## Calculating factor scores for new dataset after factor analysis

### Calculating factor scores for new dataset after factor analysis

Hello,

I want to calculate the factor scores for a new "raw" dataset after I
did a factor analysis with a "original" dataset, which is supposed to
have the same structure as the new dataset. Unfortunately I got no
idea how to do that yet and I'd be grateful if anybody could give me a
hint.

Bastian

### Calculating factor scores for new dataset after factor analysis

I haven't tried it with factor analysis, but with regression analysis
the following worked.

1) Join the old file (file 1) and the new file (file 2).
2) Make a factor analysis, but use only the cases from file 1 (with
the select subcommand)
3) The model will be build only with the selected cases, and will give
the same result
4) The factor scores will be calculated for all cases.

Hope this helps,

Gerben

Dear all,

I have a principal component analysis on the basis of a correlation
matrix of 15 variables, which -in turn- is based upon cross-sectional
individual information.

If you use the 'raw data' as the input of PROC FACTOR, it automatically
derives the factor scores. However, the factor analysis (actually, its
PCA) was based on the correlation matrix (for a very good reason, trust
me).

The question is: how do I get the individual factor scores in this case?
I know that the factor-values are the result of multiplying the scoring
coefficients with the original variables, but as far as I know, I do not
have these scoring coefficients. Or am I wrong?

The output-file contains the standardized regression coefficients for
the factors and the variables. As I have 15 variables and 3 factors, I
have 45 standardized regression coefficients. I know that the matrix of
scoring coefficients is the inverse of the matrix of regression
coefficients, so what I need to do (but don't know how to do it), is
make a matrix of 15 columns and 3 rows, invert it, and use the resulting
scoring coefficients to calculate the 3 factorscores for every
individual in my dataset.

In short, my questions are:

1. Is it true that PCA does not print the standardized scoring
coefficients if the PCA is based upon a correlation matrix?

2. If so, how can I calculate these standardized scoring coefficients?

Any help would be appreciated.

Gijs

--
Dr. Gijs Dekkers
Federaal Planbureau
Algemene Directie
Kunstlaan 47-49
B 1000 Brussel
++32/(0)2/5077413
fax  7373