Bonnie Brett posted on Wednesday, September 16, 2015 - 5:50 pm
Hello! I am currently trying to examine the factor structure of a particular scale. I am examining each subscale, one by one, as I don't have enough power to throw all items into one big CFA or EFA. I began with a CFA, as I had an a priori hypothesis about the factor structure - that these 9 items will comprise one factor. However, this approach only worked with 2 subscales - the rest evinced terrible model fit across different fit indices.
So, I chose to try an EFA approach. I remove any concerning items (e.g., items that have a skew above 3) and then put the rest in 1 and 2 factor EFAs. My next step is to examine the 2 factor solution and discard any items that don't load with the rest.
However, I am finding that there is always ONE odd variable out. So, for example, I run the 2 factor solution, and see that factor 1 loads on items a - e, but factor 2 loads on f. So I remove f (as it is operating differently than items a-e) and run an EFA again. This time, factor 2 loads on items a, b, c, and e, but Factor 1 loads on item d. So I remove d and run again, and this time, item c is alone.
So my question is thus: will 1 item always be be primarily explained by a separate factor if I request a 2 factor solution? Would I ever find a solution that has Factor 1 loading on all items and Factor 2 not loading on anything?