Online Program Home
My Program

Abstract Details

Activity Number: 438 - Missing Data Issues in Public Health Studies and Survey Sampling in the Era of Data Science
Type: Invited
Date/Time: Wednesday, July 31, 2019 : 8:30 AM to 10:20 AM
Sponsor: Section on Statistics in Epidemiology
Abstract #300052 Presentation
Title: Robust 'Squared' Estimators to Account for Selection Bias Due to Death in Estimating the Effect of Wealth Shock on Cognition for the Health Retirement Study
Author(s): Yaoyuan Vincent Tan* and Michael Elliott and Carol A.C. Flannagan and Lindsay Pool
Companies: Rutgers University and University of Michigan and University of Michigan, Transport Research Institute and Northwestern University
Keywords: Bayesian additive regression trees; Causal Inference; Censoring by death; The Health and Retirement Study; Marginal Structural Models; Penalized splines of propensity methods in treatment comparisons

The Health and Retirement Study is a longitudinal study of US adults age 50 and older, surveyed biennially since 1992 with detailed modules on financial status and health. We investigate the effect of a negative wealth shock (sudden large decline in wealth) on the cognitive score of subjects. Our analysis is complicated by lack of randomization, confounding by indication, and the fact that a substantial fraction of the sample will die during follow-up, leading to some of our outcomes being differentially censored. Common methods used to handle these problems, including marginal structural models, may not be appropriate because they upweight subjects who are more likely to die, thus making inference about an artificial “immortal” population. We propose comparing the treatment effect among subjects who would survive under both sets of treatment regimes being considered. We view this as a large missing data problem and impute the survival status and outcomes of the counterfactual. We consider a modified version of PENCOMP, using BART to improve the robustness of our imputation. We found that our proposed method worked well in various simulation scenarios and our data analysis.

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

Back to the full JSM 2019 program