Re: st: Why does streg allow both state fixed-effects and time invariant state variables?

[David Jacobs <jacobs.184@sociology.osu.edu>]

If I'm reading you right, you've entered 49 state dummies in an 
equation together with some explanatory variables.  If you estimate 
with OLS, that is precisely how fixed-effects estimation in panel 
models works (although the standard errors may a bit off).

The point is that doing this does not lead to perfect collinearity as 
long as the estimator drops one of the state dummies (or the 
intercept).  Doing this will cause the estimation routine to drop any 
explanatory variables that don't change over time, but the 
coefficients on explanatory variables that do change next can be estimated.


At 07:07 PM 2/18/2010, you wrote:
> Hi Stata users,
> I am running a Gompertz survival model with no time varying 
> variables.  I am looking at mortality within states and I include 
> indicator variables for each state. In one model, I accidently 
> included both the state indicator variable AND a state level 
> variable. In the OLS world that would be a case for perfect 
> collinearity and the coefficient on the state-level variable would 
> not have been estimated, but in this case it was! My gut tells me 
> that the estimates are wrong/uninterruptable statistical garbage... 
> Am I right ?
>
> Please an help would be great!


*
*   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/