select search filters
roundups & rapid reactions
before the headlines
Fiona fox's blog

expert reaction to latest data from the ONS COVID-19 Infection Survey

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


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

“As usual, this week’s results from the ONS Infection Survey give estimates of the way that SARS-CoV-2 infection is changing over time in the English community population.  The results are based on swab tests in a representative sample of people across the country, who are tested weekly.  This means that, unlike the daily reports of new confirmed cases, or the weekly Test and Trace statistics, the ONS numbers are not affected by changes in the general availability of tests, or in the types of people and areas of the country where Test and Trace tests are concentrated.  Therefore the ONS data are more reliable.  Broadly speaking, the same applies to results from the REACT-1 research programme, and interim results from the latest wave of that study were published yesterday.  However, an advantage of the ONS survey is that its participants have follow-up visits and tests, generally at weekly intervals.  That allows estimates to be made of the rate at which new infections are occurring.  That can’t be done directly in REACT-1, because it does not involve repeated tests.

“Looking first at the estimates of the number of new infections each day (the incidence rate), ONS estimate that for the most recent week (ending 24 September), there were 154 new infections each day for every million people in the English community population.  That corresponds to 8,400 new infections each day across the country.  But this number is estimated from a (representative) sample, so there’s inevitably some statistical uncertainty, and ONS show that uncertainty by what’s called a 95% credible interval.  They are essentially saying that their data could be consistent with a daily rate of new infections somewhere between 6,500 and 10,700.  That’s quite a wide range, because the number of new infections in the survey sample is not very high (93 in the most recent fortnight).  However, even the low end of the range is above the average rate of new confirmed cases in England, reported daily by the Government.  That’s because the ONS figures will include more people who test positive but do not show any symptoms of Covid-19.  The central estimate of daily new cases from the ONS survey is slightly lower than it was for the previous week.  That does not indicate that the rate of new infections has definitely fallen, though, because of the fairly wide margin of statistical uncertainty.  The rate of new infections definitely rose quite substantially during August and early September, and the new evidence indicates that this rise may be levelling off, though that isn’t yet by any means certain.

“On the prevalence rate, the percentage of people in the English community population who would test positive for the virus, that’s estimated by ONS to be 0.21% in the most recent week (ending 24 September).  That’s about 1 in 500 people, or 116,600 in all in the population.  The margin of statistical uncertainty around this number is relatively smaller than for the new infections – the range runs from 101,000 to 133,100.  This number is higher than the previous week, when the central estimate was 103,600 people testing positive in the population, but it’s not a great deal higher.  Looking back over a longer period than one week, there’s a clear pattern of an increasing number of infected people since August.  It’s a bit too early to say that the rate of increase has slowed, though that’s a possibility.  But the numbers do still appear to be increasing rather than decreasing, so it’s a slowing in the increase, not a reversal of the increase.  The rate of infection is higher in the North of England (North-East, North-West, Yorkshire and the Humber) and in London than in the rest of England, and is higher in teenagers and young adults than in other age groups (though there’s quite a lot of statistical uncertainty about the rates in specific age groups, because the sample sizes are obviously smaller than those for the whole population).

“Yesterday’s REACT-1 report gave considerably higher estimates for the number of infected people in the English population.  Their central estimate was 411,000, with an interval (indicating the statistical uncertainty) from 351,000 to 478,000.  The corresponding ONS estimate was 116,000, with a range from 101,000 to 133,100.  That doesn’t even overlap with the REACT-1 range, by some distance.  The REACT-1 central estimate of the percentage of people who would test positive was 0.55%, again a lot higher than the ONS central estimate of 0.21%.  These are considerable differences, given that both sets of results are based on representative samples of the English population, that the sample sizes were fairly similar, and that they relate to very similar periods of time (ONS 18-24 September, REACT-1 18-26 September).  I have to admit that I do not know all the reasons for the difference, and I hope they will become clearer eventually.

