Re: st: Fixed effect regression with and without state fixed effects

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Tue, 14/12/10, Rijo John wrote:
> I have a balanced panel from 2000-2009 on 51 states. The
> variable I am interested in is x1.
> x2-x4 are control variables and are largely state specific.
> When I ran the following command
> 
> xtreg Y x1 x2 x3 x4, fe
> 
> it returned a significant negative sign for the
> coefficient. However, the sign I was expecting was positive.
> I did the same regression without the state fixed effect
> using OLS and year dummies as follows.
> 
> reg y x1 x2 x3 x4 i.year
> 
> this regression returned a positive (as expected) and
> significant sign for the coefficient of x1 while the signs 
> and significance of other variables in the model remained 
> the same. Now I am wondering which model to go with.

Whether or not you want to control for state (i.e. use fixed
effects regression) is a substantive question. No amount of
statistics or testing or model fitting can help you with that.

One strategy that is often useful is think of your effect as
the result of a hypothetical experiment. An effect is nothing
else as a comparison of groups, say men and women. Imagine
you are allmighty and you have no moral scruples. What is the
"treatment" to which you want to subject your subjects. This 
will tell you what you do and what you do not want to control 
for.

Another useful strategy is think in terms of confounding and
intervening variables. You want to control for confounding 
variables, but you do _not_ want to control for intervening
variables.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------


      

*
*   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/