The Role of Covariates and Secondary Outcomes in Causal Studies with Intermediate Variables

• Activity Number: 397
• Type: Invited
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Fabrizia Mealli*+
• Companies: University of Florence
• Keywords: Mutiple Outcomes ; Covariates ; Principal Stratification ; Principal Causal Effects

Abstract

Confounded post-treatment variables are often present in intervention studies. Principal Stratification (PS) is a framework to deal with such intermediate variables. Due to the latent nature of the principal strata, strong structural assumptions are often invoked to sharpen inference. Distributional assumptions may also be invoked, usually leading to weakly identified models. Information on multiple outcomes is routinely collected in practice, but rarely used to improve inference. Covariates are also collected, but often used to either make the assumptions more plausible, or improve the prediction of missing potential outcomes. We show, using various inferential paradigms, including Bayesian and frequentist perspectives, the potential inferential gains from jointly modeling two (or more) outcomes, or jointly modeling outcomes and covariates. These results can also be used to assess the plausibility of structural assumptions, such as exclusion restrictions. The role of the auxiliary information is shown in two examples to evaluate the effects of a real social job training program on participants' employment and of another real job training program on trainees' depression.