Post-Instrument Bias It's a thing. It's a bit like post-treatment bias, but there are important differences, and the soc sci/econ literature on choosing covariates in IV is not very helpful osf.io/preprints/so...

Interpreting transformer's "attention" as kernel smoothing - write-up by Shalizi, extremely informative (as usual) bactra.org/notebooks/nn...

Please see here for a call for doing more to integrate preregistration into the review process: "Open Letter on the Need for Preregistration Transparency in Peer Review" -- #polisky #econsky You can sign too -- link is in the letter. docs.google.com/document/d/1...