A New Generalized Ordinal Logit Model for Multicategory Response Data

This paper introduces and illustrates a new generalized ordinal logit (GOL) model which connects the four commonly-used multicategory logit models by using two hyper-parameters. The commonly used models in multicategory models are the adjacent-categories logit model (AC), the proportional odds (PO) model, and two variants of the continuation-ratio logit (CR) models. The GOL model generalizes these four models in the sense that each is a special case of the larger GOL model, and this GOL model is used for multicategory response data. In this article, we discuss (maximum likelihood) estimation and testing related to the GOL model, providing SAS/IML computer programs for the same, and illustrating the use of the proposed model with two real datasets.