Statalist


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

Re: st: Re: Prediction after xtpoisson, fe


From   "David W. Harless" <dwharles@vcu.edu>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Re: Prediction after xtpoisson, fe
Date   Wed, 22 Jul 2009 11:34:42 -0400

david reinstein wrote:
Thanks for the suggestions, but none of this seems to work so far.

"predict yhatnu0, nu0 " gives estimates that are much too low. (When
exponentiated, they are still too low, and with virtually no
correlation to the outcomes. And it is still ignoring the fixed
effect, which I want to be part of this prediction.)

Anyway, I find that exponentiating the result of "predict yhattotxb,
xb" yields the yhatnu0 measure; so nu0 seems to be already
exponentiating, but the fixed effect seems to be always thrown away!

Anyone know how to recover the fixed effect and use it in generating
predictions? Or where to look for more information about this
procedure.

-xtpoisson- works by conditioning out the fixed effects, thus no option to use -predict- to get predictions inclusive of the estimated fixed effects.

But there is another possibility: the fixed effects Poisson model is equivalent to use of -poisson- with dummy variables for cross-sectional units (a special case among non-linear models where this is true). Hence, if the number of cross-sectional units is not too large, you could accomplish what you wish using -predict, n- after -poisson- with dummy variables.

I hope this helps,
Dave Harless
*
*   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