The Matrix Approach to Logit Models
In this chapter we again discuss logit models, but here we use the matrix approach of Chapter VI. Section 1 discusses the equivalence of logit models and log-linear models. This equivalence is used to arrive at results on estimation and testing. Because the data in a typical logistic regression correspond to very sparse data in a contingency table, the asymptotic results of Section VI.2 are not appropriate. Section 2 presents results from Haberman (1977) that are appropriate for logistic regression models. Section 3 discusses model selection criteria for logistic regression. Direct fitting of logit models is considered in Section 4. The appropriate maximum likelihood equations and Newton-Raphson procedure are given. Section 5 indicates how the weighted least squares model fitting procedure is applied to logit models. Models appropriate for response variables with more than two categories are examined in Section 6. Finally, Section 7 considers the discrimination problem.
KeywordsLogistic Regression Logistic Model Asymptotic Result Matrix Approach Case Variable
Unable to display preview. Download preview PDF.