,
7 tweets,
2 min read
Read on Twitter
Economics education is definitely moving! Here are slides from Heckman's 2019 graduate course - LOTS of DAGs!
(h/t @erinhengel) 1/6
jenni.uchicago.edu/econ312/Slides…
(h/t @erinhengel) 1/6
jenni.uchicago.edu/econ312/Slides…
Much of the derivations are way too complicated. It would be a lot easier using d-separation instead of Graphoid axioms. No idea why they are not using it. 2/6
Pinto/Heckman emphasize a lot how their idea of "hypothetical models" is better than do-operator, because the latter requires "extra-statistical" tools.
Not convincing. Hypothetical models are also extra-statistical. In fact, it is psychologically and sociologically 3/6
Not convincing. Hypothetical models are also extra-statistical. In fact, it is psychologically and sociologically 3/6
important to have a clear notation for extra-statistical concepts - because they are, in fact, very different from statistical concepts! This is why so much of econometrics runs into deep troubles, because they forget what y=xb + e actually means. See cloudfront.escholarship.org/dist/prd/conte… 4/6
They also say do-calculus is "limited", because you cannot incorporate parametric assumptions.
Quoting @yudapearl, this is like saying multiplication is "limited" because it cannot perform addition... apps.dtic.mil/dtic/tr/fullte… 5/6
Quoting @yudapearl, this is like saying multiplication is "limited" because it cannot perform addition... apps.dtic.mil/dtic/tr/fullte… 5/6
But, in sum, great to see heavy, formal use of DAGs, even with severe limitations (no use of d-separation).
Also suggests my proposal to incorporate DAGs into Chetty's Into to Econ class @ Harvard will at least get some support from Chicago. 6/6
Also suggests my proposal to incorporate DAGs into Chetty's Into to Econ class @ Harvard will at least get some support from Chicago. 6/6
BTW here are my slides on causal inference using DAGs:
