UKH

From today's Telegraph:

"Death scenarios used by the Government to justify a second national lockdown are out of date and may be four times too high, research suggests.

At Saturday night's Downing Street press conference, scientists presented graphs suggesting England could see 4,000 daily deaths early next month.

The scenario, from Cambridge University, was used as part of efforts to justify the introduction of sweeping restrictions. But data experts have questioned why the scenario – drawn up three weeks ago – was chosen to illustrate the crisis when the university has produced far more recent forecasts which are significantly lower.

The modelling presented on Saturday is so out of date that it suggests daily deaths are now around 1,000 a day. In fact, the daily average for the last week is 260, with a figure of 162 on Saturday."

https://www.telegraph.co.uk/news/2020/11/01/death-scenarios-used-government...

https://covid.joinzoe.com/post/covid-rates-are-not-surging

So what? Is 1000 a day or 365,000 per year ok with you?

Anyway, the govt published graphs showing multiple estimates so you can hardly complain it contains a range of estimates.

But screw the modelling, use your brain to spot a trend in bang up to date information.

https://coronavirus.data.gov.uk/deaths?areaType=nation&areaName=England

6th Oct 52 deaths per day (7 day average)

13th Oct 89 deaths per day (7 day average)

20th Oct 138 deaths per day (7 day average)

27th Oct 195 deaths per day (7 day average)

Increased by factor of 4 in 3 weeks. That looks like a surge in something important (deaths of my countrymen).

Well we shall see.

Inevitably the lockdown would come and on balance it was the best thing to do. OK some of Boris's decisions appear stupid, but equally he cant please even some of the people with the awful choices we are faced with. His communication of this could not be worse if he tried though.

We will see over the next few weeks and hopefully learn from whatever does happen.

Best of luck everyone and stay safe.

I'm no expert but I get a free telegraph occasionally for an alternative view (and because it has 2 codewords). In one last week the the general gist seemed to be pushing the view that Sage was following models that gave only worst case "possibilities" and that it was actually unnecessary to close down businesses. This differed from the impressions I get from Times, the "I", BBC etc. Obviously the Telegraph is still promoting this viewpoint.

Cherry picking the data again?

That was one worst case scenario in several charts used

I attached a screenshot of the relevant plot from the BBC News story [1].  This is the same plot as used in the government briefing [2] 

The point being made when this slide was given in the briefing was that all the independent models were predicting worse situations than the reasonable worst case the government had been planning to.

You and the Telegraph are cherry picking one model from many.  They are models, not set in stone.  The one that has been cherry picked is the high outlier.  The reason many different models are run is because there are so many unknowns and so many ways of formulating a model that it's not responsible to run with just a single model.   You look at many to get a feel for the statistics and sensitivity.  

Every single model in the analysis - if taken as accurate - was reasons to go for tougher covid control measures.  Focusing on one model as inaccurate - let alone the worst case of the lot - does absolutely nothing to change the conclusions that were drawn.

That just appears to be an outright lie.  The first model to hit 1,000 deaths/day does so around Nov 20th (see the guide lines I've scribbled on the screenshot).    [EDIT: I appear to be wrong on this point due to an embarrassing mistake reading the graph.  I stand by all the rest of my points]

ZOE:  The daily infection count that suggests is about half that the REACT study suggest.  This illustrates how bad current data is to predict from - a consequence of the rapidly worsening situation - and underlines why models are run over different possible parameters giving the board range of models from which you and the telegraph have cherry picked the extremal case.

What matters now is not case numbers but hospital occupancy.  Hospital occupancy today is over 3 times as high as it was when we locked down on March 23rd, and it's still rising rapidly, and we don't lock down for another 2.5 days, and last time we were coming out of flu season but this time we're going in to flu season.  This is the critical issue facing us right now.

Every single model could be thrown in the bin and we could reach the same decision on immediate, tougher risk control measures based on a linear prediction of hospital occupancy over the next month.  Except things aren't linear, they're worse.

I know you've recently espoused your fantasist's view that we've got enough healthcare to deal with a "no lockdown" scenario.  When challenged by myself and another poster to give an evidenced basis for this you instead tried to start a mass debate about the precise definition of "life expectancy" and then skulked off, not addressing that point beyond stating simply that you think it will be fine.  

I note that the title of your thread is deeply misleading - you use the plural of "scenarios" and then go on to discuss only a single scenario.  

[1] https://www.bbc.co.uk/news/uk-54767118

[2] https://assets.publishing.service.gov.uk/government/uploads/system/uploads/...

Well that explains the "cp" in their username...  Cherry picking is as easy as 1-2-3?

I remain in favour of the lockdown.  However, that argument is both overemotive and fallacious, because deaths are acceptable in other areas - influenza and road traffic being two key examples.

We actually do need to define a value for a life and an acceptable death rate.  This sounds really nasty, but it's necessary and is already done in other areas.

In your graph the purple line passes 1000 smack bang on the tick marker for Nov - its the left side of the red box you marked on.

Look you're obviously trying to con us with various dodgy statistics and made up numbers. We all see that even if you actually don't, but I think you do.

Instead of throwing mud why don't you actually put your balls and the table and start advocating alternative courses of action.

What would you actually do just now instead of locking down England? 

There was a 15-20 day lag between lockdown (23rd March) and peak in deaths (8th-12th April) which tallies with the progression of the disease for those who die (on average about three weeks after infection). Hence if we delay lockdown until Thursday we can expect a peak in deaths at about 800 per day in 18-23 days (20-25th Nov) as current quadrupling time is 21 days..

I think that suggests second wave comparable to first wave and a criminal failure to learn.

The delayed lockdown until Thursday allows deaths to grow by maybe 60% from Saturday announcement so delay costs thousands of additional deaths over November/December.

Just morbid mental arithmetic. Hopefully I'm very wrong.

Sure, I agree.

