Re: st: estimating svy nbreg with replicate weights

[Stas Kolenikov <skolenik@gmail.com>]

I would imagine that the ML estimation for one of the subsamples does
not converge. This is somewhat odd, as with Fay's correction, you
should be using every observation with non-zero weights, so if it
converges in the main sample (with weights), it should be converging
in the BRR subsamples, as well. (If you have zero weights for some
observations, then hypothetically you can run into a situation where
you may not have certain industries or regions present in a given
subsample, so Stata won't be able to estimate that particular model.)
I would suggest that you re-specify your full model as

nbreg numcat i.metro i.region i.union i.occup i.industry

and try running it with fewer factorial variables. (There will be a
caveat, though: if you do have no observations of a given industry in
a given BRR subsample, then Stata will not estimate its coefficient,
and will probably declare the results from this subsample to be
defective, showing this as e or x instead of a dot in the replication
output.)

In fact, it may suffice to run something like

svy : mean numcat, over(industry)

etc. to see if that runs into variance estimation problems. If it
does, then whatever factorial variable is being used in the -over()-
option must be causing -nbreg- to fail, too.

Now that I think about it, I had troubles with weighted ML estimates,
too: http://www.stata.com/statalist/archive/2011-05/msg01412.html. Try
to change the convergence criteria in -nbreg- options first, and if
that does not work, look for the underidentified variables.

On Mon, Apr 9, 2012 at 12:50 PM, Alison Earle <earle@brandeis.edu> wrote:
> I seek suggestions for how to successfully estimate nbreg using the svy
> command and replicate weights on a dataset (from the CPS) with approximately
> 38K observations. There are 160 replicate weights, the model has
> approximately 30 independent variables, and the dependent variable is a
> count with integer values ranging from 0 to 3 with cases somewhat even
> distributed across the categories.
>
> When I submit the following code to estimate my model, Stata will hang up
> after a few replications and never produce estimates.  I let the program run
> over night and only a total of 4 replications were completed.
>
>
> . svyset [iw=wtsupp], brrweight(repwtp1-repwtp160) vce(brr) fay(0.5)
>
>       iweight: wtsupp
>           VCE: brr
>           MSE: off
>     brrweight: repwtp1 repwtp2 repwtp3 repwtp4 repwtp5 repwtp6 repwtp7
> repwtp8 repwtp9 repwtp10 … repwtp160
>           fay: .5
>   Single unit: missing
>      Strata 1: <one>
>          SU 1: <observations>
>         FPC 1: <zero>
>
> svy: nbreg numcat metro1 metro3 metro4 region2 region3 region4 union01
> occup2 occup3 occup4 occup5 occup6 industry1 industry3 industry4 industry5
> industry6 industry7 industry8 …
>
> BRR replications (160)
> ----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
> ....
>
> This is as far as it gets.
>
> I am using Stata 12.1/SE.  Would Stata MP be able to produce estimates?  Is
> there anything else I should do differently to get this model to be
> estimated successfully?
> Thanks.
>
> Alison
>
> Alison Earle, Ph.D.
> Senior Scientist
> Institute on Child, Youth and Family Policy
> The Heller School for Social Policy
> Brandeis University
> 415 South Street
> Waltham, MA
> 781-736-3918
>
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/



-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/