Treatment vs Control graph with FE

[Paulo Andre]

Hello guys,

I'm trying to do a graph (treat x control / x-axis: year by year / y-axis: Outcome) of a DiD with a couple of FE. It is a "treatment x Post", where "treatment" has staggered adoption dummy (may starts equals 0, then turn 1 in the next period and then become 0 again) and "post" is a dummy for the last period.

I tried these possibilities:

1) Use "collapse" and "twoway" to do a simple graph but this doesn't take FE into account.

2) "Margins" it worked with something like this:

reghdfe outcome i.treatment##i.year if var1==1 & var2==1, a(firstFE secondFE third FE) cluster(group)
margins year, at(treatment = (0 1)) noestimcheck
marginsplot, xdimension(year)

But I'm not sure about the interpretation of this and if it is exactly what I want. I read the paper about it, but no quite sure about the interpretation.

3) I can do an "xtdidregress".

xtdidregress (outcome) (treatment) if var1==1 & var2==1 , nogteffects group(group) time(year)
estat trendplots

But when I do the "estat trendplots" after this. It says "treatment assignment times vary" because of the restricted sample (this is probably easy to solve). And when I use the option "nogteffects" to don't automatically create group and time FE it says that the plot it is not allowed with this option.

4) "csdid" is only valid with single adoption period for treatment, right?



Someone know how could I do a graph like this?

[Paulo Andre]

Just a sample of the data

[Paulo Andre]

And an example of the graph

[Justin Niakamal]

Here's a start

reghdfe outcome i.treatment##i.year if var1==1 & var2==1, a(firstFE secondFE third FE) cluster(group)
margins year#treatment
marginsplot, xdimension(year)

You might consider having your if call before just to not have those observations in the margins calculations (or use if e(sample)).

[Paulo Andre]

Here's a start

reghdfe outcome i.treatment##i.year if var1==1 & var2==1, a(firstFE secondFE third FE) cluster(group)
margins year#treatment
marginsplot, xdimension(year)

You might consider having your if call before just to not have those observations in the margins calculations (or use if e(sample)).

Thank you!