controlling for fixed effects

[Red Taho]

Hi,

This may be a weird question but I am trying to run a Difference in differences model in which I have two years of data and the treatment occurs in the second year. My post variable takes a value of 1 if the observation is from the second year and 0 otherwise. I was running my model and noticed something odd. Why is it that when I run the model: reg outcome DD post treat i.year i.month i.state i.year#i.month i.year#i.state and reg outcome DD post treat i.month i.state i.year#i.month i.year#i.state , I get different results? Does not the post variable account for year fe anyways?

[Clyde Schechter]

Does not the post variable account for year fe anyways?

Yes, it does. That's not the source of the different results.

The different results arise because interactions, i.year#i.month and i.year#i.state, mean different things when i.year is also present and when it is not. There is an algebraic transformation between these models so that one could calculate the results of either from the other, but it is not a simple one, and it is by no means just the identity function.