• Print |
• E-mail  |
• Bookmark

PROC LOGISTIC does not provide the KS test. However, you can obtain it by using the EDF option in PROC NPAR1WAY to analyze the predicted values in the OUT= data set from PROC LOGISTIC.

Also, although PROC LOGISTIC does not provide the KS test for a measure of predictive power, it does provide the area under the ROC curve. This is reported as the c statistic in the Association of Predicted Probabilities and Observed Responses table. The c statistic is equivalent to the Wilcoxon two-sample statistic for comparing the locations of event and non-event distributions. It can be argued that this is a better measure of predictive ability than the KS statistic because the Wilcoxon statistic concentrates its power on detecting a location shift, which is of primary interest in evaluating the logistic model. The KS test looks, in general, for any difference in the distributions, including shape differences. You can test hypotheses on the area under the ROC curve (such as testing if c=0.5, a test of no predictive power), or compare areas under the curves from competing models. Beginning in SAS 9.2, use the ROC and ROCCONTRAST statements. Prior to SAS 9.2, use the ROC macro.

In addition, PROC LOGISTIC provides a data set (by specifying the OUTROC= option) that can be used to plot the ROC curve. A plot of the ROC curve provides an excellent visual summary of the model's predictive ability. Beginning with SAS 9.1, PROC LOGISTIC can produce the ROC plot by using ODS graphics. Specify the global ODS GRAPHICS ON statement to enable ODS graphics. Beginning in SAS 9.2, add the PLOTS=ROC option in the PROC LOGISTIC statement to request an ROC plot. In SAS 9.1, add the GRAPHICS ROC statement to request the plot. Both are shown in the following example:

     ods graphics on;
     ods html;

     /* In SAS 9.2 */
     proc logistic plots=roc;
        model y = x;
        run;

     /* In SAS 9.1 */
     proc logistic;
        model y = x;
        graphics roc;
        run;

This produces an HTML file containing the PROC LOGISTIC output and plots. Rather than producing a separate HTML file, you might find it more convenient to have results displayed immediately in a browser window within SAS as follows: Instead of using the ODS HTML statement, from the Tools menu select Options ⇒Preferences. In the Results tab, deselect Create listing, and select Create HTML. Also select View results as they are generated and Internal browser.

Prior to SAS 9.1, specify the OUTROC= option and then plot the _SENSIT_ variable against the _1MSPEC_ variable in the resulting data set. Or use the ROCPLOT macro. See the example ROC Curve, Customized Odds Ratios, Goodness-of-Fit Statistics, R-Square, and Confidence Limits in the PROC LOGISTIC chapter of the SAS/STAT User's Guide.

Operating System and Release Information