Hi, I am using the csdid command.
Specifically, my data structure is like below:
That is, there are 16 time periods, and the group variable includes never-treated (g = 0), always-treated (g = 1), and ever-treated (g = 2, ..., 15) groups.
Obviously, without the never-treated group, ATT(g, 15), ATT(g, 16), and ATT(15, t) are not estimable, and the following result is consistent with my understanding:
The x marks indicate ATT(g, 15), ATT(g, 16), and ATT(15, t).
But, in this setup, when I add the control variables, the csdid command fails to estimate most of the ATT(g, t):
The global $X is the names of the (time-invariant) control variables.
If I include the never-treated group in the comparison group, the result becomes
What is the problem in the second result...??
Specifically, my data structure is like below:
Code:
. tab Gvar
Gvar | Freq. Percent Cum.
------------+-----------------------------------
0 | 10,077 27.93 27.93
1 | 2,626 7.28 35.21
2 | 2,929 8.12 43.32
3 | 2,759 7.65 50.97
4 | 3,141 8.71 59.68
5 | 2,243 6.22 65.89
6 | 2,379 6.59 72.48
7 | 2,319 6.43 78.91
8 | 1,944 5.39 84.30
9 | 1,357 3.76 88.06
10 | 989 2.74 90.80
11 | 745 2.06 92.87
12 | 669 1.85 94.72
13 | 507 1.41 96.13
14 | 513 1.42 97.55
15 | 885 2.45 100.00
------------+-----------------------------------
Total | 36,082 100.00
. tab time
| Freq. Percent Cum.
------------+-----------------------------------
1 | 1,699 4.71 4.71
2 | 2,422 6.71 11.42
3 | 2,642 7.32 18.74
4 | 2,745 7.61 26.35
5 | 2,767 7.67 34.02
6 | 2,695 7.47 41.49
7 | 2,652 7.35 48.84
8 | 2,538 7.03 55.87
9 | 2,501 6.93 62.80
10 | 2,346 6.50 69.31
11 | 2,276 6.31 75.61
12 | 2,082 5.77 81.38
13 | 1,999 5.54 86.92
14 | 1,690 4.68 91.61
15 | 1,638 4.54 96.15
16 | 1,390 3.85 100.00
------------+-----------------------------------
Total | 36,082 100.00
Obviously, without the never-treated group, ATT(g, 15), ATT(g, 16), and ATT(15, t) are not estimable, and the following result is consistent with my understanding:
Code:
. csdid Y if Gvar != 0, ivar(pid) time(time) gvar(Gvar) notyet long2 Units always treated found. These will be ignored Panel is not balanced Will use observations with Pair balanced (observed at t0 and t1) .............xx.............xx.............xx..... ........xx.............xx.............xx.......... ...xx.............xx.............xx.............xx .............xx.............xx.............xxxxxxx xxxxxxxxxx Difference-in-difference with Multiple Time Periods
But, in this setup, when I add the control variables, the csdid command fails to estimate most of the ATT(g, t):
Code:
. csdid Y $X if Gvar != 0, ivar(pid) time(time) gvar(Gvar) notyet long2 Panel is not balanced Will use observations with Pair balanced (observed at t0 and t1) .....xxxxxxxxxx.....xxxxxxxxxx.....xxxxxxxxxx..... x.xxxxxxxx.....xxxxxxxxxx.......xxxx..xxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxx Difference-in-difference with Multiple Time Periods
If I include the never-treated group in the comparison group, the result becomes
Code:
. csdid Y $X, ivar(pid) time(time) gvar(Gvar) notyet long2 Units always treated found. These will be ignored Panel is not balanced Will use observations with Pair balanced (observed at t0 and t1) .............................x.................... .................................................. .................................................x ..............x..............x..............x..... ........xx Difference-in-difference with Multiple Time Periods
Comment