Events and Seminars : Upcoming Seminars

OUTCOME DEPENDENT SAMPLING FOR LONGITUDINAL DATA

PAUL RATHOUZ, PH.D.
Professor and Chair, Department of Biostatistics & Medical Informatics
University of Wisconsin, School of Medicine and Public Health
THURSDAY, NOVEMBER 19, 2015
3:00 p.m.–4:00 p.m., CRB 692

Epidemiological study design is ubiquitous in public health research because it concentrates sampling on the most informative subjects, thereby, improving estimation efficiency. However, these cost and resource efficient designs have not permeated into longitudinal and correlated data settings due to the lack of methods for guiding sampling and for analyzing the observed, enriched sample. I will discuss recent developments in efficient epidemiological study designs for longitudinal data. These include designs that sample directly on longitudinal response summaries (outcome dependent sampling; ODS) and those that sample based on an auxiliary variable that is related but is not equal to the longitudinal outcome (outcome enriched sampling; OES). Because longitudinal data yield within-and between-subject variation in exposures and responses, we will discuss both designs that sample informative subjects and those that sample informative times within subjects. I will try to emphasize study design, but will also consider methods of analysis. Due to differences in the operating properties of associated estimators, I will mention both likelihood-based and semi-parametric estimation procedures.