We get values for constant term and residuals in Stata after using xtmixed command. However, how do we know whether these values are significant or not?
The random effects model variance has a lower boundary of zero as variance cannot be negative and therefore chi square distribution values won't be applicable. Is there any other way to check the significance level of random effects model parameters using xtmixed command?
The issue is that the sampling distribution of the variance estimate is likely to be positively skewed as the estimate cannot go below zero so therefore asymptotic normality is unlikely except if the variance estimate is a long way from zero.
Other software does give standard error estimates and allows testing, albeit with a lot of warning and caveats.
You can run MLwiN from within Stata and this will do the testing, you can also us mcmc estimation so as to get a 'picture' of the posterior distribution of the variance estimates which may give asymmetric credible intervals reflecting the skewed distribution.
This give detail on MLwiN's testing and mcmc estimation
Book Developing multilevel models for analysing contextuality, he...
and this has details on the MLwiN Stata link
http://www.bristol.ac.uk/cmm/software/runmlwin/
- Badges
- Science topic
- Similar topics
- Condensed Matter Physics
- Zero