As always, these plots are based on data from the UK government dashboard . For the cases data especially I encourage people to look at areas that interest them on the dashboard as well to include the raw data in their interpretation. By clicking on "United Kingdom" at the top of the page, other areas can be selected.
The case numbers for the most recent days are a bit odd looking - there's a sudden reduction in the decay rate for England which looks to have bottomed out, and Scotland and Northern Ireland appear to have turned to growth in the last few days. Decay in Wales looks to be slowing.
As always, the most recent week on my plots is provisional - the trend line is a polynomial filter on the data with a symmetric window about each data point that becomes asymmetric (using only data from before the data point) at the far right of the plot. This means the right most week or on the trend liens are very twitchy and can respond to the next week of data - not yet published.
This is in contrast to the government dashboard's data which presents a 7-day moving average. This is always based on past data, with a consequence that the moving average represents the past by a few days, not the present.
There's a trade off here - my trendiness are more sensitive to changes right now, but also more sensitive to noise. The noise (random variation) in cases data is oddball so it's really not sensible to make a noise model to try and put a probability on a sudden change being real or not IMO.
There's always a "false shoulder" in the data, where the weekend sampling low creates a false drop in the data, then it rockets back up on a Monday. I think is might be worse than normal for the leading edge of the data, perhaps early as a result of the extreme winter weather conditions. Scotland has always had the lowest exponential decay rate, and England the highest, so it would make sense than an external event messing up the data would push Scotland to appear like growth, and England to look like levelling off. But I could be completely wrong, and cases could be rising. Or perhaps more likely, the decay rates are slacking off and there's an unusually bad "shoulder" in the data right now. I'm just an internet punter making graphs, and right now I feel I don't understand the data as well as normal - I encourage people to look at the dashboard themselves. I'll come back with a mid-week update but I think we'll have to wait another week to really know what is happening at the current leading edge of the data.
My next post digs in to the raw data, as well as the processing I do to get to the cases trendline, which includes trying to unpick the weekend sampling issues.
This are plots I look at before writing up my analysis, but don't normally share as it's all a bit esoteric. It's described on past threads. They're not presented to a standard I'd normally share - missing axes labels and simple numeric tick marks, but otherwise it gets over-crowded.
"Raw" data has 'x' marks for dashboard data and polynomial fit trendiness, with residuals (differences) between the trendiness and the raws shown below. With the coloured data markers, the effect of weekends is very clear.
"Shunted" data has some re-assignment of cases based on a qualitative understanding of people not going for tests on the festive bank holidays. This hopefully improved accuracy of analysis then and suppresses some awful "artificial" structure in the heat maps, but has no baring on analysis of data in February
"De-weekending" includes a process that reallocates some cases from Monday and fewer from Tuesday to the weekend immediately before, working to minimise error over a linear trendline encompassing a few points either side of this 4-day anomaly period, or just to the left where the Tuesday is the last day of the data - as it is for the last week of data each time I start a new thread - so another area where next week's data will refine the understanding of this week's and why the leading edge of all plots is provisional.
Looking at the trendiness by eye, without the de-weekending, the swing from decay to or towards growth is worse in all regions. It's hard to tell if the weekend anomaly is worse than usual as the residuals don't follow a very coherent pattern.
So, depending on how the data is analysed, I get quite different results about the current situation. This suggests to me there's a lot of uncertainty in my analysis this week. Ultimately, as case numbers get lower, the statistical noise becomes larger as a proportion of those numbers and it's gong to get less clear what the change is on the timescale of one week.
Not much to add on the England plots not covered already.
A reminder that the right side of all curves - particularly plot 9e - is provisional. The final wobble in the cases exponential rate curve is in the leading edge of most weeks and goes away when another week comes in and improves correction of the weekend effects.
Deaths continues to diverge to faster exponential decay than either admissions or cases - un unequivocally good thing, and increasingly tied to the vaccination effort by demographic data I feel.
I have turned off the polynomial filtering this week - I think it's important to see the "real" data at the leading edge given the uncertainty I think I see in the data. I think updating this in 4-5 days will be interesting/informative.
Plots 18.1, 18.2
One of the reasons I'm a bit confused by the turn of events in recent days cases data is the lack of any real smoking gun at UTLA or regional level for a failure of lockdown.
The exponential rate constant is becoming less negative (heading from fast decay through slower decay, eventually towards growth) in several "blue" regions. This can be identified for individual regions from the heat maps on 18.1/ or from the blue lines on the individual regional plots on 18.2
Yorkshire and the Humber stands out as potentially returning to growth. You can see raw data in . I look at that and if anything the oddity is the previous weekend's low cases - again I wonder if the snow either reduced sampling and cases, or shifted them to a later date. Speculation. It could be that the snow created a false additional decay signal which masked a more progressive slackening of exponential decay. More speculation.
Plot 22, 22r
The trendline has crossed itself at least - both measure are now less bad than during the first wave, six and a half weeks after we locked down.
Plot D1.c - to my eye, there are vertical bands of light/dark corresponding to the weekend sampling effects. The last dark band (faster decay) was particularly dark, and the final white band on the RHS is particularly whit (less decay). So, again, I wonder if this was the winter weather causing apparent faster decay which is now being corrected by a shift in the opposite direction. Speculation on my behalf and it could be way off.
Plot D1.d There's some dark purple appearing for age ~45 on the right hand side of this plot - this is basically noise arising from deaths being very low in this bin, by this point.
Plot D3 - this builds on the changes seen in last week's D3
Decay rates continue to get more negative (faster exponential decay) for the oldest bins, but are becoming more positive (slower exponential decay, heading eventually for growth) for adults of working ages.
This looks to me like a gradual weakening (but not yet failure) of lockdown to control the spread in general, combined with the vaccine working to decrease the spread in the older demographics. It could be that some of the effect of the vaccine is masked by the general rising trend.
Decay continues to be getting better in children - this is contrary to the most recent REACT round, discussed on Plotting #12. REACT is random sampling and lags symptomatic data such as the government dashboard data that I use. ZOE data is also discussed on #12 with a comment from Punter_Pro linking to a discussion on the effects of vaccine reactions perhaps biassing their data.
Plotting the rate constants for different ages against each other. This is heavily filtered (SG filter, 21-day, 1st order) to make it comprehensible and not like a twisted seismograph. This moderates brief peaks in values.
You can see the oldest ages continuing to diverge from others towards more decay on both axes, and the younger ones that are plotted starting to slack off their decay in cases towards the growth zone (x>0)
Plot D4 - Demographic Extrapolation (not a prediction)
The changing exponential rates from D3 really show through here -
the increased decay in older ages affects this little, as there are fewer people in those ages.
the decreased decay in younger ages affects this a lot, as there are more people in those ages.
There's no UTLA watch plot today - given my lack of confidence in the data I'd like to wait for tomorrow's data. That and the lack of proof reading of these posts in no way reflects the zoom-based whisky tasting I have coming up tonight...
So, these week's date leaves me both nervous and concerned at the apparent slackening of decay up to the onset of growth across all nations, and hopeful and reassured at the effects the vaccine is apparently having on the older demographics. As those jabs go in to the arms of young people with younger immune systems I'd hope to see a big shift in case rates at younger ages. So, the situation has a definite feel of a race against time to me right now, and one to absolutely hold the line on controlling case numbers.
Each dot on plot 22 is daily, isn't it? It is nice that we have finally crossed back over properly, and the dots appear more spaced as well which indicates it's decaying faster than first wave.
I have to say at a very unscientific level I had got a sense of slowing. I looked I think middle of this week at my local (Bmth, Christchurch and Poole) on the covid map, and thought that very crudely we might drop a colour on the daily update some point next week (it was on about 160, decreasing 30-40% per week, so another third off that.....). I'm now looking at it and I'm thinking it's now next weekend.
I watch that versus Dorset on the basis of having been bitten by Tier 4 late last year - admittedly not for long before lockdown - when suddenly all my climbing in Dorset moved the other side of a tier boundary. BCP has been a bit of a laggard - we peaked late and decayed late, my parents near the epicentre in Epping Forest overtook us on their way down before we had even peaked I think - and I don't like sticking out from the crowd for risk of "special measures"!
Edit: I am aware this is n=1! Decay will get lumpier as absolute numbers fall. But it's interesting to see signs of it more widely, especially since we've had no weather disruption or anything to pin it on...
> In the 'spirit' of the openness and positivity on these threads, what whiskys are on the menu?
Thanks to a dead drop, I don’t know, see below... I can’t (yet) reveal the identity of the ones we sent by return.
In reply to AJM:
I’m holding off detailed comments tonight but yes, the dots on plot 22 are daily and yes I think the big spacing is very welcome news. The possible link between low cases and an exponential slowdown is discussed a bit with Wicaomi at the end of plotting #12.
I don't have a good feeling that they won't f*ck it up again. Whilst numbers are coming down, they are still high, especially hospital occupancy, which is only just falling below the peak of the first wave. It seems very premature to be considering starting to unlock in two weeks. I fear that the hideously high numbers we've had recently are the 'new normal', and the current numbers are only 'low' in relative terms.
With the swivel-eyed loons and restaurant chain owners baying to unlocking ASAP, I worry that they will fail to react swiftly when numbers rise again.
Thank you... really appreciate this 'hobby' of yours.
Hopefuly not much will change in policy on Monday... Things like the Yorkshire data are a bit troubling and it needs to be clear wether this is noise before decisions are taken, as it seems so hard to reverse them.
(I don't think this government is very good at U-turns - I think do these before you hit the wall.... our 'leadership' seems rather to drive intp the wall and then decide to change direction)
Firstly, couple of presentation ideas, not necessarily to share, but might help give you some data insights...
