Yiu-Fai Yung <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...
> In summary...the usaul ML fitting without missing data in SEM is > indeed FIML under the normality assumption. It is only done more > efficiently using sufficient statistics. There is a parallel situation > in regression. To estimate the regression coefficients you only need > the covariance matrix of all variables. Thank you very much for this good explanation. Perhaps the poster's point would be that FIML of raw data gives one, in principle, the opportunity to reject a model for distributional reasons: that is, it combines a test of MV normality assumptions with the test of the SEM. To follow up on the regression analogy, note that mixture-of-regression models are done via FIML on raw data, not on covariances. The same would be true of mixture-of-SEMs. But how to assess model with raw-data FIML? One option is to use parsimony indices like the AIC and BIC. -------------------------------------------------------------------------------- John Uebersax, PhD (858) 597-5571 La Jolla, California (858) 625-0155 (fax) email: [EMAIL PROTECTED] Statistics: http://ourworld.compuserve.com/homepages/jsuebersax/agree.htm Psychology: http://members.aol.com/spiritualpsych -------------------------------------------------------------------------------- . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
