Generalized residuals in CUB models

A relevant issue in regression studies is the model diagnostics performed by using estimated residuals and their transformations. Such tools annot be exploited in Generalized Linear Models framework where a decomposition of response into Þtted and error components is no more available; thus, some kind of eneralized residuals are derived by Þrst-order conditions of maximum likelihood quations. In this paper, we will introduce generalized residuals for CUB models, discuss heir main characteristics and test their usefulness on two real data sets. This approach rings out as a noticeable feature the possibility to detect a differential effect of ovariates on the probability of choice for ordinal categories.