Diagnostic Methods in the Apt Model for Ordered Categorical Data

Abstract

We describe diagnostic methods for assessing the influence of a single outlier or individual case on the estimates of the asymmetric power transformation (APT) model for ordered categorical response. The APT model includes a wide range of the probability curves of the response including both of symmetric and asymmetric curves. We also discuss methods for assessing the influence of covariate selection. Furthermore, we mention methods for assessing the appropriateness of combing the categories of the response. Two examples are used to illustrate the proposed method.