Modeling binary responses with stochastic covariates

Abstract

In a traditional binary regression model, covariates are assumed to be fixed by design. In practice, however, they are most likely to be stochastic and non-normally distributed. We develop modified maximum likelihood estimators for such situations. We show that these estimators are more efficient than the traditional binary regression estimators and robust to data anomalies. We illustrate our results using a real life example.

Keywords:

• Modified maximum likelihood
• binary regression
• stochastic covariates
• robustness
• outliers