• Ian Dudley

Dr Strangelove

Directed by Stanley Kubrick, distributed by Columbia Pictures - Dr. Strangelove trailer from 40th Anniversary Special Edition DVD, 2004, Public Domain

Dr. Strangelove: Riding the Bomb

Mr. President, I'm not saying we wouldn't get our hair mussed, but I do say no more than 10 to 20 million killed, tops! Uh, depending on the breaks.

tl;dr—if you want to know what is happening and likely to happen in the coronavirus outbreak listen to medics and medical epidemiologists who deal with patients and hospitals and health systems, not mathematical modellers—at least until lots more data is available.

Mathematical models are powerful, but they work on data. Given the current lack of testing in the UK it’s not so much “garbage in, garbage out”, more “thin gruel in, uncertain predictions out”. Because of its inability to implement mass testing, the UK government has turned to representative sampling to get the information it needs. Porton Down is testing a carefully chosen sample of a couple of thousand people to get a snapshot of the COVID-19 status of the UK. Absent the catastrophes in 2015 and 2017, this sort of sample-based extrapolation has been remarkably successful in calling elections for almost a hundred years.

The problem is that such tiny sample sizes are no good if you want to drill down through the data. What’s the COVID-19 status of Yorkshire? Of Yorkshire men? Of Yorkshire men between 40 and 60? The overall sample may be representative of the UK, but slice and dice it to any extent and you are extrapolating results from a small handful of people who are unlikely to be representative of the population you are interested in. The 2017 US Presidential election is an excellent cautionary tale. The polls were remarkably accurate when it came to predicting the respective national vote tallies for Clinton and Trump. What they failed to do (and were never intended to do) was measure the regional differences in support for the two candidates; these allowed Trump to rack up narrow wins in battleground states, while Clinton piled up votes in states she had won anyway—with the result we all know.

Some information is infinitely better than no information, but without widespread testing there’s no way the UK could safely exit lockdown—because of localised patterns of COVID spread. And without continued testing and contact tracing, we wont be able to maintain that exit over time. People talk as if we’ll be able to lift the lockdown if we “flatten the curve” but the only thing that will allow us to lift the lockdown—in a controlled way—is mass testing.

There’s a baseline mortality to COVID-19 of around 1%, but numbers from around the world appear to show that mortality increases as a health system comes under stress: you can’t save everyone, but with enough medical staff and equipment you can save a more lives than if you are inundated. That’s why I find the UK models that show that the majority (40 million) of the UK population could contract COVID-19 in short order (less than a year?) for the loss of a maximum of 400,000 lives pretty unbelievable. They suggest that medical care has no effect on mortality rates, and that feels wrong.

Recent Posts

See All

I worked at a data company for 30 years. I was a software developer, not a data scientist, but I implemented other people’s models in code, and I wrote a couple of utterly trivial models of my own. I’