Bonnie Brett posted on Thursday, September 24, 2015 - 9:07 am
I am working with a measure that has 54 items across 6 proposed subscales. To begin, I am examining each subscale individually.
I began with a CFA of each subscale (after removing concerning items; e.g., items that have a skew above 3), as I had an a priori hypothesis about the factor structure - that these XX items will comprise one factor. However, this approach only worked with 2 subscales - the rest evinced terrible model fit across multiple fit indices.
So, I switched to an EFA approach. For each subscale, I removed any concerning items and then put the rest in 1 and 2 factor EFAs. I then examined the 2 factor solution and discarded any items that didn't load with the rest.
However, I began finding that there was always ONE odd variable out. So, for example, I ran the 2 factor solution, and saw that factor 1 loaded on items a - e, but factor 2 loaded on f. So I removed f (as it appeared to be operating differently than items a-e) and ran an EFA again. This time, factor 2 loaded on items a, b, c, and e, but Factor 1 loaded on item d. So I removed d and ran again, and this time, item c was alone. Eventually, I was removing so many items that I could no longer run a 2 factor solution.
I am unsure how to proceed from here. Is there something odd about my data that is creating problems? Am I doing this wrong?
Any insights appreciated! Thank you for your time!