Dear Austin,
Many thanks for your answer.
Did you mean running the RE xtlogit JUST those who were included in the FE
xtlogit (given that Y was time-variant for them) and then proceeding to do
the Hausman test?
Sorry if I did not understood you well.
By the way, there is an easy, straightaway procedure to save the "estimation
sample"?
Many thanks again
Luis Ortiz
-----Mensaje original-----
De: [email protected]
[mailto:[email protected]] En nombre de Austin Nichols
Enviado el: viernes, 25 de enero de 2008 14:25
Para: [email protected]
Asunto: Re: st: random and fixed effects (when the all the outcomes are
either positive or negative)
Luis--
You've dropped time invariant X, but not time-invariant Y, to make
your comparison. Just run FE first and save the estimation sample to
restrict to when you run RE. FE may still be the better method--it is
more often consistent, anyway...
On Jan 25, 2008 5:26 AM, Maarten buis <[email protected]> wrote:
> As this is adressed to me I will just answer that I don't know the
> solution. This does not indicate that the solution is either difficult
> or easy (or non-existent), it is just an indication that I don't use
> panel data in my research so my active knowledge on the subject is
> limited.
>
> -- Maarten
>
>
> --- Luis Ortiz <[email protected]> wrote:
>
> > Dear Maarten,
> >
> > In the link you recommended to Federica, Dave Jacobs gives the
> > expected
> > explanation:
> >
> > "(...) probably because your dependant variable counts for some cases
> > don't
> > change over time".
> >
> > I've got a similar problem for a research I'm carrying out on the
> > likelihood
> > of being over-educated using the European Community Household Panel.
> > It's a
> > different research from the one I've asked before to the Statalist
> > (and
> > Stephen Jenkins has been so kind to respond).
> >
> > In this case, I'm just looking at the state, not a process of
> > transition
> > from one situation to another; in other words, I'm not doing survival
> > analysis, but just cross-sectional panel data analysis.
> >
> > I've tried to run the Hausman test, in order to decide if I may keep
> > the
> > random effects model (my initial choice) or I should go for a fixed
> > effects
> > model. I've been careful to exclude any time-invariant covariate.
> >
> > Yet, the results of output of xtlogit (fe) tell me precisely this:
> >
> > note: 6395 groups (12643 obs) dropped due to all positive or
> > all negative outcomes.
> >
> > Quite likely, there are a LOT of individuals (groups) for which all
> > the
> > observations in the panel reveal that they are EITHER over-educated
> > or not.
> >
> > What to do in this case? Does it reveal that the fixed-effects is the
> > wrong
> > model to specify?....
> >
*
* 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/
*
* 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/
