Grahpical diagnostic of the parallel trends assumption for DID

[Ali Burki]

Dear,
I am conducting research on carbon emissions. After collecting the data and running the regression and gathering the results. I am making a visual assessment of the common trends assumption. Below is the graph which shows the trends of the treatment and control group.

Looking at the trends of the treatment and control groups in the pretreatment time period I would say that they are not parallel and that there are some obvious differences when comparing the two trends in the pretreatment time period, especially after the year 1999. Should I based on my visual assessment of the common trends conclude in my case that the parallel trends assumption is not satisfied/violated or "weakly satisfied"?

[William Lisowski]

This seems to be a follow-on from a topic previously posted and discussed at

https://www.statalist.org/forums/for...rding-stata-17

[Enrique Pinzon (StataCorp)]

Hi Ali,

I think the post below might be of help. It explains the intuition and computation behind the graph and the plot.

https://www.statalist.org/forums/for...r-trends-model

[Ali Burki]

I have gone through the post: https://www.statalist.org/forums/for...r-trends-model. But I am dealing with Longitudinal data and doing a panel study. And here is a example of the regression coefficients and estat ptrends value.

Looking at the coefficient(-0.03342) I see that it is only significant at the 10 % significance level (i.e., 0.080 above) and I have a estat ptrends value of 3.30 with a p-value of 0.0907. Since I have a p-value (0.0907) lower than the 10 % significance level for the estat ptrends, I will reject the null hypothesis of parallel trends assumption or do I have to see the estat ptrends p-value in the context of a 5 % significance level? Because my result above is only significant at the 10 % and not 5% significance level.

[Ali Burki]

William Lisowski Enrique Pinzon (StataCorp)

[George Ford]

Angrist/Pischke (Mostly Harmless Econometrics) have a discussion about non-parallel trends (add a trend variable). Might look into that.