Influence diagnostics for generalized linear mixed models: Applications to clustered data

The Cook's distance for generalized linear mixed models is investigated, with applications to clustered data. In particular, first-order approximations are derived for the best linear unbiased predictor of the parameters due to cluster deletion. A small-scale simulation study shows that the method provides an efficient way to identify influential clusters. The notion of joint and conditional influence is also considered to address the masking effects of cluster-wise deletion. A data set on maternity length of hospital stay illustrates the usefulness of the proposed diagnostics. © 2002 Elsevier Science B.V. All rights reserved.