CRITERIA Subcommand (ROC ANALYSIS command)

The CRITERIA subcommand allows the user to decide whether or not the cutoff value is included as the positive test result value, whether or not the larger or smaller value direction of the test result variable indicates the positive test result, the confidence level of the ­asymptotic confidence interval produced by /PRINT = SE, and the estimation method for the standard error of the area under the curve.

CUTOFF=INCLUDE

Positive classification of test result values includes the cutoff values. Note that the positive test result direction is controlled by the TESTPOS keyword. This is the default setting.

CUTOFF=EXCLUDE

Positive classification of test result values excludes the cutoff values. This distinction leads to the different sets of cutoff values, but none of the output (chart or table) is affected at this moment because the user cannot choose the cutoff values for the classification.

TESTPOS=LARGE

The user can specify which direction of the test result variable indicates increasing strength of conviction that the subject is test positive. The larger the test result value is, the more positive the test result is. This is the default setting.

TESTPOS=SMALL

The smaller the test result value is, the more positive the test result is.

CI=n

Confidence level in the (two-sided) asymptotic confidence interval of the area. The user can specify any confidence level in (0, 100) for the asymptotic confidence interval created by /PRINT = SE. The default setting is 95.

DISTRIBUTION=FREE

The method of calculating the standard error of the area under the curve. When FREE, the standard error of the area under the curve is estimated non-parametrically—that is, without any distributional assumption. This is the default setting.

DISTRIBUTION=NEGEXPO

The standard error is estimated assuming the bi-negative exponential distribution. This latter option is valid only when the number of positive actual state observations equals the number of negative actual state observations.