Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 434 - Robust and Efficient Inferences in Observational Studies and from Nonrandom Samples
Type: Invited
Date/Time: Thursday, August 6, 2020 : 10:00 AM to 11:50 AM
Sponsor: Government Statistics Section
Abstract #309437
Title: Variable Selection in Causal Inference with a Binary Outcome
Author(s): Brandon Koch* and Laura Johnson and Hank Green and Erin Pullen and Karla Wagner
Companies: University of Nevada, Reno and University of Nevada, Reno and Indiana University and Indiana University and University of Nevada, Reno
Keywords: Average treatment effect; Causal inference; Variable selection; Lasso; HIV testing
Abstract:

Estimating the causal effect of an exposure on an outcome using observational data typically involves modeling the outcome as a function of said exposure and covariates, modeling the exposure as a function of covariates, or both. No matter the approach, efficient effect estimation is generally obtained by including only covariates related to the outcome in the model. Traditional variable selection approaches (e.g. lasso) tend to exclude confounding variables weakly associated to the outcome, so using such approaches to select covariates may bias the causal effect estimator. We propose a novel approach for variable selection in causal inference with a binary outcome that uses a modified adaptive group lasso to penalize a sum of outcome and exposure model loss functions. Simulation results reveal our approach is more likely to select confounding variables, less likely to select covariates unrelated to the outcome, and yields a more efficient effect estimation compared to numerous alternative approaches. We illustrate the proposed approach by estimating the effect of perceived risk for HIV on likelihood of being tested for HIV in the future among a racially diverse, at-risk sample.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program