But... the points both Elsewhere and you make miss what I consider the key the point - the issue front and centre for both lockdowns is not death but hospital occupancy and the ability to deliver life saving care to anyone who needs it, not just Covid patients.  With widespread prevalence of Covid compromises our ability to deliver that healthcare for a whole bunch of reasons - not just direct use of beds but outbreaks, staff exhaustion, resource (e.g. PPE) exhaustion.

As the options are basically very low prevalence of Covid or rapid exponential growth to the brink (twice so far in the UK...) the need to keep healthcare going for all goes on to define an acceptable death rate for Covid by proxy - the death rate has to be very low because cases have to be very low because that's the only way to avoid exponential growth to the point we have to lockdown to preserve healthcare.  

So indirectly a death from covid has a far higher cost than a death from cancer or from a car crash, because the underlying processes that caused that death are exponentially toxic to healthcare.  

The Telegraph have been suckling the Great Barrington teat for some time.

I am still hoping the Barclays civil war will result in it being sold off to someone who will turn it back into a respectible Conservative leaning paper again instead of the rag it currently is.

You might be right - I read that as Nov 20th but now I see that is likely the year...  If the grid line is the 1st of the month that's embarrassing for me.

I stand by all my other points, and by the comment that this is an outlier from the models they are presenting.

Also, a key point from the slide "THESE ARE SCENARIOS - NOT PREDICTIONS OR FORECASTS"

You are attacking scene setting data from the presentation.

The critical side was the last one shown.  I've attached it below.

This shows actual hospital usage levels, the previous peak, the surge capacity now, and all predictions in the short term busting through there surge capacity.

The onus is on you to explain either why this extrapolation of the immediate term is wrong, or why the surge capacity is wrong, or why going in to lockdown won't help improve this.

Everything else is noise.  

LSTHM predicted 264 deaths at this point and Warwick 234........so these seem so far to be fairly reliable so far.

Even these projections for the next few months looked shocking compared to the previous reasonable WCS.

So why didn't the Telegraph focus on this rather than go on the attack on the outlier in an attempt to discredit the case for a further lockdown.....we all know the reason why.

And as if by magic...

https://www.manchestereveningnews.co.uk/news/greater-manchester-news/live-u...

Also worth mentioning that the ambulance service has upped the Resource Escalation Action Plan (REAP) to level 4, the highest level denoting extreme pressure, due to over 500 jobs outstanding at one point. I can't remember service overstretch ever reaching this level and we're still nowhere near peak flu season.

This doesn't bode well.

I saw this and wondered if it was something that might occur now and again....so you saying it is almost unheard of?

Do you work in the NW? 

I see the Czech Republic is now recruiting volunteers with no medical experience to do a training crash course in order to help its health service which is at breaking point...do you want this to happen here before we actually did somethung? 

Thanks for the link.  Worrying indeed.

If I read that link having been crapposting misinformation over the last few months, I’d feel like a right shit.

Why not spend a couple of months planning your life entirely around the best case scenario?

Let us know how it goes!

Hospital capacity is beyond my mental arithmetic.

Flu cases were dramatically reduced in Australia this year because of the precautions people were taking because of Covid.

https://www.cdc.gov/mmwr/volumes/69/wr/mm6937a6.htm

Let's hope we have the same.

If your objective was to get me to really focus my planning on how to eliminate every single person who works at, funds, or reads the Telegraph, well done! You're all on the list.

Rare as hens teeth.

Yes, though thankfully I'm off until Friday. Not that I'm expecting things to be much improved by then.

Worth bearing in mind that the Australian authorities were pretty hardcore in enforcing infection control measures, an attitude not much in evidence over here.

The plot in my 18:29 post summaries it nicely.  It’s not looking good.

My wife was talking to another consultant at work, they have 28 full beds on one ward, 26 are ready to be discharged but there is no secondary care for them to go to.  Covid is not the only problem for the nhs.

Cheers

Toby

Strange how quiet this 'cp' character has become in the last couple of hours. I suppose he has no need to come back now, toxic job of sowing biassed disinformation done.

I'm not sure I follow your reasoning for that pointing out disinformation is called disinformation, unless The Telegraph is now considered fake news?

I made a picture.

It's the hospital occupancy level either side of the March lockdown and the November lockdown.  I have done an exponential fit to the data for October to project it forwards to the lockdown because the last few days of data are subject to reporting lag, and because there's a few more days to go until November 5th.  This seems justified as these extrapolate cases were "locked in" before the lockdown was announced.

I've left a lot of white space at the top of the picture so you can experiment with drawing the red curve into the weeks ahead.  Last time there was a linear rise for about 15 days after lockdown before things started to level off.  This time round the curve is rising slower (longer doubling time, lower R) but it's also starting from a much higher level at lockdown.  Doing a linear prediction with a straight edge against my monitor over 15 days I get to about 28,000 people in hospital.  I didn't do this on the graph itself because it's phenomenology not science!  Still, have a go at tracing the curve ahead with your finger and see where you end up.

The plot from the government briefing on Sunday gives a capacity of 20,000 and a "Surge capacity and postponing some hospital services" level of about 26,000 people so 28,000 people is bad.  Very, very bad.  The kind of bad that sees the death rate from Covid going up quite a lot, and that sees a lot more urgent, non-covid, hospital work postponed.

This time round, lockdown is a lot less rigid.

Agreed, which is why I put the message of hope at the end of my post.

Would I be right as interpreting that as several times greater demand for hospital beds?

You're not pointing out disinformation.  You're pointing at an article that cherry picks one piece of information from many, when many are used to give a range of opinions in the face of uncertainty.  

You are failing to engage - at all - with the actual issue at hand, hospital occupancy.  Pointing at something else to try and sow uncertainty when the real problem is staring us in the face.

You are ignoring the key context on the slide the government presented - I copy and past it here (their capitals not mine, they must think it's important) "THESE ARE SCENARIOS - NOT PREDICTIONS OR FORECASTS".  

