Online Program Home
My Program

Abstract Details

Activity Number: 491 - Causal Inference Within Reach: Pragmatic Approaches to Model Construction and Validation
Type: Invited
Date/Time: Wednesday, August 1, 2018 : 10:30 AM to 12:20 PM
Sponsor: Health Policy Statistics Section
Abstract #326910 Presentation
Title: Model Assessment in Causal Inference Using Explicit Validators
Author(s): Booil Jo*
Companies: Stanford University
Keywords: sensitivity analysis; explicit validator; principal causal effect; posterior class probability

In addition to the difficulties in defining causal models and necessary identifying assumptions, complexities and subjectivity in sensitivity analysis and model validation have greatly reduced the practicality and accessibility of causal inference approaches. In this study, we explore possibilities of accelerating model validation using a prediction framework. We will use simple examples in identifying local average treatment effect (principal causal effects) in the presence of treatment or program noncompliance. Specifically, we will focus on the use of posterior latent compliance class probability, an individual level estimate that reflects the causal model under consideration. The model validity will be assessed by how well the posterior class probability is predicted by baseline variables that are supposed to be its predictors (antecedent validators), how strongly it is associated with other intermediate outcomes (concurrent validators) and how well it predicts future outcomes it is expected to predict (predictive validators). In each validation context, we will monitor both prediction/classification accuracy and its variability.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program