Department of Statistics, The Ohio State University
Statistics and Biostatistics Colloquium Series

Data analysts often encounter missing data in practice, and when data are not missing at random in the sense of Rubin (1976), standard analyses typically lead to biased estimators. The problem of missing data is especially common in longitudinal studies, as subjects are frequently lost to follow-up or miss regularly scheduled visits. When analyzing longitudinal data, we often model unmeasured individual characteristics using random effects. It is sometimes plausible that these individual characteristics influence the missingness probabilities. On the one hand, such random-coefficient-based missingness--also referred to as informative missingness--is an instance of missing not at random and estimators from mixed effects models ignoring the missing mechanism are biased. On the other hand, this type of missing data mechanism provides special opportunities for consistent estimation because, intuitively, some information on the random effects can be obtained from the observed outcomes. In this talk, I will review various methods for dealing with random-coefficient-based missingness, make some extensions, and discuss the relationships among the methods. Results from simulation studies will be used to compare the performance of the methods and to shed light on the strengths and weaknesses of each.

Meet the speaker in Room 212 Cockins Hall at 4:30 p.m. Refreshments will be served.