Re: st: grouping, negative binomial regression, and margins

["Roger B. Newson" <r.newson@imperial.ac.uk>]

I would second Nick's advice. If you want to fit a model for each 
region, then you should use a loop over regions. As in:

levelsof region, lo(regions)
foreach RE of num `regions' {
 nbreg NUMTA i.NUMDS if region==`RE', nolog
 margins NUMDS
}

It might be even better if you used the -marglmean- package, 
downloadable from SSC, to get confidence intervals for marginal means. 
-marglmean- is a wrapper for -margins- that adds a Normalizing and 
variance-stabilizing log transform, which is a good idea with 
negative-binomial marginal means. As in:

levelsof region, lo(regions)
levelsof NUMDS, lo(nume)
foreach RE of num `regions' {
 nbreg NUMTA i.NUMDS if region==`RE', nolog
 foreach NDS of num `nums' {
  marglmean, at(NUMDS=`NDS') eform
}

This will probably give more sensible confidence intervals. I myself 
always use -marglmean- for marginal means after a negative-binomial 
regrassion.

I hope this helps.

Best wishes

Roger


Roger B Newson BSc MSc DPhil
Lecturer in Medical Statistics
Respiratory Epidemiology and Public Health Group
National Heart and Lung Institute
Imperial College London
Royal Brompton Campus
Room 33, Emmanuel Kaye Building
1B Manresa Road
London SW3 6LR
UNITED KINGDOM
Tel: +44 (0)20 7352 8121 ext 3381
Fax: +44 (0)20 7351 8322
Email: r.newson@imperial.ac.uk
Web page: http://www.imperial.ac.uk/nhli/r.newson/
Departmental Web page:
http://www1.imperial.ac.uk/medicine/about/divisions/nhli/respiration/popgenetics/reph/

Opinions expressed are those of the author, not of the institution.

On 10/08/2013 23:54, Nick Cox wrote:
> It is standard that when you run
>
> by <varlist>: <command>
>
> Stata runs through the possibilities but only results for the last
> model fitted remain in memory. In your case the results in memory are
> for the last region fitted and those are what -margins- uses.
> Nick
> njcoxstata@gmail.com
>
>
> On 10 August 2013 23:04, Ahmed Al Attar <ahmedwalattar@gmail.com> wrote:
> > Hi,
> >
> > I am working on a project where I have a bivariate model. All of my data are counts and I have verified using countfit that the best model to use is the Negative binomial regression.
> >
> > My Dependent Variable is called NUMTA (with values one of 0,1,2,3,4,5,6)
> > My Independent Variable is called NUMDS (with values either 0, 1 or 2)
> > The study takes place over 10 different regions, each with counts of NUMTA and NUMDS which brings us to a total of 870 observations.
> >
> > The commands that I have entered into stata are:
> >
> > sort region
> >
> > by region: nbreg NUMTA i.NUMDS, nolog
> >
> > The output is a negative binomial regression by region as you can see below. This is for 1 of my 10 regions:
> >
> > Negative binomial regression                      Number of obs   =         87
> >                                                    LR chi2(1)      =       0.50
> > Dispersion     = mean                             Prob > chi2     =     0.4810
> > Log likelihood = -34.128129                       Pseudo R2       =     0.0072
> >
> > ------------------------------------------------------------------------------
> >         numta |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
> > -------------+----------------------------------------------------------------
> >       1.numds |  -12.98646   1298.721    -0.01   0.992    -2558.434    2532.461
> >         _cons |  -2.044954   .3103769    -6.59   0.000    -2.653282   -1.436627
> > -------------+----------------------------------------------------------------
> >      /lnalpha |  -.7772627   3.388113                     -7.417842    5.863317
> > -------------+----------------------------------------------------------------
> >         alpha |   .4596625   1.557389                      .0006004    351.8895
> > ------------------------------------------------------------------------------
> > Likelihood-ratio test of alpha=0:  chibar2(01) =    0.11 Prob>=chibar2 = 0.367
> >
> > However, when I follow this up with the command:
> >
> > margins NUMDS
> >
> > then I get the following output:
> >
> > Adjusted predictions                              Number of obs   =         87
> > Model VCE    : OIM
> >
> > Expression   : Predicted number of events, predict()
> >
> > ------------------------------------------------------------------------------
> >               |            Delta-method
> >               |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
> > -------------+----------------------------------------------------------------
> >         numds |
> >            0  |   .0657895   .0310757     2.12   0.034     .0048822    .1266967
> >            1  |   .1111111   .1214722     0.91   0.360    -.1269701    .3491923
> >            2  |   .4999992   .6852692     0.73   0.466    -.8431037    1.843102
> > ------------------------------------------------------------------------------
> >
> > The above output says 87 observations, when I know it should show predictions for 870 observations instead. Which leads me to believe that it has done the margins command for only 1 of my 10 regions.
> >
> > Is there a way to do the margins command using (by region:) in order to get 10 different margins outputs? When I try to run the command by region: margins NUMDS state says I cannot run it. If not, is there any other course of action you would advise?
> >
> > Thank you very much for your support,
> >
> > Ahmed Al-Attar
> > RCTC-Organization
> > *
> > *   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/