Statistician: + evidence, risk, probability, chance, uncertainty. Emeritus Prof of Stats. Author Art of Statistics, co-author Covid by Numbers. Own opinions.
Polite request to PM and others: please stop using my Guardian article to claim we cannot make any international comparisons yet. I refer only to detailed league tables-of course we should now use other countries to try and learn why our numbers are high
appalling misleading headline from the Guardian: I said I didn’t take the pandemic ‘seriously enough’, but this is turned into ‘didn’t take Covid seriously’. Unacceptable.
Latest data from ONS 30 mins ago Just drawn this extraordinary, and shocking, graph. Why the massive rise in non-covid deaths? How much is judgement of registering doctor? How much is collateral damage of lockdown???
@JuliaHB1
If you test 1000 people at random, latest ONS figures estimate 1 will have the virus, and let’s assume you find them. But with an FPR of 0.8%, that’s 8/1000, and so you expect to find 8 false positives. That’s 9 positive tests, only one of which has the virus . Hope this is ok
Diplomatically-worded but strong public rebuke from Chair of UK Stats Authority to Matt Hancock about the test numbers quoted each day - to 'support trustworthiness'
shouldn’t
@BBCNews
report this, instead of grossly misleading statements like "the UK has hit 10,000 new cases for the first time since mass testing began”
Nearly 13K COVID-19 positives reported for the UK yesterday.
Technical issues delaying the reporting of data (now resolved) means announcements for the next few days will be inflated to clear the backlog.
This chart shows most of yesterdays reported positives by specimen date.
For deaths and admissions, big gap opening up now between over and under 65s. For over 75s, deaths each day have been essentially halving every *week*. Extraordinary.
‘no safe level of alcohol’, but Lancet study means 25,000 light drinkers need annual equivalent of 400,000 bottles of gin for one serious health problem among them each year
Back-of-envelope. If 5 events *were* caused by AZ vaccine, expect 1 event in 2 mil jabs. If 2 mil people in 50s put off their jab by 1 week, expect ~2,000 to get infected, 20-30 seriously ill, 5 to die. So benefit/risk ratio high even for *1* week. But risk goes on in unjabbed!
PHE's new definition of Covid deaths leads to massive reduction, but BBC 10 pm News still reports that 18 people ‘died from Covid in the last 24 hour period’. No they didn’t - these are reported deaths. Why can’t the BBC just get this right?
If now no need for confirmatory PCR after asymptomatic + LFD, why is PCR necessary after *symptomatic* + LFD? When chance of false positive is even lower? Genuinely unsure what the logic is.
Back-of-envelope sums:
@ONS
survey suggests ~ 7% infected in England = ~ 4 million. ~ 40,000 deaths linked to COVID, so very roughly a 1% average mortality rate. Given multiple uncertainties, impressively close to much-disputed 0.9% estimate by Imperial team back in March
If you feel like a chortle over my (lack of) musical taste, listen to me talking to the wonderful
@laurenlaverne
on DID at 11am tomorrow, Radio 4. Rather rare to have a life’s ambition fulfilled.
In the room yesterday at No 10 was Johnson, Whitty, Valance, and an enormous unmentioned elephant with ‘Centralised Test and Trace’ written on its belly. 6 weeks ago Sage said it was probably having ‘marginal impact’ and it’s been downhill since then
“the most consistent predictor of decreased susceptibility to misinformation about Covid-19 was numerical literacy” - great study (COI, by my colleagues) Data literacy is vital
Isn't there an option missing? Truly awful survey question from Office for Product Safety and Standards at BEIS. But it does provide great material for teaching about manipulative phrasing
To put recent Covid death figures in perspective, note that in Week 3 (ending Jan 21), fewer people were registered as dying due to all respiratory diseases (including Covid) than in the same week in 2015, 2017 and 2018.
@BBCNews
10pm 5th Aug said “On average, in the last week, 58 people have died everyday from coronavirus”. This is grossly misleading - PHE is reporting deaths of people who once tested positive. True number dying with Covid is perhaps 15-25 a day.
No medical intervention is 'safe', and balance of benefit to risk is crucial. MRHA has identified 30 serious thrombotic events with low platelets out of 18 million AZ vaccinations so, if causal, average risk of less than 1 in 500,000 is low.
