[Jared Greathouse]

SDID is quite simple in the sense that it only need 4 variables from the user. The outcome, the indicator treatment, the numerical/categorical id and the time.

year: simply shows 2002-2020 post_2005: dummy variable (1 if 2005 or after, 0 if before 2005)
year_post: 0 if before 2005, 2005-2020 if 2005 or after

My treatment_2005 variable shows 1 if the bvdid has ever been treated and 0 if never treated.

You have a code problem, then. SDID doesn't do the interaction term of the post period and treated indicator, that isn't how the syntax works. Suppose we have 50 treated states. California is treated in 1989. My treated variable would be 1 if state is California and year is greater than or = 1989, else 0.

[Jeff Wooldridge]

I think this is what you want:

 
 sdid RD bvdid year impact_2005, vce(placebo)

[George Ford]

For sdid, you want treated*post.

g did = treated*post

sdid Y id year did , vce(placebo) method(?) //where ? is sdid (default), sc (synth counter) or did (regression fixed effect).

[Jeff Wooldridge]

Yes, and I believe Anja's treated*post is how she defined impact_2005.

[George Ford]

You might also consider fect

https://yiqingxu.org/packages/fect/stata/fect_md.html

.

[Anja Jean-Mairet]

Thank you very much for all your help! Truly appreciate it. This has solved a big knot in my code.

On to the next now...

[Anja Jean-Mairet]

Dear All,

Thank you for your previous help on my sdid struggles.

I came across another question mark that I have regarding fixed effects for which I would be very interested in your expertise on the matter. The sdid model accounts for both unit-specific effects and time-specific effects according to the introduction by Arhangesly et al (2021). What I am unsure of is whether the SDID command in stata automatically accounts for unit and time fixed effects or whether these fixed effects variables need to be added to the covariates() option?

When I run my model with and without fixed effects in the covariates option I receive identical outputs which seems surprising to me which is why I want to double check whether the sdid command potentially already accounts for them and when I add them it simply drops them due to multicollinearity (help sdid says the model still runs if multicollinearity exists).

In case you know anything about this I would highly appreciate your input.

Thank you for your time!

[George Ford]

it does it automatically, so there should be no difference (you've just added a bunch of colinear variables)

[Anja Jean-Mairet]

Awesome, thank you for the clarification George!

[Dimitriy V. Masterov]

You might consider synth_runner. It's a wrapper for synth that allows multiple treated units without aggregating them. It does not do DID, however.

[Jared Greathouse]

You may also consider this new SDID event study command by Diego Ciccia