Interaction term (Moderation Effect) in SDID

[Mansour Mohseni]

Hi
I am wondering whether there is a way to obtain the moderation effect using sdid. I want to interact did coefficient with another variable and study the moderation effect. Currently, my code is:

sdid y id time d, vce(placebo) seed(1213) graph

and this only gives ATT. How can I add another variable to study how it moderates ATT?

[Jeff Wooldridge]

What are your data dimensions (N and T and number of treated units)?

[Mansour Mohseni]

What are your data dimensions (N and T and number of treated units)?

I have 421512 Observations, 2347 treated units, 17725 control units, and T is 21 (10 weeks before and after the intervention). The Panel data is balanced. My moderators are at the unit level and time-invariant.

[Mansour Mohseni]

Other than subsampling (which may not be the perfect solution for incorporating continuous moderators), it there any other way? My units are some local stores and that's why I have many treated and control units.

[Zhiya Zuo]

Dear Mansour, have you found any solution to your question on the use of `sdid` command with moderators?

[Jeff Wooldridge]

I see I failed to respond to the earlier post. In Mansour's case, I don't really see that the setup is suitable -- although I've found SDID to be pretty resilient. Still, it assumes large T0 and T1 (number of control and treated periods time periods) for valid inference, and I'm not sure T0 = 10, T1 = 11 does it. SDID is mainly meant to provide just a single average treatment effect on the treated with a small number of treated units. Zhiya, can you say more about your setup?

[Zhiya Zuo]

I see I failed to respond to the earlier post. In Mansour's case, I don't really see that the setup is suitable -- although I've found SDID to be pretty resilient. Still, it assumes large T0 and T1 (number of control and treated periods time periods) for valid inference, and I'm not sure T0 = 10, T1 = 11 does it. SDID is mainly meant to provide just a single average treatment effect on the treated with a small number of treated units. Zhiya, can you say more about your setup?

Dear Prof. Wooldridge, in my dataset, there are 10 panel units in total, in which 7 are control, and 3 are treated over a 22-month period. For each treated units, there is a 10-month pre-period and 12-month post-period).