Influence diagnostics for multivariate growth curve models

Abstract

Research has shown that in mixed effect longitudinal models, influential observations can have a large effect on the estimates of subject-specific parameters. Furthermore, they cannot always be detected by the classical Cook’s distance due to potentially large between subject variation. Thus, influential observations should be approached by conditioning on the subjects. However, no rigorous approach has been developed for influential observation detection for multivariate longitudinal mixed models where more than one response is measured for each subject at each time point. We propose a multivariate conditional Cook’s distance for this more general situation. Examples are given to illustrate how the influential observation in one characteristic changes the effects of both characteristics.