This seems to be a very basic question, but I was wondering if there's any difference between using factor scores or composite scales in statistical tests like regression analysis.
I have 47 items from a large dataset (N=1733), and I was able to derive 11 distinct factors for these items. I wish to conduct a regression analysis with these factors and I was wondering if it's better to compute a scale (average) of these factors or do I use their factor scores.
The factor analysis was performed using principal axis factoring with promax rotation on SPSS. I further conducted a confirmatory factor analysis for the items and derived very good fit indices, so the factors are relatively distinct.
Any advice or suggestions will be very much appreciated. Thanks.
do I understand it correctly that the options are (1) to use factor scores for 11 different factors or (2) an average score derived from the 11 factors? My thought is that if you are doing a factor analysis, why not using the factor scores? There is some statistical theory behind factor analysis allowing for a clear interpretation of factors and factor scores. Admittedly, the data is easier to handle when using one average score, but they should be more difficult to interpret in substantive terms.
You are right about option (1); for option (2) I meant computing an average score for the items that make up each factor. Thanks a lot for the responds. I gather from your advice that using factor scores may produce better results. However, I've noticed also that factor scores differ for different procedures (e.g. exploratory factor analysis vs confirmatory factor analysis) and sometimes for the software used to perform the factor analysis. This makes me wonder what best procedure to adopt assuming I decide to use factor scores for my subsequent analysis.
Building averages across the items would leave behind the theory behind factor analysis and strikes me as the weaker approach (whether the factor scores produce "better" results is another matter). Different results across software packages can happen and is likely to be due to different routines underlying the calculations. The choice between different approach - EFA vs CFA vs principal component analysis - is not an easy one. If you have clear expectations about what items load on how many factors, you should use CFA. If you just want to reduce the dimensionality of your data, you could go for principal components analysis. Unfortunately, I do not know a good and easy text on this topic.
Thanks a lot, your comments have been very helpful.