select search filters
briefings
roundups & rapid reactions
Fiona fox's blog
The Office for National Statistics (ONS) have released the latest data from their COVID-19 Infection Survey.
Prof James Naismith, Director of the Rosalind Franklin Institute, and University of Oxford, said:
“The ONS data report data up to 14th August. The prevalence of covid has levelled off in England (perhaps a slight decline) at 1 in 80 (75 to 85, 95% CI) compared to 1 in 75 (70 to 80, 95 % CI). A similar trend is seen in Scotland.
“This is not evidence for a steep decline in infections. Rather the number of infections in the third wave peaked late July and is falling slowly.
“The ONS estimates the number of infections per day per 10, 000 people. This data only report to the end of July, this suggests that for England the peak in cases occurred around 17th July with 76 000 cases a day and this has fallen to 54, 000 by 31st July.
“These numbers should serve to remind us that the daily positive cases are at best only a guide to where we are and can mislead.
“The discrepancy between the numbers and those entering track and trace are one of the reasons it is ineffective.
“Where are we today? There is no value in following day to day variations. The rolling 7 day average is more useful, this suggests we are seeing a rise, most likely due to relaxation in restrictions. I expect a rise in cases as schools go back and people return in doors. I do not know how large any fourth wave will be.
“It is welcome to see the caveats attached to models being communicated and accepted. It is profoundly depressing and ultimately self-defeating for us to pillory scientific modellers for being “wrong” or “changing their mind”. The UK has excellent modellers, in their work the uncertainties and assumptions are always detailed. Such things are hard to explain in a 20 second sound bite. We should try to report the uncertainties as clearly as headline numbers.
“The sure sign of charlatan in viral epidemiology is a claim to have always been right. Good scientists are always revising their ideas with more data and careful to distinguish between what they think and what they “know”. This behaviour is so different than politics, where we are used to mistake free actions and consistency of belief despite new evidence.
“The good news is the vaccines have been shown to be safe and effective, with Moderna shown to be safe and beneficial for 12 to 17 year olds. Reaching the vaccine hesitant adult population, should be priority.”
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“The latest bulletin on prevalence of infection with the virus that can cause Covid-19, from the ONS Covid-19 Infection Survey (CIS), takes the data up to the week 8-14 July (last week). In terms of very recent trends in infections, the picture is rather mixed, with different trends in different parts of the UK. In terms of the overall level of infection, compared to last summer, the picture remains rather depressing. For England last week, ONS estimate that 1 in every 80 people in the community population in England would have tested positive for the virus. That’s a somewhat better position that the week before, when the estimate was 1 in 75, and ONS estimate that the numbers infected continued to decrease in the latest week, as they have now done since late July. However, a year ago from last week, ONS were estimating that only 1 person in every 2,200 in the English community was infected, so roughly 25 times as many people in England were infected last week as they were a year before.
“It’s true that vaccination has very much decreased the chance that a person will be hospitalised, or die, if they become infected, so it’s obviously wrong to say that things are 25 times as bad in terms of Covid-19 as they were a year ago. But 1 in 80 people testing positive is a lot of infection, and despite the best efforts from the NHS and the governments of all four UK countries, there are still quite a lot of unvaccinated people. In estimates from the CIS, the infection prevalence rate in England has been higher than 1 in 80 only for six weeks in late December 2020 and January 2021, and for four weeks from early July to early August 2021. (The infection rate was very probably also higher during the peak of the first wave in spring 2020, but the CIS had not begun back then.) It’s now long enough after the removal of restrictions in England on 19 July for any effect of that change on infections to have begun to feed through into the CIS figures, so it’s somewhat encouraging that the English prevalence is still falling (slowly) – but schools in England are still on holiday, so the mixing associated with schools isn’t happening. I’m not going to try to predict what will happen next. There’s a lot of uncertainty.
“The daily counts of new confirmed cases in England on the dashboard at coronavirus.data.gov.uk fell from mid-July till roughly the start of August, but since then have been pretty well static, with some hint of an increasing trend. That’s similar to what the CIS figures for England show, but certainly not identical. There are at least two reasons why dashboard and CIS short-term trends can differ. First, the daily confirmed case figures depend on people testing positive when they have a Covid-19 test because they have symptoms, or because it is needed for work or school, or some similar reason. So they can change because the numbers and types of people turning up for testing change, and that can lead to biases. Those biases don’t apply to the CIS, which tests a representative sample of people regardless of their job or symptoms, simply to track infections across the country. But also the daily confirmed case figures include new cases only, while the CIS prevalence estimates count everyone with an infection regardless of whether it has just begun or has been there for some days. This means that, if the trend in new infection changes direction – say it starts increasing after it had been decreasing for a few weeks – that change in trend will take longer to show up in the CIS prevalence data.
