You are not logged in. You can browse but not post. Login or Register by clicking 'Login or Register' at the top-right of this page. For more information on Statalist, see the FAQ.
I am currently employing the sdid approach and am trying to find an answer to see based on which criteria covariates should be included.
Jason Acomb, in #5, you mentioned, "I will also try to better understand how covariates work within the SDID model, and see if I can include some (e.g., skills, population age demographics, GDP per capita, etc.)" Since you posted this a year ago, I assume you might be ahead of me and have found an answer by now. I would highly appreciate if you or someone else would be willing to share their knowledge.
My ideas on additional resources to consult for an ultimate answer, unfortunately, have come to an end.
I have panel data, and my level of analysis is at the household level. However, the original dataset also contains some variables at the individual and community levels. The data includes four waves collected every three years. I have 1,058 treated households.
Does this fully answer your question? Please feel free to correct me if I have misunderstood.
That answers my question. Why are you using sdid with such large sample sizes and many treated units? Have you tried other staggered intervention methods? It's easy to use xthdidregress in Stata 18, or the user-written jwdid, to allow covariates in a very flexible way. In xthdidregress you can choose twfe or aipw (Callaway-Sant'Anna).
As far as I understand, xthdidregress is for staggered treatment adoption (only), right? In the setting of my treatment, this is not applicable. Moreover, in the first place, I have inter alia tried Callaway-Sant'Anna (csdid); however, I could still not convince myself of a holding PT assumption. This is one of the reasons why I am now exploring sdid.
I will explore jwdid further. Thank you for your recommendations!
Are you saying you don’t have a staggered intervention? Any method for staggered interventions can be used for common timing. If you have at least two pre-treatment periods you can easily allow for heterogeneous trends using flexible regression. regress and xtreg will do what you need.
Comment