Do I need to control additonal time-invariant characteristics in DID?

[Gary Sean]

All,

Today I received the comments that I need to run the models e.g., teacher- or school-fixed, in my DID specification. To my knowledge, the coefficient of interaction term (variable of interest) in DID should be not changed, and teacher or school-fixed model will only affect the intercepts and SEs. Is this common approach? or is it redundant?

Looking for your advice. Thank you!

[Clyde Schechter]

You do not say enough about your situation for anyone to comment whether the advice to add teacher or school-fixed effects is appropriate or ill-advised. Most, but not all, DID analyses are estimated with fixed effects.

To my knowledge, the coefficient of interaction term (variable of interest) in DID should be not changed, and teacher or school-fixed model will only affect the intercepts and SEs.

That is absolutely not true. Anything and everything can change. The changes can be small or large, and even involve a change of sign. In fact, it is impossible to predict what the output of an FE regression will be from knowing what the regression without FE showed.

[Jeff Wooldridge]

Gary: Do you have two-period panel data? So you observe the same schools before and after the intervention? If so, the simple DID estimator is identical to the fixed effects estimator.

[Dilini Jayasinghe]

Hi,

I'm new to Statalist and I hope it's alright to post a follow-up question. I am using xtdidregress command to identify the causal effect of a policy on school achievement. My control group is students in Victoria (Australia) and treated group is students in New South Wales. I want to know if I should include time invariant controls in this did regression model as they would anyway get wiped out by the did regression

My command is as follows:

xtset hicid year
xtdidregress (schl_achieve post exposed x1) (did), group(state) time(year)

where x1 = time varying controls

Kind regards
Dilini

[Clyde Schechter]

#4 is a duplicate post. Asked and answered starting at #18 in https://www.statalist.org/forums/for...s-in-did/page2.

Please do not do this. Post your question and then wait an appropriate amount of time for a response. In most situations 24 hours would be appropriate, longer perhaps on weekends and holidays. If you need a response in a matter of minutes, while you might get lucky and get one here, that is not usually the case--Statalist is not the right venue for urgent questions.

If you do not receive a response after a reasonable amount of time, then it is reasonable to re-post. But reposting the same question will probably result in getting no response yet again. Most questions that go unanswered are unclearly worded or presented with insufficient information, or, less commonly, nobody else on the Forum knows the answer, or nobody else on the Forum is interested in it. So rather than just reposting the same question, you should try to understand why it drew no response and modify it accordingly before reposting.