Overlap in observational studies with high-dimensional covariates

DOI10.1016/J.JECONOM.2019.10.014zbMATH Open1471.62501arXiv1711.02582OpenAlexW3081149207MaRDI QIDQ2658764FDOQ2658764

Authors: Alexander D'Amour, Peng Ding, Avi Feller, Lihua Lei, Jasjeet Sekhon 

Publication date: 24 March 2021

Published in: Journal of Econometrics (Search for Journal in Brave)

Abstract: Estimating causal effects under exogeneity hinges on two key assumptions: unconfoundedness and overlap. Researchers often argue that unconfoundedness is more plausible when more covariates are included in the analysis. Less discussed is the fact that covariate overlap is more difficult to satisfy in this setting. In this paper, we explore the implications of overlap in observational studies with high-dimensional covariates and formalize curse-of-dimensionality argument, suggesting that these assumptions are stronger than investigators likely realize. Our key innovation is to explore how strict overlap restricts global discrepancies between the covariate distributions in the treated and control populations. Exploiting results from information theory, we derive explicit bounds on the average imbalance in covariate means under strict overlap and show that these bounds become more restrictive as the dimension grows large. We discuss how these implications interact with assumptions and procedures commonly deployed in observational causal inference, including sparsity and trimming.

Full work available at URL: https://arxiv.org/abs/1711.02582

Recommendations

• The role of covariate balance in observational studies
• Causal statistical inference in high dimensions
• Dealing with limited overlap in estimation of average treatment effects
• Robust estimation of causal effects via a high-dimensional covariate balancing propensity score
• Approximate residual balancing: debiased inference of average treatment effects in high dimensions

zbMATH Keywords

overlapcausal inferencecurse of dimensionalityinformation theory

Mathematics Subject Classification ID

Statistical aspects of information-theoretic topics (62B10) Applications of statistics to biology and medical sciences; meta analysis (62P10)

Cites Work

• The prognostic analogue of the propensity score
• Double/debiased machine learning for treatment and structural parameters
• Observational studies.
• Some theory for Fisher's linear discriminant function, `naive Bayes', and some alternatives when there are many more variables than observations
• Dealing with limited overlap in estimation of average treatment effects
• The central role of the propensity score in observational studies for causal effects
• Inference on treatment effects after selection among high-dimensional controls
• Balancing covariates via propensity score weighting
• Approximate residual balancing: debiased inference of average treatment effects in high dimensions
• Efficient Estimation of Average Treatment Effects Using the Estimated Propensity Score
• Semiparametric efficiency in GMM models with auxiliary data
• Robust inference on average treatment effects with possibly more covariates than observations
• Irregular identification, support conditions, and inverse weight estimation
• On Divergences and Informations in Statistics and Information Theory
• Title not available (Why is that?)
• Title not available (Why is that?)
• Subsampling inference for the mean in the heavy-tailed case.
• Estimation and inference with weak, semi-strong, and strong identification
• Learning with Finite Memory
• Robust confidence intervals for average treatment effects under limited overlap
• Maximum likelihood estimation and uniform inference with sporadic identification failure
• On estimating regression-based causal effects using sufficient dimension reduction
• Title not available (Why is that?)
• Lower Bound on the Error Probability for Families with Bounded Likelihood Ratios
• Asymptotic inference of causal effects with observational studies trimmed by the estimated propensity scores
• Instrumental variables as bias amplifiers with general outcome and confounding
• Information-type divergence when the likelihood ratios are bounded

Cited In (17)

• Estimating population average causal effects in the presence of non-overlap: the effect of natural gas compressor station exposure on cancer mortality
• The costs and benefits of uniformly valid causal inference with high-dimensional nuisance parameters
• Prior and posterior checking of implicit causal assumptions
• The role of covariate balance in observational studies
• Subgroup analysis and adaptive experiments crave for debiasing
• Benefits and costs of matching prior to a difference in differences analysis when parallel trends does not hold
• Nonparametric identification and estimation of heterogeneous causal effects under conditional independence
• Causal inference of general treatment effects using neural networks with a diverging number of confounders
• Towards optimal doubly robust estimation of heterogeneous causal effects
• Identifying surrogate markers in real-world comparative effectiveness research
• Addressing positivity violations in causal effect estimation using Gaussian process priors
• Semiparametric estimation of long-term treatment effects
• Mahalanobis balancing: a multivariate perspective on approximate covariate balancing
• In Nonparametric and High-Dimensional Models, Bayesian Ignorability is an Informative Prior
• The designed bootstrap for causal inference in big observational data
• More efficient policy learning via optimal retargeting
• Augmented minimax linear estimation

This page was built for publication: Overlap in observational studies with high-dimensional covariates

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2658764)