Very roughly speaking, fixed effect models are called "fixed" because there is some unobservable individual characteristics that don't change over time which you want to rule out from your estimation. So you are asking Stata to drop anything which is constant over time (either an outcome or an independent variable). This is the bad news. The good one is that you are allowing the independent variables to be correlated with the (fixed) effect.
I suspect you may go on with the Hausman test between RE and FE.
Please remember that conditional fixed effects estimated by Stata (maybe something is version 10 or 11 has changed, I have access to 9 only) are presumed to be not truly fixed, as reported in Allison & Waterman, 2002. ?Fixed effects negative binomial regression models.? Ross Stolzenberg (ed.), Sociological Methodology 2002. Boston: Basil Blackwell.
Nicola
P.S. I'll NOT receive/read any email but the Digest.
At 02.33 24/06/2009 -0400, Yee Kyoung Kim wrote:
>Dear statalists,
>
>I estimate my data with xtnbreg with -fe option and found that all
>zero outcomes were automatically dropped.
>Why does it drop such outcomes? Is it different from conventional fixed effects?
>I would like to compare the results of OLS, FE and RE negative
>binomial using hausman test.
>But, since OLS and RE keep all observations with all zero outcomes, it
>seems to be unreasonable to do it.
>In addition, if I exclude all zero outcome, shouldn't I doubt for the bias?
>
>Thank you in advance.
*
* 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/
