Re: st: How to interpret results from gllamm

[Rebecca Pope <rebecca.a.pope@gmail.com>]

Hi Jurijs,
To add to Stas's comments... If you haven't done so, I advise reading
the GLLAMM Manual (Rabe-Hesketh, Sophia; Skrondal, Anders; and
Pickles, Andrew, "GLLAMM Manual" (October 2004). U.C. Berkeley
Division of Biostatistics Working Paper Series. Working Paper 160.
http://biostats.bepress.com/ucbbiostat/paper160). The authors detail
the interpretation of the output of -gllamm- for many different
models. Even if you can't find your exact model there, you can usually
find an example of something similar that will help. Note that
-gllamm- is a user-written command. Depending on what version of Stata
you have, much of what it accomplishes is now done within official
Stata.

Also, since you are new to multilevel models, I highly recommend
Multilevel and Longitudinal Modeling Using Stata, 3rd Ed. by Sophia
Rabe-Hesketh and Anders Skrondal. It is available from Stata Press at
http://www.stata.com/bookstore/multilevel-longitudinal-modeling-stata.
I used it in a course last semester & found the text quite
approachable. It will also help you decide when to use the Stata
command and when to use -gllamm- since sometimes one outperforms the
other.

Hope this helps,
Rebecca



On Tue, Nov 6, 2012 at 12:33 PM, Stas Kolenikov <skolenik@gmail.com> wrote:
> The parameters var(1) and var(2) are your random intercepts and
> slopes. The individual level coefficients have the same interpretation
> as they would in a regular logistic regression: an increase of the
> explanatory variable by 1 causes the linear prediction shift by {the
> value of the regression coefficient}, and the change in probability
> depends on the particular constellation of variables quantifiable via
> marginal effects (and -gllamm- may not work very well with -margins-
> that otherwise provides a great interface to describe and visualize
> these marginal effects).
>
> --
> -- Stas Kolenikov, PhD, PStat (SSC)  ::  http://stas.kolenikov.name
> -- Senior Survey Statistician, Abt SRBI  ::  work email kolenikovs at
> srbi dot com
> -- Opinions stated in this email are mine only, and do not reflect the
> position of my employer
>
>
>
> On Tue, Nov 6, 2012 at 12:01 PM, Jurijs Ņikišins <jurijs.nikisins@lu.lv> wrote:
>> Hello,
>> I'm a newcomer to both Stata and multilevel analysis and I have some general understanding of theory but implementing it in practice is a real challenge for me so far, so I'd be really grateful for help on interpreting results I get from gllamm.
>> Using the European Social Survey 25-country dataset, I'm studying the relationship between dichotomous outcome variable demonstration_rec (whether a person took part in a demonstration last year)
>> and the following independent vars: gender, education, index of attitudes to gender, cultural and income equality (resp. geq_mean, ceq_mean, ieq_mean) and 4-rank democratic history variable new_demhist denoting period that a country has been a stable democracy.
>> I treat new_demhist as a country-level variable, allowing it to vary at a country level (i.e. trying to build a random-coefficient model):
>>
>> gllamm demonstration_rec Gender_rec Education_rec geq_mean ceq_mean ieq_mean i.new_demhist, family(binomial) link(logit) i(country_rec) nrf(2) eqs(cntry_cons cntry_democr) nip(8)
>> i.new_demhist     _Inew_demhi_1-4     (naturally coded; _Inew_demhi_1 omitted)
>> -----------------------------------------------------------------------------------
>> demonstration_rec |      Coef.          Std. Err.           z        P>|z|     [95% Conf. Interval]
>> ------------------+----------------------------------------------------------------
>>        Gender_rec |   .1956246          .0404499     4.84     0.000     .1163443    .2749049
>>     Education_rec |   .2037315        .0162273    12.55    0.000     .1719265    .2355365
>>          geq_mean |    .2762476         .0232011    11.91    0.000     .2307744    .3217209
>>          ceq_mean |     .1077089         .0102165    10.54     0.000     .0876849    .1277329
>>          ieq_mean |      .3145431          .0246306    12.77    0.000     .2662681    .3628181
>>     _Inew_demhi_2 |  -.4142843    .1031448    -4.02    0.000    -.6164443   -.2121243
>>     _Inew_demhi_3 |   .7189537    .0954098     7.54     0.000     .5319539    .9059535
>>     _Inew_demhi_4 |  -.0171796    .0929596    -0.18    0.853     -.199377    .1650178
>>             _cons |       -5.994235          .1476284   -40.60    0.000    -6.283581   -5.704889
>> -----------------------------------------------------------------------------------
>>  Variances and covariances of random effects
>> ------------------------------------------------------------------------------
>> ***level 2 (country_rec)
>>      var(1): .20405822 (.10364398)
>>     cov(2,1): .00757171 (.01157615) cor(2,1): .03609139
>>      var(2): .21568821 (.02322925)
>> ------------------------------------------------------------------------------
>> My questions are:
>>
>> 1) How actually should I interpret var(1) and var(2)? Are they individual- and country-level variance, or variances of intercept and slope?
>> 2) How do I interpret individual-level coefficients together with level 2 variances and covariances?
>>
>> Thanks a lot in advance,
>>
>> Jurijs Nikisins
>> Sociology PhD student, University of Latvia
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/faqs/resources/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/faqs/resources/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/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/