Susan Athey Profile picture
Economist
Phil Kastelic Profile picture 2 subscribed
May 17, 2022 8 tweets 2 min read
In most econ applications, clustering std errors is a problem of design, not sampling. What does that mean? Clustering is needed if treatment assignment is correlated within clusters, not because some clusters were sampled and others were not. 1/8 @jmwooldridge After 5 yrs of iterating on math, simulations, writing, we updated our working paper on clustering. Lots of time working on single formulation that captures intermediate cases, simple after-the-fact, math matches intuition. arxiv.org/abs/1710.02926
Mar 3, 2021 34 tweets 6 min read
Just published in @ScienceMag

We estimate the value of vaccine capacity, and suggest policies and investments that governments could make right now to accelerate vaccine availability and end the COVID-19 pandemic sooner.

science.sciencemag.org/lookup/doi/10.….

Thread/ 1/ Each month, COVID-19 costs hundreds of thousands of lives and hundreds of billions of dollars in short-run GDP. That’s before accounting for missed schooling and other healthcare.
This means there are huge benefits to accelerating vaccination.
Nov 9, 2020 16 tweets 5 min read
We (@ArkhangelskyD, David Hirshberg, Guido Imbens, Stefan Wager and I) posted a major revision of our paper, Synthetic Difference-in-Differences, to arxiv. The paper estimates treatment effects when units are followed over time, and some receive a treatment in later periods. 1/n Our SDID method combines weighting and outcome modeling for a form of double-robustness that improves on synthetic controls and standard DID in many settings. arxiv.org/abs/1812.09970 2/n
Sep 14, 2020 13 tweets 6 min read
@otis_reid Matrix factorization: Panel data can be thought of as a matrix. A necessary condition for being able to do prediction is that there is some structure --something about the row and column of an entry is informative about the outcome. 1/n @otis_reid One way to describe the amount of structure is the quality of approximation you can get with a low rank matrix. An NxT matrix of rank k can be written as product of two latent factor matrices with k factors: [Nxk] X [kxT]. 2/n
Mar 7, 2020 15 tweets 5 min read
Starting an open-ended conversation for faculty who are taking their courses online in a hurry. What are tips and tricks? What challenges are you facing? #econtwitter I recently had to give some lectures remotely. I used powerpoint, taking screen shots of pdfs and pasting onto ppt where needed. Ppt has very, very easy to use feature where you record a video w/ webcam, separate videos for each slide, and they autoplay in present mode.