Very Computer

We are confused. As we understand, SPSS and Systat are both products
of SPSS Inc. So, we thought that their way to do analysis is identical
also. It seems that this was a stupid thought.

Problem: In Principal Component Analysis (Varimax-rotation) both
programs produce seven factors and all loadings are identical
regarding their power. But, one factor produced by Systat has inverse
loadings compared to what SPSS produces:

Systat          SPSS

.787            -.787
-.701           .701
.666            -.666

This is a real problem because we use factor scores based on these
loadings.

I would be very grateful, if someone can tell us what is going
(wr)on(g)....

AV

: Problem: In Principal Component Analysis (Varimax-rotation) both
: programs produce seven factors and all loadings are identical
: regarding their power. But, one factor produced by Systat has inverse
: loadings compared to what SPSS produces:

: Systat                SPSS

: .787          -.787
: -.701         .701
: .666          -.666

: This is a real problem because we use factor scores based on these
: loadings.

  "This is a real problem"?  -- the serious problem is that you
believe that the DIRECTION of the scoring in Factor Analysis
has some meaning.  But:  it is 100% permissable to change the
signs on all the loadings of a factor.  So, do it.  See, the results
are the SAME!

If the algorithm does not leave coefficients that are mostly "+",
then some computer programs count the pluses and will reverse signs,
to make the NUMBER of loadings to be mostly positive; or get a sum
in order to make the TOTAL positive ... or, whatever.

It does not matter, mathematically, statistically, or clinically.

It is nice, in a way, that this problem comes up occasionally, because
there is a particular misunderstanding of Factors, arising from a
clinical rather than a mathematical perspective, which the explanation
should help to illuminate.  But I forget how to describe it.

If this is still a problem for someone, please ask.

--

http://www.pitt.edu/~wpilib/index.html   Univ. of Pittsburgh