|
Abstract:
|
Several authors have suggested prediction rules for use in logistic regression models. However, no agreed standard for selecting prediction rules from a logistic regression model has been given. Neter et al., in his text Applied Linear Regression Models, suggests three rules: 1. "Use .5 as the cutoff." 2. "Find the best cutoff for the data set on which the multiple logistic regression model is based." 3. Use prior probabilities and costs of incorrect predictions in determining the cutoff. Halpern, Gallop, and others have suggested finding an optimal operating point (OOP) from the ROC curve. The discussion from this paper suggests that there is no "gold" standard for selecting the prediction rule for all applications. Three datasets and logistic models will be presented representing market research, health care, and banking applications.
|
