How to add a reference period when I using the csdid package

[Zichun Zhao]

Hi everyone,

I am using the csdid package and event_plot comment to draw an dynamic change graph. But I did not find the way to choose the reference period, like period -1. I found this problem is common for other new DID methods like did_imputation. So, is there any method to fix it?

Best,
Zichun

[FernandoRios]

Hi Zichun
CSDID uses T-1 as the base period by default. You cannot choose other periods (for example allowing for anticipation), except by changing gvar manually.
If you see an effect on T-1, is probably because you are using default option, which relies on short gaps. Add the option -long2-, soyou can see t-1 as blank in the plot (and table)
F

[Zichun Zhao]

Hi Zichun
CSDID uses T-1 as the base period by default. You cannot choose other periods (for example allowing for anticipation), except by changing gvar manually.
If you see an effect on T-1, is probably because you are using default option, which relies on short gaps. Add the option -long2-, soyou can see t-1 as blank in the plot (and table)
F

Thank you so much!

[Lucy Lin]

Hi Zichun
CSDID uses T-1 as the base period by default. You cannot choose other periods (for example allowing for anticipation), except by changing gvar manually.
If you see an effect on T-1, is probably because you are using default option, which relies on short gaps. Add the option -long2-, soyou can see t-1 as blank in the plot (and table)
F

Hi Fernando!

I am having the same question. In my case, adding the option -long2- does not make the coefficient of g-1 disappear.

I suppose this is because csdid uses t-1 vs. t (t is varying) for pre-treatment estimation, and the coefficient of g-1 represents the difference of g-2 vs. g-1 between treat group and control group.

I was wondering whether csdid has a way to use universal reference period for pre-treatment estimation, like in R's did (the option "base_period"). That is, to use g-1 as the reference period for all the pre- and post-treatment estimation. I see the help file says the option -long2- "is similar to base universal" but I still don't get how this option works.

Could you kindly explain a bit? Thank you so much!

[FernandoRios]

Can you show the results?

[Lucy Lin]

Sure. Here is the data example:

* Example generated by -dataex-. For more info, type help dataex
clear
input float Y long gvkey float(year first_treat clustervar)
1  14002 2000    0 46
1  14002 2001    0 46
1  14002 2003    0 46
1 140760 2003 2001  8
1 140760 2004 2001  8
end

And here is the result:

. csdid Y, ivar(gvkey) time(year) gvar(first_treat) agg(event) method(reg) rseed(1) notyet cluster(clustervar) long2

Units always treated found. These will be excluded
Panel is not balanced
Will use observations with Pair balanced (observed at t0 and t1)
.........x.....x.......x.......x
Difference-in-difference with Multiple Time Periods

Number of obs = 6,212
Outcome model : regression adjustment
Treatment model: none
(Std. err. adjusted for 46 clusters in clustervar)
------------------------------------------------------------------------------
| Coefficient Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
T-5 | -.0241309 .0186482 -1.29 0.196 -.0606807 .0124189
T-4 | -.0845919 .0289547 -2.92 0.003 -.1413421 -.0278417
T-3 | -.01381 .0184488 -0.75 0.454 -.0499689 .022349
T-2 | -.0266007 .0151541 -1.76 0.079 -.0563023 .0031008
T-1 | -.013177 .0089546 -1.47 0.141 -.0307278 .0043737
T+0 | -.0362986 .0071993 -5.04 0.000 -.0504089 -.0221882
T+1 | -.029889 .0096612 -3.09 0.002 -.0488247 -.0109534
T+2 | -.0250312 .0105008 -2.38 0.017 -.0456125 -.00445
T+3 | -.0041936 .0122462 -0.34 0.732 -.0281956 .0198085
T+4 | -.0158408 .0189659 -0.84 0.404 -.0530133 .0213316
T+5 | -.0457156 .0307329 -1.49 0.137 -.105951 .0145198
T+6 | -.4788718 . . . . .
------------------------------------------------------------------------------
Control: Not yet Treated
end

[FernandoRios]

Are you sure you have the latest versions of csdid and drdid?

[Lucy Lin]

Are you sure you have the latest versions of csdid and drdid?

I updated csdid and drdid to the latest version and it works as below. Thank you for reminding me!

. csdid Y, ivar(gvkey) time(year) gvar(first_treat) agg(event) method(reg) rseed(1) notyet cluster(clustervar) 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
Difference-in-difference with Multiple Time Periods

                                                         Number of obs = 6,212
Outcome model  : regression adjustment
Treatment model: none
                            (Std. err. adjusted for 46 clusters in clustervar)
------------------------------------------------------------------------------
             | Coefficient  Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
     Pre_avg |  -.0324621   .0140837    -2.30   0.021    -.0600657   -.0048585
    Post_avg |  -.0908344   .0095774    -9.48   0.000    -.1096058    -.072063
         Tm6 |  -.0241309   .0186482    -1.29   0.196    -.0606807    .0124189
         Tm5 |  -.0845919   .0289547    -2.92   0.003    -.1413421   -.0278417
         Tm4 |    -.01381   .0184488    -0.75   0.454    -.0499689     .022349
         Tm3 |  -.0266007   .0151541    -1.76   0.079    -.0563023    .0031008
         Tm2 |   -.013177   .0089546    -1.47   0.141    -.0307278    .0043737
         Tp0 |  -.0362986   .0071993    -5.04   0.000    -.0504089   -.0221882
         Tp1 |   -.029889   .0096612    -3.09   0.002    -.0488247   -.0109534
         Tp2 |  -.0250312   .0105008    -2.38   0.017    -.0456125     -.00445
         Tp3 |  -.0041936   .0122462    -0.34   0.732    -.0281956    .0198085
         Tp4 |  -.0158408   .0189659    -0.84   0.404    -.0530133    .0213316
         Tp5 |  -.0457156   .0307329    -1.49   0.137     -.105951    .0145198
         Tp6 |  -.4788718          .        .       .            .           .
------------------------------------------------------------------------------
Control: Not yet Treated
end

One more question here. Can I wrap up all periods before g-3 (that is, collapse Tm6 Tm5 Tm4 Tm3 into one line)?

I really appreciate your help!