Robust binary regression with continuous outcomes

The authors propose a robust bounded-influence estimator for binary regression with continuous outcomes, an alternative to logistic regression when the investigator's interest focuses on the proportion of subjects who fall below or above a cut-off value. The authors show both theoretically and empirically that in this context, the maximum likelihood estimator is sensitive to model misspecifications. They show that their robust estimator is more stable and nearly as efficient as maximum likelihood when the hypotheses are satisfied. Moreover, it leads to safer inference. The authors compare the different estimators in a simulation study and present an analysis of hypertension on Harlem survey data.