The output from a principal components analysis performed on the dental calculus data is given in
Figure 1.34.

Figure 1.34

EIGENVALUES OF WITHIN CELLS CORRELATION MATRIX

EIGENVALUE PC? OF VAR CUM PCT

2 .02696 67 .56528 67 .56528

2 67398 22.46611 90.03139
3 - 29906 9.96861 100.00000

NORMALIZED PRINCIPAL. COMPONENTS

COMPONENTS

VARIABLES l 2 3.
RCAN -.73816 -.65197 -.17338
RLI -.90389 08961 41827
RCI -.81551 49081 -.30667
DETERMINANT - 40855

BARTLETT TEST OF SPHERICITY -  87.87194 WITH 3 D. F.
SIGNIFICANCE - .000

F(MAX) CRITERION - 3.07466 WITH (3,100) D. F.

MANOVA also enables the user to rotate the principal components loadings. The keywords
for specifying the type of rotation are VARIMAX, OUARTIMAX, and EOUIMAX (see SPSS,
Second Edition, pp. 484-485, for a description of these three rotations). NOROTATE inhibits
rotation. For example, if

PRINT<PRINCOMPS ( COR, ROTATE(VARIMAX) ) /

is specified, a principal components analysis is performed on each error correlation matrix and the
varimax method is used to rotate the component loadings. By default, all components are rotated.
Fewer components may be rotated by specifying the number of components to be rotated, in

parentheses, after the NCOMP keyword or by specifying a cutoff value for the eigenvalues, in
parentheses, after the MINEIGEN keyword. For example, specifying

PRINT<PRINCOMPS ( COR, ROTATE( VARIMAX) , NCOMP(2) ) /

causes only the first two components to be rotated. If
PRINT-PRINCOMPS ( COR, ROTATE( VARIMAX) ,MINEIGEN(1.5)) /

is specified, only those components associated with eigenvalues greater than 1.5 will be rotated.

1.35 Discriminant Analysis

MANOVA can be used to perform discriminant analysis for each effect in the model. The PRINT
subcommand reguesting discriminant analysis has the format

PRINTsDISCRIM(output list)/
The output list may include reguests for
1 The raw discriminant function coefficients. These are obtained for each tested effect by
specifying
PRINT<DISCRIM (RAM) /

2 "The standardized discriminant function coefficients. If
PRINT-DISCRIM( STAN) /

is specified, the standardized discriminant function coefficients (obtained by multiplying each
raw coefficient by the corresponding standard deviation of the variable) will be printed.

3 The effect estimates in the discriminant function space. To obtain the estimates of each effect
for the canonical variables, specify

PRINT<DISCRIM( ESTIM) /
The canonical variables are defined here as the canonical variates associated with the response
variables.

4 The correlations between response variables and canonical variables. These are obtained by
specifying
PRINT<DISCRIM( COR) /

MANOVA 39