Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: meglm


From   Stas Kolenikov <skolenik@gmail.com>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   Re: st: meglm
Date   Wed, 16 Oct 2013 10:39:48 -0500

If this is a fractional response, then -meglm- is probably using the
convention of -probit- that a zero is a zero, and everything else is a
one. The likelihood of 0 indicates exactly that. -probit- gets around
it with a complicated heuristics of perfect prediction; -meglm- may or
may not be doing that, and numeric integration just blurs the things
by less-than-perfect calculation of the likelihood.


-- Stas Kolenikov, PhD, PStat (ASA, SSC)
-- Senior Survey Statistician, Abt SRBI
-- Opinions stated in this email are mine only, and do not reflect the
position of my employer
-- http://stas.kolenikov.name



On Wed, Oct 16, 2013 at 7:17 AM, Alfonso Sanchez-Penalver
<oneiros_spain@yahoo.com> wrote:
> I forgot to mention that ysm is a fractional response variable, which is why I'm trying -meglm- instead of -meprobit-.
>
> Alfonso Sánchez-Peñalver
>
>> On Oct 16, 2013, at 8:00 AM, Alfonso S <oneiros_spain@yahoo.com> wrote:
>>
>> Hi,
>>
>> I am trying to get my head around using the -meglm- command in Stata 13 for the mac, and I must be doing something wrong. The following are the results I get when running it:
>>
>> ----------------------------------------------
>> . meglm ysm ldis_totcurexpppa sch_enrlunsh lsch_enrtotal y_dum2 y_dum3 y_dum4 y_dum5 y_dum6 y_dum7 y_dum8 mldis_totcurexppp mlunch mlenrol || sprp_sch:, covariance(exchangeable) family(binomial) link(probit)
>>
>> Fitting fixed-effects model:
>>
>> Iteration 0:   log likelihood = -2347.0075
>> Iteration 1:   log likelihood =          0
>> Iteration 2:   log likelihood =          0
>>
>> Refining starting values:
>>
>> Grid node 0:   log likelihood = -8.835e-06
>>
>> Fitting full model:
>>
>> Iteration 0:   log likelihood = -8.835e-06  (not concave)
>> Iteration 1:   log likelihood = -1.147e-12
>> Iteration 2:   log likelihood = -1.142e-12
>>
>> Mixed-effects GLM                               Number of obs      =      6856
>> Family:                binomial
>> Link:                    probit
>> Group variable:        sprp_sch                 Number of groups   =       857
>>
>>                                                 Obs per group: min =         8
>>                                                                avg =       8.0
>>                                                                max =         8
>>
>> Integration method: mvaghermite                 Integration points =         7
>>
>>                                                 Wald chi2(0)       =         .
>> Log likelihood = -1.142e-12                     Prob > chi2        =         .
>> -----------------------------------------------------------------------------------
>>               ysm |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
>> ------------------+----------------------------------------------------------------
>> ldis_totcurexpppa |  -.6370373          .        .       .            .           .
>>      sch_enrlunsh |   3.177517          .        .       .            .           .
>>     lsch_enrtotal |   .0229953          .        .       .            .           .
>>            y_dum2 |    .699251          .        .       .            .           .
>>            y_dum3 |   .8369304          .        .       .            .           .
>>            y_dum4 |   .7515347          .        .       .            .           .
>>            y_dum5 |   .6936224          .        .       .            .           .
>>            y_dum6 |   .6023263          .        .       .            .           .
>>            y_dum7 |   .6499135          .        .       .            .           .
>>            y_dum8 |   .6696426          .        .       .            .           .
>> mldis_totcurexppp |   .3410811          .        .       .            .           .
>>            mlunch |   6.249509          .        .       .            .           .
>>           mlenrol |   .0017366          .        .       .            .           .
>>             _cons |   12.76935          .        .       .            .           .
>> ------------------+----------------------------------------------------------------
>> sprp_sch          |
>>         var(_cons)|   .5950633          .                             .           .
>> -----------------------------------------------------------------------------------
>> LR test vs. probit regression:       chi2(0) =     0.00   Prob > chi2 =      .
>>
>> Note: LR test is conservative and provided only for reference.
>> -----------------------------------------------
>>
>> To see that the variables were doing fine I ran the -xtgee- estimation and got normal results
>> -----------------------------------------------
>>
>> . xtgee ysm ldis_totcurexpppa sch_enrlunsh lsch_enrtotal y_dum2 y_dum3 y_dum4 y_dum5 y_dum6 y_dum7 y_dum8 mldis_totcurexppp mlunch mlenrol, family(binomial) link(probit) corr(exch)
>>
>> Iteration 1: tolerance = .21314375
>> Iteration 2: tolerance = .00124994
>> Iteration 3: tolerance = .00001481
>> Iteration 4: tolerance = 3.957e-07
>>
>> GEE population-averaged model                   Number of obs      =      6856
>> Group variable:                   sprp_sch      Number of groups   =       857
>> Link:                               probit      Obs per group: min =         8
>> Family:                           binomial                     avg =       8.0
>> Correlation:                  exchangeable                     max =         8
>>                                                 Wald chi2(13)      =    113.63
>> Scale parameter:                         1      Prob > chi2        =    0.0000
>>
>> -----------------------------------------------------------------------------------
>>               ysm |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
>> ------------------+----------------------------------------------------------------
>> ldis_totcurexpppa |   .2161359   .3463714     0.62   0.533    -.4627396    .8950114
>>      sch_enrlunsh |  -.0725626   .2464216    -0.29   0.768    -.5555401    .4104148
>>     lsch_enrtotal |  -.0349141   .1043114    -0.33   0.738    -.2393606    .1695323
>>            y_dum2 |  -.1418307   .0822838    -1.72   0.085    -.3031039    .0194425
>>            y_dum3 |  -.2154219   .0750023    -2.87   0.004    -.3624236   -.0684201
>>            y_dum4 |  -.1581733   .0651644    -2.43   0.015    -.2858932   -.0304533
>>            y_dum5 |  -.1202846   .0596719    -2.02   0.044    -.2372394   -.0033298
>>            y_dum6 |  -.0590736   .0576424    -1.02   0.305    -.1720506    .0539034
>>            y_dum7 |  -.0919327   .0537733    -1.71   0.087    -.1973266    .0134611
>>            y_dum8 |  -.1063621   .0504099    -2.11   0.035    -.2051636   -.0075606
>> mldis_totcurexppp |  -.0413327   .4116867    -0.10   0.920    -.8482237    .7655583
>>            mlunch |  -1.001581   .2810579    -3.56   0.000    -1.552444   -.4507177
>>           mlenrol |   .0000284   .0003227     0.09   0.930    -.0006039    .0006608
>>             _cons |  -.3508665   2.187324    -0.16   0.873    -4.637943     3.93621
>> -----------------------------------------------------------------------------------
>>
>> ---------------------------------------------------------------------------------------
>>
>> Can someone tell me if I am specifying the meglm command wrong? If not, why does it not reproduce the results from xtgee?
>>
>> Thanks,
>>
>> Alfonso.
>>
>>
>> *
>> *   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index