February 19, 2021

expert reaction to latest data from the ONS Infection Survey

The Office for National Statistics (ONS) have released the latest data from their COVID-19 Infection Survey.

Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:

“There’s been a lot of good news about the pandemic this week, and the regular bulletin on the ONS infection survey provides some more. ONS estimate that the percentage of people in the community population who would test positive for the virus fell in the week of 6-12 February, compared to the previous week, in each of the four UK countries, and in every region of England. In England, the percentage who would test positive fell in all seven of the age groups that ONS provide estimates for. The data by age in the other three UK countries are rather more uncertain, statistically, because the numbers of people swabbed are smaller, but the numbers in the ONS data set indicate that the percentage probably fell at all ages in Scotland, Wales and Northern Ireland as well. The decreases over that single week are pretty large too – the numbers fell by almost a third in England and in Wales, by over a quarter in Northern Ireland, and though the fall was a bit smaller in Scotland, 18%, the numbers were already lower in Scotland than in the other UK countries.

“In England, ONS estimate that between 1 in 105 and 1 in 120 people would have tested positive on any day in the most recent week. That’s down from between 1 in 75 and 1 in 85 the week before, so roughly 210,000 fewer people testing positive. The numbers for the most recent week in Wales and in Northern Ireland are roughly similar to the English numbers, allowing for the statistical uncertainty that’s inevitable in making estimates from a survey. But in Scotland the level of infection is quite a bit lower, and ONS estimate that somewhere between 1 in 150 and 1 in 215 people would test positive.

“Infection levels across the UK are therefore a good deal lower than they were at the start of 2021. But, though the levels are lower, they are still not low. In all four countries, they are roughly at about the level of early October, when there was considerable concern about how high they were getting. The numbers are going down now, not up as they were in October, but they certainly need to go down quite a lot further.

“Looking at the data for different age groups in England, what’s interesting is that the estimated infection levels really don’t differ all that much between the groups, allowing for the statistical uncertainty, except in the over 75s where the rate is lower (under 1 in 200 would test positive). There have been particularly big falls in infection rates in children of secondary school age, where the estimated rate is now lower than every other age group except the over 75s.

“There’s also much less difference in infection rates between different English regions than there was a few weeks ago. ONS estimate the infection rate as highest in the North West, where it’s 1 in 85, and London, where it’s 1 in 100, but that’s not so different from the lowest regional estimates of 1 in 135 in the North East, South East and South West. There’s quite a lot of statistical uncertainty about these regional estimates, though, particularly for the North East where the number of swabs tested is smallest, and the ONS give the plausible range for testing positive there as 1 in 100 to 1 in 200.

Additional information

It’s notable that there are several differences between these ONS results and the latest interim results from the Imperial College REACT-1 survey that came out earlier this week. REACT-1 is carried out only in England, not in the other UK countries, but both it and the ONS survey aim to take representative samples from the community population in England, and to take swabs from them to measure whether they are infected. So you might expect them to give broadly similar results.

The main news from this week’s REACT-1 preprint was that infection levels had fallen by over two-thirds since their previous report in January. The fall in the average of the ONS estimates over the same period is rather less, though the ONS numbers still fell by over half. The same trend, then, even if the detailed numbers differ.

Also, the REACT-1 estimate for the percentage who would test positive for their most recent period, 4-13 February, is 0.51%, or about 1 in 195 of the community population, though their range to take account of the statistical uncertainty goes from 1 in 170 to 1 in 220. The ONS estimate for their latest period, 6-12 February, is 0.88%, quite a bit higher. That’s 1 in 115 of the population, with a range to indicate the uncertainty from 1 in 105 to 1 in 120. Those ranges don’t even overlap. (The ONS range is narrower mainly because they use different statistical modelling methods.)

However, this different doesn’t concern me too much. The time periods covered are slightly different. More importantly, the two surveys draw the people that they test from different lists (REACT-1 uses lists of people registered with GPs; ONS mainly uses a list of residential addresses), the numbers who agree to take part are not the same, the age groups covered are different (REACT-1 is age 5 and up, ONS is age 2 and up), the way that the swabs are taken is not exactly the same, different labs are used, and the details of the numbers of people tested in different places are not exactly the same. Those non-overlapping ranges for the estimates can take into account only part of the uncertainty in the results. The position is much as can happen with political opinion polls, where different polling companies use different methods and approaches, which can lead to systematic differences in their findings that go beyond what you’d expect simply from random variability. I’m certainly not saying that these infection surveys are exactly likely opinion polls – the number of people that they test is far, far larger than most opinion polls for one thing – but there are certain features in common. The key thing for me is that both surveys are showing a consistent trend downwards, falling fairly rapidly, even though they differ in the actual estimates and in the speed of that decrease. We’re very lucky in England to have two of these major surveys, so that one can compare them and so check that neither is giving a misleading impression.

What is interesting, though, is that the estimates from the two surveys on the position in the English regions do differ rather more. These latest ONS estimates put the positivity rate in the North East as the lowest in the country, though as previously mentioned there is quite a lot of statistical uncertainty about it because the sample size in that region is small (because the population of that region is relatively small). ONS estimate that infections have been falling pretty fast in the North East since mid-January. Despite the overall REACT-1 estimate of positivity being quite a lot lower than the ONS estimate, their estimate of the rate in the North East is higher than the ONS estimate, they reported a slower rate of decrease of infections there than in all but one of the other English regions, and they estimated that the R number for the North East was probably currently greater than 1. There are other differences between the regional estimates from the two surveys, but this is the largest. The sample size in the North East for REACT-1 is also relatively small, so there is a lot of uncertainty, and so this does not concern me too much. But it does illustrate that, even with these large surveys giving estimates for regions with a population of millions – the North East population is about 2.7 million – there’s still a considerable and unavoidable level of uncertainty involved.

https://www.ons.gov.uk/releases/coronaviruscovid19infectionsurveyuk19february2021

All our previous output on this subject can be seen at this weblink:

www.sciencemediacentre.org/tag/covid-19

Declared interests

Prof Kevin McConway: “I am a Trustee of the SMC and a member of its Advisory Committee.  I am also a member of the Public Data Advisory Group, which provides expert advice to the Cabinet Office on aspects of public understanding of data during the pandemic.  My quote above is in my capacity as an independent professional statistician.”