David Jacobs replied to Ghislain Dutheil:
> Economists often use -xtgls- and manually enter the dummies for cases (and
> years if you need 2-way fixed effects).
>
> Another possibility is to use xtreg, fe and cluster on case id. This will
> correct the t-values for serial correlation, but you probably don't have
> enough cases in each cross-section. This approach might work if you enter
> less than about 10 explanatory variables, however.
>
> By the way, the -xtreg, fe- routine in Stata 10 will warn you if you've
> exceeded the limits on the number of explanatory variables when you cluster
> on case id and the matrix is not full rank. But Stata 8.2 will not.
Ghislain didn't make clear how her dependent variable is measured; I
shall assume it's continuous and without bounds.
She reported that in her dataset, N=14 and T=57. If by this, she means
that she only has 14 panel cases measured across 57 time-points
(assuming these are equally spaced), then I'm pretty sure that any
cross-sectional GLS estimates will likely be pretty unreliable as
compared to those reported in OLS-PCSE models. Indeed, this is the
very reason one would fit the latter kind of model to such data.
Nevertheless, I would suggest fitting all three models mentioned by
David to your data, -estout- the model output and then eyeball the
estimates across the columns to judge which model appears closest to
your theoretical expectations.
Hope that helps.
--
Clive Nicholas
[Please DO NOT mail me personally here, but at
<[email protected]>. Please respond to contributions I make in
a list thread here. Thanks!]
"My colleagues in the social sciences talk a great deal about
methodology. I prefer to call it style." -- Freeman J. Dyson.
*
* 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/
