This morning’s so-called 'React study' — an attempt by Imperial College to estimate the prevalence of current Covid-19 infection in Britain — has aroused much interest thanks to its suggestion of a sharp fall in the R number. Its central estimate for R is 1.06, but it applies a range of between 0.74 to 1.46, with a 63 per cent chance that R is above the critical level of 1.
It was a rare piece of good news to lead BBC news bulletins — as well it might given that it was the previous report from the React study three weeks ago that led to the government introducing the rule of six.
But there are a couple of interesting details about the React study that have gone unreported. Firstly, its estimate of the number of people in England who are currently infected with Covid-19 is vastly higher than is suggested by official figures for confirmed cases. React, which tests swabs from a randomised sample of the population, estimates that between 351,000 and 478,000 people in England are currently infected, with a central estimate of 411,000.
Over the past fortnight, by contrast, 61,617 people have officially been diagnosed as having the virus. Even since the beginning of the crisis only 388,342 people have officially tested positive (the vast majority of whom tested positive more than a fortnight ago and should not still be carrying the virus). If the React estimate is anything close to reality, it suggests that the vast majority of infections are being missed, in spite of the huge increase in testing. If the government is only succeeding in catching a small percentage of cases there is zero possibility of controlling the virus through test and trace.
The React estimate, however, is out of kilter with the estimate arrived at by the similar ONS infection study. That, too, takes swabs from a randomised sample of the population. Its latest estimate, published last Friday, puts the number of people infected with the virus in England at between 85,600 and 123,400, with a central estimate of 103,600. React and the ONS are trying to study the same thing via the same means, so why are they so far apart?
There is, it has to be said, a difference in timing. The ONS swabs were taken between 13 and 19 September and the React swabs between 19 and 26 September. Given that the number of cases seems to have been doubling around every 10 days, this might account for some of the difference, but not all. There should not be an issue with sample size: React tested 84,610 people and the ONS 248,030.
According to Steven Riley of Imperial College, who leads the React team, there is a difference in how he and the ONS recruit volunteers to be swabbed: React recruits a different sample each time whereas ONS keeps people in its sample and swabs them repeatedly. But it is hard to see why that should make much of a difference. React and the ONS use different methods of testing the swabs, and it is possible that React might be more sensitive, picking up people with lower viral load. There is also the issue of false positives. But then for a period in the summer React was detecting lower levels of infection than was the ONS.
It is something of a mystery — and underlines how unreliable much of the data on Covid-19 still is. Yet the fact that the last React study was a big influence on the government’s introduction of the rule of six shows just how hard it is for advisers and ministers to resist responding to the latest set of figures.
Watch Professor Sunetra Gupta assess the government's coronavirus response on SpectatorTV