DISCRIM Subcommand (MANOVA: Multivariate command)

DISCRIM produces a canonical discriminant analysis for each effect in a design. (For covariates, DISCRIM produces a canonical correlation analysis.) These analyses aid in the interpretation of multivariate effects. You can request the following statistics by entering the appropriate keywords after the subcommand DISCRIM:

RAW. Raw discriminant function coefficients.

STAN. Standardized discriminant function coefficients.

ESTIM. Effect estimates in discriminant function space.

COR. Correlations between the dependent variables and the canonical variables defined by the discriminant functions.

ROTATE. Rotation of the matrix of correlations between dependent and canonical variables. Specify rotation type VARIMAX, EQUAMAX, or QUARTIMAX in parentheses after this keyword.

ALL. RAW, STAN, ESTIM, COR, and ROTATE.

By default, the significance level required for the extraction of a canonical variable is 0.25. You can change this value by specifying the keyword ALPHA and a value between 0 and 1 in parentheses:

ALPHA. The significance level required before a canonical variable is extracted. The default is 0.25. To change the default, specify a decimal number between 0 and 1 in parentheses after ALPHA.

• The correlations between dependent variables and canonical functions are not rotated unless at least two functions are significant at the level defined by ALPHA.
• If you set ALPHA to 1.0, all discriminant functions are reported (and rotated, if you so request).
• If you set ALPHA to 0, no discriminant functions are reported.