scoringutils was already a cool package. If you’re scoring probabilistic forecasts in R you should check it out! (In fact, I’ve heard of workflows which keep dependency on R just to use scoringutils — a strong endorsement.)

Last month, the Nucleic Acid Observatory identified an engineered virus sequence in wastewater. The underlying agent was benign; but it's a cool demonstration of our system working: We can now monitor for new, genetically engineered pathogens.

Covid brought us rapid at-home tests. But we fail to use them for mpox & bird flu. We could do better with triple rapid tests that can be used for current and future outbreaks: Tests with rapid reconfiguration, deployment, and results. @JacobSwett and I write in @statnews:

I also enjoyed talking through the last five years of thinking about situational awareness. Slides are here with lots of links out. https://t.co/KdoPZk2t02 I had some good feedback that people enjoyed the focus on mistakes so planning to keep making plenty more.

🚀 the team at @ifp has been working on this for a long time, and today, we're finally launching "metascience 101"! it's a nine episode podcast series designed to be a crash course into the big ideas motivating metascience first episodes dropping today: https://t.co/QzXYO3V56V

The package uses @mcmc_stan to estimate epidemiological delay distributions. These parameters are key for epidemic response (see e.g. https://t.co/ji03oPL0Z9) so IMO it's especially important we are well prepared with thorough testing in advance!

Most software is routinely tested to ensure it works as expected. However, testing is rarer for statistical models. In this blog post, I discuss my experience implementing tests for a Bayesian software package:

(Also, I don't think the authors of this article are presenting it as telling anyone what to do! Especially given the between-expert variation. So perhaps part of this is just me lamenting [celebrating] getting older.)

That's not to say that high-level advice isn't useful (and some of the advice in this article is very niche!) but I think most of the value is in putting advisees into a position to plausibly have impact, rather than in specifying exactly what to do.

Something I've noticed recently, as I have gotten a little closer to some of these problems, is that the shape of the issue can seem to change from what you initially expect.

This is a fun series, and it's interesting to see 5/12 experts mention biosurveillance.

A thread about how science should work, ways to cheaply save lives, peer review, and diarrhea 👇

Anyway, I'm no expert, but if people have more thoughts about statistics and meta science I'm very happy to chat about it or hear thoughts -- please get in touch. And again, thanks Matthew for doing polite and important work.

It warms me to see folks talking about splines and GLMMs in the replies. But if that's the barrier to entry for thoughtful use of methods, then should we really be suprised? Good software, documentation, and exposition can only go so far I worry.

I used to enjoy listening to podcasts or reading popular science books by Haidt (and many others). I'm gradually learning to treat this genre of content as much further towards entertainment than information.

This is very commenable work by Matthew. It does make me sad to see the quality of methods people use, but over time I am less and less suprised (especially when people have partiular incentives for things to turn out in one direction).

For tomorrow's community seminar (3pm UK time) Dr Kaitlyn Johnson will discuss the challenges and opportunities in developing models and software tools that incorporate data on viral concentrations in wastewater to forecast hospital admissions in the US https://t.co/nfX7dEE3Zc

Thanks Gavin! See also my list of things I wish I'd known before starting my PhD: https://t.co/XEaipa6YTL (And despite these flaws, I did pass.)

Although the USG had invested $700M in pathogen early warning in 2019, COVID-19 caught us off guard. At @IFP, we’ve investigated why. One answer is the map below: Biosurveillance is fragmented and siloed across jurisdictions (1/9). https://t.co/B6xcKcdUZb

Also I'm graduating with my PhD in the next few months (!) and looking for my next role. Let me know of any opportunities I might be interested in -- I'm open to both industry and academic roles. See https://t.co/E1xLnZe1nU for more information. CV available on request.