One piece of FUD bullshittery that I missed in my initial reply because I was looking at the BBC version of the figure and not the actual one the government put out...  Their actual figure includes shading on the curves for some uncertainty bounds.  It's hard to tell on the figure (the shading from most curves overlaps at this point) but I'd guess that the one cherry picked line  has about a 50% uncertainty bound that falls at around 400 deaths/day on Nov 1st, when the actuals (once reporting lag is resolved in 5-6 days time) look to be close to 300 deaths/day.  So actually the reality isn't so far removed from the confidence interval in that model.  The telegraph are claiming the model is out by a factor of 3.3x but if the confidence interval is only out by 0.3x the model isn't even provably wrong at this point.  

 I'm not very happy making predictions because I don't have anything more than a gut feel for how effective the new lockdown will be.  I'd go for about 50% more demand than the first time around if this lockdown is similar effective.

It's a different demand than last time with a lower (perhaps half?) requirement for ventilator beds.  

Pretty much. Its garbage. 

Just to back this up: 

"A union boss says he has 'never known anything like it' after the North West Ambulance Service was forced to declare a major incident due to high demand. NWAS Unison branch secretary Jeff Gorman said crews have never dealt with a situation like this before, and the situation is 'very extreme'."

https://www.manchestereveningnews.co.uk/news/greater-manchester-news/nwas-d...

But apparently not covid related,  particularly?

Having loads of crews self isolating is hardly going to help...

Not quite sure how they've arrived at that conclusion. Just spoke to a mate who was working today (12 hour shift that over ran to 13 hours). 5 of his 6 jobs were respiratory. Another friend tells me that several sectors have got sickness rates at around the 25% mark, mainly down to confirmed covid or staff needing to self isolate.

That's what's alarming about it. It might not be covid-related: it could be anything. A crash on the M62. A fire. A batch of dodgy, er, homeopathic remedies.

When any of these things can turn into 999 saying, "Don't call us, we'll call you" then the local health system won't be able to cope with those normally routine emergencies, because capacity to deal with them has been reached.

While you are here, why are Mondays busier normally anyway.

Difficult to get hard data but a few things at play, mainly people feeling unwell over the weekend wait until Monday to ring the GP and can't get an appointment, at which point they either ring 111, who send an ambulance, or they just ring 999...I've had people tell me they rang 999 after being kept on hold too long by the GP/111

Road traffic deaths is a poor example, they clearly are not accepted and never really have been, we live with countless expensive add-ons and restrictions to our motoring and road network to radically improve safety. We've had the easy wins and we're into the long tail where some residual deaths still occur but even those are still being actively pursued cause by cause.

The problem really is that at a covid death rate where we (in Britain, other countries are making much better choices) start to get twitchy we're only a few short weeks and or some bad choices from a healthcare meltdown we may not be able to stop. That has dire consequences for those in medical need and for our economy.

Running 'warm', managing the caseload near NHS covid & urgent other care capacity, resulting in say 500 deaths a day, that has a number of obvious downsides. However it is theoretically an option, one which requires us to live within a tightly managed control loop of strict surveillance and constantly varying, strictly enforced (or willingly obeyed...) rules. It requires excellent and timely scientific data (which we could do better things with) and submitting to exactly the sort of overbearing state control of our day to day business nobody wants.

jk

They *are* accepted, because banning road traffic or severely restricting it where other options exist would be comparable measures to a lockdown, but those measures are not taken.  The lockdown-equivalent would be that private cars would be totally banned from our large cities unless you were entitled to a Blue Badge, say.  Yes, it'd hit the economy, but so do lockdowns.

There's even a value for a life (I recall it being about 1.6 million quid or thereabouts) used for deciding if given road safety improvements are a good idea or not.

It already did on the quiet, Guardian ran a piece last week (Monbiot IIRC). Serco (with the nod from government) running T&T has been bumping minimum wage school leavers into roles they were supposed to be recruiting band 6 clinical staff for. These are skilled contact tracing and tele-medicine roles being given to teens reading scripts and ticking boxes.

jk

Press ‘5’ if you would like to talk to a computer.

I'd argue the lockdown 'equivalent' is more like temporary speed limits in roadworks or a road closure to safely run a marathon.

Edit: If you really believe road deaths are widely accepted why not have a go at making the rural economy case for allowing drink driving, see what response that gets. I might be wrong.

jk

Much like postmanpat last week, if they’re getting paid for this rather than just doing it as part of their membership duties, I don’t think cp will be getting their bonus for this one.

Its notable that the OP has basically failed to engage the key points as raised by many posters.  Normally they at least put up a presence of wanting an informed, 2-sided debate but now they’re just crapposting and then either leaving it or trying to start sidetrack debates (two days ago, over the definition of “life expectancy”).

If anyone is influenced in to taking more risks because of their posts, I consider that they will have blood and economic harm on their hands.  

As I said last time, they can either substantiate the key points of their argument or they should have the decency to pack it in.

There are real debates to be had about the lack of democracy in the measures - rushed every time after leaving it too late, in the level of state control, in the effectiveness (or otherwise) of lockdown measures, over the current state of immunity and its persistence (or otherwise).  Yet the press, the “flipped” scientists like Yeadon and Gupta, the usual suspects on UKC, they all avoid the real issues and keep banging the same limited set of misinformed and misleading drums.  They claim to be providing balance but really they’re all leaning hard to port trying to capsize the boat.

I doubt I’m the only person watching the new “combined” PHE surveillance report with keenness.

It looks from the latest one (figure on Datamart positivity) like the common cold causing rhinovirus is spiking much in line with recent past years.  This tends to happen before the influenza viri start rising.  So the covid risk control measures aren’t stopping Rhinovirus.  Hence I’m not taking an optimistic view on influenza.  The lockdown may stop it but potentially that’s just creating a double whammy effect when lockdown is released.

News just then: 15% Covid related, worsened by a few factors included staff self isolating. 

