I'm conducting a FA using 17 dichotomous variables from a larger study of cancer patients. I believe these 17 items represent an underlying construct of "waiting distress" while caregivers wait for a loved one to die. Eyeballing the variables, I expected them to load on 4 factors related to this underlying construct. My goal is to combine the items into a scale that I can then use as a predictor for outcomes like caregiver bereavement.
I'm not sure how to interpret my EFA findings. Some questions:
1) One factor loading is greater than 1. I've read previous threads and it seems this is acceptable?
2) Two error variances are negative. When I removed these 2 variables and re-ran the EFA, I got a negative error variance for a third variable. All 3 variables feel important, I'm not sure what to do.
3)I previously conducted EFA using SPSS, but then read that MPlus was the better program to use for EFA with dichotomous variables - the problem is, the SPSS results were more consistent with what I'd been expecting! i.e. variables loaded on 4 expected factors, whereas with my MPlus EFA, it's a little murkier.
May I send you my files to get your thoughts? I'd also welcome your thoughts on my basic plan, i.e. pulling these items together to create a scale "after the fact", and using scale scores to predict other outcomes.
Thank you so much, I'll read up on ESEM and give it a shot.
My plan has been to perform an EFA with half my sample (N=135) and then a CFA with the other half (N=136). As I'm unfamiliar with ESEM, is it appropriate to keep with this same plan, i.e. ESEM with half the sample and then CFA with the other half? Or is ESEM enough?
Everyone thanks you for your patience, but I am a total novice and should extend an extra hearty thanks.
Thank you, Linda. I am slogging my way through Bengt's 2009 ESEM article, which is mostly over my head but I think I'm getting the main points.
My professors had suggested I split the sample, so I'm glad I asked. Given that the sample is too small to split, does performing an ESEM on the full sample limit the conclusions I can make? It sounds from the article that ESEM can replace CFA in some instances.
If I can save face and solve the problem on my own, I'd like to - - I added "Analysis: parameterization=theta;" to my input and the output ran successfully.
Was this the correct move?
If so, my new problem is that I'm having trouble discerning the results. Do I read the "Lambda" table as my factor loadings? If this is the case, is there a way to tell if the loadings are significant? A couple of my items load on more than one factor.
I promise once I get this figured out I'll be out of your hair for a while!
Hello! I've successfully run a series of EFAs on my 17 items, as well as an ESEM to provide more information about potential problems with my EFA. Achieved what appears to be a nice 13-item, 4-factor model.
My question now is, which fit statistics are appropriate to use in an analysis of this type?
I've read the Yu dissertation, which suggests the following: probability associated with the chi-square greater than 0.05; CFI greater than or equal to 0.96; TLI greater than or equal to 0.95; RMSEA less than or equal to 0.05, RMSR less than .05.
Which of these are appropriate in my situation? Also, in reporting my findings, I was going to report my EFA fit statistics, with supporting evidence of factor loading significance from my ESEM. Should I be reporting my fit statistics from the ESEM?