Ordinal probit: A generalization

Abstract

Probit models for the analysis of oidinal qualitative vaiiables are discussed. A generalization of the conventional method of McKelvey and Zavoina (1975) is proposed. Ihe geneial model entails the elimination of the assumption of constant category thresholds for the oidinal latent index. Instead the thresholds are assumed to be linear functions of obseived exogenous vaiiables whose values aie observation specific. A few typical examples aie discussed and the details of maximum likelihood estimation of the general model aie given. Ihe method is applied to an analysis of the bond ratings of electric utilities and is thereby shown to be computationally efficient and conducive to hypothesis tests that are precluded by conventional methods.

Keywords:

• categorical variables
• maximum likelihood estimation
• corporate bond ratings
• threshold parameters