On any daily "quantity v date" graph, line join all the Mon points, all the Tue points, all the Wed points etc. Either with straight lines or some kind of curve fitting. This might give you some extra view of the weekly profiles in terms of weekend lag etc, helping to spot real outliers etc. You might end up with obvious "layers" similarly to a 75%ile curve always being above the median always being above the 25%ile.
On your initial plots, the number 6s. Draw lines connecting the outside points to form a sort of tunnel. The shape of this might just help you with determining whether any change in slope or reversal in direction is noise driven or real.
Secondly, what is the purpose of all this? When the numbers are coming down all over the place then your analysis and others don't really add much for most purposes, as long as the numbers are REALLY coming down.
What your stuff is really needed for, is early detection that things are going wrong and maybe being able to determine why; then it becomes very important. I don't think you are, but remember not to lose that focus on what it's for.
> With the swivel-eyed loons and restaurant chain owners baying to unlocking ASAP, I worry that they will fail to react swiftly when numbers rise again.
I heard that guy on the radio this morning. What will it take for these people to understand: no one is ignoring the downsides of lockdown. We all know - all the policy makers, all the scientists, all the people who've thought about it in just a tiny bit of detail - that lockdown is absolutely brutal. This story that the PM and his advisors haven't noticed the downsides of lockdown is just deliberate bullshit. Of course they've noticed it. Have they got a reliable quantitative analysis that *proves* that the brutal downsides of lockdown are worth it in the face of the risk of failing to control the virus?
Well, no. No such analysis if feasible. The risk that is being mitigated by lockdown is another lockdown. Has John Vincent realised this yet? That politicians can't say "we'll have a few more deaths and trade that off for better outcomes in catering and hospitality, better wellbeing for those employees and customers". It doesn't work like that you idiot, Vincent! How long have you had to get your head around what the problem is, and why it isn't anything like the cost-benefit analysis of opening a new branch of your shit food chain?
This is a pandemic not a f*cking overpriced organic Mackie D's in a train station. Wanker!
As well as the possible case upturn on Monday/Tuesday, which could yet be noise, there is also a notable change in the testing numbers this week. Unfortunately it is in the wrong direction - ie there is a definite reduction in both LFTs and PCRs being undertaken at national level - so it doesn't easily explain the behaviour in the cases curve.
Even when everything is coming down (or going up) there is interest and probably some insight and learning to be had by looking at geographical variations in rates and how they change, as well as looking out for problem areas. It's also important and interesting to be looking at hospitalisations and deaths at the moment, especially to understand vaccination effects. So while an early detection of something going wrong is important, it's far from the only real purpose.
I agree that the detail is useful for getting insights etc for possible (but unfortunately almost certain) future use, it's why I put "most purposes". But I reckon most of us simpletons are simply thinking "great, it's carrying on getting better" and only look for better understanding when things get worse again.
For the clarification of anyone's doubt - I am not decrying WT's work at all, he's doing great, but the thing I would really like to see is some of his work/presentation being more widely accepted/used by those with/in power, which whenever things go the wrong way again, would allow them to more quickly both implement necessary action and be able to communicate the need for it.
Thanks as ever. The plateauing is not entirely unexpected. There is a certain minimum level of transmission which is going to be associated with the residual level of social interaction. As you say, we need another week’s data to check the trend but it would be unfortunate if this plateaus at 10k recoded cases a day. That’s a lot and it’s not going to reduce much until the vaccine is rolled out to younger people. I suspect that the roll out to date has had limited impact on transmission as most of those vaccinated have been old people who have limited social interaction.
Interesting report this week, thanks. Definitely something strange going on, no one seems to be sure what it is though!
I guess we shall see over the next week.
My London friends mentioned in our group chat that so far this weekend, the local parks near them have been manic with people out and about doing stuff, basketball courts packed, coffee shops really busy etc.
There were also a couple of houses in my Cul-De-Sac yesterday who had people over, I think complacency/lockdown fatigue has now set in and we will see more of that with this warm spell that we are currently having.
Monday's announcement will certainly be interesting....
> There were also a couple of houses in my Cul-De-Sac yesterday who had people over, I think complacency/lockdown fatigue has now set in and we will see more of that with this warm spell that we are currently having.
I did wonder whether it could be some sort of weather effect. The really freezing weather the couple of weeks before could have kept people at home more, then more may have gone out and about this week as it has been quite pleasant. Certainly the local playpark was pretty busy when I took my son this week, although it was also half term so key workers kids and some nursery kids will also be off.
I'm not convinced in the theory though given that the extra interactions will be almost all outside and normally warmer weather is supposed to help.
Edit: thinking more this is garbage as the infections recorded on Monday would have happened in the previous days when it was still really cold.
Plot 16.1 - UTLAs are ranked horizontally by their case rates when we entered lockdown on Jan 5th
Plot 16.2 - UTLAs are ranked horizontally by their case rates two weeks before the date of the plot
Plot 16.2 gives insight in to more recent changes in the data.
A reminder of the annotations that not UTLAs showing concerning trends:
▲ - Cases have risen more than 1.1x above their minimum since the ranking date, suggesting that cases could be on the rise - or it could be noise. The minimum level is shown by a horizontal line under the data marker.
▲▲ - Cases have risen above their level on the ranking date
▼ - Cases have fallen by less than 0.1x their level on the ranking date
Rutland is back this week, there was discussion of a prison outbreak on plotting #12, there's now a second spike in its dashboard data , and it's reporting a case rate above that at lockdown. Again, it looks like a spike and not a sustained rise on the dashboard...
Calderdale and Bury were flagged up as rebounding and this looked to be real last week , they're still not clearly dropping and are flagged up again.
A lot more UTLAs are flagging up as on a rebound. This could be partly the effects of noise and weekend jitter in the data - but if these UTLAs were falling as fast as they previously were, a bit of noise wouldn't be causing them to get flagged up as concerting.
A lot of the UTLAs are ones that have cropped up before. I'm going to try and make a heat map of how many weekly updates each UTLA has been flagged in combined with the severity of the flag.
It's quite concerning how many places are suddenly showing; we'll see what the next update brings. It's also broken the formatting of my plot.
As I said in the OP: Or perhaps more likely, the decay rates are slacking off and there's an unusually bad "shoulder" in the data right now. I ran out of time to actually waffle about what I think is behind the slacking off...
I'm worried that we're hitting some sort of floor in decay rates.
I've commented in plotting #10  and plotting #12  that the exponential decay has been was faster in UTLAs with higher prevalence; now the prevalence is dropping the exponential decay rates seem to be slowing in line with the pattern seen in the data.
Under a simple model far from herd immunity thresholds, the exponential case rate is independent of case levels in a decay phase. With herd immunity emerging, the effect would be in the opposite direction to what we see - over time more people are infected (even as cases decay, new people are being infected) and so the decay rate should get faster, not slower.
Way back in plotting #5  I noted a similar mechanic at the UTLA level where the exponential rate constant was more positive (faster growth) when cases rates were low, and the exponential rate constant would become more negative (slower growth) as cases rose. Again, this should have been happening far from a point where immunity makes a significant difference.
It's wide open to interception what this means, I think the most common in the various discussions is that the more infection there is locally, the more carefully people work to control infection - so there is perhaps a large social/behavioural aspect behind this. Likewise, perhaps the worst hit UTLAs had the best compliance, and as the ambulances fade from sight, the emails from schools about classroom exposures slow down, and the buzz on various local grape vines subsides, people are not so committed to the reduction of transmission.
This is rampant speculation; lots of individual anecdotes out there towards it. It suggest that clear, consistent and coordinated messaging over the importance of sticking to restrictions even (especially) when cases are low is the urgent way forwards.
I remain concerned that having cases rise in younger people combined with a reduction in control measures and a mostly immune older population presents the ideal environment to breed and then amplify an immune evading variant.
I started doing plots because I flat out disagreed with what some other people were saying, and often found the reporting on the news to be confused or confusing. I started putting them all in one place by request from another poster, and it's just sort of rolled on from there. Generally I only add one plot a week or so, and the rest are made by computer programs I keep, so don't take much time.
> When the numbers are coming down all over the place then your analysis and others don't really add much for most purposes, as long as the numbers are REALLY coming down.
I think the last few weeks of decay showed some really interesting phenomenology over local level cases/100k vs exponential decay rates. So, I think it gives me a bit more context to understand - in some limited way - what's going on now. Which is nice, because I feel a lot happier about big stuff going on in the world when I understand it than when I don't, and as good or bad as it is, I can normally park any and all stress about it if I don't feel totally at a loss for understanding what's going on. Some feedback on here, through private message and elsewhere suggest I'm not alone in that.
If my noddy theory on the phenomenology of cases/100k vs exponential decay rates is not too wide of the mark (and I can't think of another convincing approach now we see the same behaviour in a decay phase), it also illuminates the problem which is a step to the solution.
> What your stuff is really needed for, is early detection that things are going wrong and maybe being able to determine why; then it becomes very important. I don't think you are, but remember not to lose that focus on what it's for.
I think it's doing a bit more than that - netting me some interesting connections which is nice as I sever the link to the place I've been hanging around for 20 years, I've been told it's feeding in to some PCTs, and through another channel it's fed in to some interesting and relevant places.
In terms of an early warning for when it's going wrong, I never have managed to answer minimike's question on how to put a probability on the analysis saying "it's going wrong". The noise in the cases data is so weird that a "correct" statistical approach is beyond me - and I abhor making up claimed certainties when the noise model is not appropriate for the statistical test. I see a lot of CIs and errorbars on other plots where, if the authors looked, they'd find the noise is neither gaussian nor Poissonian... But, I think the simple answer is "if things look like they could be rising or it could be noise, they're definitely not falling fast enough".
