Recovering Marginal Treatment Effects from a Transition Model for Longitudinal Data with Drop Out Using Path Analysis

• Activity Number: 139
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Emin Tahirovic*
• Companies: University of Pennsylvania
• Keywords: direct/indirect effects ; SWEEP estimator ; path analysis ; Doob's decomposition

Abstract

We show how to recover marginal treatment estimates from a transitional incremental model with martingale random effects working assumption by using the generalized SWEEP (G-SWEEP) estimator. Conditioning on intermediate outcomes is often used in settings with generally non-ignorable dropout to facilitate unbiased estimation in such cases. Estimates recovered from such a strategy are conditional on the past outcomes. Combining this with path analysis method that can recover marginal treatment effects from the conditional ones yields a method that effectively adjusts for non-ignorable drop-out and can be used to unbiasedly estimate total effect of the treatment. The application of the G-SWEEP estimator to incremental longitudinal data performs an implicit path analysis and recovers the effect of treatment at baseline which is generally masked by conditioning on intermediate outcomes.