Because of the non-normality of the distribution of the items I'm using to measure a latent variable, I decided to use the ULS - Unweighted Least Squares method for a CFA.
AMOS doesn't produce the RMSEA index, but an editor pretends it to me.
I found an equation to calculate it based on Chi Square, DF and Numerosity of the sample (see the file in attachment).
My question is: is it correct or it doesn't work with ULS?
Any other solution?
Thank you
Diego
RMSEA is does not work with ULS estimator but you still can evaluate your model with GFI (Goodness of fit). This is because RMSEA and GFI is considered under absolute fit. Actually, there are three categories namely absolute fit (RMSEA, GFI), incremental fit (IFI, CFI, TLI, AGFI, NFI) and parsiminous fit (chisq/df). Thus, GFI is adequate for evaluate your model.
Thank you Wan for your answer.
Please, do you have any reference to prove it?
More: I found that "GFI and AGFI are affected by sample size, than the current consensus is not to use these measures (Sharma, Mukherjee, Kumar, & Dillon, 2005)."
My sample is 1117, that I can reduce to 345 (primary school) and 772 (secondary school), but they are big anyway.
FIML = log (Ɵ) + tr(S Ɵ-1(Ɵ)) – log ǀSǀ - (p +q), Chisquare = FIML (N-1)
The formula of chisquare is associated with maximum likelihood and RMSEA. Thats why RMSEA cannot be perform since you apply ULS. I believe there are some article from Holmes-smith " Structural equation modeling: From the fundamental to advance topics
But AMOS calculate CMIN and I supposed it was chisquare.
If I use ML method, AMOS give me both chisquare and CMIN and they are the same.
If I use USL method, AMOS give me only CMIN
Do you know why?
Thanks
Diego, ULS provides the minimum of the fit function, but this value is not distributed as a chi-square. Therefore you do not get the chi-square value and also not RMSEA.
ULS offers the advantage that it leads to a consistent estimator of θ
comparable to ML and WLS, yet in contrast to ML, distributional assumptions are not
necessary. Disadvantages are that ULS does not provide the most efficient estimates for θ and that these estimates are neither scale invariant nor scale free (Bollen, 1989). Furthermore, some software programs do not provide χ2 statistics and standard errors if ULS is applied. Others estimate standard errors and χ2 for ULS only under the assumption of multivariate normality.
References:
Bollen, K. A. (1989), Structural Equations with Latent Variables, New York: John Wiley & Sons.
Schermelleh-Engel, K., Moosbrugger, H., & Müller, H. (2003). Evaluating the fit of structural equation models: Tests of significance and descriptive goodness-of-fit measures. Methods of Psychological Research-Online, 8, 23–74.
HTH
Karin
Dear Karin,
I realy thank you for your reply: it is very enlightening.
Best,
Diego
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis