Government policy is being dictated by guesswork

Ministers don’t understand how scientific modelling works, which makes ‘the science’ far from fixed

When the Government sent us into lockdown seven weeks ago, we were told it had to be done or a quarter of a million people would die. The origin of this belief was the paper published by Professor Neil Ferguson and his Imperial College team on March 16, which claimed that a herd immunity strategy would kill 250,000 people. This immediately became “the science” and was treated rather as if the clouds had parted and divine truth had shone through.

But it hadn’t. All that happened was that one group of scientists, working with fragments of observed data and making various assumptions, ran a mathematical model which produced a frighteningly high number. It was treated with a reverence it did not deserve because too few commentators and government ministers really understand the difference between observation and scientific modelling.

When scientists count and measure things they can generally be trusted to work to objective principles. There is still room for disagreement

– as Cambridge professor of the public understanding of risk, David Spielgelhalter, wrote this week, defining what constitutes a death from Covid-19 is far from straightforward – but when you are presented with observed data it is not unreasonable in many cases to treat it as fact.

But modelling should never be treated as such. When scientists build models they are creating their own little digital worlds, making huge assumptions along the way. What you get out is only as good as those assumptions – and they are often way off-beam. Modelling in epidemiology is really no different from economic modelling – and we know how often that has gone wrong, such as with the Treasury’s forecast that unemployment would rise by between 500,000 and 800,000 within two years of a referendum vote for Brexit. Actually it fell to the lowest level in 45 years.

To take the infamous Imperial paper, it assumed that 30 per cent of people who are infected with the virus Sars-CoV-2 will show symptoms of Covid-19. Subsequent studies from Italy and China, using observed data, suggest that 50-80 per cent is more accurate. The Imperial paper also assumed that the infection mortality rate is 0.9 per cent. The Imperial team has already revised this down to 0.66 per cent, and some epidemiologists believe it is far lower still. Those two factors completely undermine the 250,000 figure.

The usefulness of the Imperial team’s modelling became apparent at the end of March when another paper from the team predicted that with social distancing measures in place, Britain was now heading for just 5,700 deaths. The toll is now over 28,000. Yet still the Government seems to be beholden to Imperial’s predictions – the latest one being that 100,000 people could die if lockdown is lifted too soon.

The original Imperial paper also suggested that almost a third of over-80s would be admitted to hospital with coronavirus, 71 per cent of whom would need critical care. As this newspaper reports today, this projection resulted in Nightingale hospitals remaining empty, as models failed to take into account that ventilator use is rarely advised for elderly people.

Epidemiology necessarily relies on some degree of modelling. It would be unethical as well as impractical to build two identical human communities, one subject to lockdown and one not, then to infect both with a pathogen and see what happens. So those communities get modelled instead. The trouble comes when scientists get carried away with the cleverness of their models and when non-scientifically minded ministers and civil servants think they are being presented with hard evidence.

It is the same with that other great hot potato of the moment, climate science. We can’t build two parallel Earths, pump in different levels of carbon dioxide to their atmospheres and then observe what happens. Yet the figures we get from crude models are treated as if they were the results of carefully controlled experiments. We don’t really know how the climate will respond to higher levels of carbon dioxide, any more than we know how many people will contract Covid-19 next week.

Over the next few months scientists will be undertaking controlled trials of potential vaccines for Covid-19. They will be meticulously observing what happens when real, human volunteers are administered either a shot of the vaccine or of a placebo. The contrast between this rigorous science and the guesswork which has governed policy on coronavirus to date could not be greater.