“One definite difference is that REACT-1 is estimating the number of infections in the country, while ONS is estimating the number of people that would test positive (if they all had a swab test).  The underlying issue here is that the rtPCR tests, that both surveys use, can result in quite a number of false negative results – that is, people who are actually infected but whose test is negative.  REACT-1 make an adjustment for this, but ONS do not.  Without that adjustment, the central REACT-1 estimate would be about 308,000 – but that is still considerably more than the ONS estimate of 116,000.  It may seem strange that REACT-1 make this adjustment and ONS do not.  The ONS rationale for not making an adjustment is that the false positive and false negative rates for the test are not known very accurately, and they state that making the adjustments would not make a huge difference to the estimates anyway.  That is true, I believe, for rates of positive tests around the ONS estimate of 0.21% – at that level of positive tests the false negatives roughly cancel out with the false positives (people who aren’t really infected but have a positive test result anyway).  However, it’s no longer true if the rate of positive tests is around 0.55%, as REACT-1 estimate.  And, adjusted or not, there’s still a big difference between the estimated rates of testing positive.

“This might be something to do with the difference in the ways the participants are obtained.  The ONS results relate to the so-called community population – that is, they omit people living in communal establishments, such as care homes, prisons or university halls of residence.  The REACT-1 sample of participants is based on list of people registered with GP practices, and that could include people living in communal establishments.  But the proportion of people living in communal establishments in England was well below 2% at the most recent Census.  Even though the infection rate in some types of communal establishments is likely to be higher than average, I think this difference is unlikely to account for much of the discrepancy between the ONS and REACT-1 estimates.  The difference in the sources of information for choosing participants may also be a factor, in some way.  REACT-1 uses lists of people registered with GPs, while ONS uses lists of addresses drawn from a standard database.  Personally I do not know of characteristics of these two types of list that would lead to the sort of discrepancy that we can see, but such a difference might exist.  Another possibility, though now I’m definitely speculating, is as follows.  The new REACT-1 data this week are interim – they are based on results from only the first part of the REACT-1 round 5.  It could possibly be the case that, though the whole sample for REACT-1 round 5 is reasonably representative of the whole population, the people in the sample aren’t spread evenly across the whole time for Round 5.  Just maybe, the people who were tested in the first 9 days (so included in the interim report) came, more than average, from areas or groups with a relatively high infection rate, and estimates based on the whole of Round 5 could perhaps be lower.  We’ll (possibly) know more about that next week, when the full report on round 5 of REACT-1 are published.

“There is also information from the ONS survey on infection in Wales and Northern Ireland.  For Wales, the ONS statisticians report that the number of infections rose over the past six weeks or so, but that that rise may be levelling off.  That’s how it looks to me too, though (as ONS point out) there’s considerable statistical uncertainty because the number of swabs taken and the number of positive test results are both relatively low in Wales, so considerable caution about the conclusion is warranted.  The survey has not been in progress for so long in Northern Ireland, and the population is smaller, so it’s too early to assess trends there.  The ONS survey is now also being carried out in Scotland, but not enough results are yet available for any findings to be published.”


Prof James Naismith FRS FRSE FMedSci, Director of the Rosalind Franklin Institute, and University of Oxford, said:

“The ONS surveys are the most reliable guide we currently have to the progress of the epidemic.  Unfortunately it is a lagging indicator, this release stops on 24th September.  The ONS studies demonstrates beyond any doubt, the virus rose rapidly from the end of August to mid September.  There is some evidence that the rate of increase may have slowed which is encouraging.  This finding is consistent with other data, but it is too early to be certain.

“As of the date of the survey, over 100,000 people most likely had covid19.  Cases have risen fastest in the 12 to 24 year old age group.  As a group, this age range is amongst the least likely to suffer serious illness from covid but some of them will and will be left with life changing complications.

“We do not have enough data to be sure, but the rate of rise in positive cases of those aged over 70s (the most vulnerable group) is only slightly lower than those aged 35 to 49 and 50 to 69.  A policy which effectively shielded the elderly to allow others to return to normal life would, from a scientific point of view, greatly reduce deaths.  However, based on the actual data, there is no evidence that we can effectively shield the most vulnerable from a rise in infections.  I would therefore advise our politicians, who quite correctly must take these decisions, that releasing social restrictions now is very likely to lead to a significant number of deaths in the vulnerable population (as well some harm in the younger generation).  I would further advise that actually demonstrating, as opposed to talking about, an effective shielding of the vulnerable policy, as measured by this ONS survey, should be a requirement before loosening of social restrictions.”


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


Declared interests

Prof Kevin McConway: “I am a Trustee of the SMC and a member of the Advisory Committee, but my quote above is in my capacity as a professional statistician.”

None others received.

in this section

filter RoundUps by year

search by tag