An Empirical Study of Marginal Structural Models for Time-Independent Treatment

Tanya A. Henneman, Department of Biostatistics, School of Public Health, University of California, Los Angeles
Mark J. van der Laan, Division of Biostatistics, School of Public Health, University of California, Berkeley

Abstract

In non-randomized treatment studies a significant problem for statisticians is determining how best to adjust for confounders. Marginal structural models (MSMs) and inverse probability of treatment weighted (IPTW) estimators are useful in analyzing the causal effect of treatment in observational studies. Given an IPTW estimator a doubly robust augmented IPTW (AIPTW) estimator orthogonalizes it resulting in a more e±cient estimator than the IPTW estimator. One purpose of this paper is to make a practical comparison between the IPTW estimator and the doubly robust AIPTW estimator via a series of Monte- Carlo simulations. We also consider the selection of the optimal choice in the class of IPTW estimators. Also included is an empirical study that examines the range of efficiency across the class of AIPTW estimators. Both the IPTW and AIPTW estimators are applied in the analysis of data collected from The Study of Physical Performance and Age-Related Changes in Sonomans.