Ambulance crews don't do covid testing and a lot of calls are for people in some form of respiratory distress but until they're tested they'll go down on mobile despatch as 'sudden worsening of breathing' or a variant on that. I strongly suspect that 15% figure is people who've received a positive test before they've called for an ambulance and that the true percentage is a fair way North of that.

https://www.bbc.co.uk/news/health-54777741

Never mind CP.

All the usual suspects will knock your post down and the figures quoted.  

You are only allowed a certain level of deviation from their norm.   

Would you like to address the questions put to him that he dodged? 

Yes, tough isn't it?  You just post a quick controversial point on modelling and statistics, and people pile in with detailed rebuttals using modelling and statistics. 

What's the world coming to when you can't sow a bit of misinformation without do-gooders objecting?   

Whereas really really really believing will override reality?

Unfortunately Billhook, it's reality that's taken a huge detour from the norm, not theories and hypothetical statistics.

Leaning hard to starboard, surely?

I thought about that and tried to get it right.  Which in any sensible world it would be because port is red is right (is wrong.) I clearly wouldn’t make a naval officer and should probably have checked my choice with google...

When you are modelling an exponential phenomena, a factor of your isn't a bit margin.  Maybe a couple of weeks.

Extraordinary claims require extraordinary evidence.

If you turn up with bullshit and cherry picking expect to be told to do better or don't bother.

jk

An absolute worst case scenario subsequently being found to be an overestimate is perhaps a good thing. Even 1000 deaths per day is a disaster and a sign that we've learnt nothing (This doesn't just apply to the UK)

For a glimpse of a possible future maybe look at Belgium

https://datastudio.google.com/embed/u/0/reporting/c14a5cfc-cab7-4812-848c-0...

They have more people in hospital and ITU than at the first peak and Germany are taking patients for them.

Now bear in mind that over a usual winter period England has about 10,000 free hospital beds. We coped with the first peak of 17,000 inpatients as it was Spring, by cancelling anything non-urgent and emptying out hospitals in preparation.  At current rates we'll be above the first peak in 2 weeks. Once hospitals are full, it's not just the deaths from covid, it's from all the other treatable conditions.

I wasn't commenting on the subject matter.  I read  the article in The Telegraph.  Did you?

I was simply replying to the OP about  what normally happens when Covid 19 comes up for discussion and the poster deviates from your opinion.  Did you read the article in the paper?

Given there are so many varying figures - death from Covid-19, for example, then I'd expect that other posts/posters must pick their figures from other more than one source.   Or are you saying that there is only one set of reference figures and that you don't cherry pick at all?

Dunno about cp but pmp isn't getting paid. He just believes in a right load of old shit - global warming was an anti-capitalist conspiracy, the post-modern neomarxits are plotting to bring down western civilisation, all that stuff. 100% predicable.

For some things, yes there is a set of pretty stable reliable reference figures though never quite set in stone.

For the epidemic there isn't, it's pretty much all provisional we're often comparing apples with pears, at least granny smiths with russets when we can only see half the russet anyway. None of that is to say what we do know, where we understand the limitations and distortions is useless. Understanding that is important, our picture is incomplete but some of it is clearer than other bits. The OP isn't about understanding that, it's using the clearly acknowledged wide range of projections to pretend nothing is knowable then to promote the absurd idea we should therefore ignore what is known (healthcare overload looms) and do... what? That's the bit I don't actually get, what exactly is the libertarian plan?

The government briefing clearly acknowledged that there were a wide range of projections, *all* of them worse than the worst case scenario they had to-date been working with yet CP focuses on the worst of the worst projections then compares it to a clearly anomalous single weekend day death count to discredit the government's whole position. It's either stupid or dishonest. If I thought it was stupid I'd be gentler in my response.

Yes we can question why out of date projections were included in the briefing but the answer isn't always conspiracy, in this case it's likely part sales pitch, partly that these things take time to update, collate and present and our government is incompetent.

jk

People have only been using that trick for, what, 6 months now?

I'm torn between shorting the £ and reducing the liability of pension funs owned by the Koch family.

I don't cherry pick.  I've gone out of my way in some analysis to explore and present a range of values where there is a "free parameter" - where I choose a number that could impart bias to the analysis.  This "sensitivity analysis" over a range of values lets everyone see how important that choice is, and people can make their own decisions about which they think is more accurate.  Below is an example where I calculate the case fatality rate for a range of lag times from detection to death.  

This is the opposite of cherry picking.

Cherry picking is when someone else comes along, takes the extreme curve in this analysis, and says "Wintertree said the CFR is 3.5% but that was proved wrong so this discredits everything he says".

Carl Heneghan was on Today this morning talking about the Cambridge graph shown in the briefing and asserting that most measures of infection, hospital admissions and deaths were 'flatlining' (an unfortunate term under the circumstances) 

52.00 minutes in:

https://www.bbc.co.uk/sounds/play/m000p0zg

To be fair, he does make an interesting point about the most recent Cambridge projections, apparently to be released today.  It will be interesting to see if they show what he says they will show.

Thanks that was interesting. I do fear his point about diminishing returns being correct, but I also fear him being wrong about the outlook on hospital beds. Hopefully he's wrong and right about those things respectively.

How very unlike Heneghan to be taking the optimistic side of things.   Unfortunately my lack of a TV licence or iPlayer login seems to mean I can't listen.

The Cambridge projections are out [1].  Their finding (5): The growth rate for England is estimated at 0.05 (0.03–0.06, 95% credible interval) per day. This means that the number of infections is growing by 5% each day. This translates into a doubling in the number of new infections approximately every 15 days.

I don't see any great support in the nowcast for much lowering of growth rates in any region except maybe London, although I don't know the point H. made as I can't access the recording.

