The Many Faces of Logistic Regression

Logistic regression has found wide acceptance as a model for the dependence of a binary response variable on a vector of explanatory variables. It can also be used, however, as a maximization algorithm for fitting a variety of other parametric models. The easy availability of logistic regression in standard packages is a major advantage; further, the regression diagnostics routinely supplied are frequently useful, even though the model being fitted is not logistic. In some cases the objective function maximized is a likelihood, but the method seems to arise especially often in the maximization of a so-called pseudolikeli-hood. Applications include models from choice theory, spatial modeling, random graph theory, and educational testing.