Issue with CSDID command

[Laurence Jones]

Hi, I am having a problem with the csdid command, it will work for years but not for age:

My dataset:

id	year	age	treatment_age	treatment_year	Outcome
5	1989	43			0
5	1991	45	53	1999	0
5	1993	47	53	1999	0
5	1995	49	53	1999	0
5	1997	51	53	1999	0
5	1999	53	53	1999	1
5	2001	55	53	1999	0
5	2003	57	53	1999	0
6	1989	42			1
6	1991	44	50	1997	0
6	1993	46	50	1997	1
6	1995	48	50	1997	0
6	1997	50	50	1997	1
6	1999	52	50	1997	1
6	2001	54	50	1997	1
6	2003	56	50	1997	0

Option 1: By year:

csdid Outcome if id == 5 | id == 6, [ivar(id)] time(year) gvar(treatment_year) agg(event)

Outcome:

No never treated observations found. Using Not yet treated data
...xxxxxxxxx
Difference-in-difference with Multiple Time Periods

                                                Number of obs     =          8
Outcome model  : regression adjustment
Treatment model: none
------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     Pre_avg |          0  (omitted)
    Post_avg |          1          .        .       .            .           .
         Tm4 |          1          .        .       .            .           .
         Tm2 |         -1          .        .       .            .           .
         Tp0 |          1          .        .       .            .           .
------------------------------------------------------------------------------
Control: Not yet Treated

Option 2: By age:

csdid Outcome if id == 5 | id == 6, [ivar(id)] time(age) gvar(treatment_age) agg(event)

Outcome:

No never treated observations found. Using Not yet treated data
xxxxxxxxxxxxxxxxxxxxxxxxxx                  mean():  3301  subscript invalid
                 aggte():     -  function returned error
              makerif2():     -  function returned error
                 <istmt>:     -  function returned error

Surely, I should be getting the same result, irrespective of whether I use age or year in this example. I need to run it by year for my dataset but it always drops all the observations and errors.

All help is greatly appreciated, many thanks in advance!

[Jared Greathouse]

Why would you declare year as your time series variable

[Laurence Jones]

Because the command doesn't appear to work with age

[FernandoRios]

The problem is that Year is common to all, Age is not
Specifically, there are only odd numbers in years . But you have odd and even numbers in age. So , because of that, CSDID is trying to work with a "delta" of 1 instead of 2.
F

[Laurence Jones]

FernandoRios thank you so much! this is a very helpful answer and I have fixed the issue with this knowledge! much appreciated!

[Laurence Jones]

many thanks again, I had a very quick follow up question FernandoRios . If I am using bi-annual data like this, with forced bi-annualised ages. (so age=37 or 38, is new_age=38, etc) I am getting delta of Tp1, Tp2, Tp3 etc in the csdid output, does this delta now reflect 2 year periods or 1 year? and if it is 1 year, how can it be calculating this using bi-annual data?

[Laurence Jones]

solved it, my treatment variable wasn't bi-annualised