My latest plot of estimated doubling times (from the 4 pm dashboard release yesterday) is below.  The deaths appear to be slackening off, but they often do on a Monday with the weekend reporting lag effects.  At a guess the deaths curve is going to end up settling down to a doubling time of ~ 20 days, as with admissions - with about the expected lag between them.  The cases data is garbage I think, as the positivity of Pillar 2 data is rising to levels that indicate it's not well sampled.    I don't see support in this plot that admissions are "flatlining" - more that the doubling time is reasonably constant so they remain quite exponential.  Deaths - the doubling time is nowhere near enough to start "flatlining" their curve, and it may well drop with data updates over the next few days.  

[1] https://www.mrc-bsu.cam.ac.uk/nowcasting-and-forecasting-3rd-november-2020/

Here is another example of not cherry picking data.  This is plotting the Covid hospital occupancy data for the UK either side of both the March and November lockdowns.

We haven't got to the lockdown yet, and the most recent 5 days of data are subject to reporting lag (different nations report in on different lags).  I'd like to know where occupancy is going to peak, so I project the data forwards with a simple model fit to it.

But, the data is "noisy" - it doesn't follow a perfect model.  So I fit a forwards projection to different 7-day windows and I look at their spread.  The date in the legend is the most recent date in that 7-day window. 

I project the curves forwards to the point where I'd expect the lockdown measures to start pushing things down below the exponential trend.   See how there is a difference of almost 5,000 patients or 20% between the projection from lockdown - 7 days and lockdown -5 days.  Yet there is no trend to where the projections land - these are the two extremes, with projections from lockdown - 9 and lockdown - 11 landing in the middle.  

So, I could cherry pick the most recent projection, that for lockdown -7, and decide that we're not going to burst through the surge capacity of 26,000 patients.  That is, after all, the most recent projection. 

But by looking at a range, it shows me that there's a lot of random variation in the data.  Exponential growth is very sensitive to randomness when projected forwards.   So what I'm actually going to do is keep updating this plot for myself every day and see if a consistent trend starts to emerge, or not.  

I only put a few different time windows on this plot so as not to over-crowd it.   The second plot shows the occupancy at lockdown + 10 days for a wider range of analysis windows.  So it seems the trend is down, but a few days ago the trend was up.  If the trend continues down over the next few days it's starts to feel a bit more real.  Right now the projections for +10 days (which is not the peak, but when we might expect the curve to start shallowing out towards the peak) are hovering between two lines I shall call "oh s**t" and "oh f**k".  

It's amazing how often a value or a projection from a single day is given in a media piece without looking at the context of the day-to-day variation in the numbers.

Thanks for continuing to post these, Wintertree- I agree with another poster’s observation, this is the best place I’m aware of for informed discussion of the situation, and your contributions are central to this.

I consider wintertree's posts the authority with which to check the credibility of the claims I hear in the media (and I stick to Radio 4, not the f*cking Telegraph).

Great job. Should really be attracting a wider audience than UKC!

The MRC nowcast/forecast is out now: https://www.mrc-bsu.cam.ac.uk/nowcasting-and-forecasting-3rd-november-2020/

Also Whitty and Vallance were at the Sci/Tech committee today, and it's worth a watch: https://www.parliamentlive.tv/Event/Index/8bbb6325-cbfe-4025-9be9-5f50bc220...

Heneghan seems to have been comprehensively wrong. I hope he's back on Today to say so asap.

Unfortunately I have friends who are taking the words of Tim Spector (ZOE) and the Centre for Evidence Based Medicine (Heneghan?) over any other scientists even or perhaps especially any linked with government.

Outside these threads it's hard to keep pushing back especially when organisations like ZOE are, I believe, doing quite a good job of recording prevalence etc. and it's the interpretation and extrapolation which differ most widely between sources.

I might have put this in the "wrestling with pigs" thread only the friend in question isn't a pig, he's in a science-based occupation and is genuinely attempting to assimilate all the numbers, projections and arguments. As so many who don't feel particularly connected (see relevant thread plus others!), he seems to have a natural bias against lockdown - and every time he veers toward caution, another statement (e.g. Spector's tweets about R value being near 1 already) comes along to tip him back the other way. The apparent good reputation of people like Yeadon makes their statements particularly believable to those without equivalent qualifications. (Like me, only I have UKC to keep me informed!)

I think Heneghan is as compromised as Gupta.  He features prominently I see in today’s UKC misinformation thread carefully written and posted from a pop up account.

If my work was being abused to try and kill people in my country - this is where things stand right now  - I would do my damnedest to clarify.

His opinion piece on false positives in PCR testing was questionable at the time.  It’s legitimacy hinged on their being no rising deaths with a lag from detected infection.  The data at the time on deaths didn’t have the “28 day cut off” and had a long tail of people dying long after their tests. If this was appropriately estimated and subtracted from the data, a rising signal was visible in deaths at the time making his stance questionable.

Since then, the data proves his point unequivocally irrelevant to our situation, regardless of its accuracy.  Has he made any effort to put clear water between his opinion piece and the current situation?  Has he f**k, despite its massive popularity in anti-lockdown conspiracy circles and the emergence of Yeadon as a new player.

Generally speaking the way Henhegan approaches the conversion of data into “evidence” appears to me to be crap science that hinges on words, blogs and not applying any of the myriad tools designed for asking questions of noisy, real world data.  

Oxford University has a serious problem.

Agreed. Heneghan, Sweden and false positives are nearly always mentioned by the Covid deniers, I normally counter with the facts on Sweden and false positives and simply say Heneghan has been discredited.

I haven’t posted about it but cases in Sweden appear to have gone exponential recently with a 7-day doubling time.  A rather important point of information.  

Nearly 500 deaths reported today.

The other thing noticeable is that tests performed per day have dropped off in the past few days without any reduction in cases.

It's a tough one this.  The last six months for me have been like getting hit round the head with a clue stick about scientists and science in universities.  Being a scientist does not give people better abilities to make judgements from confusing, conflicting and partial evidence.  There's nothing in the undergraduate training in many science degrees about this, and it's pot luck if their PhD or postdoc jobs will nurture or select for this ability, or not.  It's an environment in which people can make good progress with those abilities, but career development is a broad church.  

