The first model uses 2 of the predictors from the /VARIABLES list. The following example uses the bank.sav data that was provided with several versions of SPSS. The binary hit/miss variables are then easily created with COMPUTEĬommands and applied to the McNemar or Q test through the NPAR TESTS command. You can also save the predicted class from each analysis with the /SAVE subcommand. SPSS allows you to request analyses of separate functions in a single DISCRIMINANT command and to print the classification results (actual class by predicted class) for each analysis. Huberty suggests that the hit/miss rates of 3 or more rules could similarly be compared with Cochran's Q, which is also available in NPAR TESTS.

McNemar's test would be employed to test whether the true proportions of correct classifications ('hits') are equal for the 2 rules. Predicted the correct class as a 0, if incorrectly classified. Huberty (1994, 'Applied Discriminant Analysis', Wiley) suggests that alternate classification rules be compared by creating a variable for each rule, such that a case is coded as 1 if the rule You can perform sequential tests of classification accuracy for alternate predictor lists by using DISCRIMINANT in combination with NPAR TESTS. These two aspects of HDA are discussed and illustrated below. HDA has also been presented to us as a question of whether an additional block of predictors results in a significant change in Wilks' Lambda, i.e., further separates the groups on the resulting functions. In hierarchical (or sequential) discriminant analysis (HDA), one is generally interested in whether one or more predictors improves upon the classification accuracy of a previously entered set of predictors.

Hierarchical sequential testing