|
Activity Number:
|
50
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 8, 2004 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
ENAR
|
| Abstract - #301200 |
|
Title:
|
An Index of Discrimination for Logistic Regression Models with Ordinal Responses
|
|
Author(s):
|
Michael P. McDermott*+ and Christopher A. Beck
|
|
Companies:
|
University of Rochester and University of Rochester Medical Center
|
|
Address:
|
Department of Biostatistics and Comp Biology, Rochester, NY, 14642,
|
|
Keywords:
|
model discrimination ; proportional odds model ; ROC surface
|
|
Abstract:
|
A commonly used index of discrimination for a logistic regression model with a binary outcome is the so-called c-index, which is an estimate of concordance between predicted and observed responses. This index is equivalent to the area under the receiver operating characteristic (ROC) curve generated by varying the cutpoint for the predicted probability that is used to classify observations. An index of discrimination is proposed for a logistic regression model with an k-level ordinal outcome. The index is based on the idea of generalizing the concept of an ROC curve to that of an ROC surface when observations can be classified into three or more categories. The surface is generated by varying k-1 cutpoints for the linear predictor that is used to classify observations, with a larger value of the linear predictor being associated with an increased probability of being classified in a higher category. The volume under the ROC surface serves as an index of discrimination. Issues concerning estimation and interpretation of the volume under the ROC surface are discussed.
|