I can't say as I blame him.  Despite some of the claims from posters on here about me having hysteria or an agenda or an ideology or wanting to lock us all up, I have a strong bias against lockdown - it made my life very difficult the first time around and I'm not sure I've fully shaken off all the effects of the work burnout from that period.  But despite my strong bias against it, it seemed like the best of a bunch of bad choices at the time.    

Yes, using a minority of scientists with decent credentials to confuse the issue is a pretty well developed tactic now, and one the wider UK public is utterly unprepared to resolve - by this I mean nothing in our education systems or day to day life leaves a person off the street with much basis to resolve the issue of disagreement between scientists.  

Updated forwards extrapolations of hospital occupancy until lockdown + 10 days.

The level projected with today's data release is the lowest yet from recent days.  Plotting the hospital admissions with a doubling time shows the measured doubling time to be gradually increasing.  It would have to go to + infinity for the level to "flatline" as per Henighan.  It isn't, it 's increasing very slightly - but that is good news if the trend continue because it means hospitalisation at +10 days won't hit my "Oh f**k" line and it may even fall under my "Oh s**t" line.  I picked +10 days as that's where the hospital occupancy curve should start to level of if this lockdown is as effective as the last one.

I'm still not felling at all happy about where we are, but things appear to be moving in the right direction.

Not spent time in academia after my MSc but a lot of that rings true. In theory, historians ought to be some of the best at untangling sources of varying reliability - but then they'd need the scientific knowledge to base their judgements on...

Also agree about lockdown - most people I know are doing OK apart from small inconveniences but we're well aware that others are suffering badly. As you say, there are no good options, only bad and less bad.

The scientific credentials aspect bugs me most. As a non-biologist but with some stats background (earth scientist, actually - much exposed to fuzzy data!), I think I have a fair grasp of what looks like a solid argument and good data. But against the boss of the "Centre for Evidence-Based Science", how can I possibly claim my science is better than his? (It's a sign of the cynical times that when I first saw that name, I thought "What are they selling?", so wasn't surprised to find some of their output suspect.)

Being hesitant in this way means I can't confidently debunk the bad science, and confidence seems to be one hallmark of what gets believed. Plus as I mentioned the misinformation keeps coming so it's like swimming up a waterfall.

If anyone's interested I have found some sources (as well as the usual ONS, PHW and PHE etc. often referenced here) on youtube that look like they're giving good round-ups of the medical science. I'm judging that by a few characteristics: they talk mostly about medical results and stats, largely avoiding more subjective economic/political judgement outside their expertise (though my first link proves me a liar on that! To be fair, it's a bit of an unusual outburst from him.); they list referenced papers; they have medical backgrounds; and they're willing to say they've got something wrong when new information turns up. But who knows what invisible biases there might be despite all that...?

Anyway, links to their latest posts in case they're useful:

youtube.com/watch?v=Z27M-ekXlbI&

youtube.com/watch?v=eQO1PB8-xtg&

Thanks all for your efforts to keep the facts straight and up-to-date for us.

Just been back and corrected all my Swedish case numbers - even a day after the total is posted on worldometers it will get revised upwards. It's supra-exponential, case number percentage rise of previous day has been increasing steadily since 4th September:-

4/9 0.17%

11/9 0.25%

18/9 0.28%

25/9 0.43%

2/10 0.52%

9/10 0.76%

16/10 0.74%

23/10 1.08%

30/10 1.94%

Yesterday 2.19%

To put this in context, UK peaked at 2.77% on 24/10 and is now down to 2.22% today. Brazil hasn't been above 2% since July.  All 7 day averages.

28/9

Not good.  So much for some people’s suggestion that Sweden was achieving naturally acquired herd immunity.

One proviso on the U.K. % day-on-day; detected cases are not scaling with the rise inferred from random sampling surveys and positivity is something daft like 18% so that data is - roughly speaking - total garbage now.

Could this be a side-effect of the "only one person in a household gets tested because the others are self-isolating so can't leave the house" scenario? What's the guidance for contacts getting tested? Over here they make people self-isolate for a number of days past their 'contact date' to ensure they've built up enough virus to be detected before getting them in for a test.

I know. I don't have much hope of a mea culpa from Heneghan but I have hope nonetheless.

Oxford has always had a serious problem, but unfortunately it's also a strength for them. I can't explain Heneghan, but Gupta for example seems to carry her weight in the Zoology dept; it's just that she's now speaking beyond her remit. Which waywardness is not unknown in Oxford Zoology (a certain R. Dawkins comes to mind).

I don’t think so.  As I read it anyone self isolating due to another infection in the household should get tested if they develop symptoms.  If they actually do is another matter...

I think it’s just that cases were or are rising fast and are now 1/3rd of testing capacity or so.  The WHO I think advise sufficient tests to keep positivity below 5% for sufficient coverage and we’re well north of that.  If we were catching all cases with that positivity we might need 2 million tests a day, we’re doing 0.27 million tests a day.

For reasons I don’t know, the number of daily tests run is down 8% on last week.

Testing and by extension contact tracing just aren’t very useful when prevalence is high and rising.

Yes.  It’s not cool.  There is no UK professional standards body regulating or licensing academics.  It’s rare for a professor to loose a post unless they end up on trial for some unrelated criminal behaviour or speak out against senior management.  Yet the badge of “Professor” carries a lot of authority with the general public.  Misusing that perceived high status should not be free of consequence - with great academic freedom should come a requirement to use that freedom responsibly.

My take is that he puts up a plot and says some words but he doesn’t apply any sort of statistical tests or robust analysis to the data.  It’s words and pictures without any actual “science” to bridge the two.  The last paragraph in my link below goes through the problems I had with a particular piece from him.  To be fair I often use graphs and words on here and don’t go in to the science methods that start to quantify things - but this is UKC not an academic group calling themselves the centre for evidence based medicine...  For UKC I try and analyse a range of values or periods as a way of showing the uncertainty in the data rather than putting up residual plots and chi squared values as one is a lot more intuitive than the other to people who don’t do data all the time.

https://www.ukhillwalking.com/forums/off_belay/aberdeen_back_into_partial_lock...

