"Draw me a picture of sociological theorists working hand in hand with computer scientists to build a large number of permissible models ascertained by theory".

way to reinforce structural bias, GPT 5 🤦

@ramina-sotoudeh.bsky.social gave an AMAZING talk yesterday as part of the #metrics-and-models seminar series👏. Next (30.07), we have @klint.bsky.social on "Reproducing Expert Judgement with Shortened Surveys using Simulated Annealing"🔥! See website for more; metrics-and-models.github.io!

Hurrying to post about RobustiPy, i forgot this fun animation i made for teaching purposes (with @deevybee.bsky.social on the SABS R3 CDT) around when we started developing. Specification uncertainty is just _one_ type of uncertainty, though; our forthcoming work looks at the role of random numbers!

We also think it has some tight #dataviz, too! Here's the canonical `union.dta' dataset examined by other model uncertainty tools. See our paper and website for more examples of how to use the accessible Python app, released under an open source GNU GPL 3.0 license!

if you can out-do us with the pretention of this, i will salute you:

arxiv.org/pdf/2411.19223

FWIW

1/5 📄✨ITS FINALLY HERE!📄✨. After _five_ years, our new model evaluation metric — The InterModel Vigorish (IMV) — is now published (@plosone.org).

Paper: bit.ly/3RlqrQH
Preprint: bit.ly/41LzGjm
Code: github.com/intermodelvigorish

‘What’s the vig on this action?’ (bit.ly/4hPoZ3Z). ⬇️⬇️⬇️

🎉🚀🔬Our new model uncertainty library -- RobustiPy -- now also has joint inference tests (as per @urisohn.bsky.social et al., 2019, NHB). Near to release/preprint! GitHub:

github.com/RobustiPy/robustipy

project website:

robustipy.github.io/

/w @dvaldenegro.bsky.social & @jiani-y.bsky.social!🎉🚀🔬

In an age when so much modern scientific conversation surrounds 'Artificial General Intelligence', lets not forget that René Descartes -- in De homine (1662) -- claimed that non-human animals could be explained reductively as automata.

“you will be visited by three types of epistemic error” arxiv.org/abs/2411.19223

Now on arXiv! We design a new two-stage Individual Synthetic Control algorithm to cleanly identify causal estimates of the “caring penalty.” High-intensity unpaid carers face ~45% personal losses; households lose ~12%. 🏠📉

Link: arxiv.org/abs/2411.10314
Blog: shorturl.at/IVQ6M

Having spent much of the summer in China talking about LLMs in general, I can absolutely confirm that this is true.