Identifiability of finite mixtures of logistic regression models

Abstract We allow the intercept in logistic regression to have a nonparametric, discrete distribution and give simple conditions that ensure that the vector of fixed regression coefficients and the mixing distribution of the intercept are identifiable. For binary responses, the number of atoms in the mixing distribution must be bounded by a function of the number of covariate vectors that agree except for one coordinate. For binomial responses, the number of atoms must satisfy the same bound or be bounded by a function of the largest number of trials per response.