Absolutely phenomenal explainer Nick. I think this issue more generally that covariates within diff in diff is not doing the same thing that we think of when we were taught to include controls to address omitted variable bias is psychologically a hard lesson for a lot of people. I keep wondering -- how much difficulty have we created for people by assuming (and not saying it) constant treatment effects when introducing econometrics the first time. So many things seem to go back to that.
Thanks! And I agree - I think leaving heterogeneous effects (at the very least in terms of how we arrive at a particular ATE/ATT/etc.) is probably a mistake. It's such an abstract thing though, I can imagine it being difficult to get across in an UG classroom. Grad students heading for research should definitely be getting it though.
Absolutely phenomenal explainer Nick. I think this issue more generally that covariates within diff in diff is not doing the same thing that we think of when we were taught to include controls to address omitted variable bias is psychologically a hard lesson for a lot of people. I keep wondering -- how much difficulty have we created for people by assuming (and not saying it) constant treatment effects when introducing econometrics the first time. So many things seem to go back to that.
Thanks! And I agree - I think leaving heterogeneous effects (at the very least in terms of how we arrive at a particular ATE/ATT/etc.) is probably a mistake. It's such an abstract thing though, I can imagine it being difficult to get across in an UG classroom. Grad students heading for research should definitely be getting it though.