Fixed-Effects model and Entities with No Variation

Asked Feb 19 '19 at 12:58

Active Feb 19 '19 at 12:58

Viewed 272 times

Suppose I am interested in the effect of some treatment $T_{it}$.

I consider the following simple fixed effects model:

$y_{it}=b_0+b_1T_{it}+\gamma_i +e_{it}$ where $\gamma_i$ is individual-level fixed effects.

For some group of individuals, I observed two situations: with and without treatments.
For the other group of people, they didn't get any treatment in entire sample period.
If I include the group of people who didn't get any shock in the regression, my estimate $b_1$ seems to shrink to zero.

Could you explain why?

asked Feb 19 '19 at 12:58

user3509199

Just to confirm: you are using a fixed effect model with fixed effect transformation? I.e. not a random effect model? – Andreas Dzemski Feb 19 '19 at 13:59
Yes, this is not a random effect model. – user3509199 Feb 19 '19 at 14:11
This is strange. I think your estimate shouldn't change at all. – Andreas Dzemski Feb 19 '19 at 14:13
That is exactly what I thought.. – user3509199 Feb 19 '19 at 14:22

0 Answers0