Anonymous (Dec. 24): [links to Zvi’s post above, arguing the link above isn’t wrong but understates the real-world consequences]
David (Dec. 25): I can’t say I came away from that post believing that “we’re fucked, it’s over” (like, beyond what someone might have thought two weeks ago, at least) was a well-argued or reasonable conclusion. The view I held/hold is: “we should be concerned — perhaps very much so — and we might be fucked, but also there’s a strong chance not”. The “maybe not” doesn’t mean “don’t be concerned”, and he seems to be conflating the two, even while criticizing others for doing exactly that.
There’s just a huge amount of uncertainty right now about exactly how contagious this strain is, and he’s taking the dead center of the confidence or credible intervals from some very, very early modeling, and possibly not properly accounting for the founder effect along with many other sources of variance over time in number of cases that aren’t due to changes in the virus itself. And you can’t just say “law of large numbers” (for a complex, dynamic, nonlinear process in which LLN may not apply) or “evidence is evidence” (when there’s a ton of noise in our epidemiological data which can make it very difficult to interpret) as a way to dismiss these possibilities.
(Everything else aside, he also made at least one numerically substantial mistake in the extrapolation process: from a chart of daily new cases in the UK, going from ~305/M on Dec 16 to ~500/M on Dec 23, he called that “doubling in one week”. Then he extrapolates that, for every arrival of the new strain in the US today, that amounts to 1M infections in 20 weeks. But that’s actually a ratio of 1.64x, which extrapolates out to 20k infections in 20 weeks. (Of course the reality is much more complicated than either simplistic extrapolation.))
This is all to say that I think we should be concerned about the new strain, but he’s treating some very preliminary analysis as if it’s analogous to our state of knowledge about Covid in, say, early February 2020, as opposed to, like, late December 2019.
(Perhaps a tangent, but then again maybe not since this seems to undergird his thinking on this subject: I really don’t think it’s accurate to say that non-Bayesians are “bonkers”. In fact I find it kind of bonkers to say that. One reason is that any attempt to be a Bayesian in the real world runs into huge problems identifying prior probability distributions in any meaningful way which is also inter-subjectively useful. Another, more significant problem with his take on it is that likelihood ratios for “non-academic information”, as he puts it, are almost impossible to meaningfully pin down, so it makes them rather thin gruel for an objective discussion, as opposed to confirming pre-existing biases. Not saying such information should be ignored entirely, but nor can they just be plugged into Bayes’ Rule in some mechanistic ritual.)
Social media comments from David Schneider-Joseph (copied here so they’re part of the central discussion, not to endorse them):