Coronavirus predictions bedeviled by the details

Until last week, the UK government under Boris Johnson had been oddly relaxed in its response to the coronavirus pandemic. Unlike many other nations that had closed schools and restaurants and banned gatherings of even five people in an effort to curb the spread of Covid-19, the UK had allowed life to continue much as normal, only testing patients entering hospitals and
requesting those with symptoms to self-isolate.
That’s changed significantly, with the government now advising people to work from home and to avoid public transport and gatherings with friends and family; restaurants, gyms and movie theatres will be ordered to close. The about-face came after hundreds of scientists from around the world attacked the earlier policy as needlessly risky. And an epidemiological modeling group at Imperial College London — a key scientific resource for the government — revised its estimate of how soon serious cases would overwhelm the National Health Service, adding that the previous policy could have resulted in roughly 250,000 deaths.
It’s good to fix a policy mistake, and the task now is to move forward while learning how to avoid similar mistakes. The biggest problem, it appears, was how decision makers handled — or in this case mishandled — modeling uncertainty. Sophisticated models look impressive and have their uses, but they can be less useful than simpler models if their predictions depend on a few small details.
The group at Imperial College used a complicated model to simulate the spread of the disease, as well as the effects of various countermeasures. It included millions of individuals; realistic patterns of human movement and contact in homes, businesses and public places; and details of how the virus can spread. That’s all good. Models like these are important precisely because they allow modelers to include government responses and test the likely consequences of a range of policy scenarios.
There’s a drawback, however. Models of this kind depend on parameters for such things as the incubation period of the virus and when people either with or without symptoms can pass it on to others. The values of these parameters are uncertain. The early UK policy was based on the belief that one key parameter — the fraction of hospitalised people needing intensive care — was lower than it turned out to be. Indeed, the actual number seems to be roughly twice as large as initially expected, rendering the earlier modeling results irrelevant.
This is an issue mathematicians have been writing about for a number of years: It’s particularly easy for policy makers to misuse complicated models. “Modeling — especially complex modeling — can promote something of a fairy-tale state of mind,” says Erica Thompson of the London School of Economics and the London Mathematical
Laboratory.
In essence, the old UK policy aimed for an optimal strategy in the midst of great uncertainty, hoping to thread the needle between disasters of public health on one side and economics on the other.
—Bloomberg