In Defence of Neil Ferguson

Simon Anthony responds to Toby Young’s and Ben Lewis’s criticism of Neil Ferguson

Artillery Row

A dramatized, near-namesake of mine declared that he came to bury Caesar, not to praise him.  For a while I wondered whether I’d write a piece on Neil Ferguson on a similar theme.  It’s not turned out that way.  That’s not to say I’m going to praise him and his model (which, unlike Caesar’s wife, must not be above suspicion) but I’ll try to explain his work and why he really couldn’t have come to different conclusions.

Were these criticisms to be partly or even entirely true, they’d still be almost completely beside the point

In the two months since it was published, the Imperial College paper and computer model for Covid-19 have been widely criticized for, among other things: the poor quality of its code, its opaqueness, the variability and instability of its predictions, the lack of QA, the long delay before it was made available and the apparent polishing of the code by Microsoft engineers before it was fit to be seen in public.  As recently as the past few days, fresh attacks have been made in the Telegraph and Spectator.  Were these criticisms to be partly or even entirely true, they’d still be almost completely beside the point.  They amount to claims that the model’s predictions might have been out by ~20%.  When the paper was published in mid-March uncertainties of such a size were utterly irrelevant to the consequent decisions.

With a reproduction number, R0, (number of new infections due to each current infection) of 2.4 and an IFR (infection fatality rate – percentage of people who become infected who go on to die) of ~1% (both values, derived from Chinese data, were widely accepted as the best available in early March) – and given the well-established effects of Covid-19 infection – any realistic model would have concluded that intensive care demand would be more than 30 times greater than available facilities and, over the course of a few weeks, about 500,000 people would die.  Simple models, complex models, agent-based models, statistical models, any plausible model based on the evidence available when the paper was published: all would have led to similarly frightening conclusions.

I don’t know what other advice Boris Johnson sought but, had he spoken to any other competent epidemiologists they’d have backed up the conclusions of the IC paper as the most likely outcome.  They might have pointed to uncertainties in the R0 and IFR numbers but they’d have had no grounds at that time for claiming them to be materially wrong or for convincingly arguing for alternatives.

Johnson was faced with a situation in which, if he continued with the then current policy, half-a-million people were likely to die from Covid-19 over a couple of months.  On top of that, there would be many more non-Covid-19 deaths as the NHS effectively collapsed, with the risk that civil society and order might follow.  From mid- April to mid-July the daily death toll would have been of the same order as that on the first morning of the Battle of the Somme, repeated each and every day for almost two months.  No Prime Minister, no democratic leader could risk such an outcome if there were any better alternative (even Ghengis Khan or Tamburlaine might have thought twice, at least about inflicting such carnage on their own people).

To worry about 20% differences in mortality under those circumstances is like sitting in a beach bar, knowing a tsunami is headed your way and debating whether the incoming wave will be 12m high or only 10m.  In either case, it won’t affect your decision on whether to have another piña colada.

To worry about 20% differences in mortality under those circumstances is like sitting in a beach bar, knowing a tsunami is headed your way and debating whether the incoming wave will be 12m high or only 10m

One day someone will be given measurements of the size, location and velocity of a rock.  He’ll put that data into his computer model of solar system dynamics and he’ll be the person to tell the world that an asteroid the size of Texas (was that in “Armageddon”?) is likely to hit the Earth.  Relatively few people will be able to understand the equations which the model uses but the prediction will be so robust that the details won’t matter – simple models, complicated models, add special and general relativity, dark matter, rogue black holes – none of those things will change the unpleasant future: the asteroid is coming our way.

The trajectories of neither asteroids nor epidemics can be predicted perfectly: the position, size and velocity of the asteroid will be uncertain, just as the values of R0 and IFR are uncertain for an epidemic.  With Covid-19’s R0 of 2.4 and IFR of 1%, Ferguson was faced with a similar situation to the celestial mechanic whose central, strongly peaked prediction is that the asteroid will hit the Earth.  It’s not Ferguson’s model that predicted the course of Covid-19: it’s the all but inevitable consequence of a virus with those characteristics.  Ferguson’s model just happened to be the public bearer of bad news.  To scapegoat Ferguson for the damage done by Covid-19 is to shoot the messenger in the primitive hope that the sacrifice will somehow make reality better.

Enjoying The Critic online? It's even better in print

Try three issues of Britain’s newest magazine for £5

Subscribe
Critic magazine cover