The PHE week 44 report says this:- "*Positivity data was previously deduplicated across the course of the pandemic to prevent persistent infections being counted as new cases. Since week 40, positivity is calculated as the number of individuals testing positive during the week divided by the number of individuals tested during the week. This approach accounts for the increasing number of individuals who will have been tested multiple times as the pandemic progresses."

so that could account for the increase in positivity. Why they had to change it I don't know.

We just have way to many cases.  I was out with 18%; it’s closer to 12% for males and 9% for all looking at last week’s PHE report.  Who knows what it is by now, testing numbers recently took a big dive...

I assume you want to know how many positive cases there are currently in a population. New cases charts the growth, but current cases is important too. Especially if some are getting better in 2 weeks and some aren’t. 
 

Different people will be using different data for different applications. 

I’m not surprised. I suspect the number of false positives wasn’t representative of a random sample as lots of ‘well’ people were clogging up the system so that they could get the all-clear to fly. 
 

I’ve also heard that there is also a large number of people being tested in the film/TV industry. 

Thanks for link. I do see the problems with the "misinformation" there...Basically the data's OK within its limits, so as long as you know what those limits are e.g. de-duplicated or not (can't see why they'd change that, either). It's once (some) people put words and policy proposals to those data that it goes haywire. But the problem of kicking against "authority" remains unless you're speaking to people who can make sense of the numbers, in which case you probably wouldn't be having to argue...

Accountability for professors: great idea, though I can see pitfalls and possible dangers in the implementation of any system to achieve that.

Lastly, seems like a fair balance of technical material in what you post here - after all people can always ask questions if they'd like specific details. Thanks.

You have a key point to the argument in your post- if this was genuinely someone looking too provide an unbiased view of the evidence, why are they mixing up policy advocation and data interpretation?  It smacks of an agenda.

You're welcome; I try to keep an objective view of the data although my personal preference - openly stated - has long been to take our lumps early when cases are low, not late when they're high.   I'm not claiming to be a centre for evidenced based decision making, so I don't have a problem with my approach.  Some do...

I just modified one of my plots to fit linear and exponential models to the specimen date data from the month of July from July 2nd to July 28th inclusive (dates chose to be a multiple of the 7-day aliasing period from the "weekend effect")

This is approaching the standard of analysis I would expect from a scientist to back up their claim that the data is linear.  I have fit linear (left) and exponential (right) models to the data by linear least squares fitting; this is shown on the top plots.  The middle plots are the residuals between the fit and the data.  The bottom plot is normalised residuals estimated by taking the statistical noise as the square root of the model fit - assuming that poisson statistics are at work and that the numbers are >> 1 (which they are).  The methodology on that last one might be a bid dodgy like; I'm used to working with data points that have experimentally determined uncertainty, where-as here the noise is in the random and structural processes with going for a sample to be taken, not with the result of the sample processing and using the model fit to estimate the noise is a bit cheeky.  Still, it's good enough for me...

The "sigma" value in the legend is the standard deviation of the normalised residuals.    This is a crude test of how well the model fits the data, with a value of 1 indicating a good fit.  For both models the value is 5.1 (no units).   This indicates that both models fit the data equally badly - there is no support in the data for H's claim that it is linear, not exponential.  For that claim to be proved true, the sigma value for the linear fit would need to be a fair chunk smaller than for the exponential.  They're both very bad because the models do not account for the (highly erratic) 7-day periodicity in the data associated with the "weekend effect".  

A more thorough approach would fit the curves by minimising Chi squared (square normalised residuals) and would give the reduced chi squared of the normalised residuals, not the standard deviation (this accounts for different models having different degrees of freedom allowing them to better or worse fit to noise as well as the trend).  That's a bit too much effort for a UKC post though and will basically show the same thing, as the fits are so similar and as they both have the same number of degrees of freedom.

The data as presented simply did not evidence that the growth was linear not a low-rate exponential.  If Heneghan understood either mathematics linking the two (e.g. Taylor series) or basic undergraduate hypothesis testing he would have understood this from just looking at the plot on the link I gave.  I didn't have the code set up to do this analysis when I criticised his post at the time, but felt perfectly confident doing so without having done the analysis.  (note the plot on their website has since updated to include more data than just July 2020 as it's some auto-generated widget...).   

I expect someone who works in the field would deploy somewhat different tools to the ones I have done but to the same ends.  The point being, if they just put a graph like that up and say "it's linear", that's propaganda not science.  If they deploy appropriate tools to show that it's linear, that's science.

Perhaps.  Just because people know how to test a hypothesis doesn't mean that they tend to, or indeed that that spot BS when its being deployed.  Sometimes you have to guide them through what they already know to get them there.

Nice methodological overkill!

My approach would be to eyeball the scatter plot, decide that you can fit nothing or everything to such data, conclude that cases per day are rising (especially as there is a good reason why the low values may be systematic outliers), and leave it at that.....

That’s what I did the first time around.  Eyeball it and think “what a crock of crap, this data isn’t going to support one view or the other”.  

To close the loop on that methodological overkill - the data does not support a distinction between the two models.  This means it does not support a distinction between H’s claim - false positives rising with testing capacity - and the alternative of exponential growth with R just a little over 1.

So for H to claim otherwise he’s either professionally incompetent or being deliberately dishonest.  I suppose it could always be both...

Should we be placing more emphasis on stats at A Level? Ok, if you are going into engineering mechanics is useful, but it seems that more and more people in positions of power/influence are making decisions based on statistics. I don't think many A level students will be able to follow your "methodological overkill" but they should be able draw the same conclusion as you and cb did just by looking. Could Matt Hancock? Is NZ in a different position because Ayesha Verall would understand your methodology?

