I have done EFA in SPSS with a sample of 305; only one factor was extracted in PCA. It indicated unidimensionality. Is this contract good to go ahead for CFA? What does it indicate?
Construct had 5 subscales; 14 items.
Thanks.
Hello Yumna,
Sure; corroboration of an obtained EFA factor solution with additional data is never a bad idea! So, carry on with the CFA.
One note about your EFA, however:
Among the extraction methods available for EFA, PCA (principal components analysis) is probably the least desirable, unless you can be sure that there is absolutely no specific or error variance present in the observed score set. Principal axis or maximum likelihood extraction factoring would be better choices, especially with small numbers of variables (PCA tends to bias estimated loadings upwards in strength compared to the common factor analysis methods).
Good luck with your work.
Hi Yuma,
beyond what David noted (although I would not describe PCA as an extraction method for EFA but as a complete different animal),
I would not only do an CFA now but add further variables that could not only as validation criteria but also increase the testability of the proposed factor structure.
Testing a CFA on the same data set as the formely explored EFA is a bit of a problematic thing as you test the very structure that you just explored. By adding other variables that are supposed to cause your facor or are caused by the factor, you add new restrictions which have to pass the test and were not involved in the EFA. These restrictions mean that the factor model has to explain all the correlations between each of the indicators and the variables.
Best,
Holger
Hey Yuma
You've had some great advice already. I'd add that you might want to consider testing a few alternative, or competing, CFA models. You could test a one-factor model and also a multi-dimensional model based on the 5 subscales - so you'd have the 14 items measuring 5 latent variables, with the pattern of loading dictated by how the subscales are supposed to be scored. There are some useful and straightforward ways to compare CFA models such as chi-square difference test and the use of BIC or AIC.
Mark
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts