csdid and aggregations for each period - see estimates for the pre-treatment periods

[Stefan Sliwa]

Dear Statalist,

I am applying csdid to estimate the ATT for each period, across all cohorts, with the option "agg(calendar)".

Below an exemplary output:

"

csdid Y, ivar(i) time(t) gvar(gvar) long agg(calendar)
...............
Difference-in-difference with Multiple Time Periods

                                                        Number of obs = 31,368
Outcome model  : regression adjustment
Treatment model: none
------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
    CAverage |   .0102665      .0027     3.80   0.000     .0049747    .0155584
          T4 |   .0130415   .0026837     4.86   0.000     .0077815    .0183015
          T5 |   .0193123   .0033636     5.74   0.000     .0127198    .0259048
          T6 |  -.0015542   .0032392    -0.48   0.631    -.0079029    .0047945

Why is it that the output does not show the pre-treatment periods, in my example T1-T3? Is there a way to produce these coefficients even when aggregating on the period/calendar level (and not on the default cohort level, for which the command produces estimates for all available time periods)?

I hope my inquiry is comprehensible, if not please let me know. Thank you very much!

[FernandoRios]

Because you are requesting CALENDAR aggregation
Those estimates are literally the impacts for those units treated during period T4 T5 T6
What you may want its EVENT

[Stefan Sliwa]

Hi Fernando,

Thank you very much for your answer.
I understand that I am aggregating effects on the calendar, i.e. on the period level. I am doing this because I want to analyze the effect of a policy on election results, I am less interested in cohort-specific effects but rather I want to know the aggregate effect that a policy had on an event (here: elections) across all treated cohorts.

Perhaps I have some issues conceptualizing it, but why is it not possible to estimate coefficients for events prior to treatment implementation, to use them as placebo events? One option could be to aggregate the placebo effect across all cohorts, since they are all equally not yet treated. Am I thinking wrongly here?

Thank you for your help.

[Stefan Sliwa]

Apologies, I read too quickly! I will read up the differences between EVENT and GROUP!