Anonymous posted on Monday, January 24, 2005 - 11:19 am
I was under the impression that the chi-square test statistic is normally defined as (N-1)*F, where F is the value of the maximum likelihood discrepancy function at the final estimates. However, in Mplus, the formula used seems to be N*F. I realize that for larger samples this will result in trivial differences. But for smaller samples, I wonder which expression is more accurate?
bmuthen posted on Monday, January 24, 2005 - 5:57 pm
I think it is correct to say that both n-1 and n have only large-sample rationale. Using n-1 is based on assuming that S (the sample cov matrix) is Wishart distributed, while using n is based on assuming the variables are normally distributed - a normal distribution assumption results in working with n. I am not aware of studies that have compared the two.
suprasith posted on Tuesday, January 25, 2005 - 6:05 am
Dear Prof. Muthen:
According to Bollen (1989), GFI can be computed when using ML, GLS, or ULS to estimate the model.
Is it appropriate to compute GFI as one of the fit indices when the MLM is used in the estimated model? If yes (any refs), how? Mplus does not report the GFI.
BMuthen posted on Tuesday, January 25, 2005 - 2:02 pm
I think GFI can be computed using MLM. Mplus does not report GFI because it appears that investigations in the literature do not favor this fit index.
Anonymous posted on Thursday, March 17, 2005 - 9:49 pm
AIC and BIC (along with loglikelihood values) should now be reported in version 3.12 for MLM. These were not computed in earlier versions.
Anonymous posted on Tuesday, April 19, 2005 - 8:31 am
I am currently running 3.11; does this mean that the AIC and BIC are not available for MLM or that syntax must be written to produce these estimates? If so, what syntax is needed? Thanks for your help.