General Departmental Seminar Series
Robust Methods to Analyze Longitudinal Data with Informative Missingness
Soomin Park, Graduate Student
Department of Statistics, University of Wisconsin
Friday, Feb 16, 2001, 12:00-1:00 p.m.
3285 Medical Sciences Center - 1300 University Ave.
Missing data are common in longitudinal studies, and the missing pattern is often related to the outcome through random effects representing unmeasured individual characteristics such as health awareness. The recent biostatistical literature contains a number of methods for handling the bias caused by such "informative missingness" which can be monotone or intermittent. Because of intractable integrals, estimation based on the full likelihood is difficult to implement except for in some special cases. We propose an estimation method, which can be easily employed in practice, based on grouping the data according to the missingness pattern. The resulting estimator is consistent under both selection and pattern-mixture models, and its efficiency can be improved using some suggested modifications. Simulation studies conducted to evaluate performance of the proposed method demonstrate its robustness to the missing data model and competitive efficiency for practical sample sizes. Methods to detect informative missingness are also discussed. We apply the proposed procedure to data from the Wisconsin Diabetes Registry Project, a longitudinal study tracking glycemic control.
Back to General Departmental Seminar Series