4
$\begingroup$

After doing semantic differential experiment with two different group of people, there is need to do Factor Analysis, and after that to compare group's coordinates in factor space. I did factor analysis separately for two groups, and unfortunately the result is their factor structure are little (or not little) different. What than should I do for compare? Need I to "merge" two groups at one file and make Factor Analysis? Or it would be wrong? I use SPSS.

$\endgroup$
4
$\begingroup$

Rotations in factor analysis are arbitrary. You cannot simply look at the loadings for two groups and determine that they are similar or different, because even identical factor structures may come somewhat rotated relative each other in space (even if you didn't rotate your factor solutions at the analyses). This link will teach you how to compare factor loadings. Using procrustean rotation and congruence coefficient is a nice way to go.

But before you start comparing factor loadings: do you have a good reason why you did factor analysis for each group separately?

$\endgroup$
7
  • 1
    $\begingroup$ That's so. Procrustes is one (and the simplest) way to go. Confirmatory FA is another. On my web-page there is an abundant SPSS macro doing procrustes rotations for two groups. $\endgroup$ – ttnphns Sep 28 '13 at 7:34
  • $\begingroup$ I added a line to your response, please check if it will do for you. $\endgroup$ – ttnphns Sep 28 '13 at 7:42
  • $\begingroup$ @ttnphns, i did factor analysis for each group separately, because one group from China, and second -- from Russia. Both group evaluated some objects on a set of scales: in their native language. And my goal is to check out how differently these objects are perceived in China and in Russia. Is it a good reason for doing at first separately Factor Analysis? $\endgroup$ – user30809 Sep 28 '13 at 9:51
  • $\begingroup$ And what is the better: Procrustes or Confirmatory FA? $\endgroup$ – user30809 Sep 28 '13 at 9:59
  • $\begingroup$ @user30809, you seem to reply to Hotaka's last line - it is not this line which I added. Yes of course, in your study it is first required that examine whether the two factor structures in whole or individual factors there are similar enough, to compare the populations. This task is not very easy. You must be quite experienced in FA to do it. CFA is better relative to Procrustes in that it gives you several ready-made significance tests. But these, of course, imply assumptions not always met in real data. Apart from this, both approaches tend to concur in the conclusions they induce. $\endgroup$ – ttnphns Sep 28 '13 at 10:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.