Alberto Abadie Guest Talk - Synthetic Controls

[@lucasng9617]

Your channel is awesome, Brady! The course website is amazing as well. Thank you for sharing! I found your material to be more "user friendly" than other traditional books in the field, like Imbens and Rubin's Causal Inference for Statistics, Social, and Biomedical Sciences. Congratulations!

[@the_causal_mindset]

Amazing video! Trhanks a lot! I was just puzzled by the graph with all the donor in the pool for california. It seems to me that on average the you can more or less fit pre-period, while of course, mechanically, afterwards variance explode. So, what if the divergence with California is just random, how do we assess that? (like what is the risk that it's random deviation).

[@michelspeiser5789]

Concerning the regression counter-arguments: how about using a Lasso regression restricted to positive coefficients?

[@chanansh]

Alberto claims the Diff-in-Diff will have a bias because groups are different even before intervention (left panel 9:00). However, as I understand it, diff-in-diff does NOT assume groups to be exchangeable before the treatment. It just assumes parallel trends for the no-intervention potential outcomes.

[@danailvasilev8565]

I think he's referring to West Germany having a different trajectory before intervention. You see the GDP per capita line is steeper than for the OECD, i.e. the parallel trends assumption does not hold.

[@alessiogarau7948]

This lecture is amazing. Anyway, once you have followed the lesson it may be more useful to review the slides. There is any chance to obtain them?