Residual variance PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Dmitriy Poznyak posted on Thursday, July 28, 2011 - 10:38 pm
Dear Linda and Bengt,

I am fitting a multilevel model with moderation effects on the 2nd level. Because of a small number of clusters (17) and complexity of the model I use Bayesian estimator. The model seems to fit fine (based on traceplots), but there's one question that bothers me. Some of the residual variances are significant, although I can't see why this is the case.

For instance, for the observed (2nd level) indicator X, I get the fully standardized variance estimate = .021, S.D. = .026, 95% C.I. = [0.001 - 0.096], yet the p-value is .00. In another case, S.D. is also actually larger than the variance coef., and again, the p-value is significant.

This is only the case with the observed dependent variables on the 2nd level which seem to have very high loading into the. between-level factor (around .989 and .992). For the other 2 indicators with lower factor loading everything seems to be fine.

Please advice if I am missing something here.

Thanks in advance for your feedback,
 Bengt O. Muthen posted on Thursday, July 28, 2011 - 11:24 pm
Note that the Bayesian p-value that is given is the part of the posterior distribution that is below/above zero for a positive/negative estimate. So your zero p-value for the variance means that there is no part of the posterior that is below zero. I would simply report the 95% CI as you mention.
 Dmitriy Poznyak posted on Friday, July 29, 2011 - 4:43 am
Dr. Muthen,

Thank you for your prompt reply. This makes perfect sense now.
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message