Compute Post_avg-Pre_avg

[Wenzhi-Ding]

Hi,

Thanks for this great package for implementing CS DID. When I try to use it in my project, I found after estimating ATT, there is no command to calculate post_avg_att - pre_avg_att.

For example, if the allocation of treatment group and control is not random, and such factor affects dependent variable (as the following generated data shows)

The estimated results will be

I may overlook something, but is it possible to generate a line using the pre-treatment ATT as benchmark, so that it will be clearer that whether the treatment effects are significantly different from zero. Same problem also applies to the results in table.

Attached are code and data I used:

data.csv

library("did")
library("tidyverse")

df = read.csv("data.csv")

out = att_gt(
  yname = "y1",
  gname = "treat",
  idname = "id",
  tname = "year",
  xformla = ~x,
  data = df,
  est_method = "dr"
)

dyn = aggte(out, type = "dynamic")
summary(dyn)
ggdid(dyn)

Thank you so much!

Best regards,
Dave

[Wenzhi-Ding]

I cross check with Sun and Abraham (2021)'s command in Stata (eventstudyinteract), it can produce correct event study plot and obtain ATT unaffected by the non-random treatment allocation.

Using the same data I uploaded above.

gen control = treat == 0
gen T = year - treat
replace T = . if control == 1
tab T

forvalues l = 0/6 {
	gen L`l'event = T == `l'
}

forvalues l = 1/7 {
	gen F`l'event = T == -`l'
}

eventstudyinteract y1 L*event F*event, absorb(id year) cohort(treat) control_cohort(control) covariates(x)

event_plot e(b_iw)#e(V_iw), default_look stub_lag(L#event) stub_lead(F#event) together