Read on Twitter
At the risk of being real dumb and starting an unnecessary flame war, a couple thoughts on the annoying causal inference debate:
I ultimately come down closer to the side of team PO. Econs have death wars over ideas and justifying your arrow will inevitably devolve into an intractable war on hot button issues. I think it's right probably to focus on areas where you can get cleaner(ish) identification.
Where I side with DAG - like a lot of econs I think it's useful for forcing people to write down their assumptions. Big one - which variables are you controlling for and why. Much of applied finance/acct/law is just a paint by numbers approach of throwing in equilibrium generated
variables in a linear regression with no justification. That's bad, show your work. Another area of confusion in the last day of debate: PO people seem to say that applied econ folks focus on a small number of tools where we understand the model really well (paraphrasing here).
Really? Alwyn Young just showed that a lot of (most) IVs published in the top econ journals have serious specification issues. Not sure why Team PO is going to bat for a tool that, if we're being honest, most of us don't really trust outside of the RCT context.
What about DiD? @pedrohcgs, @agoodmanbacon, @a_strezh and others have shown how messed up most DiD applications are too. Not like clustering on the wrong level wrong, but *wrong sign* wrong.
A shit ton of RDD work with overfit polynomials creating facially implausible estimates. The list goes on. Can we really argue that as a field we *really understand* any of these designs?
I spend way too much time thinking about these issues from a career perspective if we're being honest. I consistently hear how it would be better to use some bad off the shelf technique on a novel problem or dataset rather than expend bandwidth on getting the estimates righter.
And that's surely right. I'm not going to stop, bc if I wanted to just jerk around with data all day I'd go back into consulting and get paid more with higher job security. If we care about the issues we're studying we need to be doing a better job justifying what we're doing.
