores. (JEL C21, C31, C51, H75, I21, I28)

Citations

• Identification and Estimation of Local Average Treatment Effects
• Dealing with limited overlap in estimation of average treatment effects
• Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects
• Recent Developments in the Econometrics of Program Evaluation

Cited by

• What Do We Get from Two-Way Fixed Effects Regressions? Implications from Numerical Equivalence
• Leniency Designs: An Operator's Manual
• Linear Regression in a Nonlinear World
• What climate policies facilitate sectoral mitigation? Global evidence from three decades
• A randomized controlled trial on the effect of administrative burden and information costs on social inequalities in early childcare access in France
• Estimating Treatment Effects Under Bounded Heterogeneity
• A Unified Framework for Spatial and Temporal Treatment Effect Boundaries: Theory and Identification
• Interactions Between Multiple Environmental Markets: Addressing Contamination Bias in Overlapping Policies
• The purpose of an estimator is what it does: Misspecification, estimands, and over-identification

Discussions

• Our paper (with @instrumenthull.bsky.social and Kolesar) shows three things… (ungated here: arxiv.org/abs/2106.05024) [bsky, 21 points, 1 comments]

Related

• Two-way Fixed Effects and Differences-in-Differences Estimators with Several Treatments
• Demystifying and avoiding the OLS "weighting problem": Unmodeled heterogeneity and straightforward solutions
• On the implied weights of linear regression for causal inference
• Exact Bias Correction for Linear Adjustment of Randomized Controlled Trials
• Contamination Estimation via Convex Relaxations