That's just about what I did, quick look, they're all over the place but it looks like the trend is upwards.

I've not been looking at Wintertree's plots in detail, but that's because I haven't really needed to.  I can see that it's getting significantly worse just from the basic Government plots. Wintertree's "mission" - and I'm glad somebody has the motivation and ability to do it - is to combat with solid evidence those who are claiming that we're over-reacting etc.

UKC is fortunate in that it has lots of experts and intelligent people (you are allowed to be both but it's not mandatory 😁) in its midst. 

Covid: Regulator criticises data used to justify lockdown:

https://www.bbc.co.uk/news/health-54831334

Its not just the 'fake news' telegraph who want the government to use accurate data when deciding when to remove people's civil liberties on mass.

I’m done with you since you claimed to be pushing one sided stuff (most egregiously through the pub) to get a rise trolling people.  No doubt you’ll be back with some more clever debate to claim victory.  Go for it.  It’s yours.  It’s hollow.

It wasn’t data.  It wasn’t claimed to be data.  It was clearly presented as a scenario.  It was the extremal scenario.  Scenarios are not data.  

The data that motivated lockdown was actuals on hospital occupancy.  The rest is scene setting.

Precisely.

CB

... and if you have looked at more of such plots in your life than is good for your sanity you will spot an pattern in the outliers (here, low recording on weekends, presumably), mentally subtract them, and STILL conclude that there is no way you can tell whether the rise is linear growth or at the beginning of an exponential rise. No need to even fit a curve.

Usually this only hacks off your students when you tell them that they have way too few data points to support their conclusion, and need to go back to the bench.

Here it is much more sinister, when the statistical illiteracy of the general population is deliberately exploited to push a dangerous policy.

I agree that wintetree's efforts are absolutely heroic!

CB

It would be fascinating to sit down with cabinet members, present a series of slides with data and conclusions drawn, and ask them to rank the conclusions in terms of how well supported they are or aren’t. 

Part of the problem may be the presence of a SPAD who portrays themselves as a data science person, but whose last blog entry (just as this was kicking off in early 2020) sent a lot of science types into disbelief.  Perhaps there should be a requirement to have someone qualified in interpreting data on the advisory team for cabinet, but I don’t know how you even go about deciding what counts as a qualification.

Diversity I think is a key step forwards - having MPs and hence cabinet drawn from a diverse range of backgrounds, professional experiences and so on.  As of about 3 years ago about 3% of MPs had a PhD and very few of those were in areas that would cover how you statistically test data of any sort.  But we don’t need someone who has been helped through it by a supervisor / colleagues for a PhD and then moved on, we need someone who has done it for a living for a long time.  

Although from what I gather, if the current lot didn’t like what this advisor is saying, they would just get ignored.

[1] http://virtualstoa.net/2016/08/28/doctors-in-the-house/

This is one of the few things that is good about Merkel. As much as I dislike her, she is not stupid and has an extremely quick grasp of mathematics and quantitative relations (according to people who have personally negotiated with her as opposition party state ministers).

Of course this would be expected of someone who has a proper research PhD in quantum chemistry (not some thin sociology / history crap like Helmut Kohl's dissertation on the history of his party through the last three state elections).

CB

Margaret Thatcher also studied chemistry and then became a barrister !

...and she was not stupid either. As much as I disagree with her politics, I am not sure that she would have allowed CV19 to spiral out of control that badly.

CB

Shit no.

I think rather more robust steps would have been taken by all of the PM’s in my lifetime. Maybe:

thatcher>Blair>major>Brown>May>Cameron >>>>>>>>>>>>>>>>>>>>>>>twatface 

Not sure I agree with all of your ordering but you've certainly got the last bit right ☺️

I certainly like to think she wouldn’t let it spiral out of control close to the healthcare brink a second time.

In reply to Dr.S at work:

You’d put Blair above Major?  Interesting...  I would take either of them right now mind you.

Models that predict a scenario then, but really you are splitting hairs. You use this to work out what is the best plan of action is. Is it too much to ask that the govt use the most up to date ones available?

Context is important, don't pretend that it isn't.

Is it too much to consider that they did? Anyway, as has been pointed out over and over, it's the high and growing hospitalisation rate that finally forces Johnson's blood soaked hand.

It is but in context I'd say the regulator's criticism is valid, limited and mild. https://osr.statisticsauthority.gov.uk/correspondence/ed-humpherson-to-sir-...

jk

Today's Swedish calculations give a 2.67% rise in cases, live case number doubling time of 11 days, and 46,000 lives cases, as many as Germany did on the 14th October, France at the end of August, and the UK on 21st September.  More ammo to hit the herd immunity squad with, they're going to run out of things to argue with soon!

Virus is getting weaker - disproved

More people die of RTAs - disproved

More people die of flu - disproved

Sweden has herd immunity - disproved

Sweden proves lockdown isn't needed - about to be disproved

Also look at New York, who've done really well until now - 95% remote schooling and so forth, but cases have gone exponential even so. Deaths seem to be lagging a lot though.

https://www.nytimes.com/interactive/2020/us/new-york-coronavirus-cases.html

Yup.  Various posters have tried to argue for the superiority of the Swedish approach and that they had achieved more herd immunity; I didn't see credible evidence for it then and it certainly doesn't look true now, does it?

Likewise there've been passionate suggestions that NY was holding cases steady due to enhanced immunity, as I said repeatedly the data and maths did not imply that at the time, nor did they disprove it but the holding of R~1 despite the strongly adhered to risk control measures in NY state certainly suggested against it.

Disappointing but not surprising to see the data for both locations now making the point.

I would be interested to see a geographic breakdown of NY though - is the rise happening in the urban areas hard hit the first time around, or is it now elsewhere?

The more I read about how Donald Trump has politicised and weakened the CDC the more it beggars belief.  I start to see why so many non-hyperbole prone Americans refer to his administration as a death cult.