Online Program

A Simplifying Reformulation of the Likelihood-Ratio Binormal Distribution

*Stephen L Hillis, University of Iowa 

Keywords: binormal, ROC, proper binormal

I will present a simplifying reformulation of the binormal likelihood ratio (binormal-LR) model (a.k.a. “proper binormal model”) that Dr. Metz and Dr. Pan spent many years working on. A great advantage of this model is that it never shows any hooks in the fitted ROC curves. A disadvantage is that it is not intuitive because the corresponding normal and diseased distributions are unfamiliar; for example, the domains are dependent on the model parameters. It will be shown how that the binormal-LR model can be expressed as a bi-variable model that results in familiar distributions for the normal and diseased populations with domains that do not depend on model parameters. This equivalent definition makes the binormal-LR model easier to understand and simplifies derivations of properties