Re: st: Goodness of fit of xtpcse

[Gordon Hughes <G.A.Hughes@ed.ac.uk>]

A somewhat belated answer to this question. The coefficients 
generated by -xtpcse- are identical to those generated by pooled OLS 
if autocorrelation is suppressed and identical to those generated by 
the pooled Prais-Winsten estimator if autocorrelation is 
permitted.  In each case the routine generates a corrected set of 
standard errors.  It follows that you can use either -regress- or 
-prais- to generate any measures of goodness of fit which depend upon 
the coefficients rather than their standard errors.

But ... you shouldn't rely upon log-likelihoods or associated 
measures for this purpose, because the statistical assumptions for 
-xtpcse- do not correspond to the assumptions used to construct the 
likelihood function for OLS or Prais-Winsten models..

A further qualification: the R^2 value reported by -xtpcse- can be 
greater than 1, though only, I think, when autocorrelation is 
included.  Hence, you should be careful about relying upon that as well.

Gordon Hughes
g.a.hughes@ed.ac.uk

==============
Dear in Stata-list

I am looking for a goodness of fit measure for pooled time
series models using xtpcse other than R^2.

Unfortunately, Stata does not provide log-likelihood values (for 
calculating BIC or AIC), am I right?


Is there a way to get log-likelihood valus when using xtpcse?


Regards

Talal

University of Southampton


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/