Fixed effects estimation cannot estimate the effects of variables that are constant over time for all cases. That includes the constant term in a regression. These variables are tossed out. Yet, statistics packages that do fixed effects estimation typically report a constant term. What is it? It is the mean of the individual parameters for the fixed effects. William Greene's Econometric Analysis, 4th edition, points out that often one may want to know about these fixed effects, whether they are significantly different from one another. (Packages more commonly report whether the effects themselves make a significant contribution collectively, a question that may be of less interest). This can be tested by estimating a dummy variable regression with a constant term in the equation, and one case omitted. Comparing the R-square for this equation with the R-square for the restricted equation without the dummies tests whether the fixed effect parameters are significantly different!
from one another. David Greenberg, Sociology Department, New York University.
----- Original Message -----
From: RonDorsey <[email protected]>
Date: Friday, August 9, 2002 8:26 pm
Subject: st: xtreg fixed effects
> Dear Statalist(ers)
>
> Can you put me right on something re: fixed effects on panel data?
>
> I've always thought (probably wrongly) that running fixed effects
> (xtreg,fe) is the same as running OLS (including group dummies) on
> the same model.
>
> If I run both separately, I do get the same results except that:
>
> 1. dummy coefficients from OLS are not the same as the fixed
> effects ui
> from xtpred
>
> and
>
> 2. the constant terms (and their significance) are different.
>
> I presume this may be because in OLS I have to exclude one group
> dummy.
> My questions are:
>
> a) Are the group dummies and fixed effects coefficients
> equivalent in some
> way?
>
> b) Should I really use the ui from xtreg, fe and if so how can I get
> standard errors so I can test significance for each group?
>
> c) Whether or not the constant term is significantly different
> from zero
> has important implications for my hypothesis testing (it is in OLS
> but not
> for fixed effects. Can you confirm which I should be using?
>
>
> Many thanks in advance for your assistance.
>
> Ron Dorsey
>
>
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
