Seem the British, who, while generally charming, talk funny, have renamed Rube Goldberg ‘Heath Robinson‘ and did it a decade before anybody had heard of Goldberg. Sneaky, that.
Just ran across a Twitter thread where Katya, English woman with a background in economics and finance – kind of like me! – looked at Neil Ferguson’s model in excruciating detail:
When this whole panic started, I looked a model linked to a petition to the White House to lock us down for 2 weeks to ‘flatten the curve’ and then ease back to normal, or else 11 million Americans would die! Ah, those days of innocence. I didn’t get into the actual math of the model, I just played with the 4 most immediate inputs and observed how the model reacted. Conclusion: Garbage in, garbage out. Simply assume a death rate of 2.5, and R0 of 2.5, a population of 330 million, and nobody washing their hands or staying away from grandma if he is felling ill., and – Boom! 200M infections and 11 million dead! I saw no place to put in ‘died out in the spring’ or ‘maybe a bunch of people are already immune’ or ‘maybe it doesn’t spread the same in Montana as it does in Greenwich Village’ – nope. Buried in the math, I suppose.
To top it off, there was a magic mitigation slider: just decide how effective your mitigation efforts are, and watch the numbers change! Toward the end of the article, after the panic porn where you are told, effectively, that failing to sign the petition made you personally responsible for all the inevitable deaths, was the admission that nobody actually knows how effective any of the mitigation efforts are (or how they amplify or defeat each other, like how a lockdown in a nursing home or hospital very possibly raises the chances of the people in those places getting it. But I quibble).
Even I, with my feeble math chops, could see that no amount of fancy calculations was going to smooth over the insane unstated assumptions of this model and the weird lacunas.
The Ferguson model I never checked out, although it seems clearly the spiritual sibling of the one I did check out. There’s this little thing called ‘reality’ that models, if they are to be useful, are measured against. I don’t trust the Communist Chinese any farther than I can throw a pagoda, but: if Fergy’s models is remotely correct, the Chinese government is hiding millions of dead and hundreds of millions of sick people and perfectly suppressing the subsequent unrest – and that strains credulity. They’re every bit that evil, but nowhere near that competent.
And the Diamond Princess. And Japan, Singapore, and now Sweden.
So I’m grateful to Katya for her analysis, but I think the model fails the sniff test before you even get into the 450(!) variables and poor code.
Meanwhile, another guest post on William Brigg’s fine blog attacks some of the other issues I’ve covered here, essentially, the inescapable messiness of the data. Poorly defined or undefined terms, different standards of data gathering and reporting, different reporting schedules, lots of room for judgement calls such as: 99 yr old nursing home prisoner dies, not of old age, but of COVID 19, as does a 37 yr old drug overdoes victim.
Adding up causes of death is not going to tell you much beyond 1) how deaths are reported in aggregate; and 2) lots of people die, like pretty much all of them sooner or later. Since most people now days die when, by historical standards, they are very old and subject to the many afflictions of advanced age, it makes assigning a cause of death other than ‘the decedent was very old’ a reach.
But we do.