John and Stefan- Thank you both for your comments and advice.
I agree that fixed effects may not be the most efficient method
because as you say few people do actually experience marital
transitions, or changes in the other predictors, over the relatively
short time span of 5 years.
I ran the Hausman and it came out statistically significant (p<0.05)
which means that endogeneity is a indeed a problem in the model.
I got hold of the STATA multi-level book by Rabe-Hesketh and Skrondal
(2008, page 119-120) and i think I will go with their suggestion to
use a random-intercept model (using xtmixed or xtreg, mle) and include
all original time-varying variables plus the cluster means for all
time-varying variables. This way I will be able to look at both within
and between variation for those variables, while dealing with some of
the endogeneity problem, as I understand it.
Thanks again for your help,
I find this stuff quite confusing and appreciate being pointed in the
right direction.
Cheers
Anna
On Tue, Jan 20, 2009 at 12:49 AM, Stefan Kreisel <[email protected]> wrote:
> Try something like:
>
> xi: xtmixed Attitude i.Mar_status*Wave || Id:
>
> Here you are using a mixed-effects model as John pointed out, with intercepts (the || part and what follows to the right) being allowed to vary randomly within each individual (you could also include random coefficients . Allowing marital status to vary as time passes (i.e. Wave) would be what the interaction term does (i.e. Mar_status*Wave expands to include the main effects).
>
> That's an answer, now here's a question to those listening:
>
> What does one do with random effects that vary over time (i.e. change in cluster membership)?
> Taking the above example, say city of residence is something one wanted to look into.
>
> I'd could do something like (may not be the best example...):
>
> xi3 i.Mar_status*Wave*City
> xi: xtmixed Attitude Wave City _I_* || Id:
>
> but wouldn't below be the correct alternative?
>
> xi: xtmixed Attitude Wave City _I_* || City: || Id:
>
> Regards
>
> Stefan
> *
> * 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/
>
*
* 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/
