# Help System (web edition)

```   Factor analysis can be performed in following steps:

1) Compute the correlations of selected variables and observations
by the CORR operation.
This gives the correlation matrix CORR.M and the matrix MSN.M
of means and standard deviations.
The correlation matrix can be loaded by the sucro command
The significant correlations are then highlighted.

2) If you like to save these results more permanently, copy them by
MAT R=CORR.M
MAT M=MSN.M
for example.

Factor analysis (continued)
3) Initial solution:
Select the number of factors (say k) and activate
FACTA CORR.M,k,L
This gives the maximum likelihood solution as the factor matrix
FACT.M with k factors. The result is written from line L onwards.

Another alternative is:
MATRUN PFACT,R,k
gives the principal axes solution as factor matrix PFACT.M .
The difference in estimated and computed communalities is displayed
and the computed communalities are placed on the diagonal of R.
MATRUN PFACT,R,k can be reactivated several times until
the process has converged.
The communalities and efficiencies of the factors can be computed
afterwards by MATRUN SUM2,PFACT.M

Factor analysis (continued)
4) To rotate the maximum likelihood solution FACT.M (or the principal
axes solution PFACT.M), activate
ROTATE FACT.M,k
This gives the Varimax solution AFACT.M and the corresponding
rotation matrix TFACT.M with k factors.
Other methods (graphical rotation etc.) can be selected by the
ROTATION specification (See ROTATE?).
The communalities and efficiencies of the factors can be computed
(also) afterwards by MATRUN SUM2,AFACT.M .

The (rotated) factor matrix is displayed in a more legible form by

Factor analysis (continued)
5) The factor scores are computed in two steps.
At first the factor score coefficients are computed by
/FCOEFF,AFACT.M,MSN.M .
This gives the matrix of coefficients FCOEFF.M .
If an oblique rotation has been used, the loadings for the factor
scores are computed by /FTCOEFF,AFACT.M,TFACT.M,MSN.M .

6) Finally, factor scores are obtained by
LINCO <data>,FCOEFF.M(F1,F2,...)
where <data> is the original (or corresponding) Survo data (file)
and F1,F2 are (possibly new) variables for factor scores.
Thereafter the factor scores can be analyzed like any other data
in Survo.