It might be useful to do a variance decomposition to see how much of the variance is due to the cross-sectional dimension and how much is due to the time dimension.
If much of the variance comes from the cross-sectional dimension, you could use the between estimator or simply pooled OLS with clustered standard errors.
Patrick Gaul�
________________________________________
De : [email protected] [[email protected]] de la part de yjh jsh [[email protected]]
Date d'envoi : samedi 2 ao�t 2008 14:20
� : [email protected]
Objet : Re: st: a simple panel data question: FE and RE
This does helps. thanks
For me, it seems that different "levels" of Y in units are assumed to
be caused by those unobserved time-invariant variables. What if they
are not?
In my hypothetic case, for example, for both unit 1 and 2, we can't
see relationship between x and y if we only consider within-variation.
But this conceals the fact that Y takes higher value in unit 2 when 2
takes higher value in unit 2. That is, we do see a relationship that
higher x causes y. so, if FE a bad choice if we have a large between
variation compared to within variation in the data?
thanks
*
* 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/
