how to correctly interrupt the results produced by csdid Callaway and Sant’Anna 2021 DD)?

[Saleh sharmah]

Dear members

I am running the csdid regression as a robustness check for my staggered did results. However, I am not sure how to interpret the results produced by the csdid post-estimate commands. It is noteworthy that I have 18,951 firm-year observations, and when I ran the cddid regression, the number of observations dropped to 12,223. what caused this drop?

Here are the results of the csdid post estimation results:

1-estat pretrend
Pretend Test. H0 All Pre-treatment is equal to 0
chi2(66) = 1262.6064
p-value = 0.0000

Question 1) what does the p-value tell me about the treatment above?

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
2- Average Treatment Effect on Treated
------------------------------------------------------------------------------
| Coefficient Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
ATT | -.0205827 .0242701 -0.85 0.396 -.0681512 .026985

Question 2) does having insignificant p-value as indicated above suggest that is no concern of earlier treated firms having an impact on the later treated?

----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
3- estat event
ATT by Periods Before and After treatment
Event Study:Dynamic effects

------------------------------------------------------------------------------
| Coefficient Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
Pre_avg | .0905568 .0280406 3.23 0.001 .0355982 .1455153
Post_avg | -.0150769 .0264718 -0.57 0.569 -.0669606 .0368069
Tm13 | -.2472415 .0541207 -4.57 0.000 -.3533162 -.1411668
Tm12 | -.1979252 .0367847 -5.38 0.000 -.270022 -.1258284
Tm11 | -.0110969 .044918 -0.25 0.805 -.0991346 .0769407
Tm10 | 1.027638 .0637626 16.12 0.000 .9026659 1.152611
Tm9 | .4128842 .221467 1.86 0.062 -.0211832 .8469515
Tm8 | .0482778 .1091105 0.44 0.658 -.1655749 .2621304
Tm7 | -.0006784 .0396747 -0.02 0.986 -.0784393 .0770826
Tm6 | -.050637 .1602308 -0.32 0.752 -.3646836 .2634096
Tm5 | .298256 .0836639 3.56 0.000 .1342778 .4622341
Tm4 | -.0624409 .0473203 -1.32 0.187 -.155187 .0303052
Tm3 | -.0016701 .0300503 -0.06 0.956 -.0605676 .0572274
Tm2 | -.018172 .018596 -0.98 0.328 -.0546195 .0182755
Tm1 | -.0199566 .0218704 -0.91 0.362 -.0628217 .0229086
Tp0 | -.0254588 .0161276 -1.58 0.114 -.0570684 .0061507
Tp1 | .0139025 .0237889 0.58 0.559 -.032723 .060528
Tp2 | -.0172645 .0241829 -0.71 0.475 -.0646622 .0301332
Tp3 | -.02495 .0364027 -0.69 0.493 -.096298 .046398
Tp4 | -.0542365 .0341698 -1.59 0.112 -.1212081 .0127352
Tp5 | -.0217989 .0336454 -0.65 0.517 -.0877426 .0441449
Tp6 | -.042868 .0440775 -0.97 0.331 -.1292584 .0435224
Tp7 | -.0465764 .0360253 -1.29 0.196 -.1171848 .0240319
Tp8 | -.0087139 .0332699 -0.26 0.793 -.0739218 .056494
Tp9 | .0031924 .0312295 0.10 0.919 -.0580163 .0644011
Tp10 | -.0214046 .0406333 -0.53 0.598 -.1010445 .0582353
Tp11 | .0467994 .0410415 1.14 0.254 -.0336405 .1272393
Tp12 | -.0299339 .1034559 -0.29 0.772 -.2327039 .172836
Tp13 | .018235 .0980668 0.19 0.852 -.1739724 .2104423

question 3) can i conclude that the policy implantation has a negative effect of on the treated posted treatment?

thank you immensely for your insights.

[George Ford]

1. not good. you've got a lot of difference in the pre-treatment.
2. you got no treatment effect.
3. you got nothing.

[Saleh sharmah]

Thanks George for your response. I’m just wondering if the results are impacted by the drop of observations from 18,951 to 12,230 ? What possibly caused the drop ?

[FernandoRios]

The loss in observations wouldn’t explain. Your lack of significant treatment effects
also the sample is smaller because not all attgts were estimated or the data was not available for all cases

[Saleh sharmah]

a follow-up question when running the csdid regression, is there a restriction on noun variant control variables to be excluded from the model?