Correcting bias due to misclassification in the estimation of logistic regression models

Abstract

This paper describes several properties of some bias correction methods in the estimation of logistic regression models with misclassification in the binary responses. The observation error model consists of a primary data set plus a smaller validation set. The large sample properties of different bias correction methods are compared under various situations, and the asymptotic relative efficiencies of some important methods are derived. Our small sample simulation studies conclude that the semiparametric estimation method considered by Pepe (Biometrika 79(1992)355–365) is quite reliable under a reasonable surrogate classifier.