“The positions in the other three UK countries are different, both from the position in England (in some cases) and from one another. An issue is that, statistically, the CIS estimates in the other countries are less precise than the English estimates. That’s because the statistical accuracy depends on how many people provide samples for the survey, and the sampled numbers are smaller in the other three countries than in England. (They are chosen to be roughly proportional to the populations of the countries.) ONS estimate that the number of people testing positive in Scotland is still falling, as it has been since mid-July, and now they estimate that 1 in 200 people there would test positive. That’s well under half the infection rate in England, even taking into account the statistical uncertainty. That does roughly match the position on new confirmed cases on the dashboard, which have been falling in Scotland since about the start of July. However, there’s some evidence of an increase in confirmed cases in the Scottish dashboard numbers, since the start of August – quite a slow increase so far, but if it’s a real increase and not just an effect of who is being tested, it hasn’t yet fed through to the CIS numbers.
“It’s harder to interpret the results for Wales and Northern Ireland, but the position looks worse than in England and Scotland. In Wales, ONS estimate that the number of people testing positive increased in the most recent week. The increase looks quite substantial – the official estimates are that 1 in 130 people would test positive in Wales in the most recent week, compared to 1 in 220 the week before. However, the statistical margins of error are rather wide, and the estimated positivity rate is considerably lower than in England even after this increase. In very broad terms, the increase isn’t surprising given what we’ve seen in new confirmed cases on the dashboard. In Wales, they fell from mid-July to the start of August, but since then have risen again to round about the mid-July level. In Northern Ireland, ONS say that the trend in infections was ‘uncertain’ in the week ending 14 August, and the sample size in Northern Ireland is the lowest of all the UK countries so there is a lot of statistical uncertainty. ONS estimate that 1 in 50 people would have tested positive there in that week. Although the margin of error runs from 1 in 40 to 1 in 70, even allowing for that, the rate of infection is pretty clearly higher in Northern Ireland than in the other UK countries.”
Further information
“ONS, as always, give prevalence estimates for subgroups of the English population, divided up into regions and into age groups. They provide some similar estimates within the other UK countries too, but the numbers of people samples in those groups are usually very small, so that the estimates have very wide margins of error and hence very difficult to interpret sensibly. Even in England, there is a lot of statistical uncertainty about the regional and age group estimates, so that trends are often not very clear. For the most recent week, ONS consider that the rate of testing positive fell in the West Midlands, East Midlands, and North East, and possibly rose in the East of England, but they also warn against too detailed interpretation of trends in regions (and age groups) because of the small number of people swabbed in these groups for the survey. In age groups ONS report a falling trend in those of secondary school age (school years 7 to 11, so ages 11/12 to 16/17), those aged 25-34 and those aged 70 and over. Trends in all the other age groups that they consider are said to be ‘uncertain’. The overall pattern of which age groups have the highest rates of infection haven’t changed. Prevalence is highest in those up to age 24, with 1 in 45 testing positive from age 2 up to school year 11 (age 16/17), and 1 in 35 for school year 12 to age 25, and very much lower in older people, 1 in 310 for people 70 and over.
“As usual, I’ll mention briefly that ONS also provide incidence estimates – that is, estimates of the daily numbers of new infections (so not including infections that started earlier). In principle these should be a better match to the counts of new confirmed cases on the dashboard – but in statistical terms, incidence estimates from a survey like the CIS are inevitably less precise statistically than prevalence estimates, and also they take longer to produce, so that they are two weeks behind the CIS prevalence estimates and their latest week is 25-31 July. In very broad terms, the recent trends in incidence from the CIS in England and Scotland do match the dashboard new case trends – falling England from mid-July to the end of July, and falling in Scotland throughout July. But they do not match so well in Wales or Northern Ireland. New confirmed cases on the dashboard in Wales fell from mid-July to the end of July (and since then have risen), but that fall didn’t really show up clearly in the CIS prevalence estimates and the broad trend was upwards in July. In Northern Ireland, trends are harder to distinguish both in the CIS and on the dashboard, because of the relatively small population and the relatively small CIS sample size. But the trend in July in the ONS incidence estimates for Northern Ireland was quite strongly upwards, different from the dashboard new cases which were relatively level. It’s not really possible to say whether these differences in trends are due to biases in the dashboard estimates because of who is being routinely tested, wide statistical margins of error in the CIS estimates, or some combination of both.”
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.”
None others received.