Dif in Dif with controls and fixed effects

[Manuel Marcone]

Hello Everyone,

I'm trying to replicate a Dif in Dif of a paper and my results are similar but not equal to the paper's.
The paper speciffied that they use controls, and fixed effects. Huber-White robust errors. and a weighted regression.

So the code I made is the following:


reg NotWorking treat policy inter gage man head couples level children_tot childrengage childrenlevel gagelevel i.sem i.dpto i.semdpto [pw=pesosem], robust

like..
(Y control_group treatment control_variables control_variables_interactions i.fixed_effects_variables i.fixed_effect_variables_interactions [pw=weighted], robust)


Please, correct me if I'm commiting any mistake.

Thank you!

[Kye Lippold]

It is hard to say anything for sure without knowing the exact details of the original model. But one possibility is that i.semdpto should not be included; a typical DiD includes fixed effects for unit and time, but not their interaction (it has the treatment indicator instead). What do -sem- and -dpto- represent?

[Manuel Marcone]

Thanks for replying.

I'll cite textually for the paper
"At and Bj are a full set of semester and state fixed effects. Finally Cjt are state-by-semester effects, introducted to control for state-specific shocks over the period that might be independient of national economic conditions"

I have a lack in Stata nomenclature.

Cause my regression have: treatment-variables, control-variables, fixed-effect-variables, interaction-between-variables

What`s the speciffication difference to include all them in the same regression to control for a DiD reg??

i don't really know which of this is correctly used by i.

[Kye Lippold]

I see, there are indeed unit-by-time interactions in your case. The standard way to specify these in Stata is with the "#" operator (type help fvvarlist in Stata):

reg NotWorking treat policy inter gage man head couples level children_tot childrengage childrenlevel gagelevel i.sem i.dpto i.sem#i.dpto [pw=pesosem], robust

rather than including a "semdpto" variable equal to sem * dpto.

If you are getting similar but not identical results, a common possibility is that the authors include different control variables (or interactions with those variables). But to help more, I would need a link to the paper in question, an example of your data (see help dataex) and the output of your regression.