We are working on a project using panel data, using broadly a DID approach. However, our problem consists of the following:
Ideally, we would want to implement one of the new strategies (i.e., from Baker et al (2021), Gardner (2021), Woolridge (2021) and Callaway & Sant'Anna (2021)) to ensure we do not run into a late v early comparison. However, many of the inbuilt stata functions (i.e., stackedev, did2s, csdid, eventstudyinteract, xthdidregress) do not allow for the exploitation of increasing treatment intensity.
Our question is specifically for stackedev: can MSAs be treated multiple times, by utilizing different stacks?
E.g., the treatment group of one stack would be all MSAs who are all treated in a certain year (whether or not they have been treated before) with the control group being the MSAs that are never treated.
For treated MSAs that occur in multiple stacks, only the treatment intensity and (potentially) unemployment would change.
Given that the control group will remain constant, it seems the late v early comparison is not there. However, by construction, treated MSAs will be repeated through stacks.
Even though we conceptualized this question for stackdev, we are open to suggestions on how we can exploit increasing treatment intensity using other econometric methods.
- Staggered treatment adoption
- Continuous treatment
- Increasing treatment intensity
Ideally, we would want to implement one of the new strategies (i.e., from Baker et al (2021), Gardner (2021), Woolridge (2021) and Callaway & Sant'Anna (2021)) to ensure we do not run into a late v early comparison. However, many of the inbuilt stata functions (i.e., stackedev, did2s, csdid, eventstudyinteract, xthdidregress) do not allow for the exploitation of increasing treatment intensity.
Our question is specifically for stackedev: can MSAs be treated multiple times, by utilizing different stacks?
E.g., the treatment group of one stack would be all MSAs who are all treated in a certain year (whether or not they have been treated before) with the control group being the MSAs that are never treated.
For treated MSAs that occur in multiple stacks, only the treatment intensity and (potentially) unemployment would change.
Given that the control group will remain constant, it seems the late v early comparison is not there. However, by construction, treated MSAs will be repeated through stacks.
Even though we conceptualized this question for stackdev, we are open to suggestions on how we can exploit increasing treatment intensity using other econometric methods.
