Techniques for Discriminant Analysis with Discrete Variables 145

An inspection of the result indicates that the variable "sex" does not contribute very much to the discrimination. If only the variable "education" is used, one obtains the following result: -

I ^4

1

2

3

1

0

2

27

2

5

24

47

3

173

217

41

к

1

2

3

^3

^3

^^2

I X.

1 + 2

3

1 + 2

74 / 536

31 / 536

3

41 / 536

390 / 536

This classification into A + В and С + D compares favourably with the classification obtained in the original study. For comparison, the estimated probabilities of classification of that study are given here:

I >v

1 + 2

3

1 + 2

66 / 540

37 / 540

3

62 / 540

375 / 540

( There are slight differences in the total figures of the original study and those obtained in this study, but they should not invalidate a comparison.)

For a discrimination between A and B, a continuous test variable — e.g. "cube construction test" - should be used.

The best procedure would of course be to discriminate at first between A + В and С + D and administer the test only to subjects who have been classified into A + B. In spite of that, the analysis using both the discrete and the continuous variable simultaneously will be carried out here in order to illustrate the results obtained in the previous sections.

Within each of the 3 cells, the subdivision into regions R R , R has to be done. One is here in the fortunate position that only a single continuous

19 Metnka 2