Biao Zhang

Department of Mathematics

University of Toledo


Fitting and Testing Proportional Odds Models Based on Case-control Data

Proportional odds models are commonly used in modeling multicategory ordinal responses. In this talk, we consider fitting and testing (prospective) proportional odds model under a case-control sampling plan. By indicating the equivalence between the proportional odds model and a semiparametric selection bias model after reparameterization, we develop maximum semiparametric likelihood estimation for the unconditional response probabilities and the parameters in the proportional odds model. We also establish an optimal property for the maximum semiparametric likelihood estimator and point out a close connection between the semiparametric likelihood approach and the pseudo-likelihood approach under discrete-choice models. In addition, we propose a goodness of fit test for the proportional odds model based on case-control data. The proposed test statistic is the weighted average of the absolute differences between two estimated distribution functions in each response category. Finally, we present a bootstrap procedure along with some numerical results.


Back to Colloquium Series