Statalist


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

Re: st: mi estimate and xtmixed: reporting of groups


From   ymarchenko@stata.com (Yulia Marchenko, StataCorp LP)
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: mi estimate and xtmixed: reporting of groups
Date   Thu, 01 Oct 2009 12:54:08 -0500

Fred Wolfe <fwolfe@arthritis-research.org> asks if -mi estimate- intentionally
omits the group information when used with -xtmixed-:

> I wonder if I am missing something obvious.
>
> xtmixed haq_ duration i.cocat ib3.q4_onsetyear fage i.sex ib2.edcat
> ib45000.totinco if _mi_m==2 || patkey: duration , cov(uns) mle
> 
> reports  the Group variable," "Number of obs," and the "Number of groups"
>
> However, mi estimate, cmdok dots):
> xtmixed haq_ duration i.cocat ib3.q4_onsetyear fage i.sex ib2.edcat
> ib45000.totinco if _mi_m==2 || patkey: duration , cov(uns) mle
>
> reports the number of obs, but not the group variable or the number of
> groups. I suppose the number of groups could differ according to each
> imputation. But it would be helpful to have some information about
> groups, for if I come back to the data later with -mi estimate using,
> I may have forgotten the group N. So is it intended that we should
> calculate the group N before running the mi estimate, or is there some
> simple way I have overlooked to get this information?

Fred is not missing anything obvious.  -mi estimate- is an estimator on its
own and produces generic output, containing information specific to
multiple-imputation analysis, for all estimation commands.  When used with
survey estimation commands, -mi estimate- does report additional information
about the survey design.  We are considering customizing the output of -mi
estimate- to include some additional information specific to other supported
estimation commands in the future.

Fred also asks if there is a way to get information specific to the used
estimation command after -mi estimate-.  In general, the answer is 'yes' as
long as the results are constant across imputations.  For certain estimation
results ( e.g., -e(N)-, -e(N_strata)-) you can also access their varying
values.  See the section 'Saved Results' in '[MI] mi estimate' for details.

Having said that, there are estimation results which cannot be retrieved after
-mi estimate- for technical reasons.  The group information saved by -xtmixed-
in the -e(N_g)- matrix is one of them.  We will add an option for saving
requested estimation results from each imputation in a future update.
Meanwhile, Fred needs to access this information manually.  He can specify the
-noisily- option with -mi estimate- to see the output from each -xtmixed- run.
To access -xtmixed-'s estimation results Fred can use the -mi xeq- command to
run -xtmixed- on a chosen imputation.

Estimation results specific to estimation commands are preserved after -mi
estimate- as long as they are constant across imputations.  Estimation results
which vary across imputations are set to missing and their names are saved in
macros -e(names_vvs_mi)- (for scalars), -e(names_vvl_mi)- (for macros), and
-e(names_vvm_mi)- (for matrices).  Values of certain varying estimation
results (e.g. -e(N)-, -e(N_strata)-) are returned as matrices.  Also, some
estimation results, (e.g., -e(ll)-, -e(deviance)-) are always reported as
missing after -mi estimate-.


-- Yulia
ymarchenko@stata.com
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index