Akaike test and random effects

[Pasquale Pazienza]

Dear Statalisters,

I am analyzing a panel data (25 years x 30 countries) through OLS, FE, RE and mixed model (xtmixed). The results I get from the LM test and the Hausman show that I have to focus on the RE model. For the hoice of the best-fit model between the RE and the mixed model I have learned that the Akaike test should be implemented. Now, when I run the AIC / BIC test (through the use of commands "est stats" or "fitstat") for OLS, FE and mixed model I have estimated I get the result. I do not get any information when I run the AIC / BIC test for the RE model (apart from the no. of observations and the degree of freedom). I also tried various options to run the AIC test on the RE model but again I receive zero information. Do I miss to do something?
I would be grateful if you could help me to catch the right route. Thanks in advance.

[Carlo Lazzaro]

As per - help estimates_stats - AIC and BIC are appropriate for maximum likelihood models only (this is not the case for - xtreg, re - if - mle - option is not selected). Hence Pasquale can get both AIC and BIC information criteria if he types - xtreg depvar indepvar1 ...indepvarn, re mle - . Unfortunately, - mle - option does not support vce(cluster id), A possible solution is to impose vce(bootstrap), specifying a reasonable number of replications (e.g. 200 are usually enough for standard errors).
Hope this helps.
Kind regards,
Carlo