DID - event study analysis

[Marry Lee]

Hello,
I received a question about my work using the DID approach that I couldn't answer. I hope you can help me.
My DID model is of the following form:

y = a + b treat*post+ cX + u

The event study to check for the parallel trend assumption is the same except for replacing the indicator for post period 'post' by 'year'

y = a + sum bi treat*yeari + cX + u

The event study shows that in the post-period, some years show statistically significant increase in the outcome y, however, my DID estimate shows no statistically significant effect on y.

The question I received is that these two results are confusing so which result should we believe? How should I comment the two results?

Thank you.
Best

[Jeff Wooldridge]

Marry: As per the FAQ, you should show us Stata output. It’s hard to provide guidance otherwise.

[Jeff Wooldridge]

Generally, the event study estimates are compared with a single pre-intervention period — usually the one just before the intervention. The usual DiD regression averages across all pre-intervention periods. Are the pre-treatment estimated effects large?

[Marry Lee]

Dear Jeff Wooldridge, thank you so much for your interest in my question.

I have changed my sample by adding more years in the post-period, hence now I do not see the significant increase in Y in the event study analysis anymore.
But if possible, can you please comment also on the possibility that we do not see a significant increase in any year in the event study but that we can find a significant effect in the DID regression? How should we see those results?

Thank you so much!


Here is the result from the DID regression:

reg Y i.treat##i.post  X    `provinceXyearFE'    i.coun, cluster(coun)
note: uni_M omitted because of collinearity.
note: 622003.provinceXyear omitted because of collinearity.
note: 141127.coun omitted because of collinearity.
note: 152921.coun omitted because of collinearity.
note: 211421.coun omitted because of collinearity.
note: 321322.coun omitted because of collinearity.
note: 341723.coun omitted because of collinearity.
note: 419001.coun omitted because of collinearity.
note: 429006.coun omitted because of collinearity.
note: 431382.coun omitted because of collinearity.
note: 445381.coun omitted because of collinearity.
note: 500237.coun omitted because of collinearity.
note: 512021.coun omitted because of collinearity.
note: 533103.coun omitted because of collinearity.
note: 621121.coun omitted because of collinearity.

Linear regression                               Number of obs     =      9,044
                                                F(179, 284)       =          .
                                                Prob > F          =          .
                                                R-squared         =     0.2450
                                                Root MSE          =     .36781

                                                (Std. err. adjusted for 285 clusters in coun)
---------------------------------------------------------------------------------------------
                            |               Robust
                 Y         | Coefficient  std. err.      t    P>|t|     [95% conf. interval]
----------------------------+----------------------------------------------------------------
                      1.treat|  -.0116747   .0362744    -0.32   0.748    -.0830755    .0597261
                 1.post     |   .2081641   .1263802     1.65   0.101    -.0405966    .4569249
                               |
               treat#post |
                       1 1   |  -.0113878   .0194134    -0.59   0.558    -.0496001    .0268246

Here is what I have for the event study analysis: (please note that the post-period include 1998 and years afterwards)