It's important to understand that the advisory apparatus to government has privileged access to a lot of non-public data, including longitudinal data and things like the SGTF data. Early detection and early action on that early detection are the critical steps to having a data>policy driven approach to controlling case rates that is a stable control loop, and not an unstable oscillatory one. When optimising a control system with a lot of latency in you, you start by looking at the worst latency that is within your ability to fix, and fix that. (Some latency like policy change > infection > detection is beyond all control). I don't think that the analysis of data is the biggest latency in the control system.
> Rutland is back this week, there was discussion of a prison outbreak on plotting #12, there's now a second spike in its dashboard data , and it's reporting a case rate above that at lockdown. Again, it looks like a spike and not a sustained rise on the dashboard...
It is still all going on in the single MSOA with that prison in it (rate over 700, other MSOAs in Rutland all below 150 I think) so I think a prison outbreak is a fair assumption.
I'm not convinced the behaviour we've seen of rate vs exponential constant is mostly down to behaviours changing with time due to risk perception changing. I definitely think it's a factor.
Theorising a bit.
Let's say we are in a rising situation with r=2, so every 10 infected people are infecting 20 new people. Then let's say we introduce lockdown measures universally overnight that drop r to 0.5. Then every 10 infected people will only infect 5 new people.
In an idealistic situation where all people are identical and do identical things, that situation will continue to play out independent of infection rate.
Now let's consider a situation that is still extremely simplified but a bit more representative. We are going to assume that 50% of the population act as per the above, but 25% were only ever passing it on to one other person to start with because they don't go out much, and the the 25% who have people facing jobs started off passing it on to 3 (so r was still 2). Then post lockdown the public facing people continue passing it on to 1 person each while the 25% who didn't get out much before are now shielding and pass it on to no-one. So overall r is still 0.5 (5 people pass it on to 2.5, 2.5 pass it on to another 2.5 and 2.5 pass it on to 0.)
If we start lockdown at a time when there is equal infection throughout, with 10 infected people in the middle group and 5 in the other groups (20 in total), then after one generation of virus there will be half as many people infected. However, now let's assume that the three types of people mostly socialise within their own group. Let's assume that each group passes 75% of infections to their own group and 25% to the other group that isn't shielding. What that means is, after one generation there is now no-one with the virus in the shielding group, there are (10*0.5*0.75) + (5*1*0.25) = 5 infected in the main group and there are (10*0.5*0.25) + (5*1*0.75) = 5 people infected in the public facing group.
So in the first generation we have gone from 20 infected to 10 infected, as expected with r=0.5.
However, in the next generation, with the exact same lockdown and behaviours, the 5 infected people in the main group will go on to infect 2.5 new people and 5 infected people in the public facing group will go on to infect 5 new people. That means, the 10 infected people have now passed it on to 7.5 new people. So r has increased to 0.75.
Obviously this is a very over simplified and extreme example. But I think it illustrates how, with no change in restrictions, weather, behaviours or anything else as a function of time, r can/will change in a realistic society where some types of people are more likely to catch and pass on the virus to others. And that will of course change the exponential rate.
Essentially, as infection rate is driven down, it is likely concentrate cases in the groups who are most likely to catch it and pass it on, and therefore the rate of decline is likely to slow.
I half expect someone to identify an error in this, because otherwise I'm not sure why we haven't thought of it in these terms before.
I also haven't worked through the logic in a rate increasing scenario.
Yes - separate sub-populations as I'd call it is another candidate for the floor; one I'd discussed with Wicamoi on plotting #12 . I think it needs quite a specific set of circumstances to fit with the current situation though.
"A" and "B" are separate sub-populations
We measure the cases for A+B without insight in to their separate identities
A is going to go in to exponential decay during lockdown,"B is going to continue having moderated but at least slightly +ve exponential growth
In order to see rapid exponential decay in (A+B), absolute case numbers for A have to be much bigger than for B for A's falling cases to mask the rising ones in B for some weeks.
Eventually, the cases in A drop below the cases in B and we see growth in (A+B), with no other changes.
You are right I think that this mechanic produces a change in R or the exponential growth rate or whatever when you look at the aggregate number, without any actual changes happening other than separate exponentials going their different ways, before being added together by data capture or reporting. It's particularly in play when there's strong demographic differences to the rate constants.
This has been a motif of the data from the start - I recall a colleague telling me it was all under control in early 2020 because the Worldometer headline was decaying - but it was only decaying because the large but falling number of cases in China outnumbered the small but rising number in Italy, so the sum drops. Until the moment it doesn't.
I've only thought this through qualitatively but I'm not sure how well this fits what we're now seeing. Some inferences:
When we look at the drop in total cases between peak and what may be the handover of control of total cases from A to B, I'd estimate the size of B to be < 1/6th that of A.
Membership of groups probably hasn't changed that much during the pandemic,
Perhaps 1/5th of people in the UK have now had Covid and have some level of naturally acquired immunity
Infections are likely concentrated in B both in and out of lockdown, so past infections and immunity are likely to be more concentrated in B
Vaccination has been going to people working in medical and care sectors who likely make up a proportion of B
So, whilst I agree entirely with your description of the mechanic as in general viable, and whilst I think it can be seen at times in geographic and demographic data, I think the situation now with naturally and vaccine induced immunity leaves relatively little wiggle room for it to be underlying what we see now. Given the size constraints on B it's hard to imagine there are that many people left in a susceptible state to keep driving exponential growth.
The other thing pulling me towards a behavioural aspect is that we see the same link between case rates and exponential rate constants in recent growth phases, and a behavioural link fits.
It could well be a mixture of both effects - a bit from one and a bit from the other, and really A and B aren't totally separate, so a failure of lockdown in B starts to anchor A away from 0 cases/day at some point.
Either way, it seems that some urgent attention is needed from people who can actually do something with the results of their speculation...
There’s no particular geographic pattern to the UTLA heat map, which suggests it’s not a new strain. Or, if it is, it’s already been seeded all over the place.
I think it’s just a function of the lockdown effect plateauing and noise becoming more significant as the numbers reduce. I’m not a statistician but to my mind a random local outbreak at a workplace or some other communal setting which impacts say 50 people would have more of an impact on the decline rate with current case numbers compared to a few weeks ago. So the litmus test will be whether those areas flip back into growth or at least steady state next week.
I've just looked at the raw daily data here . Annotated screenshot below.
The value for Monday 15th Feb is higher than any of the proceeding 11 days, so cases are rising from the minimum in the period since baselining in my plot. This change will making it through the de-weekending and mild filtering I use for the plot.
The absolute numbers are quote low so there's probably a fair bit of statistical noise. Is it real or is it noise? Tune in next week to find out... The ▲ annotation is twitchy to noise in areas with low case rates, but as discussed in other posts it serves as a warning that things aren't in clear decay...
Yes, agree the theory I described doesn't explain this week's particular data without something else going on. It's more just another explanation why r will likely moderate over time. I think that can partially explain some of the things we have seen, including this week, but there undoubtedly other things going on too. It could possibly be this plus noise and a weather effect on testing last week. Alternatively I agree some behavioural change is certainly plausible with all the headlines about exit plans.
I think the best way to look at the potential behavioural aspect is to pick a region you know and devour the media - local news articles, local facebook groups, local business stats (like Ian can give from his garages). Walk the streets. That will give you a good flavour of the behavioural changes - are more people wearting masks or less? Are the roads/shops busier? In my experience people behave more strictly when they perceive things as being bad, and slacken off 'when they're out of the woods'. It's human nature. The thing is that the prevalent variant is the Kent strain now, and the effect of slackening is going to be magnified by it. What people 'got away with' coming out of lockdown 1 they won't now. There's a lot of 'out of the woods' talk going on - moving onto younger people in the vaccination process, good news on single dose effectiveness, good news on transmission reduction, and the talk of summer holidays. Nice weather will encourage people to look positively or to go out and socialise. Rock climbers will be thinking of hitting the crags again!
Your theorising, am I right in saying that unless you have a homogenous group with homogenous interactions, then you can effectively construct/postulate groupings/interactions that will produce effects that are (for a time) hidden by the overall R factor, but may then burst out of obscurity in surprising ways?
In which case I can see that it may be possible to construct something that gets closer to the observed numbers - presumably once any "noise" has been suppressed.
Thanks for a great discussion both of you, and the post to which I am replying is particularly fine. I'm sure that both distinct sub-populations, intentional population behaviour, accidental population behaviour (not forgetting sub-population behaviour nested within that), virus transmission, virus mutation, weather conditions, vaccinations, and a lot more than six other impossible-to-know things that I should have thought about before breakfast, are all involved, and all interacting.
My background is behavioural ecology where everything is interaction, and nothing is law, so perhaps I'm biased, but population behaviour seems likely to be one of the key factors to me, and given that it's really the only one over which government has any influence, I'd say it's certainly one to think pretty hard about.
Thanks for the plotting update as always wintertree - I find my interest in it increasing rather than decreasing.
I think it's important to recognise that my 3 group model with one group taking themselves out of society was only used for the purposes of illustrating the concept. In reality to get anywhere close to approximating the truth I'm sure you would need to account for many many groups and lots of different ways they behave like Wicamoi lists. Ultimately there are 65 million of us who are all a bit different. So personally I don't think it is worth trying to think too much about exactly what the effect of this should be on any given date, but it does give an indication that the trend over time without any changes in behaviour or restrictions (you might call it a baseline) should not necessarily be expected to be a constant exponential rate like we have been looking for.
> Yes, agree the theory I described doesn't explain this week's particular data without something else going on. It's more just another explanation why r will likely moderate over time. I think that can partially explain some of the things we have seen, including this week, but there undoubtedly other things going on too. It could possibly be this plus noise and a weather effect on testing last week. Alternatively I agree some behavioural change is certainly plausible with all the headlines about exit plans.
Given the roll off in decay seems pretty sharp and countrywide (where much of the response and data collection is devolved) I suspect it's really there and we're back in growth. The fact it's apparently countrywide and synchronised suggests something we've all been exposed to unlike Kent last time. The unlocking messages differ across the nations but the mood is changing. I suspect it's predominantly the cold snap in combination with the sense it's all over.
