The LOGISTIC Procedure

ROC Statement

ROC <'label'> <specification> </ options> ;

The ROC statements specify models to be used in the ROC comparisons. You can specify more than one ROC statement. ROC statements are identified by their label—if you do not specify a label, the ith ROC statement is labeled ROCi. Additionally, the specified or selected model is labeled with the MODEL statement label or Model if the MODEL label is not present. The specification can be either a list of effects that have previously been specified in the MODEL statement, or PRED=variable, where the variable does not have to be specified in the MODEL statement. The PRED= option enables you to input a criterion produced outside PROC LOGISTIC; for example, you can fit a random-intercept model by using PROC GLIMMIX or use survey weights in PROC SURVEYLOGISTIC, then use the predicted values from those models to produce an ROC curve for the comparisons. If you do not make a specification, then an intercept-only model is fit to the data, resulting in a noninformative ROC curve that can be used for comparing the area under another ROC curve to 0.5.

You can specify a ROCCONTRAST statement and a ROCOPTIONS option in the PROC LOGISTIC statement to control how the models are compared, while the PLOTS=ROC option controls the ODS Graphics displays. See Example 54.8 for an example that uses the ROC statement.

If you specify any options, then a ROC Model Information table summarizing the new ROC model is displayed. The options are ignored for the PRED= specification. The following options are available:

NOOFFSET

does not include an offset variable if the OFFSET= option is specified in the MODEL statement. A constant offset has no effect on the ROC curve, although the cutpoints might be different, but a nonconstant offset can affect the parameter estimates and hence the ROC curve.

LINK=keyword

specifies the link function to be used in the model. The available keywords are LOGIT, NORMIT, and CLOGLOG. The logit link is the default. Note that the LINK= option in the MODEL statement is ignored.