TIL both Oren Cass and Lina Khan were in the political economy major at Williams designed for a small handful of pre-law dweebs to avoid all calculus and statistics and larp as economists.
Institutions have consequences.
"you won't always have numpy" is the grown-up version of "you won't always have a calculator in your pocket". I have both a calculator and numpy in my pocket.
3 years ago, I took my methods comprehensive exam just as everything was shutting down.
To prep, I made a cheatsheet that quickly got out of hand and now stands at ~130 pages of notes on econometrics / causal inference / machine learning.
apoorvalal.github.io/methods/tex/no…
My team is hiring interns! Applications welcome from PhD students with expertise in causal inference and experimentation and an interest in developing and/or applying methods to industrial applications.
jobs.netflix.com/jobs/306716846
IPWRA is great; screenshot from Imbens (2004) review article, py and R implementations.
Don't be beholden to statacorp; write in a programming language that lets you implement the math almost verbatim
You can do almost all of this, with the PS estimated by MLE, using teffects in Stata. Unfortunately, while IPW uses normalized weights, AIPW does not. And, for some reason, ATET (ATT) is not an option with AIPW. You won't get moderating effects using teffects, though.
Statistics is a funny field: it so spectacularly fails at its "building tools for scientific enquiry" remit that most applied fields have to invent their own versions of it (data science, biostatistics, econometrics, psychometrics, pol methodology,...) and yet it persists.
As someone who has spent good entire adult life in academic statistics departments I can say with certainty that we have lost the Mandate of Heaven. I would certainly prefer hiring someone trained as an economist or a political scientist if my goal was finding truth from data