Exploring the Finite-Sample Properties of Inverse Probability Weighted and G Estimation of a Structural Nested Failure Time Model Under Positivity Violations

• Activity Number: 416
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Ashley Isaac Naimi*+ and Stephen R. Cole and Erica E. M. Moodie and Jay Kaufman
• Companies: McGill University and The University of North Carolina at Chapel Hill and McGill University and McGill University
• Keywords: G-estimation ; Structural Nested Model ; Causal Inference ; Positivity ; Finite-Sample Properties

Abstract

When seeking to estimate the effect of a time-dependent exposure on a failure time outcome, time-dependent confounding may be a concern. The inverse probability weighted (IPW) estimator is consistent for the parameter of a structural failure time model under conditions of time-dependent confounding when positivity holds. This assumption requires that, conditional on relevant confounders, the probability of being exposed is bounded away from zero and one. The semiparametric G-estimator is consistent for the parameter of a structural nested failure time model, even under nonpositivity. In this talk, we explore the finite sample properties of the G-estimation of a structural nested failure time model and compare them to IPW estimators of an equivalent model under different degrees of nonpositivity.