Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 572 - Addressing Complications in Causal Inference
Type: Topic Contributed
Date/Time: Thursday, August 6, 2020 : 3:00 PM to 4:50 PM
Sponsor: Section on Statistics in Epidemiology
Abstract #313676
Title: A Comparison of Singly and Multiply Robust Methods for Causal Inference in Failure Time Analyses
Author(s): Lan Wen* and Miguel Hernan and James Robins
Companies: and Harvard University and Harvard T.H Chan School of Public Health
Keywords: Causal inference; Inverse probability weighting ; Iterative conditional expectation ; Multiply robust estimators; Survival analysis

Estimating the effect of treatment strategies from longitudinal observational data often requires g-methods, such as inverse probability weighting (IPW) or the iterative conditional expectation (ICE) g-formula. These estimators are singly robust in the sense that there is only one opportunity to get valid estimates. Multiply robust estimators that combine IPW and ICE offer more than one opportunity for valid estimation. This is important because some degree of model misspecification is almost always expected in practice. Though several multiply robust estimators exist, they have never been compared to singly robust methods in the context of survival analysis and so, it is unclear whether the increased complexity of these methods is worthwhile. Via simulation studies we show that multiply robust estimators confer more protection against model misspecification than singly robust estimators, and that certain multiply robust estimators offer more protection than others. We also compare these methods in an analysis of a large epidemiological study and provide guidelines for practitioners interested in implementing these methods in estimating the treatment effect in survival analyses.

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

Back to the full JSM 2020 program