How to construct a composite measure using several items in SPSS?

For my master thesis, I have to do a regression analysis. But, as an independent variable, I have to construct a composite measure, being perceived importance (of interest groups). I have 9 variables which are: high ability of interest groups in gaining public opinion, legitimacy of interest groups, power of interest groups (which form should form the component selectivity), interest groups using direct lobbying, interest groups using letter-writing campaigns (which should form the component actions), interest groups focussing on customer issues, interest groups focussing on environmental issues, interest groups focussing on minority rights, and interest groups focussing on economic issues (which should form the component issue focus), on a 5-point Likert-scale ranging from not important at all to very important.

When I run a factor analysis, spss tells me that all variables cluster together into one factor/component. No problem, but I have to get three components ultimately. So, do I now have to run three principal component analyses (for each component seperately)? Or is there a way in which I am able to specify which variables should cluster together beforehand? And what statistics do I need to report/check in order to find out if everything is allowed and valid/reliable? And of course, how do I finally make up the composite measure out of the three components?

As my laptop doesn't allow me to use any better version, I use SPSS version 18.0.

• This doesn't really sound like a software / code question to me. I think the real issue is understanding FA better. IMO, this is on-topic. – gung Jun 10 '15 at 14:19
• ":I have to get three components ultimately." Why use PCA then? Factor analysis (partly) tells you the number of components you should have - if you already know how many you need, there's no need for PCA. – Jeremy Miles Jun 11 '15 at 23:06
• You can also do factor extraction and specify 3 factors (opposite to the default estimate of number-of-factors with eigenvalue>1). Then a varimax/promax rotation could possibly locate the 3 factors meaningfully in the 9-variable-space. If this comes out to give indeed a meaningful structure you can then do the regression using the factorscores. – Gottfried Helms Jun 12 '15 at 9:28