[Luca Toni]

I'm still looking for answers to my 2 questions above. If anyone can help me figure it out that, it will be highly appreciated.
Thanks.

[George Ford]

1. Do you mean whether the treatment is endogenous? Do you have some reason to believe receiving the treatment is based in some way on the outcomes?

2. After eventdd, you are supposed to be able to -estat lags- to test all the lag coefficients (or -estat leads- for the leads). I can't get it to work. Instead, you can run the reghdfe command (using keepdummies from eventdd) and test the lags (testparm lag*, or testparm lead* for parallel paths).

clear all
use http://www.damianclarke.net/stata/bacon_example.dta, clear
gen t2treat = year - _nfd
save junk, replace
eventdd asmrs pcinc asmrh cases , timevar(t2treat) method(hdfe, absorb(stfips year) cluster(stfips)) keepdummies

reghdfe asmrs pcinc asmrh cases  lead* lag*,absorb(stfips year) cluster(stfips) 
testparm lag*
global L lag0
forv i = 1/27 {
    global L $L + lag`i'
}
lincom ($L)/28

[Luca Toni]

1. Do you mean whether the treatment is endogenous? Do you have some reason to believe receiving the treatment is based in some way on the outcomes?

2. After eventdd, you are supposed to be able to -estat lags- to test all the lag coefficients (or -estat leads- for the leads). I can't get it to work. Instead, you can run the reghdfe command (using keepdummies from eventdd) and test the lags (testparm lag*, or testparm lead* for parallel paths).

clear all
use http://www.damianclarke.net/stata/bacon_example.dta, clear
gen t2treat = year - _nfd
save junk, replace
eventdd asmrs pcinc asmrh cases , timevar(t2treat) method(hdfe, absorb(stfips year) cluster(stfips)) keepdummies

reghdfe asmrs pcinc asmrh cases lead* lag*,absorb(stfips year) cluster(stfips)
testparm lag*
global L lag0
forv i = 1/27 {
global L $L + lag`i'
}
lincom ($L)/28

Thank you.

1) Yes, I mean checking whether the treatment is endogenous. I don't necessraily have a particular reason, I just want to check it formally. How can I get any indication, if at all?

2) I can run estat lags after eventdd and it works, but I wanted to know how I can find the ATT, not whether the lag coeffficients are significant.

[George Ford]

1. I don't think there's a test in this framework.
2. The code provide in #33 will get you the average effect after treatment and test it.

[Luca Toni]

1. I don't think there's a test in this framework.
2. The code provide in #33 will get you the average effect after treatment and test it.

2. Thank you! I got a significant ATT coefficient, although pvalue is 0.027 (not very significant but acceptable).

[Luca Toni]

1. I don't think there's a test in this framework.
2. The code provide in #33 will get you the average effect after treatment and test it.

Hey George Ford,

I would like to plot the control group pre-post trends along with the treatment group. Is there any built-in command for that? what's the most effective way to code it in the staggered DiD approach?

[George Ford]

clear all
use http://www.damianclarke.net/stata/bacon_example.dta, clear
gen t2treat = year - _nfd
g treated = !mi(t2treat)
eventdd asmrs pcinc asmrh cases , timevar(t2treat) method(hdfe, absorb(stfips year) cluster(stfips)) 

lgraph asmrs year, by(treated)

[Luca Toni]

clear all
use http://www.damianclarke.net/stata/bacon_example.dta, clear
gen t2treat = year - _nfd
g treated = !mi(t2treat)
eventdd asmrs pcinc asmrh cases , timevar(t2treat) method(hdfe, absorb(stfips year) cluster(stfips))

lgraph asmrs year, by(treated)

Thanks.

I'm not satisfied with the plot when I use year variable. My dependent variable is volatile so that's the reason I guess:

[George Ford]

Year makes no sense, does it. You have a lot of different "treatment" groups.

Look over this.

https://causalinf.substack.com/p/step-3-in-pedros-diff-in-diff-checklist

I recommend subscribing to Scott's substack.

[Luca Toni]

Year makes no sense, does it. You have a lot of different "treatment" groups.

Look over this.

https://causalinf.substack.com/p/step-3-in-pedros-diff-in-diff-checklist

I recommend subscribing to Scott's substack.

I tried registering for the free trial but still can't access to this.
I agree that year doesn't make sense, but can I use the period? t2t gets missing values for untreated, so not sure if it's correct to plot it with the period. Or maybe I'm completely wrong and there is another approach that I'm missing out.
By the way, if there is another source regarding this issue, I would like to read.

[George Ford]

clear all
use http://www.damianclarke.net/stata/bacon_example.dta, clear
gen t2treat = year - _nfd
g treat = t2treat>0
save junk, replace
eventdd asmrs pcinc asmrh cases , timevar(t2treat) method(hdfe, absorb(stfips year) cluster(stfips)) keepdummies

collapse (mean) asmrs , by(_nfd year)
xtset _nfd year
xtline asmrs , ov

[Luca Toni]

clear all
use http://www.damianclarke.net/stata/bacon_example.dta, clear
gen t2treat = year - _nfd
g treat = t2treat>0
save junk, replace
eventdd asmrs pcinc asmrh cases , timevar(t2treat) method(hdfe, absorb(stfips year) cluster(stfips)) keepdummies

collapse (mean) asmrs , by(_nfd year)
xtset _nfd year
xtline asmrs , ov

After running it, I got this plot:

Although my goal was to see how, for a given treatment year/period, the control group behaves compared to the treatment group (both in the same plot).

[George Ford]

need to get the missings in there (the controls). I'll fiddle with it in a bit.

[Luca Toni]

need to get the missings in there (the controls). I'll fiddle with it in a bit.

Ok, thanks. Please let me know if you find out.

[George Ford]

Here's a sketch of something that may help guide you.

I create a control mean of the not-yet-treated (you could also create one of never treated), and then means for each treatment group.

clear all
use http://www.damianclarke.net/stata/bacon_example.dta, clear
gen t2treat = year - _nfd

egen controlmean = mean(cond(t2treat<0,asmrs,.)), by(year)
tab _nfd
foreach t in 1969 1970 1971 1972 1973 1974 1975 1976 1977 1980 1984 1985 {
    egen treatmean`t' = mean(cond(_nfd==`t',asmrs,.)), by(year)
    g diffmean`t' = treatmean`t' - controlmean
}

foreach t in 1969 1970 1971 1972 1973 1974 1975 1976 1977 1980 1984 1985 {
    twoway connected controlmean year , sort || connected treatmean`t' year , sort xline(`t') name(group`t', replace)
    twoway connected diffmean`t' year , sort xline(`t') name(group`t'_diff, replace)
}