STRATIFIED TESTING FOR TREATMENT EFFECTS WITH MISSING DATA

Dawson and Lagakos (1993, Biometrics 49, 1022-1032) proposed a stratif ied test for repeated measures data that contain missing observations. They recommended stratification based on missing data patterns and co nsidered sufficient conditions under which the size of the test is pro perly retained. In this paper, we point out some practical problems wi th these conditions and illustrate them with their CD4 count example a s well as a simulation study. We give a less stringent condition and d elineate its merit. We also discuss what to do when none of the condit ions are met.