PM says airport test would only pick up 7% of cases, and this means "93% of the time you could have a real false sense of security”. This is hopelessly wrong. Even if 7% were true (evidence?) the vast majority of people will have *correct* negative tests
Blog on fatal risks of Covid to children and younger adults eg women aged 30–34, around 1 in 70,000 died from Covid over peak 9 weeks of epidemic. Over 80% pre-existing medical conditions, so if healthy less than 1 in 350,000, 1/4 of normal accidental risk
My calculations of risks to the population based on yesterday’s ONS data Extraordinary absolute risk gradient by age. Relatively, Covid risk has been around 50% extra than normal risk for over 45s, much less for under 45s
Another strong rebuke from
@UKStatsAuth
to Matt Hancock for the way data on tests is reported: 'The aim seems to be to show the largest possible number of tests, even at the expense of understanding.' Data should help the public understand both the epidemic and testing programme
My BMJ paper on using the idea of ‘normal’/background/actuarial risk to communicate the huge range of risks from COVID-19 experienced by people of different ages
*Personal Tweet* Extraordinary feat by part of my family climbing 3 Peaks (Snowdon, Scafell Pike, Ben Nevis) over 6 days last week in bad weather, with 3- and 6-year old grandchildren walking all the way. In aid of a fine charity working in Cameroon -
Completely unacceptable that UKHSA put out absurd statistics showing case-rates higher in vaxxed than non-vaxxed (Fig 2), when just an artefact of using hopelessly biased NIMS popn estimates. Feeding conspiracy theorists worldwide.
For Nerds. Politicians put up curves showing cumulative counts of covid-19 cases and deaths. Journalists want to know if we are flattening off. They are asking about a third derivative.
This paper motivates the call for the end of significance. A 25% mortality reduction, but because P=0.06 (two-sided), they declare it ‘did not reduce’ mortality. Appalling.
Our new interactive site displaying wonderful ONS data on all-cause, COVID-19 and excess deaths in England and Wales. Shows that ‘excess’ can be very sensitive to choices
My blog on the COVID risks faced by *population* during the peak of the epidemic (great ONS E+W data)
1 in 1800 died over 5 weeks. Risks doubled if a man, and for each 6–7 years extra age: compared to 20-year-old, an 80-year-old had ~ 500 times risk
COVID deaths within 28 days of +ve test may reach 100,000 today. But ONS data show that over 100,000 people in the UK had died with COVID on death certificate by *Jan 7th*, nearly 3 weeks ago. 108,000 by Jan 15th, and total now will be nearly 120,000
Yesterday PM said things were worse than Reasonable Worst Case Scenario, and yet this RWCS has not been published, just leaked to some media, eg Spectator
So much for transparency. Need fixed timetable for releasing Sage docs, not just at whim of No 10
Rather than a queue, The Queue is perhaps better thought of as a pilgrimage, in which the journey is as important as the goal. Yesterday reminded me of walking the Camino to Santiago – dogged continuity made hugely enjoyable by meeting wonderful people who look out for each other
very thoughtful piece on risk-based shielding from Raj Bhopal "No one is advocating “letting the infection rip”, a phrase only heard from the lips of people who do not want this discussed"
Another day, another awful IHME study published in the Lancet Remember all the confident 'all alcohol harmful' headlines in 2018? Well now they have completely changed their minds. 1/3
But they call any risk above non-drinkers ‘harmful’, and so say young people should not drink. By this logic, they should not drive, or even get out of bed. No absolute risks provided, which breaks Lancet guidelines and provides no basis for policy recommendations. Awful. 3/3
excellent opportunity to see the staggeringly linear relationship between log-death rates and age! Meaning risks increase exponentially between 5 and 95, doubling for each 5-6 years older
We have just released a number of new features on the
#UKCovdi19
Dashboard (🧵)
1. We now have an option to view graphs in logarithmic scale - I know some of you really wanted this! Feedback are welcome.
Many thanks to
@statsgeekclare
and
@DaveJephson
for the QA.
1/🧵
How have Covid-19 fatalities compared with other causes of death? For each age-band, compares with ‘normal’ flu and accidents. Extraordinary gradient with age.
Modelling excellent data from
@ONS
on COVID deaths in March Clear pattern: men have double the risk of women the same age, and death rates around 11% higher for each year older, ie they double for every 7 years older.
Awful misleading BBC News headline y'day “highest number of deaths recorded in a single day since the pandemic began” implying 1,610 died in last 24 hrs. But Tuesday’s *reported* deaths always high: truth probably nearer to 1,000 (which is high enough) 1/2
what a great demonstration of the need for
@Nat_Numeracy
!! I hope every maths class is using this in their lessons today. Remember, this man has 144K Twitter followers...
My new blog on whether Covid raises everyone’s relative risk of dying by a similar amount. Latest ONS data shows that, of recent death registrations, the proportion linked to Covid does not depend on age.