If so it's far from over, it suggests we have very little freedom to relax restrictions without going back to growth.
Here's hoping it's a broken machine somewhere distorting the usual lag pattern but I'm not hopeful.
I think its unfair to look at this either way as yet, but sensible to be cautious. In a week it will be more obvious and by March 8th quite clear. If Boris is good to his data driven word and problems are real he may well need to U turn yet again.
Johnson being good to his word is always a big ask...
And, even if he does stay 'data driven', what datum is he going to use to make the decision? A case rate of 50k/day? 10k/day? 100/day? 'Data driven' still gives plenty of scope for arbitrary decisions.
If it's coldsnap driven (more social meeting indoors and office windows closed perhaps) it may remain a blip up until March 8th and the sensitivity will remain quite well hidden from those prone to optimism. Time will tell though as you say.
> The value for Monday 15th Feb is higher than any of the proceeding 11 days, so cases are rising from the minimum in the period since baselining in my plot.
Squinting at that raw plot, it appears to have plateaued. The moving average seems to agree. If you're baselining/comparing against 1st Feb (two weeks prior), then 15th Feb is lower. 31st Jan was a nice, low, Sunday value...
The weekend values for 13/14 Feb are pretty low; maybe a bigger reporting lag than usual.
But using the 'if it's not obviously falling, we need to worry' maxim, we need to worry...
Where Boris used his view on data to ignore the SAGE advice in September and December it became publicily obvious that it had happened and it will again. He has the blood of tens of thousands on his hands already (,and all the resultant extra unnecessary economic damage), so I don't need any extra cynicism about his likely actions.
> Given the roll off in decay seems pretty sharp and countrywide (where much of the response and data collection is devolved) I suspect it's really there and we're back in growth.
Just some partly linked random musings...
I’ve been pondering the issue of Covid inequality, specifically why post industrial areas often struggle to keep their rates low and the issue of BME communities and high Covid + low vaccine uptake.
During the January lockdown, many parts of the SE had about triple the case rate of most of Grter Manc. Currently, many parts of SE have about half the case rate of most of Grter Manc. In the last ten days, 3 (of 10) GM LA’s have seen cases rise, most other areas easing up. Whilst there are some outlier areas, it does appear that the post industrial areas really struggle to get cases down. We experienced this in August/September so should come as no surprise.
Tower Hamlets is the most deprived LA in London and has 55% BME population (32% Bangladeshi). Yet it’s death rate is 139 and a case rate of 56. This strikes me as much lower than average. So I went on the govt dashboard and counted. Out of 343 local authorities only 47 have a lower death rate.
Wigan has death rate of 282, case rate of 175 and is 97.3% white and is less deprived.
I also looked at some other recent research - traffic usage is significantly up on Lockdown 1 and less people are following the rules .
My opinion: this tailing off/growth is not driven by bad weather. It is more about lockdown fatigue, especially in areas of perpetual lockdown/measures. Social inequality / deprivation is (possibly) an over simplistic view - probably more about types of employment and bad management, exacerbated by low income meaning people have to work. A Tiered system is almost a given. This stuff depresses me, but is more than counter balanced by recent news on vaccines.
> PS jkarran - the bump this week isn't universal. It shows up in all regions but far from all UTLAs or even all city regions.
Would you expect it to? The smaller regions are much noisier, it's much easier for an emerging trend (or temporary blip) to hide or be exaggerated over the short term in smaller regions. If it is just a weather related blip it may have been all but cancelled by noise (or regional behaviour differences) in some cities and UTLAs.
Hopefully it'll just vanish as the weekend reporting lag clears through.
> My opinion: this tailing off/growth is not driven by bad weather. It is more about lockdown fatigue, especially in areas of perpetual lockdown/measures. Social inequality / deprivation is (possibly) an over simplistic view - probably more about types of employment and bad management, exacerbated by low income meaning people have to work. A Tiered system is almost a given. This stuff depresses me, but is more than counter balanced by recent news on vaccines.
I'm sure there's an element of that. Close to home I can see it in my Mrs agitating to book a March holiday and my neighbours getting more use from the hot-tub again, people are tired of all this and the government is giving the impression we're well down the home straight where in reality we just don't know yet.
I'm sure fatigue will play a part in rolling off the rate of decrease but I'd be surprised (in the absence of another Barnard Castle gaffe) if the effect were so sharp, I'd expect to see the it smeared out over time and place producing a much less abrupt flattening to the cases curve. That said I'm rubbish at people and live in a bubble, I could be misreading the public mood.
I still suspect the change is real and it is a transient event but I'll not be surprised if I'm wrong.
I believe more people are being asked to ditch WFH and go back to the office. I know one example where they have been called back - 10 out of 12 staff caught Covid!! I guess this is more likely to happen in the ‘usual suspect areas’ of what feels like perpetual lockdown. Thinking aloud: lockdown fatigue extends beyond getting bored so dropping your guard - it’s about having to work and dodgey work practices etc.
A lot of really good input from a lot of posters; sorry I’ve not engaged with it all. Busy days...
Offwidth’s link to comments from a SAGE member and a specific comment there on workplaces, misha’s comment on individual outbreaks being more likely to show against low numbers and the way some UTLAs seem to pop up at random in the watch plots, Si dH’s comments on Rutland and an outbreak, and on sub populations, your comment - it all fits well I think with increased workplace transmission now being exposed prominently by falling cases elsewhere.
I agree with jkarran about the timing of this standing out as not compatible with gradual failure.
I think we must have several very different things coming together right now, and a lot of confusion here.
As wicaomi says, behaviour is one aspect where the government can tale action, so regardless of what the problems are that’s where some of the solution lies.
> Thinking aloud: lockdown fatigue extends beyond getting bored so dropping your guard - it’s about having to work and dodgey work practices etc.
I agree (For example, I'm partially back in the office this lockdown vs WFH last time) but the effect of that 'fatigue'* should have been smeared out over weeks, either reducing the rate at which cases fell and or reducing the degree of freedom to relax restrictions when they have fallen.
* in some cases it's that businesses now have more information and have adapted to operate quite safely with more staff 'in'.
> If you're baselining/comparing against 1st Feb (two weeks prior), then 15th Feb is lower
Check the criteria for the different warning flags in the original post. ▲ - Cases have risen more than 1.1x above their minimum since the ranking date, suggesting that cases could be on the rise - or it could be noise. The minimum level is shown by a horizontal line under the data marker.
So, it's flagged because cases are higher on the 15th than the minimum level between the 1st and the 15th. Deweekending and a 13-day 2nd order SG filter is applied to the data before measuring for these warning flags, although this is still more noise prone at the leading edge f the data, which the plot baselined two weeks in to the past is.
> They usually point to something that's not right, or some misunderstanding.
Exactly - this is why I encourage anyone I work with to keep pressing because until we either find the problem or the misunderstanding, one of us is wrong somewhere which isn't good. Likewise I expect people to be happy being pressed on things where there's misunderstanding or scope for error. In this case I hope I've made it clear it's a misunderstanding - and I can revisit how the plot is explained in the future to make the clearer.
I've attached a plot of my filtered time series (bars) and raw data ('x' marks) for the region for the last 4 weeks. The two weeks used to fire my ▲ criteria are shown in red. As the description says "[... cases could be on the rise - or it could be noise".
In past weeks, very few places have fired this warning off - a lot have this week. It could well be a depression in figures caused by the winter weather around the 12th/13th. Whilst that would explain away the apparent bottoming out and rising that triggers the criteria, it would also have added a false decay to the data, perhaps masking a more gradual levelling off for a while.
Re the slowdown in decay/plateau/regrowth we can also look at some other countries that are a few weeks ahead of us to see their experience (with all the usual caveats about such comparisons), and it does look like due possibly to the timing of the lockdown earlier in winter, more workplaces trying to stay open after previous lockdowns, or transmissiblity of uk variant other countries haven't crushed the numbers to as low a level despite early rapid decay.
Denmark was mostly open throughout September - November, pubs and gyms shut 7th Dec and full lockdown started on 24th Dec and is still in force (though inside private socialising is allowed as it has throughout the pandemic, recommended max 5 people and the same group). Case numbers have now come down a lot but reached a plateau at a higher rate than it did previously post march lockdown (multiply by ~12 for population to compare to uk case numbers, so christmas spike reached ~55000 cases a day, lockdown came in 2 weeks earlier than uk, plateau at ~6000 last 3 weeks):
The Indie sage report from Friday is really good. 15 mins on the data linking to a worrying slowing in cases probably due to failure of workers to self isolate in deprived areas. 10 minutes of a sensible opening strategy (that the government will not be following) and then a very sobering look at what its been like inside the NHS and why 'let it rip' would be very bad.
> > .....but the effect of that 'fatigue'* should have been smeared out over weeks,
I’ve often pondered whether there is a ‘tipping point’, so rather than a linear ‘regression’ we get a sudden dip. For example, most people are OK home working for 4 weeks but then 75% of home workers go ‘I’ve had enough, I’m back in the office.’ I’ve no evidence for a aying this, but my gut feeling is our government doesn’t seem to grasp (and factor in their planning) how people/communities function. Being driven by data is key, but they need to factor in the realities of life/human behaviours. Strange how the SE went from mega Covid rates to below half of their post industrial counterparts. In other words, the ‘usual suspect areas’ reached their tipping point sooner because of being in restrictions for much longer. Not a clue really tbh.
> Strange how the SE went from mega Covid rates to below half of their post industrial counterparts.
