I am using a SEM for my PhD dissertation but I am rather new to this methodology. I have data from a panel of 399 respondents. I am interested in ego-Involvement, and included a 15-items scale to capture it. When I run EFA with all items (varimax rotation with principal component extraction), though, the scale seems to be bi-dimensional.
Do you have any suggestions on the best way to deal with it to use in my SEM?
First, both varimax rotation with principal component extraction are problematic for social science data. If you are going to continue from EFA to SEM, I recommend maximum likelihood extraction with correlated factors (e.g. oblivion rotation).
If you still get two factors, that is not a problem for SEM. You would simply specify your two "latent" variables according to which items load on each factor, with a correlation between the two latent variables.
Hello Giulia,
As Professor Morgan recommends, I'd urge you to consider either maximum likelihood or principal axis extraction (and _not_ principal components extraction), using parallel analysis to help inform your choice of the number of factors to extract. If more than one factor, then oblique rotation (e.g., promax or oblique) of some sort would likely be preferable to orthogonal (e.g., varimax, quartimax, etc.).
If this affirms a two-factor structure then you'll have to decide if both factors are equally germane to your target construct of ego-involvement. If "yes," then use factor scores on both in your model. If "no," then select the one factor that is most germane. If "neither" (unlikely, I suspect), then it's back to the drawing board to work on your measure.
Good luck with your work.
Thank you, Bilal Mustafa Khan what I typed was "oblimin" which, as you say, is a form of oblique rotation, but the spell checker messed it up..
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling