Callaway and Sant'Anna Diff in Diff with continuous Treatment

[Emmanuel Ngoy]

Hello,
I want to estimate the effect of childhood exposure to conflict intensity on adult height using a DID with continuous treatment.
I have used cohort groups to capture exposure to conflict at different ages, with 3 treated groups : age 0-5, age 6-11 and 12-17 at the onset of conflict (1996).
individuals between the ages of 18 and 33 at the beginning of conflict are the control groups.

Here is how I generate the gvar
gen first_treat=.
replace first_treat=0 if ages>18
replace first_treat=7 if ages>=1 & ages<=7
replace first_treat=13 if ages>=13 & ages<=18

Note ages correspond to the age of the individual, which I include as time.

runing the Callaway and Sant'Anna Diff in Diff

csdid mtheight conflict, time(ages) gvar(first_treat) method(dripw), The model cannot be identified since I have Units always treated.

I guess the problem is in the way I define the gvar. Is this the right way to generate the gvar variable in a repeated cross section dataset ?

Thank you in advance.

Emmanuel

[FernandoRios]

Hi Emmanuel
two points
1. CS cannot be used for cases with continuous treatment. At best you can adjust it so you estimate effects for treatments with selected intensities, assuming the intensity doesn't change. And using different models to do so.
2. the best way to see if your gvar is correctly identified is:
tab ages first_treat

If you do not see observations for every combination of ages and first treat, then your gvar is incorrectly identified.

Also, if you have repeated crossection probably your "time" variable will be the "year" of the survey. not the age of the person.
for your gvar

gen gvar1 = 1996 if inrange(age_in_1996,0,5)
replace gvar1 = 0 if inrange(age_in_1996,18,33)

gen gvar2 = 1996 if inrange(age_in_1996,6,11)
replace gvar2 = 0 if inrange(age_in_1996,18,33)

gen gvar3 = 1996 if inrange(age_in_1996,12,17)
replace gvar3 = 0 if inrange(age_in_1996,18,33)

I think that would work for your case
F

[Emmanuel Ngoy]

Hi Fernando.
Thank you for your comment.

I have used age of individuals because the years of the survey are 2007 and 2013.
Yet, the individuals in my sample are born between 1963 and 1996. Could year of birth used as time?


I have also checked if gvar is not correctly identified and I realized that is not


tab ages first_treat

| first_treat
ages | 0 1 7 13 | Total
-----------+--------------------------------------------+----------
1 | 0 742 0 0 | 742
2 | 0 851 0 0 | 851
3 | 0 654 0 0 | 654
4 | 0 995 0 0 | 995
5 | 0 996 0 0 | 996
6 | 0 939 0 0 | 939
7 | 0 0 999 0 | 999
8 | 0 0 871 0 | 871
9 | 0 0 971 0 | 971
10 | 0 0 1,191 0 | 1,191
11 | 0 0 1,194 0 | 1,194
12 | 0 0 1,125 0 | 1,125
13 | 0 0 0 958 | 958
14 | 0 0 0 1,000 | 1,000
15 | 0 0 0 1,000 | 1,000
16 | 0 0 0 651 | 651
17 | 0 0 0 767 | 767
18 | 0 0 0 721 | 721
19 | 752 0 0 0 | 752
20 | 686 0 0 0 | 686
21 | 661 0 0 0 | 661
22 | 710 0 0 0 | 710
23 | 653 0 0 0 | 653
24 | 599 0 0 0 | 599
25 | 656 0 0 0 | 656
26 | 349 0 0 0 | 349
27 | 522 0 0 0 | 522
28 | 436 0 0 0 | 436
29 | 509 0 0 0 | 509
30 | 384 0 0 0 | 384
31 | 374 0 0 0 | 374
32 | 421 0 0 0 | 421
33 | 477 0 0 0 | 477
34 | 163 0 0 0 | 163
-----------+--------------------------------------------+----------
Total | 8,352 5,177 6,351 5,097 | 24,977


I have also tried to generate gvar the way suggested, but the model cannot be identified since I still have units always treated.

What could be right way to construct the gvar given the structure of my data? Are there any stata packages suitable for DID continuous treatment, Csid cannot be used for continuous treatment?

Thank you