Question on testing differences between factor loadings

[Mike Donatello]

I hope someone can point me in the right direction.

I’m looking to test differences between factor loadings on two iterations of
the same scale (not an issue of test-retest reliability per se).  That is, I
have a set of items measuring media use motivations for one medium, and the
same for another medium (i.e., repeated measures).  I want to determine
whether the differences in loadings on specific items -- the actual factors
themselves come out nearly identical for both media -- are significant, but
I can’t find any description of a test to do this.

Can I use the Pearson-Filon test for comparing correlated correlations to
determine whether individual loadings on each of the factoring iterations
are significantly different?  If so, is there supporting literature that I
overlooked?  If not, is there another relevant test?  Should I treat this as
a reliability issue?

Thanks.

Mike Donatello

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Art Kendall]

A few suggestions.
1) see this book, the page has an email link to the author to see if she will try to answer the question.
http://www.amazon.com/Structural-Equation-Modeling-AMOS-Applications/dp/0805863737
However, you might want to clarify first why you are asking this question. I have been working with factor analysis since 1971 and this is the first time I have heard of this question.


2)To get a clearer picture for him the old fashioned way:
How do plots of the eigenvalues from the two sets versus the eigenvalues from a parallel analysis look?
How do the scoring keys for the summative scales compare?
How well do the summative scale scores correlate 4 scores:
set 1 by key 1
set 1 by key 2
set 2 by key 1
set 2 by key 2?

How do the Cronbach's alphas for the 4 scores look?

What do the canonical correlations look like?


3)Explore the data with INDSCAL (available e.g., in SPSS).  It takes multiple matrices of correlations, finds "factors" in a common space, and gives "saliences" i.e., how much is the dimension used is the different sets.  This would be particularly if you work on data for multiple media.

Art Kendall
Social Research Consultants
301-864-5570

On 11/8/2011 12:53 PM, Mike Donatello wrote:

I hope someone can point me in the right direction.

I’m looking to test differences between factor loadings on two iterations of
the same scale (not an issue of test-retest reliability per se).  That is, I
have a set of items measuring media use motivations for one medium, and the
same for another medium (i.e., repeated measures).  I want to determine
whether the differences in loadings on specific items -- the actual factors
themselves come out nearly identical for both media -- are significant, but
I can’t find any description of a test to do this.

Can I use the Pearson-Filon test for comparing correlated correlations to
determine whether individual loadings on each of the factoring iterations
are significantly different?  If so, is there supporting literature that I
overlooked?  If not, is there another relevant test?  Should I treat this as
a reliability issue?

Thanks.

Mike Donatello

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD