🚨 A few months ago, a paper published in Science reported that “LLM adoption is associated with a large increase in researchers’ scientific output.”
In a comment released today, we show that the reported effects are driven by a methodological issue in the empirical design. arxiv.org/abs/2605.17979
They’ve got to get one of those 70s game show mics for Wemby
This note brought to you by a recent replication I did of a 20 cluster study where CR2 and wild bootstrap SEs were like double the published Stata defaults…
Thomas Renault
Another one of those unbelievably cruel and pointless things most people won't understand because they have no contact with our country's immigration system.
it low key tickles me that folks have been really bad about citation practices wrt packages — much less understanding what those packages do — and yet LLMs spook everyone out. sirs, did you know how the black box magic package worked in ye olde days? no? ahh, well, see…
Participated in “Perspectives on External Validity” workshop yesterday, organized by @melodyyhuang.bsky.social at @ispsyale.bsky.social
isps.yale.edu/events/2026/...
Hi All. Perhaps something useful from me. I'd like to announce that drlate, a Stata module for doubly robust estimation of the local average treatment effect (LATE) and the local average treatment effect on the treated (LATT), is now available in SSC.
Number 1 thing I ask for any time I get student questions about DiD. Great to see the package updates!
We handwave that we’re in asymptopia but in so much of polisci we have *very few* actual observations!
The default robust inference methods really break down even in like the 50 state setting once you start estimating separate effects by treatment cohort.
The “sample size” point is really important - not just for precision but for the validity of our inferential techniques (see @amandaweiss.bsky.social here osf.io/preprints/os... )
