Heather Schwartz <[email protected]>:
You can see what is in Stata 10 at http://stata.com/help.cgi?xtmixed
but weights are not there. -gllamm- (http://gllamm.org/) can do what
you want, as can other software packages, but have you considered that
you many not want it?
webuse nlswork, clear
g age2 = age^2
g tenure2 = tenure^2
xtmixed ln_w grade age* ttl_exp tenure* || id:
est sto ri
xtreg ln_w grade age* ttl_exp tenure*
est sto re
xtreg ln_w grade age* ttl_exp tenure* [iw=w], mle
est sto mle
xtreg ln_w grade age* ttl_exp tenure*, fe
est sto fe
ssc inst estout
esttab *
shows that the random intercept model gives essentially the same
results as xtreg, re or xtreg, mle (the latter allows iweights and
estimates the same variances as -xtmixed- but not on the log scale).
However,
hausman fe re
rejects that the assumptions for RE are justified. The FE model is
also much easier to use with pweights, for good reason.
If you do
xtmixed ln_w grade age* ttl_exp tenure* || id: grade
est sto rc
xtmixed ln_w grade age* ttl_exp tenure* || id: grade, cov(unstruct)
est sto rcu
esttab mixt mle rc rcu
you can see that the supposed improvements of the RC model over the
random intercept model are washed away when allowing for correlated
random effects. The ever more complicated models are identified by
ever stronger distributional assumptions that are rarely satisfied in
practice, IMHO.
On Feb 11, 2008 4:25 PM, Heather Schwartz <[email protected]> wrote:
> I would like to use sampling weights in a multi-level model (where the
> pweights are associated with observations in level 1 of the model). My
> understanding is that pweights is not an option for multilevel models in
> Stata 9. Is this still the case for Stata 10?
>
> Thanks,
> Heather
*
* 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/
