Abstract
Binomial regression is commonly used to construct a binary classification rule from a set of observable covariates. The accuracy of the rule is summarised by empirical type 1 and type 2 misclassification rates as the threshold of the rule varies. However, the empirical misclassification rates are not the maximum likelihood estimators of the true rates if the binomial regression model is assumed true. We give a simple algorithm for the maximum likelihood estimators of the misclassification rates for any given threshold. Statistical calibration of these estimates is best provided by bootstrap.

This publication has 0 references indexed in Scilit: