Model 1 is nested in models 2 and 3. You may have your setup backwards. You should run model 1 last. If you are doing this and still having problems, send your input, data, output, and license number to email@example.com.
I am using DIFFTEST with WLMSV to compare the goodness of fit of 3 models based on 17 categorical items. (There are some theoretical reasons for comparing these different models.) In the first model the 17 all load on 1 factor; in the second model, 9 items are on 1 factor and 8 on a second. In the third model, also with 2 factors, 7 of the original 9 load on the first factor while the remaining 2 of Factor 1 now load on the second factor in addition to the 7 already specified in the second model. Is there any way I can compare the Goodness of fit between the second and third models since the message I get is that these are not nested. Can I use the difference between the two Chi-square difference measures, the first from comparing Models 1 and 2 and the second from comparing Models 1 and 3 as a further Chisquare difference measure?
Your second and third models are not nested so using a difference test to compare their fit is not appropriate.
Sanja Franic posted on Wednesday, November 18, 2009 - 5:26 am
I cannot use the "difftest" option in MPlus when also using the "model constraint" option. Is there any other way to calculate the chi-square difference, given that I am using the WLSMV estimator? Thanks, Sanja
Sanja Franic posted on Wednesday, November 18, 2009 - 6:05 am
I think I just found the answer to my own question. I use WLSM instead of WLSMV and use the scaling correction factors I get in the output to perform a shi-sq difference test.
I have a second-order latent variable in CFA, and I would like to see if the third-order latent factor is a better alternative of the construct. Both second-order factor and the third-order alternative have the same number of 34 ordinal indicators, so I am using WLSMV. The only difference is whether a 2nd or a 3rd-order construct is used. Hence I would like to ask if the DIFFTEST is appropriate in this situation? Or should I simply compare values of fit indices of both models and select a better fitting one?