I gained’t make this lengthy. Can I clarify why parallel tendencies breaks down utilizing a easy instance about compositions of teams? I’m going to strive. I flipped a coin and it got here up heads twice.
Thanks once more everybody on your help! It is a labor of affection and I hope you take pleasure in this put up. It’ll be paywalled except you’re a subscriber as a result of randomization of the paywall I exploit for my non-Claude code sequence. Take pleasure in!
Damaged Parallel Developments
When there are two teams, one handled and one untreated, and two time intervals, then we all know the work of the management group in difference-in-differences. It’s to impute the untreated potential final result (ie the counterfactual) for the remedy group. And it does that utilizing two issues:
-
It makes use of its personal first distinction to interchange the counterfactual with its personal noticed worth.
-
And it’s correct if parallel tendencies holds.
All the remainder is particulars. They matter but when you must distill it to one thing memorable that’s it. Take the primary distinction, and impute, which is authorized to do if parallel tendencies.
So when is that not going to work? Nicely it gained’t work for those who don’t have a management group after all. That’s one factor. It gained’t work for those who don’t have two intervals. And it gained’t work if parallel tendencies is just not true.
Let’s concentrate on the final one. Why is parallel tendencies damaged within the first place?
You may actually over clarify this. Or let me say it one other approach. There are numerous methods to speak about this, all fruitful and wanted, however I wish to discuss it utilizing covariates. I’ll use this as my instance: organic males and organic ladies.
