On 16/05/07, Ghita, Mihaela <[email protected]> wrote:
I have a similar problem. I am working with panel data and I have discovered
problems of heteroskedasticity and autocorrelation in my data. Thus, I have
used xtgls..........panel (hetero) corr (ar1). However, I am not very sure
this is the model I should use. I have also tried using random effects with
robust standard errors as well as the Mundlak approach (including the within
group mean of the independent variables). When using the last 2 approaches,
the Hausman test (for FE and RE) allows using RE models.
Yet, the results from FGLS are better. But can I use those results or is
better to rely on RE models?
Much of it depends on how many observations (or, perhaps more
accurately, degrees of freedom) you have to play with.
In general, the larger your df, and the more confident you are that
your data was drawn at random from the population of origin (or, even
better, it _is_ the population!), the more reliable your parameter
estimates, standard errors and model diagnostics are.
So it goes for any kind of regression model, really.
--
Clive Nicholas
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