This is really correlated with the rise to provenance of the new variant. Two factors I think include local response to high local infection rates once the hospitalisations hit the grapevines, and significant immunity emerging in the sub-populations most at risk of catching and spreading the virus during T4/lockdown conditions. Both of these lead to a self-moderating effect; I think I see that looking at the plot 18 heat map; some of the changes in the data at a regional level that tie to the data not the messaging/lockdown; this I think tells us that people and behaviour are one of the key inputs. So, when you say...
> Being driven by data is key, but they need to factor in the realities of life/human behaviours.
The key is understanding how to link the two, I think.
> I think the best way to look at the potential behavioural aspect is to pick a region you know and devour the media - local news articles, local facebook groups, local business stats (like Ian can give from his garages). Walk the streets. That will give you a good flavour of the behavioural changes - are more people wearting masks or less? Are the roads/shops busier? In my experience people behave more strictly when they perceive things as being bad, and slacken off 'when they're out of the woods'. It's human nature. The thing is that the prevalent variant is the Kent strain now, and the effect of slackening is going to be magnified by it. What people 'got away with' coming out of lockdown 1 they won't now. There's a lot of 'out of the woods' talk going on - moving onto younger people in the vaccination process, good news on single dose effectiveness, good news on transmission reduction, and the talk of summer holidays. Nice weather will encourage people to look positively or to go out and socialise. Rock climbers will be thinking of hitting the crags again!
I may as well reply to this post since i've popped up in it, and it will be my last meaningful contribution as we have over the last few weeks been megabusy handing over the garages to the new owners, so no more access to data for me
Anyhoo, january is always a quieter month - nobody has any cash left after xmas, and very weather dependent, but generally continued in the vein of december in that the lockdown didnt have a major dampening effect on fuel sales, and indeed the effect, as with previous lockdowns, lessened as time went on. People are tired of it now, and my take is that if restrictions are lifted any more than very gradually, people will go nuts, and we'll be in another lockdown by may. Vaccinations or not. There was a significant dip in demand when the cold snap hit, but that always happens in winter. Up to that point, demand for fuel was around 85% of january's expected volumes- it started the month at about 75%. S not really a national lockdown, and if the rate of decay of cases is not as great as it was, this would anecdotally back that up. One thing that people around these parts do appear to be doing is mask wearing and social distancing. Whether thats because those who insist its "not for them" have moved on elsewhere i dont know (we do still get some, and those remaining do seem to get abusive more quickly when asked to mask up or use the night hatch).
It included a few things I had missed, like only 5% of secondary school kids in school in England (why risks are bigger there for a complete opening). It was a very good public summary of the situation on Friday. Most of Wintertree's excellent work needs a bit more thought to follow, some plots quite a bit more thought.
> this I think tells us that people and behaviour are one of the key inputs. So, when you say...
I think it also tells us that when people understand a clear message (from whatever source), they comply, and do the right thing.
It's why clear messaging us so vital, and why I am worried by the government giving the message that 'things are okay now'. I was even a bit dismayed by the 'good news' in the Indie SAGE video. We need to keep stressing we are not out of the shit yet, not by a long way. Falling numbers are what we expect; that's no reason to celebrate. The reactions to the numbers at the moment should be either continued caution (falling), or grave concern (flattening, or rising).
A different way of looking at the cases data to try and get a feel for what's going on with cases over the last week.
I've wrapped each week around so every Monday falls in the same place, and every Tuesday etc. Historic weeks are grey, the current week (in the lagged data set) is blue, and the week before it is red.
We see two sorts of fall on the plot - week on week fall which can be seen in the "steps" of an individual bar, and the day-on-day fall which is visible in the change in height of a single colour across the plot. You can see that the day-on-day fall is part "real" and part the structure from the weekend sampling anomaly - pick a week (colour), look at cases on the last days (right side) of that week and you'll see some days (sometimes including Friday) of data reported in the end of that week are lower than the first few days of data in the next week (by colour).
The important plot for interpretation is the right hand side one which has a log scale on the y-axis. If cases are falling on a fixed exponential rate, then regardless of a specific "day of week anomaly" overlaid on cases, within any single bar, each step should have the same physical size, and within a single colour, each step should have the same size on every day of the week - the anomaly only manifests as different vertical offsets.
This pattern associated with exponential decay and a day-of-week anomaly seems pretty consistent until, wham, out of the blue, Monday 15th of February. The first "red" step is tiny showing much less exponential decay, week-on-week over the last few Mondays. The red bars are getting bigger size then, so there's some sign of more exponential decay returning - but not enough to restore the previous decay rate. So it looks a bit like a mix of unusually strong over-reporting on a Monday (but, from where?) and lowered decay rate.
I've also put a plot for Scotland below. The case numbers are much lower so the statistical noise is a lot worse (as a fraction of cases), I'm not sure much meaning can be found starting at it.
Thanks - it was interesting trying to understand these graphs. But when you say...
> So it looks a bit like a mix of unusually strong over-reporting on a Monday (but, from where?) and lowered decay rate.
Log-scaling doesn't come naturally to me, and you have demonstrated yourself to be nobody's fool on more than enough occasions, so I'm likely missing something.... but what makes you think there's an 'unusually strong over-reporting' element involved? I'm not seeing any particular signal in the tail of the red week preceding Blue Monday (as you say: "but from where?"), and the succeeding slight increase of red Tues-Thurs over red Mon feels like it could be just noise as lower numbers are involved. Speaking of which...
> I've also put a plot for Scotland below. The case numbers are much lower so the statistical noise is a lot worse (as a fraction of cases), I'm not sure much meaning can be found starting at it.
Agreed - apart from general gloom - but many thanks for including it anyway.
Poor messaging... Yes. WTAF is Johnson doing with the phrases "irreversible" and "one-way street to freedom"? Short-term incitement and long-term hostage to fortune. Where is this language coming from?
> but what makes you think there's an 'unusually strong over-reporting' element involved? I'm not seeing any particular signal in the tail of the red week preceding Blue Monday (as you say: "but from where?"), and the succeeding slight increase of red Tues-Thurs over red Mon feels like it could be just noise as lower numbers are involved.
I haven't put a p-value on this being real vs noise, nor do I think I have a sound basis to try and do so.
The effect (if real) is nowhere near enough to explain most of the drop in the decay rates I think. But the intersection of a real decay and a weather effect is a possibility...
The trend is that the unusually small red zone is getting larger over the week, which is what I'd expect for delayed reporting. The drops from grey to red for Saturday and Sunday look in line with all the others, but, if the decay rate of cases has actually slackened off somewhat - as we can infer from the blue days, then it shouldn't have dropped that much for the Red Weekend.
So, I think idealised cases dropped ~2/3rd as much as normal from genuine but reduced exponential decay from grey to red on the weekend, and the other 1/3rd of the actual drop (on both Sat and Sun) is due to the bad weather effects shifting cases towards the following (red) days, in particular the Monday. That weekend is when the deep powder snow was drifting over roads up North and when East Anglia has actual proper snow. If I'm write, the drop from red to blue will be biggest mid-week as blue is "artificially" raised on the left, and red is "artificially" lowered on the right.
I'm not sure it's worth this much thought really!
> And you're wasted in England.
Was that literal or metaphorical? Certainly it was true in a literal sense...
> Unfortunately, the government’s messaging has been so poor for so long I think a lot of people will take today’s government announcements as signalling the end of the pandemic.
A 4ᵗʰ wave feels almost inevitable, and with the government progressively relaxing restrictions, I think a lot of employers will take their cue from government and consider the risks to staff mitigated. R<1 was notably absent from the reopening criteria.
The next 4 months seem like a particular risk for adults in their 40s and 50s as the adherence to restrictions collapses in younger adults. Whilst hospitalisations are currently limited by more careful behaviour and vaccination in > 60s and falling cases, relax restrictions and the exponential growth passes a threshold and bam, suddenly there's enough people waiting to be hospitalised in the age range 40-60 to put healthcare at risk, again.
How inevitable it really is is going to depend on the battle between fewer restrictions increasing R and ever more and younger vaccination decreasing R. If no new, immune evading variants emerge then vaccination wins in the long term. With the current vaccination rate, the data emerging on efficacy against transmission after the first jab and current infection rates and exponential rates, it feels too close to call on the medium term. Certainly too close for comfort.
> The drops from grey to red for Saturday and Sunday look in line with all the others, but, if the decay rate of cases has actually slackened off somewhat - as we can infer from the blue days, then it shouldn't have dropped that much for the Red Weekend.
This is what I was missing. These are quite subtle changes though, and for me it is difficult to judge how to interpret them.
> That weekend is when the deep powder snow was drifting over roads up North and when East Anglia has actual proper snow. If I'm write, the drop from red to blue will be biggest mid-week as blue is "artificially" raised on the left, and red is "artificially" lowered on the right.
Well done for making a (conditional) prediction - let's see. I thought about the snow when you posed the question "but where from?" ... then got more interested in the signal for your asking.
> > And you're wasted in England.
> Was that literal or metaphorical? Certainly it was true in a literal sense...
It's the justification for why the unlocking is so slow, isn't it? It's the words with which he is trying to buy off the CRG and their allies.
"Yes, it's slower than you'd like, but that's to really make sure it's the last"
(For clarity, my view is that it is frustrating that more in the way of low risk outdoor activity isn't allowed from the 8th - but otherwise I'm not arguing myself that it is too slow. But plenty will be)
Edit: and if on cue, my Google feed suggests this in the Torygraph (I can't read the article, but the title says it all...)
> > Unfortunately, the government’s messaging has been so poor for so long I think a lot of people will take today’s government announcements as signalling the end of the pandemic.
> A 4ᵗʰ wave feels almost inevitable, and with the government progressively relaxing restrictions, I think a lot of employers will take their cue from government and consider the risks to staff mitigated. R<1 was notably absent from the reopening criteria.
