ROC / AUC Analysis

UNISTAT’s Logistic Regression procedure features a comprehensive implementation of ROC (Receiver Operating Characteristic) analysis. It is possible to compute AUC (area under the curve) and plot ROC and sensitivity-specificity curves with multiple covariates. Multiple comparisons can be made between AUCs.

The output includes the difference between AUCs, their standard errors, tail probabilities and confidence intervals (asymptotic normal) and a further chi-square test.

All statistics for diagnostic tests and their confidence intervals are computed for all fitted Y values (classification threshold probabilities). By default, only the sensitivity and specificity values and their confidence intervals are displayed.

You can choose to display any statistic with or without confidence intervals. The case (row) corresponding to the classification threshold probability (the best cut-off point) is marked by an asterisk.