Wow, this sounds bad! Better read the article theguardian.com/education/2023…
Oh, it's nitrates and nitrites. I don't really know how bad that is, I'll follow the link. Ok, so it's "in mice" and 15% of their diet being bacon and sausages. That's going some.

I know it's bad to dunk on paid blue-ticks to give them engagement they can monetise, but this guy is so egregiously wrong and there's a danger that people with anxiety disorders might believe his "calculations" that I think it's important. Sorry 🧵

https://twitter.com/drseanmullen/status/1696229699846832172

I've written a lot about the Russian roulette model - constant chance of Long COVID per infection - and have explained how that doesn't fit with real world data? Well, this guy has stepped it up, and done a similar calculation if the chance goes up with each reinfection.

Can I just talk about this, from a member of @IndependentSage, because it's been kind of annoying me? It essentially comes down to what you mean by "doesn't decrease". And it's true that the ONS do say this: ons.gov.uk/peoplepopulati…

Ok, I think this may be me more or less done here, COVID-wise. Having this variant data so easily on hand (thanks in a huge part to @OliasDave and @AlexSelby's processing of it) has made nowcasting an absolute breeze since the start of 2022.

https://twitter.com/TWenseleers/status/1638910529996562435

For all the value of the ONS, variant percentages have been pretty much unique lately as a genuinely *leading indicator* - letting us forecast the timing and scale of likely waves with solid confidence two or three weeks before they started (receipts here

https://twitter.com/BristOliver/status/1604101423112781824

)

This is not a quote tweet to dunk on anyone, but I'm unconvinced that ONS (random sampling) is the right model to track prevalence in the really important initial exponential growth phase of a new disease, from a statistical point of view.

https://twitter.com/nick5michaels/status/1633874405523861505

Say H5N1 did take over in humans (and goodness knows we hope it doesn't) .. The lowest COVID prevalence that ONS ever estimated in England was 14,000 (late June 2020). But this came with a massive confidence interval (proportionately speaking) from 5,000 to 31,000.

One of the things I talk about in Chapter 2 of Numbercrunch is "orders of magnitude" errors in the media and online. Brief thread of some of my favourites, do please let me know the ones I missed! Third favourite: This all-time classic, which was one of the things that got me thinking there was a need for this material in the first place.

It's 12 days until Numbercrunch is published, and there are 12 main chapters in the book, so it's probably a good time to tell you a bit more about what's actually in it. (And if you didn't see them already, there are comments here⬇️ from some great people).

https://twitter.com/BristOliver/status/1618987008801783810

Chapter 1 discusses something very familiar - how plotting data on graphs can be both useful and deceptive. In particular, I talk about *why* we are doing it, to find patterns and trends and make predictions, but how overfitting to data can make apparent structure lead us astray.

One very exciting (if daunting!) thing about Numbercrunch amzn.to/3O9vX6L was having it read by some of my absolute favourite number-focused writers. I'll share their comments in the next few weeks, starting with this from @TimHarford - huge thanks to all of them. Another Numbercrunch endorsement that I was really happy about was this from @Ananyo, whose own von Neumann biography was the best thing I read last year. Might have to add "maths-whisperer on a mission" to my Twitter bio!

There's talk of mandates and electability so, because I take the old-fashioned view that elections are won by convincing floating voters, here's why I voted for a Boris-led Conservative party in 2019 but never could again. (I suspect everyone will be pissed off by some of this). I voted Remain in 2016. I could see the arguments both ways, but I was personally more convinced by the arguments about frictionless trade (which I couldn't see happening) than by the arguments about Brussels becoming an unaccountable superstate (though I believed it was).