Robust Outcome-Dependent Sampling for Continuous- and Counted-Response Longitudinal Data

• Activity Number: 394
• Type: Invited
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Paul J. Rathouz*+ and Lee McDaniel and Jonathan S. Schildcrout
• Companies: University of Wisconsin School of Medicine and Public Health and University of Wisconsin-Madison and Vanderbilt University
• Keywords: Bias ; case-control ; missing data ; two-phase sampling

Abstract

In outcome dependent sampling (ODS) designs for longitudinal data, the subjects and/or the observations are sampled as a stochastic function of the longitudinal vector of responses. The sampling may for example be a variant on case-control sampling for extreme subjects, or may alternatively sample individual observations from subjects as a function of a surrogate process. ODS results in a type of missingness-by-design, and important questions of optimal design and robust analysis ensue. Methods that condition on "being sampled" can be highly efficient, but rely on distributional assumptions, either at the observation or at the subject level, to carry out the conditioning, thus begging the question of robustness. Often non-ODS methods will be robust to such distributional assumptions, but this may not be the case when accounting for ODS. In this talk, we will explore the use of various approaches to the analysis of continuous or count data under outcome dependent sampling designs, with a particular focus on flexible methods that are robust to distributional assumptions.