st: RE: RE: -xtmixed- and multilevel data [Was: Grouping income variables- RECODE COMMAND]

["Antonio Rodriguez Andres" <Antonio.Andres@emu.edu.tr>]

Dear Prof. Jenkins

Thank you very much for your valuable input. In my empirical application,
what matters is to explore the effect of children on depression. That is,
the effect of one of the individual predictors of depression. It seems to me
that I should estimate FE models or OLS regressions for each country.

Regards

Antonio

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of
S.Jenkins@lse.ac.uk
Sent: Thursday, February 06, 2014 11:32 AM
To: statalist@hsphsun2.harvard.edu
Subject: st: RE: -xtmixed- and multilevel data [Was: Grouping income
variables- RECODE COMMAND]

"Antonio Rodriguez Andres" <Antonio.Andres@emu.edu.tr>:

You are indeed stuck with 23 countries. The cited papers make 2 suggestions
that may be relevant to you:

(1) consider whether you are really interested in getting good estimates of
the effects of the individual-level predictors, or whether the country
effects are integral to your research project. If the former, then you can
exploit the large number of persons per country. And there are several
modelling approaches at your disposal, including FE or separate regressions
for each country.

(2) If country effects are integral, consider whether Bayesian approaches
are feasible in your case: there is some Monte Carlo evidence that the
estimates of country effects derived using these perform better in the small
number of countries case. See the references cited in the paper. As a Stata
user, you might combine, say, -runmlwin- (on SSC) with MLwiN. Or similar
wrappers that call WinBugs.

Good luck
Stephen

------------------------------

Date: Wed, 5 Feb 2014 13:30:48 +0200
From: "Antonio Rodriguez Andres" <Antonio.Andres@emu.edu.tr>
Subject: st: RE: -xtmixed- and multilevel data [Was: Grouping income
variables- RECODE COMMAND]

Dear Stephen,

Your feedback is much appreciated. Based on your research paper, the results
are no longer valid with 23 countries. I am stacked with this issue and how
to proceed. Maybe a good starting point is to replicate your table 3. My
dependent variable is the depression score and the key variable of interest
is having children in home and see how its effect differs across gender.