> The next 4 months seem like a particular risk for adults in their 40s and 50s as the adherence to restrictions collapses in younger adults. Whilst hospitalisations are currently limited by more careful behaviour and vaccination in > 60s and falling cases, relax restrictions and the exponential growth passes a threshold and bam, suddenly there's enough people waiting to be hospitalised in the age range 40-60 to put healthcare at risk, again.
> How inevitable it really is is going to depend on the battle between fewer restrictions increasing R and ever more and younger vaccination decreasing R. If no new, immune evading variants emerge then vaccination wins in the long term. With the current vaccination rate, the data emerging on efficacy against transmission after the first jab and current infection rates and exponential rates, it feels too close to call on the medium term. Certainly too close for comfort.
I agree with all this. However I have to say that it was always what I had expected to happen and tentatively thought was probably the right way forward myself, until with the new variant the risk of vaccine evasion became more greatly apparent and it seemed they might take a more cautious approach to limiting case rates. It now appears that we over interpreted the change in Govt language. It seems to me that they will allow cases to rise between steps as long as hospitalisations don't rise quickly. 4-5 weeks between steps gives them time to see the hospitalisation effect of a step change in restrictions whereas you can see the effect on cases changing within about =< 2.5 weeks (based on historical experience from the previous lockdowns and openings.)
The crucial hope for me is that either no new variant comes along inside the next six months that evades the vaccine or that our test/trace organisation is focused enough to catch one if it does. If we get lucky on that front and vaccine effectiveness stays high then in the long run then I think allowing cases to rise will be seen in hindsight as the right decision to allow schools, social interaction and the economy to gradually open up. There is certainly a lot of risk though! If it goes the wrong way...
Next couple of weeks will be crucial in cases data to see whether we actually have any margin in r before schools open.
> Also, the ONS claims that PCR testing has a FPR of 0.8 - 4.3%
Do they? That would be well high for a well run PCR lab.
I think you have severely mis read something due to the ONS presenting the FOIA request very badly. That range you give comes from the “You asked” verbatim reproduction of the request sent to the ONS from an interested parted, and is not factual.
Linked to from your link is the already publicly available material they point the interested party to:
We know the specificity of our test must be very close to 100% as the low number of positive tests in our study means that specificity would be very high even if all positives were false. For example, in the most recent six-week period (31 July to 10 September), 159 of the 208,730 total samples tested positive. Even if all these positives were false, specificity would still be 99.92%.
We know that the virus is still circulating, so it is extremely unlikely that all these positives are false. However, it is important to consider whether many of the small number of positive tests we do have might be false. There are a couple of main reasons we do not think that is the case.
That puts an upper bound on the false positive rate of 0.08%, and that’s for a random sampling survey. The running average of daily PCR tests is about 310k/day at the moment. 0.08% x 310k = 248 false positives a day.
But that’s an absolute upper bound; reality could well be an order of magnitude lower.
You’re right that it’s another candidate for a “floor”, but assuming all the labs are running control negatives to test for cross-contamination, it shouldn’t be an issue.
> The crucial hope for me is that either no new variant comes along inside the next six months that evades the vaccine or that our test/trace organisation is focused enough to catch one if it does
My gut is that test/trace is only much use with cases < 1000/day for total elimination of a variant. There were messages coming out from advisory members along those lines but any such caution seems gone now. It may be that they think they can do it with 15x the numbers, going off a perceived success at keeping the SA variant from spreading exponentially with the enhanced tracing efforts. Reported numbers aren’t clearly winding down, however.
With evasion it looks like all the variants close to the original have enough co immunity to prevent most serious illness and hospitalisation; as the vaccine rolls out to everyone, we then have the possibility of a full blown uncontrolled pandemic that passes as little more than a cold or at worst a bad flu season, and leaves most people with some naturally boosted immunity, the numbers declining the vaccine hopefully being low enough that, with the smaller R from mass vaccination of others, prevents a wave of hospital overload. Is this a good or bad thing? I don’t know how statistical bounds can be put on the probability of new variants changing further towards increased evasion or increased lethality. Most of the changes that are showing a selective advantage seem convergent, so it feels like there’s a convergent nexus ahead for the virus. What’re the next variants from there? Short of swapping the spike protein (again?), total loss of immunity seems unlikely.
I’m all for keeping mild restrictions as the vaccine rolls out until there’s a bit more clarity around all this - the world will provide answers soon enough - and until more of the promising candidate therapeutics are out of trials to disarm infections earlier on. I’ll not get what I want I suspect. Still, of all the risky approaches to controlling cases since last February, this is the first one to feel like it’s got better than even odds of paying off...
> > Also, the ONS claims that PCR testing has a FPR of 0.8 - 4.3%
> Do they? That would be well high for a well run PCR lab.
Are we sure that all the labs are well run though? Because last year the woman who runs my local racecourse had some of the PCR Lab contracts... the footage on the BBC Panorama documentary did not make for easy viewing with leaking samples chucked in the same cardboard box.
I hope since then that things have been tidied up with regular audits now in place.
> I think you have severely mis read something due to the ONS presenting the FOIA request very badly. That range you give comes from the “You asked” verbatim reproduction of the request sent to the ONS from an interested parted, and is not factual.
You are right, apologies, on further digging it would appear that those figures originally came from the Lancet and a couple of studies linked within the article.
> Edit: and if on cue, my Google feed suggests this in the Torygraph
Reporting of the newspaper headlines went as expected: the right wing press complaining everything was too slow.
It very much bothers me that the mutation issue is not being discussed, even by the experts. Tildesley made no mention of it this morning.
It also very much bothers me that the BBC reporting talks about the government approach, and then only mentions dissenting voices calling for a faster unlock, but does not mention dissenting voices calling for a more cautious approach.
The lancet article is not a primary source for that claim, but references... The government document and the non reviewed medrxiv ”pre print” (I’ve seen no sign it was intended to be published anywhere...?).
The government document does not give a likely false positive rate - if notes a hard upper bound (positivity when testing during a pandemic) that would in no way be taken by any sane person to be representative of the false positive rate.
The medrxiv document is a non peer reviewed meta review of other labs doing RT-qPCR on a bunch of other, largely random viruses going back up to 16 years. From a quick glance I’m not sure this would pass peer review... I’m not convinced a meta review even of the same virus would have much value as it *all* comes down to sample integrity and QC in the specific labs involved.
So I’m afraid I see absolutely no support in these documents for the false positive rate, it’s all smoke and mirrors and the simple answer is “we don’t know” and despite some attempts to paint a credible looking counter factual in the literature, it just isn’t know. Understanding the false positive rate of the most sensitive diagnostic test isn’t an easy nut to crack I think.
I assume the lighthouse labs run control negative and aren’t engaging in some giant conspiracy to cover up a false positive floor revealed by these. I also think - assuming QC hasn’t changed since last September - the PCR positivity now vs then tells us we’re at least 6x higher than an absolute worst case bound given by the positivity then.
> I guess we will just have to see what the data looks like in 1 - 2 weeks time.
It seems clear from the unlocking document that it’s the data on hospitalisations that guides unlocking from here on, and my social media this morning is a sea of people making plans around each nominal unlocking data - the sense of urgency is gone from most people and I suspect with it their full commitment in the meantime.
> A different way of looking at the cases data to try and get a feel for what's going on with cases over the last week.
> This pattern associated with exponential decay and a day-of-week anomaly seems pretty consistent until, wham, out of the blue, Monday 15th of February. The first "red" step is tiny showing much less exponential decay, week-on-week over the last few Mondays. The red bars are getting bigger size then, so there's some sign of more exponential decay returning - but not enough to restore the previous decay rate. So it looks a bit like a mix of unusually strong over-reporting on a Monday (but, from where?) and lowered decay rate.
I wonder if the post was significantly disrupted by the weather causing a backlog (possibly even just a glitch in the typical delay between suspecting symptoms and dropping the home test swab in the postbox). We didn't get the snow too bad here but it's Britain, somewhere always grinds to a halt when it snows and I guess the postal tests are going a long way to one of a handful of big hubs. Maybe also coupled with a softening of the decay rate caused or exacerbated by the weather. The fact the pattern seems to repeat in Scotland suggests a large scale external influence, not just one broken down lab or a new door to door screening program somewhere but then we might expect Scotland to cope a bit better with snow and presumably they mostly use their own labs. Maybe it's not complicated, just sick people understandably not wanting to walk to the post box in bad weather.
What was everyone else's take away? To me it sounded like 50+ will get vaccinated in time, under 30 you're getting the live unattenuated vaccine, ~35-50 you'll be peer-pressured into taking your chances under the bus. That what everyone makes of it?
F**ing press today... it's like they never forget when a politician says something they later contradict, but they have free reign to change their bloody tune every 30 seconds. "What are we waiting for?" comes not 2 weeks after all the crap on the front pages about how they government was going to risk everything by unlocking too soon. For the sake of living f*ck it makes me want to shove their incessant clickbait and won't-take-no-for-an-answer cookie nags where murdoch's head is.
> What was everyone else's take away? To me it sounded like 50+ will get vaccinated in time, 0 you're getting the live unattenuated vaccine, ~35-50 you'll be peer-pressured into taking your chances under the bus.
Yep, we're going under the bus so long as hospital numbers stay sub-critical. Looks like the vaccine program for <50s is being raced against another 'rip' of covid, presumably hoping the vaccine slows the growth and ultimately wins the race but tolerating the losses (it'll still be tens of thousands) if not. Most of us will survive, lots of us won't.
Perhaps worse, it looks like we could be going into autumn-winter again with very high prevalence, probably more new strains and immunity of unknown longevity (Manaus should be a warning here). I wouldn't bet against the virus at this point.
> F**ing press today... [...] For the sake of living f*ck
I haven't looked. I'm not going to for a while. Had a successful maths home schooling lesson this morning, got a mountain of non-Covid data to pour through, and got some now-established Hawthorne hedge that needs its tree-protectors removing. Even without any decent gardening gloves left I'd rather do that than read the news today.
