Hi everyone. I spent the day coding up a basic tutorial on the two step synthetic control method (SCM). Perhaps those of you who use Python may find it useful.
In short, people who use methods like difference-in-differences don't usually consider the parallel trends assumption they're making. They sort of throw coefficients into a nice event study graph and move on with their day. Indeed, the same is true with SCM in many cases. People sort of shove their data into synth or scul or whatever your drink of choice is, without thinking about the exact nature of the parallel trends assumption for SCM.
Well... this matters quite a lot for practitioners. My post attempts to explain why and how, in a less technical sense than you'll get in econometrics journals. It's still being drafted, so in the future I'll add much more, but hopefully this would even help future synth writers for Stata work through these kinds of issues.
In short, people who use methods like difference-in-differences don't usually consider the parallel trends assumption they're making. They sort of throw coefficients into a nice event study graph and move on with their day. Indeed, the same is true with SCM in many cases. People sort of shove their data into synth or scul or whatever your drink of choice is, without thinking about the exact nature of the parallel trends assumption for SCM.
Well... this matters quite a lot for practitioners. My post attempts to explain why and how, in a less technical sense than you'll get in econometrics journals. It's still being drafted, so in the future I'll add much more, but hopefully this would even help future synth writers for Stata work through these kinds of issues.
