,
10 tweets,
3 min read
Read on Twitter
Harvard's econ star Raj Chetty getting a lot of praise for his new "empirical" intro to econ class, "Using Big data to Solve Economic and Social Problems". Great idea:
1. Focus on important problems (inequality, health)
2. Illustrate credible empirical research
BUT... 1/8
1. Focus on important problems (inequality, health)
2. Illustrate credible empirical research
BUT... 1/8
When you look at the slides, you may wonder whether the course does more harm than good.
Here is the motivation for "big data", from lecture 1.
I thought we had erased this myth! 2/8
Here is the motivation for "big data", from lecture 1.
I thought we had erased this myth! 2/8
Here are key concepts from the class. Notice it's all "theory", and data / causal inference not mentioned. (Ok, it say's "economic concepts", but still) 3/8
Here's one of the few (relatively) explicit notion of causal effect - note the counterfactual statement - illustrated with a bad example of post-treatment adjustment. No assumptipons mentioned. 4/8
Anything about causality is not formalized at all. On the other hand, econ theory and regression approaches are somehwat formalized, so this cannot explain why causality is not. Here is a labour supply diagram... 5/8
....and here a formal regression model. 6/8
And here's the second time causality is discussed, and it's bad IMO.
Really, we want to know the ATE of college - because apparently we want to *force* people to go to college? Isn't the ATT more relevant?
Oh wait, the ATT cannot be identified using experiments alone... 7/8
Really, we want to know the ATE of college - because apparently we want to *force* people to go to college? Isn't the ATT more relevant?
Oh wait, the ATT cannot be identified using experiments alone... 7/8
So all in all, *much* room for improvement. Not gonna surprise anyone by using that causal graphs should be used. I think Raj may actually have the person for this in his dept (@maxkasy) 8/8
Here is the link to the slides: opportunityinsights.org/course/
And here are my slides on causal inference using DAGs:
