Object Scores in CATPCA

[o***@hotmail.com]

I've performed a CATPCA in SPSS and wish to use the components as explanatory variables in a logistic regression.

I have finalised the CATPCA and am happy with the (5) principal components computed. However, when I attempt to extract the object scores to use in the regression model, I'm surprised to see that the values are either blank or zeros in each row.

For reference, the 14 variables used in the CATPCA are a combination of nominal, numerical and ordinal. I've defined scales, discretized and, for missing data, defaulted to exclude (and impute a mode after for correlations after quantification).

Can anybody shed some light on what, if anything, I'm doing wrong?

Am I correct in interpreting object scores as component scores?

Should I be extracting something else from the CATPCA to use in the logistic regression?

Thanks in advance.

[Rich Ulrich]

On Sat, 4 Jul 2020 08:06:54 -0700 (PDT), ***@hotmail.com
wrote:

I will start by saying -
[cribbed from an older message by Bruce]

The SPSS mailing list (http://spssx-discussion.1045642.n5.nabble.com/)
is a lot more active than this forum these days. You might want to
join that list (if not already a member) and post your question there.
Via the page given above, click on -more options- near the top for
info on how to subscribe. HTH.

Post by o***@hotmail.com
I've performed a CATPCA in SPSS and wish to use the components
as explanatory variables in a logistic regression.
I have finalised the CATPCA and am happy with the (5) principal
components computed. However, when I attempt to extract the
object scores to use in the regression model, I'm surprised to see
that the values are either blank or zeros in each row.

That sounds like you should have a Warning or Error message.
Or, all cases have been excluded because of Missing?

Post by o***@hotmail.com
For reference, the 14 variables used in the CATPCA are a combination
of nominal, numerical and ordinal.

Begging your pardon - I know nothing else about your data, and
my own experience is not with CATPCA - but that sounds like a mess.

In FA and PC, I've always wanted to have (a) one domain, or
(b) a set of variables with "similar uniqueness" (if I may coin a
phrase). Mixing nominal and scale does not seem promising for
the goal of getting interpretable latent factors. - If they are not
interpretable, what do you have?

Is your N of cases at least 10 or 20 times the number of d.f. that
you have for the 14 variables? - That will say something about
how robust a solution is, but if there are more cases than d.f., it
does not explain the absence of solution.

Post by o***@hotmail.com
I've defined scales, discretized and,
for missing data, defaulted to exclude (and impute a mode after for
correlations after quantification).
Can anybody shed some light on what, if anything, I'm doing wrong?
Am I correct in interpreting object scores as component scores?
Should I be extracting something else from the CATPCA to use in
the logistic regression?