Variable selection for marginal longitudinal models

Abstract

Model selection is an essential part of any statistical analysis. A generalized version of Mallows's Cp (GCp) (Cantoni et al., 2005) was developed as a model selection criterion for longitudinal data. GCp, just like Mallows's Cp (Mallows, 1973), provides an estimate of a measure of a model's adequacy for prediction. In this study, the performance of GCp is compared with that of Cp by using longitudinal data of 137 urinary incontinent elderly patients from 38 medical practices