Message/Author |
|
Sanjoy posted on Thursday, April 21, 2005 - 8:53 pm
|
|
|
Prof. Muthen Can we run Melton and Liang's GEE approach In MPlus, (mine is MPlus version 3.12), the reason I'm asking so because in your 1997 article, u have done a comparative study of ur proposed estimator in comparison to their GEE ... if we can then how In my model, indicator variables are all ordered categorical? Thanks and regards |
|
Sanjoy posted on Thursday, April 21, 2005 - 9:16 pm
|
|
|
Dear Proffesor ... In connection to my earlier message ... I forgot to mention the main reason why I'm looking for Melton and Young's(M&Y) GEE approach ... in ur 97 article, page 24, you have mentioned M&Y's approach performed better than Robust WLS particularly when the sample size is small ... I have two sample size, one is 240 and the other one is 139 (my models are SEM with covariates and categorical indicator outcomes variables) thanks and regards |
|
Sanjoy posted on Thursday, April 21, 2005 - 9:20 pm
|
|
|
sorry ... the word "Professor" has been misspelt :-) |
|
|
No, Mplus does not use the GEE approach. We don't feel there is enough of an advantage over robust WLSMV. We also have full-information maximum likelihood for categorical outcomes which theoretically should perform as well if not better than GEE. |
|
Sanjoy posted on Friday, April 22, 2005 - 9:40 am
|
|
|
Thanks... one more quick question what should be our estimator choice in order to run full-information ML for Categorical Outcome ... is it ML, MLR or MLF (page 366, MPlus manual) I suppose my understanding could be wrong, please correct it if it so.... isn’t the case that we use Limited Information ML (I mean separate equation wise) for the first two-stages of 3-stage estimation procedure propounded by Dr. Muthen (1983,1984) thanks and regards |
|
|
Our default estimator for maximum likelihood is MLR. Your understanding is correct. |
|
Sanjoy posted on Friday, April 22, 2005 - 5:51 pm
|
|
|
Thanks Madam ... have a nice weekend |
|
Back to top |