Message/Author |
|
Jaume posted on Friday, May 14, 2004 - 4:22 am
|
|
|
Hi Is there a reference about chi-square difference testing with the WLS estimator M-Plus uses? Thank you |
|
|
I don't know of any. It behaves the same as difference testing for maximum likelihood. |
|
Jaume posted on Tuesday, May 18, 2004 - 8:49 am
|
|
|
Thanks for answering. Just another question: What would you recomend to choose between nonnested models with categorical outcomes? Just the CFI, TLI and RMSEA or is there anything else? Thanks |
|
|
There is a dissertation by Yu that can be downloaded from the homepage of www.statmodel.com that discusses the merits of different fit statistics for categorical outcomes. |
|
JI posted on Saturday, October 15, 2005 - 10:36 am
|
|
|
Hi, I need to run a series of latent pairs to ascertain discriminant validity. I've done it in AMOS but I'm relatively new to MPLUs. I've read the explanation in the user's guide. Is the procedure in the user's guide uses the S-B scaled Chi-Square? I'm currently running it under WLMSV. Thanks |
|
JI posted on Saturday, October 15, 2005 - 12:25 pm
|
|
|
Hi, I need to correct the last sentence. WLMSV should be changed to robust weighted least squares (the default for CFA with categorical observables). Btw, If I conduct an EFA with WLSMV, should I use the same estimator for CFA and the full latent variable model? Does it matter? Thanks once again. |
|
JI posted on Sunday, October 16, 2005 - 8:53 am
|
|
|
Hi, I've just ran the chi-square difference test by following the MPlus user's guide. Is it normal for the Ho model to be non-positive definite when the path between two latents is constrained to 1? Also, is it normal for the Ho & H1 model to have equal d.fs? I encountered this weird situation for only one of the latent pairs that I ran for discriminant validity. Thanks. Your answers will be deeply appreciated. |
|
|
Yes to your question in the first paragraph. WLSMV does not calculate degrees of freedom in the usual way. See formula 110 in Technical Appendix 4. With WLSMV, the value to look at is the p-value not the chi-square value or the degrees of freedom. |
|
Back to top |