The need to conduct repeated classifications in a logistic regression model with misclassification in the dependent variable

Abstract

Maximum likelihood estimators for the logistic regression model with misclassification in the response variable are extremely biased when error probabilities are ignored. If misclassification parameters are incorporated in the likelihood function, the bias of the estimators will be satisfactorily reduced, however, there would be a considerable increase in variability, which would reduce the quality of the decision-making process. In order to minimize the problem, there is a need to introduce additional information or to have knowledge of the magnitude of misclassification rates. Since such knowledge may be unavailable, it will be demonstrated that the realization of repeated measures in the response variable, or in part of it, can reduce bias and variability of the estimators, simultaneously.