Accessing a Model's Ability to Classify Subjects: The Importance of Considering Marginally Accurate Classifications

Abstract

The purpose of this paper is fourfold: (a) discuss the importance of considering “marginally accurate” classifications, which are predicted probability values whose confidence limits contain the cut-value used to classifying subjects, (b) present a six-step calculation procedure used to identify the "marginally accurate" classification values, (c) illustrate how the identification of these “marginally accurate” values are important in the evaluation of the model, and (d) discuss how a review of the “marginally accurate”
values can used to access the differential effectiveness of various modeling procedures with respect to their replicability and stability.