Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 191 - Causal Inference
Type: Contributed
Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
Sponsor: Biometrics Section
Abstract #314063
Title: A More Powerful Test of the Composite Null Hypothesis Arising in Mediation Analysis
Author(s): Caleb Miles* and Antoine Chambaz
Companies: Columbia University and Universite Paris Descartes
Keywords: Causal inference; Mediation analysis; Composite null hypothesis; Power

The most common estimator for an indirect effect in mediation analysis is what is known as the product method estimator. This is the product of regression coefficients from two models: the mediator coefficient in the outcome model and the exposure coefficient in the mediator model. While the product method has been shown to estimate the natural indirect effect under linear models, many other indirect effect estimators amount to a product of two independent, asymptotically normal variables. Thus, the null hypothesis of no indirect effect is a composite null hypothesis, as the null holds if either regression coefficient is zero. A consequence is that existing hypothesis tests are either severely underpowered near the origin (i.e., both coefficients being small) or invalid. We propose a numerical approximation-based hypothesis test that (i) is guaranteed to preserve level alpha type 1 error, (ii) meaningfully improves power when both true underlying effects are small relative to sample size, and (iii) preserves power when at least one is not.

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

Back to the full JSM 2020 program