The ROCCONTRAST statement compares the different ROC models. You can specify only one ROCCONTRAST statement. The ROCOPTIONS
options in the PROC LOGISTIC statement control how the models are compared. You can specify one of the following contrast specifications:

REFERENCE<(MODEL  ’roclabel’)>

produces a contrast matrix of differences between each ROC curve and a reference curve. The MODEL keyword specifies that the
reference curve is that produced from the MODEL statement; the roclabel specifies the label
of the ROC curve that is to be used as the reference curve. If neither the MODEL keyword nor the roclabel label is specified, then the reference ROC curve is either the curve produced from the MODEL statement, the selected model
if a selection method is specified, or the model from the first ROC statement if the NOFIT
option is specified.

ADJACENTPAIRS

produces a contrast matrix of each ROC curve minus the succeeding curve.

matrix

specifies the contrast in the form row1,row2,…
, where each row contains the coefficients used to compare the ROC curves. Each row must contain the same number of entries as there are ROC curves being compared. The elements of each row refer to the ROC statements in the order in which they are specified. However, the first element of each row refers either to the fitted model, the selected model if a SELECTION=
method is specified, or the first specified ROC statement if the NOFIT
option is specified.
If no contrast is specified, then the REFERENCE contrast with the default reference curve is used. See the section Comparing ROC Curves for more information about comparing ROC curves, and see Example 72.8 for an example.
The following options are available:

E

displays the contrast.

ESTIMATE <= ROWS  ALLPAIRS>

produces estimates of each row of the contrast when ESTIMATE or ESTIMATE=ROWS is specified. If the ESTIMATE=ALLPAIRS option
is specified, then estimates of every pairwise difference of ROC curves are produced.
The row contrasts are labeled "ModelLabel1 – ModelLabel2", where the model labels are as described in the ROC
statement; in particular, for the REFERENCE contrast, ModelLabel2 is the reference model label. If you specify your own contrast
matrix, then the ith contrast row estimate is labeled "Rowi".

COV

displays covariance matrices used in the computations.