You're spot on with their pogoing in attitude in a way they refuse to accept from a politician; particularly toxic when there may be a need for data driven policy changes in the coming weeks and months.
> ~35-50 you'll be peer-pressured into taking your chances under the bus.
The vulnerable adults are being offered a vaccine in some places (my daughter, age 21, got the invite - she’s had a few accidents/ A and E visits so thinks she’s ‘on a list’. So yes, there is a rush to vaccinate over 50s and vulnerables
Discussions upthread about data and bad weather: the parts of NW England which have been showing increase in cases were not hit by snow etc so possibly reading too much into the weather factor
I’ve found myself increasing looking at the Daily Mirror - says a lot about both me and what I feel about other media. The BBC is best left for articles about bears and toilets and woods.
Earlier today, BBC news had a countdown ticker to lockdown ending. Yawn.
Has it actually been confirmed that the vaccine strategy is still basically going to be working their way down the age bands? I'd have thought there was a point where starting as much as possible to prioritise people with a high risk of infection / transmission - essentially, people who are still working in places with other people - would become a more optimal strategy...
You're quite right, it hasn't been mentioned explicitly and to be honest the longer that debate is delayed the happier we'll all be. But you're right. By age is only the clear choice up to a point. After that the "me me me first"s start in earnest. The teachers have sort of opened the bidding there with an arguably very strong case, but then everyone else gets to have a moan about why they should be next as soon as the floodgates open.
Dunno, as someone who works from home, gets shopping delivered, doesn't have kids at school and is generally leading a very low-risk lifestyle I'd be extremely happy with a strategy that gets overall infections down as quickly as possible, even if it meant that I had to wait a bit longer for my own shot. I'm not convinced it wouldn't actually work out better for me in the long run, too - I'd rather infections were low enough that things could open and feel relatively safe anyway than be vaccinated myself but have things closed for longer.
> I'm not convinced it wouldn't actually work out better for me in the long run, too - I'd rather infections were low enough that things could open and feel relatively safe anyway than be vaccinated myself but have things closed for longer.
I agree - once hospitals are out of the woods, using the vaccine capacity strategically to drive cases down as fast as possible is the most effective way to protect everyone. This of course depends on a policy of keeping R<1 for the next 3 months, which is the first apparent stumbling block.
However, the problem I see is in rapidly determining and implementing that strategy. Even without the cracks of division appearing in terms of prioritising specific groups, this is a non trivial task; I expect vaccination will be widespread before everyone with a say could agree.
As there is some age stratification in society, I’ve wondered about a random approach for 18-60 year olds, that’s a skew distribution towards older people to begin with, aiming to finish one year of age in ranked decreasing order every 4 days or so. This balances moderating rates across all ages (for the protection of all) and eliminating transmission and disease in risk ranked order. Objective, not anointing some professions and balancing societal and individual wellbeing. I can’t see it happening for a moment though... I think this can also adapt quite organically to capacity and 2nd doses, although I’m having a slow day today so perhaps it’s nonsense.
> However, the problem I see is in rapidly determining and implementing that strategy. Even without the cracks of division appearing in terms of prioritising specific groups, this is a non trivial task; I expect vaccination will be widespread before everyone with a say could agree.
Trying to optimise it on an individual level is presumably impractical, but surely there are broad brush-strokes approaches that could be used? We should be able to identify high-risk jobs by this stage (I'd assume food service, maybe manufacturing, transport, call centre staff etc) so just going through major employers should get some proportion of needles in more effective arms than just mindlessly working down the age categories?
I believe the first 9 bands (down to age 50 and including all vulnerable adults younger than that) are fixed in priority order and to be done by April 15th. They should therefore have some immunity before any indoor hospitality opens again. After that, the roadmap document issued yesterday on gov.uk explains briefly that the government has asked the JCVI to advise on how to best meet the objective to further bring down deaths, serious illness and hospitalisations. It leaves open what the best vaccination strategy to do that is and doesn't say when a decision will be made, but obviously they'll be starting just after Easter at the latest if things go to plan, so it must be fairly soon.
> Dunno, as someone who works from home, gets shopping delivered, doesn't have kids at school and is generally leading a very low-risk lifestyle I'd be extremely happy with a strategy that gets overall infections down as quickly as possible, even if it meant that I had to wait a bit longer for my own shot. I'm not convinced it wouldn't actually work out better for me in the long run, too - I'd rather infections were low enough that things could open and feel relatively safe anyway than be vaccinated myself but have things closed for longer.
My son goes to pre school three times a week but I'd definitely still agree with this as well. No particular interest in my own vaccination. Once vulnerable people are as safe as they can be I just want them to be used to keep infections low if possible so the risks for others more vulnerable than myself are less and we can eventually go back to semi-normality. I'd definitely prefer them to vaccinate front line workers before me.
While we're sharing, my thoughts on this are that it will be a dot in history really soon. We're talking about a part of the process that is aiming to go from the listed priorities, through everyone else, down to having a last few to mop up, in about 3 months. Dart gun starts to look like the best strategy when you think about it in those terms. Realistically nobody in these groups is getting jabbed more than 10-ish weeks before anyone else, so... meh. But, that's not how front pages work.
> Trying to optimise it on an individual level is presumably impractical, but surely there are broad brush-strokes approaches that could be used? We should be able to identify high-risk jobs by this stage (I'd assume food service, maybe manufacturing, transport, call centre staff etc) so just going through major employers should get some proportion of needles in more effective arms than just mindlessly working down the age categories?
It quickly throws up awkward divisive questions though. Prisoners or sex workers for example, higher or lower priority than checkout staff? These decisions should of course be dispassionate, evidence lead with a net harm reduction objective but simply allowing them to be is a political decision with consequences when the electorate don't like who they're queuing behind.
If the delivery rate can keep accelerating then age stratified delivery and progressive relaxation of social controls shouldn't go too badly wrong so long as there is a feedback mechanism to unwind pause then retry any failed changes, so long as they're not to echo Johnson 'irreversible'. We'll see.
If the vaccine delivery falters or can't keep pace with case growth in the young to middle aged and the government won't row back relaxations then something like Wintertree's idea of an older-weighted random distribution of shots across the remaining age range might be a good idea. I think if it goes wrong (or right for that matter!) in reality it'll all happen too fast for refinements like this, we'll either scrape through with a hefty bump in spring-summer infections that does tolerably little harm to people and confidence or we'll end up re-tightening restrictions.
> Trying to optimise it on an individual level is presumably impractical, but surely there are broad brush-strokes approaches that could be used? We should be able to identify high-risk jobs by this stage
You'd hope soo; the Guardian article Offwidth posted from a SAGE member up thread suggests that employers have been asked not to inform PHE of outbreaks under many circumstances which I presume limits the hard data available to base this on. Their own reports have a lot of appearances of the category "Other"...
> (I'd assume food service, maybe manufacturing, transport, call centre staff etc) so just going through major employers should get some proportion of needles in more effective arms than just mindlessly working down the age categories?
My worry is than any prioritisation effort is going to turn in to a public bun fight based on the lobbying and emotional firepower of different unions and that what we get is toxic division which undermines the good will need to keep the wheels on the transmission control efforts.
As Si dH says, the JCVI are working on it so we'll see what they come up with. Once vaccination has covered the more medically vulnerable and adults down to age 52 or so (I'm a fair bit younger than that) I think that it's best to use it to minimise transmission rather than protect individuals directly. But the devil is in the details. As Longsufferingropeholder notes, the vaccination is proceeding at such a pace that a bit of prioritisation here or there isn't going to make that much difference. I had hoped that keeping R<1 through progressively weakening restrictions as the limiting of transmission gradually slid from control measures to immunity would protect people throughout the process. That is not the explicit claim of policy but I hope it still goes that way.
We defined a 'covid age' at my work to remove those at most risk from face to face classroom teaching. Using such techniques in combination with role asessment that looks at say the difficulty of employees in being able to social distance at work or for vulnerable people in their care environment, and could give a numerical risk factor. I really doubt trade unions would object to such a system. It could be run independently to the selection of the vaccination efforts, based on the oldest and most vulnerable (normally in self isolation). This all could have been worked out months back. Highly vulnerable people like Jo Whilley's sister should have been vaccinated before any safer 70 year olds living at home.
> We defined a 'covid age' at my work to remove those at most risk from face to face classroom teaching. Using such techniques in combination with role asessment that looks at say the difficulty of employees in being able to social distance at work or for vulnerable people in their care environment, and could give a numerical risk factor. I really doubt trade unions would object to such a system. It could be run independently to the selection of the vaccination efforts, based on the oldest and most vulnerable (normally in self isolation). This all could have been worked out months back. Highly vulnerable people like Jo Whilley's sister should have been vaccinated before any safer 70 year olds living at home.
I would rather trust the JCVI to determine who is at most risk than see individual employers taking their own pop at it. And while I have a great deal of sympathy for someone in the situation that it seems the Whiley family is, if the JCVI or PHE have not identified someone with that particular condition as at higher risk of death than an average 70yo then they should not be vaccinated before them.
> My worry is than any prioritisation effort is going to turn in to a public bun fight based on the lobbying and emotional firepower of different unions and that what we get is toxic division which undermines the good will need to keep the wheels on the transmission control efforts.
I don't know - I feel like we've done reasonably well so far at seeing vaccination as a collective solution to a collective problem, and I've not seen much grumbling about the current approach from, say, younger folks who just want to get their shots and get out partying. So I'd optimistically hope that if the government are reasonably good at showing their working and emphasising that this is about keeping infections down so it's safe to open stuff up for everyone and not just a question of giving the selected people a magic ticket to fun-land then they might be able to manage it.
> As Si dH says, the JCVI are working on it so we'll see what they come up with.
