Abstract

Logit and probit are regression models for binary outcomes that allow one to avoid the problems associated with the linear probability model, such as nonconstant error variance and the unrealistic assumption of linearity in the parameters. Logit and probit also serve as building blocks for more advanced regression models for other categorical outcomes. In this entry, the focus is on logit and probit models for binary and nominal outcomes. Binary outcomes are dichotomous-dependent variables coded as 0 or 1. Nominal outcomes are dependent variables with three or more unordered categories. The entry considers several topics related to binary and multinomial logit/probit models, including motivation for the models, estimation, interpretation, hypothesis testing, model assumptions, and connections to ordered regression models. The features of the models are illustrated with examples using Stata.