Likelihood Inference in a Correlated Probit Regression Model
Y. Ochi, Ross L. Prentice · Biometrika · 1984
Equicorrelated binary observations are modelled using a multivariate probit regression model. Log likelihood derivatives are reduced to simple linear combinations of equicorrelated multivariate normal probabilities, which are approximated using the method of Mendell & Elston (1974). A data set with overdispersion illustrates the use of this model.