The predict command following nbreg will give you the expected number
of events, or counts. So, with y as the response term and xvars as the
predictors, type the following on the command line:
. nbreg y xvars
. predict mu
where mu is the term I used for the fitted or predicted count. If you
want to use all of the residuals and fit tests that come
with using the -glm- command, then following the use of -nbreg-, type
glm y xvars, fam(nb `e(alpha)')
predict mu2 /// or whatever postestimaton
statistic you wish; this is same as mu above.
The `e(alpha)' captures the return code saved by -nbreg- for the value
of the ancillary parameter. If you forget, just use the value of alpha
displayed in the -nbreg- output. So
glm y xvars, fam(nb .33333)
The log link is the default link with the negative binomial family with
-glm-, which is the same parameterization as -nbreg-. It does not need
to be typed as an option.
Joseph Hilbe
[email protected]
========================================================
Date: Sat, 20 Jun 2009 22:20:50 -0400
From: [email protected]
Subject: st: computing expected values for GLM with nbreg
Hello,
I am running a negative binomial model - because my data are time
series I
switched to running the model through the GLM command so I could
control for
autocorrelation and heterosk. w/ HAC variance estimates. I want to
compute
expected values to make the interpretation a little more tranparent.
When using
the nbreg command previously I used clarify to get expected values.
Clarify
won't run after GLM, and looking through STATA help I see no
postestimation
command for getting expected values. Does anyone know how I might go
about
getting expected values through STATA with this set up -- would I be
better of doing hand calculations?
Thanks for any help,
Ashlie Delshad
*
* 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/
