Treatment anticipation in difference-in-differences with csdid

[Guy Low]

Dear Statalist,

I am working with the csdid package to perform a difference-in-differences with multiple time periods. I am mainly interested in the dynamic effects results and plotting them in event-plots, since treatment is given throughout the time-frame of the dataset (15 year-periods).

However, I can clearly and consistently see what appears to be treatment anticipation two periods prior to treated groups receiving treatment.

The original paper on which csdid is based (Callaway & Sant'Anna, 2021, Difference-in-differences with treatment anticipation) clearly indicates that limited treatment anticipation does not need to violate the identification assumptions of ATT. I have gone through the documentation of csdid and online resources to find how I can allow for limited treatment anticipation using csdid, but to no avail.

My csdid command is as follows, looping for dependent variables v and specific treatments k, and without covariates:

csdid `v', ivar(ID) time(YEAR) gvar(treatyear_`k') cluster(NUTS2_cat) method(reg) level(99)

I ask for suggestions on whether it is possible to include treatment anticipation in the above command (and how). If not, what can I - reasonably - do to address anticipatory behaviour in the treated groups?

Many thanks,

Guy Low

[FernandoRios]

The only way is to do it manually
Specifically, change your gvar so it accounts for anticipation
gen gvar2 = (gvar-1)*(gvar>0)

[Guy Low]

Thank you, Fernando. Will try that.