Create a group variable first_treat indicating the first year when each unit becomes treated

[Jules brenda]

I am trying to create a group variable first_treat indicating the first year when each unit becomes treated, to perform a DID analysis. There is variation in treatment timing, and therefore I need to define a variable "first-treat" reporting for each ID the year when they first became treated (treatment = 0 if not treated, 1 otherwise). In case the units are never treated, the value of first_treat will be zero. I report below a simplified dataframe: I have the variables ID, Year, and Treatment. I need to create the variable first.treat as follows.

a	2016	0	2017
a	2017	1	2017
a	2018	1	2017
b	2016	1	2016
b	2017	1	2016
b	2018	1	2016
c	2016	0	2018
c	2017	0	2018
c	2018	1	2018
d	2016	0	0
d	2017	0	0
d	2018	0	0

What is the best way to do this in stata? Thankyou in advance.

[Carlo Lazzaro]

Jules:
welcome to this forum.
You may want to consider something along the following lines (the variable in the rightmost column was not considered as I do not get its role here):

. bysort id (Year): g wanted=sum(Treatment) if Treatment==1

. g first=1 if wanted==1

. drop wanted

. list

     +------------------------------+
     | id   Year   Treatm~t   first |
     |------------------------------|
  1. |  a   2016          0       . |
  2. |  a   2017          1       1 |
  3. |  a   2018          1       . |
  4. |  b   2016          1       1 |
  5. |  b   2017          1       . |
     |------------------------------|
  6. |  b   2018          1       . |
  7. |  c   2016          0       . |
  8. |  c   2017          0       . |
  9. |  c   2018          1       1 |
 10. |  d   2016          0       . |
     |------------------------------|
 11. |  d   2017          0       . |
 12. |  d   2018          0       . |
     +------------------------------+

.

[Nick Cox]

clear
input str1 id year treatment want
a    2016    0    2017
a    2017    1    2017
a    2018    1    2017
b    2016    1    2016
b    2017    1    2016
b    2018    1    2016
c    2016    0    2018
c    2017    0    2018
c    2018    1    2018
d    2016    0    0
d    2017    0    0
d    2018    0    0
end

bysort id : egen wanted = min(cond(treatment == 1, year, .))
replace wanted = 0 if wanted == .

list, sepby(id)

     +--------------------------------------+
     | id   year   treatm~t   want   wanted |
     |--------------------------------------|
  1. |  a   2016          0   2017     2017 |
  2. |  a   2017          1   2017     2017 |
  3. |  a   2018          1   2017     2017 |
     |--------------------------------------|
  4. |  b   2016          1   2016     2016 |
  5. |  b   2017          1   2016     2016 |
  6. |  b   2018          1   2016     2016 |
     |--------------------------------------|
  7. |  c   2016          0   2018     2018 |
  8. |  c   2017          0   2018     2018 |
  9. |  c   2018          1   2018     2018 |
     |--------------------------------------|
 10. |  d   2016          0      0        0 |
 11. |  d   2017          0      0        0 |
 12. |  d   2018          0      0        0 |
     +--------------------------------------+

[Jules brenda]

Thank you very much, this was really helpful!

[Jules brenda]

I have an additional question, does anyone have a suggestion on how to perform a DID analysis with multiple treatment periods? Because when I generate a variable: relative treatment time (generated by subtracting the variable wanted from the survey year), I have the treatment group alligned where 0 is the time where the shock took place. But for the control group, just the survey year is available.

Thank you very much in advance for helping, I really appreciate it.

[Nick Cox]

That's quite a different question. Please start a new thread with a suitable title.

[Jon Thomson-Settle]

Hi all, apologies for posting this thread, but this was a question similar to mine. I am creating minimum years for first_treat for house transactions to apply csdid. The grouping is the same house sold in different transactions.

I applied the formula below, but it worked sporadically, I thought at first it was where there was only one group transaction, but that doesn't seem to be it. Could anyone help?


bysort objectid : egen ftreat_10km = min(cond(treated_10km == 1, year, 0))

Thanks!

Data example below

objectid	addedyear	treated_10km	ftreat_10km
1	1997	0	0
1	2005	1	0
2	2012	1	2012
3	2000	0	0
5	1996	0	0
5	1998	0	0
7	1993	0	0
13	2008	1	2008
14	2022	1	2022
15	1999	0	0
15	2022	1	0

[Nick Cox]

The main idea in #7 seems right, but I can't explain your result for objectid 1.

* Example generated by -dataex-. For more info, type help dataex
clear
input byte objectid int addedyear byte treated_10km int ftreat_10km
 1 1997 0    0
 1 2005 1    0
 2 2012 1 2012
 3 2000 0    0
 5 1996 0    0
 5 1998 0    0
 7 1993 0    0
13 2008 1 2008
14 2022 1 2022
15 1999 0    0
15 2022 1    0
end

bysort objectid : egen wanted = min(cond(treated_10km, addedyear, .))

list, sepby(objectid)

     +----------------------------------------------------+
     | objectid   addedy~r   treate~m   ftreat~m   wanted |
     |----------------------------------------------------|
  1. |        1       1997          0          0     2005 |
  2. |        1       2005          1          0     2005 |
     |----------------------------------------------------|
  3. |        2       2012          1       2012     2012 |
     |----------------------------------------------------|
  4. |        3       2000          0          0        . |
     |----------------------------------------------------|
  5. |        5       1996          0          0        . |
  6. |        5       1998          0          0        . |
     |----------------------------------------------------|
  7. |        7       1993          0          0        . |
     |----------------------------------------------------|
  8. |       13       2008          1       2008     2008 |
     |----------------------------------------------------|
  9. |       14       2022          1       2022     2022 |
     |----------------------------------------------------|
 10. |       15       1999          0          0     2022 |
 11. |       15       2022          1          0     2022 |
     +----------------------------------------------------+