WITHDRAWN: A Coding Model for Time-Dependent Post-Randomization Covariates in Regression and SEM

• Activity Number: 230
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Thomas Templin
• Companies: Wayne State University
• Keywords: individual causal effects ; post-randomization covariates ; coding individual trajectories

Abstract

In this presentation I describe how to determine a coding scheme for a time varying post-randomization covariate, self-selected treatment dose, which yields an efficacy interpretation. In addition to the casual issues involved, coding a time dependent covariate so that the predicted trajectory reflects the model implied under different levels of compliancy is a challenge. Using the principle that the predicted trajectory from partial compliance should converge to the efficacy trajectory under full compliance, I developed a coding scheme and general equations for linear and piecewise linear treatment effects. This involves coding design variables differently for different patterns of compliance. The coding scheme is described both graphically and algebraically and can be used with linear mixed effects models or with longitudinal structural equation models. The adjusted treatment effect gives the expected change in the intervention outcome for a one unit change in intervention dose. Model assumptions and limitations are discussed.