Very detailed slide pack from PHE and NHS Test and Trace with reasons for every Tier decision
But it doesn’t seem to turn up on a Google search! Need a good website with all this info there
BBC: "Grant Shapps confirms repatriation of Thomas Cook customers...will cost taxpayer around £100m. He tells BBC's Today: "We know that the Monarch rescue was about £50m. We know that this is about 50% bigger.” "
I suggest a
@Nat_Numeracy
course
@IsabelOakeshott
says Covid was “a disease with a low mortality rate among all but the very elderly”. Around 1% of men my age (66) who caught it in 1st wave died. This is about the risk of coronary bypass graft surgery, 1,000 parachute jumps, or 70,000 miles on a motorbike.
‘Obscure’?!! Some of us have spent much of our working lives exploring Bayes theorem. (Great article - I suspect this is the sub-editor’s rather than Tom’s label)
In the 10 days up to feb 6th, deaths were falling at twice the rate in over-85s compared to under-65s. This is good news. [sorry, just put up yesterday’s graph!]
Just to show I was not exaggerating on
@DesIsDiscs
, here is my Dad doing some escaping: note the 1950s Health and Safety comprising a man with a bucket of water
Just finished this Christmas beast of a jigsaw. Grey, gloomy and very difficult. But got to the end eventually. There’s a contemporary lesson there somewhere...
Well let’s see how this goes down in Germany, France, Austria, Poland, Sweden, Denmark, the Netherlands and Spain, who are all currently denying their older citizens this protection
Significant move from
@WHO
- they are following Britain in recommending the Oxford AZ vaccine be given 8-12 weeks apart, saying a longer gap "associated with greater vaccine efficacy" and will reach more people.
And recommend for over 65s too
For time of year, total and excess deaths are without recent precedent. But deaths without Covid as primary cause are at lowest level for 10 years. Minimal flu, and some brought forward by first-wave?
Let’s look at this claim again. ONS now says 221 Covid deaths on 11th January, 77% ‘caused’ by Covid that week, so that’s ~ 170. Less than half 379. So this claim was untrue (which was clear at the time). But it got 3K retweets.
Two years ago, could you have imagined a pandemic causing 379 deaths in a single day (UK), with politicians insisting the disease was “mild” and nobody in the media reporting it?
Isn’t there an essential mismatch in policy now? Opening up by relaxing measures, and yet (predictably) causing massive disruption to economy and families by telling millions each week to self-isolate. And to what benefit?
I am so sad about this. David was a hero to statistics in general, and me in particular. He took me apart, in the nicest possible way, in my PhD Viva 43 years ago, probably after skimming it on the train. 1/
Big piece finally installed! Designed by my friend for Norfolk Broads house, reflecting her view of the reeds. Made in three panels, but with rather nice flow between them. Small picture shows the wonderful green/brown glass better. Other work here
Blog by Sebastian Walsh on asking the right questions "Furthermore, it has never been “safe” to attend school. In the 2017-18 flu season, in which (unlike now) children were super-spreaders, 16 school children died. Were schools safe then? "
Vaulable statement from OSR "where models are referred to publicly, particularly to inform significant policy decisions, the model outputs, methodologies and key assumptions should be published at the same time" (COI: I am a non-exec on the UKSA board)
@trishgreenhalgh
sorry, this is not true. These are ‘reported’ deaths, always high on Tuesdays, it was 77 the day before. True count of deaths caused will be less than half 379 (still high, of course). Politely suggest deleting tweet to slow spread of misinformation
Latest PHE data: Out of 257 Delta deaths, 118 (46%) fully vaxxed, 92 (36%) unvaxxed. What we would expect with high coverage by very effective – but not perfect – vaccine.
Tweet below is for mass testing. Very different if test 1000 people *with symptoms*. then from PHE data expect 40-50 covid, detect say 30-40. So most positive tests are correct. The current rise is real.
@JuliaHB1
If you test 1000 people at random, latest ONS figures estimate 1 will have the virus, and let’s assume you find them. But with an FPR of 0.8%, that’s 8/1000, and so you expect to find 8 false positives. That’s 9 positive tests, only one of which has the virus . Hope this is ok
Our attempt at summarising Bayes theorem in words: "If you pick what looks like a needle from a haystack, then your confidence that it really is a needle depends on how many are in the haystack in the first place."
Today,
@d_spiegel
and I write about rapid testing in schools.
1️⃣ Test stats can be confusing.
2️⃣ The false-positive rate is not the probability a positive test is false.
3️⃣ Looking for needles in haystacks, your confidence depends on the number of needles.
Lots of talk about a 'peak', but this term can give a misleading impression of an obvious maximum. Reporting practices mean highly volatile daily counts, and underlying shape is more a rounded hill than a peaked mountain: look at Italy’s reports over last 2 weeks