RC test based on predicted probabilities from the new model | |||
Statistically significant | Not significant | ||
RC test based on predicted probabilities from the old model | Statistically significant | A. Both new and old models are miscalibrated or use incorrect functional form | B. New model provides improved calibration across risk classifications |
Not significant | C. New model is miscalibrated or uses incorrect functional form. Old model is preferable. | D. Reclassification is not choosing between models. |