I can't say it any better than Si, so "while I have a great deal of sympathy for someone in the situation that it seems the Whiley family is, if the JCVI or PHE have not identified someone with that particular condition as at higher risk of death than an average 70yo then they should not be vaccinated before them."
I think the JCVI got things wrong...why change now if they didn't? The bigger error though was not planning these wider issues well in advance; BAME and care home problems and extra risks were well known from the start. My top priority is that in doing this is you don't slow down the vaccination process.... provide slots early with a different assessment process and always use any remaining vaccine from no-shows before its use-by time occurs. Where post code differences occur, don't slow the fast areas, build more capacity and resolve problems in the slower areas (unless supply is so bad you need to ration).
> Sound reasoning but you're going to be in a tiny minority with that position.
Probably because no-one has properly explained the 'stick with a hard lockdown for just a bit longer, to hammer the case rate as low as possible, then we can unlock more safely, and manage the outbreaks better' approach. The government haven't even tried, as it would show their previous efforts to be utterly inadequate.
> So I'd optimistically hope that if the government are reasonably good at showing their working and emphasising that this is about keeping infections down so it's safe to open stuff up for everyone and not just a question of giving the selected people a magic ticket to fun-land then they might be able to manage it.
That's it - hospitalisations drive NHS pressure and deaths; whereas cases which don't cause hospitalisations (partly due to vaccination and partly due to a lot of cases being in the less vulnerable demographic) aren't a big issue per se, except that they make the appearance of new variants more likely. So it depends how likely new variants are to emerge / how lucky we get.
A lot depends on schools as you say. I guess we should get an initial indication in the case reporting in the second half of w/c 15 March and a pretty good idea a week later. A lot of those cases will be among parents, so people in their 30s, 40s and 50s. Not the most vulnerable (though still accounting for a lot of admissions) but not vaccinated yet either...
> > So I'd optimistically hope that if the government are reasonably good at showing their working and emphasising that this is about keeping infections down so it's safe to open stuff up for everyone and not just a question of giving the selected people a magic ticket to fun-land then they might be able to manage it.
> I hope your view is right and mine isn't.
I think it has to be large groups like "old", "vulnerable " etc that people don't really identify with individually. As soon as you get to jobs or races or something, it's just different groups special pleading. If we are all to be done by July, the effort of identifying small groups just seems a distraction anyway.
> my social media this morning is a sea of people making plans around each nominal unlocking data - the sense of urgency is gone from most people and I suspect with it their full commitment in the meantime.
I think it's positive to make tentative plans for x y z dates - it gives people something to look forward to (eg I'm looking forward to climbing outdoors again from Easter and then indoors if it's not busy from 12 April, though I suspect that date might slip). People just need to be aware that the dates might slip and so they might be disappointed.
The commitment point is important though. There will certainly be some people who will start doing things they shouldn't be doing yet. Equally, there will be some people whose resolve to follow the rules may be strengthened by the fact that 'freedom' is in sight. Others might just carry on as they are (whatever level of adherence that might be) and adopt a 'believe it when I see it' attitude. I suspect it might balance out overall but I'm not a behavioral scientist.
I wonder if the ZOE numbers are consistently understated by a factor of x. Their numbers are pretty much the same as the official case count, which won't include all symptomatic cases as not everyone gets tested. Obviously ZOE won't pick up asymptomatic cases and nor will the official case count for the most part (other than a bit through local asymptomatic testing).
Agree but the issue is a practical one - how to do define priority occupations and how do you make sure you invite the right people? I suspect GPs don't know what their patients' occupations are but they will know their age.
ZOE was running at about half the “new daily infections” estimate the ONS used to put out from their weekly random sampling updates.
> Obviously ZOE won't pick up asymptomatic cases and nor will the official case count for the most part (other than a bit through local asymptomatic testing).
A subject that’s come up on here a couple of times is about where pillar 2 falls on the spectrum between random testing and symptomatic. It’s not a simple question I think.
> The commitment point is important though. [...] I suspect it might balance out overall but I'm not a behavioral scientist.
I agree with all of that, lots of scope for competing effects in different directions. Behavioural science (*) is far from being a science yet; they let to me is not to over think what might happen but to use effectively basic marketing psychology models to guide the messaging, to measure the effects through systematic random surveys etc and to update the messaging. Given the choice of behavioural scientists sourced from academia into SPI-B or a decent marketing agency, I think I’d have been on the phone to Saatchi & Saatchi to run the public messaging campaign.
(*) a quote from The Hunt For Red October springs to mind.
> A lot depends on schools as you say. I guess we should get an initial indication in the case reporting in the second half of w/c 15 March and a pretty good idea a week later. A lot of those cases will be among parents, so people in their 30s, 40s and 50s. Not the most vulnerable (though still accounting for a lot of admissions) but not vaccinated yet either...
Not vaccinated, although a fair number in that demographic have had covid by now in the harder hit areas. We did some calcs a few weeks back, can't remember the numbers...
> I think it has to be large groups like "old", "vulnerable " etc that people don't really identify with individually. As soon as you get to jobs or races or something, it's just different groups special pleading. If we are all to be done by July, the effort of identifying small groups just seems a distraction anyway.
As Voltaire wrote “Il meglio è l'inimico del bene” - perfect is the enemy of good. Sometimes you have to pick a plan and then make it the right one.
A point that Offwidth raised that I have more time for is it feels like some of these decisions could have been made sooner; however it seems the efficacy of the vaccine against transmission (as opposed to infection) is surpassing expectations considerably and so there is cause to rethink historic plans. Vaccinating those at greatest risk of spreading the virus is a hard political sell I think when faced with support groups focused on immediate well-being of their members. Which is another reason why I think we should control R<1 for another couple of months. It’s the only way to protect everyone. The choice of who to vaccinate first is very public and contentious, but dropping controls on R implicitly makes similar choices without the scrutiny or emotion. I’d rather we were all in this together for another couple of months; I believe it to be the safest bet for a return to health - human and economic. I know at this point I’m beating a broken drum.
That group were issuing outputs a while ago that I thought were a bit weak. However if I take their results at face value, they are suggesting population average attack rates of between 19 and 35% (outside of the south west). Case counts in the 30-50 age bands are much higher than (maybe up to double, very ROM?) those in other age bands on average, 4th or 5th graph down:
So that would imply somewhere between 30 and 50% of people in those age bands might already have some immunity, of unknown duration. Possibly more in local areas that always struggle with high infection rates. I'm being very lax with numbers here but it's certainly enough to help a bit.
The Voltaire quote suits my position perfectly. I don't care if things are a bit messy around the edges as long as they stay fast. I really do believe it was possible to coordinate vaccination of more of the vulnerable at high risk of infection and likely to be a cause of further spread, especially vulnerable people in multigeneration family homes or in care and residential support homes. Outside care homes older BAME contract workers who cant afford to self isolate and have additional factors like obesity and underlying health problems are the at risk groups that arguably top that list. I really fail to see why especially vulnerable disabled groups in residential settings who either could not social distance or did not even understand the concept were not higher priority. Middle class recent pensioners tucked up safely at home were low risk of infection and spread so should have been lower priority but needed to be very careful as they were obviously high mortality risk if they caught it.
Some degree of state system protection is also very important, especially the NHS and care workers but there is a system risk in other occupations as well. It's all well and good saying all those secondary schools have gone back if they can't function properly (as a minority of teachers are so at risk they can't come to work and a larger number will soon be off sick or self isolating). The unions get accused of special interest bleating all the time but a lot of the return arangements for secondary schools look near impossible in practical terms (mass testing topping that list). Vaccination of all secondary teachers helps (inefficient as it is) as the ministry seem incapable of alternative solutions like making the high risk workplace safe enough.
Hospital deaths down 40% compared to last Wednesday. Yesterday, total reported deaths down 48% in two weeks and Mondays hospital death toll down 20% (I think). I know we shouldn’t over analyse on such a small sample etc, but sod it, I am ! Got to be the vaccine/immunity kicking in. I’ve got everything crossed.
Hope we can get the vaccine supply sorted: Van Tam admitted supply shortage.
”A fall in the number of Britons being vaccinated against COVID-19 each day is down to "supply fluctuations", England's deputy chief medical officer has told Sky News.
Professor Jonathan Van-Tam said it "will take a few months" before vaccine manufacturers are able to produce doses in a "steady routine", adding that "global supply restraints" have also hampered the UK's vaccine rollout.
He said: "There are always going to be supply fluctuations. These are new vaccines, by and large the manufacturers have not made them or anything like them before."
Supply would continue to be unpredictable as the manufacturing process is "a bit like beer-making", he said - the end product is not always the same and the yield might be different each time - so "you do get batch-size variations".
But he said the number of inoculations would pick up again.’
> He has been spending to much time in the same room as the prime minister. Just shows how your thoughts get subverted.
I think it's not too bad an analogy if you think of large commercial brewers.
I suppose it'll be a cell culture with various constructs added, rather than a yeast fermentation, but both are basically industrial scale shake and bake operations that follow detailed protocols and procedures with complex supply chains of organic compounds, and where it takes a lot of work to capture all the factors actually affecting the target yield and control of unwanted contaminants in the inputs and output. With a multi-week timescale from the start of a batch to the end, it's a long wait to find out something went wrong and to try and identify what previously unknown factor had changed and affected yield.
It's like battery manufacture or semiconductor fabrication I imagine - you don't just build a new factory and then have consistent, high yields. It can take months or years to really zero-in production at a specific site. All of which makes me all the more grateful for both the rate the AstroZeneca output has ramped at, and for the diversity of suppliers we have signed up and increasingly approved.
Edit: Although if I was a senior bod addressing the public, I wouldn't use the analogy...
> Hospital deaths down 40% compared to last Wednesday. Yesterday, total reported deaths down 48% in two weeks and Mondays hospital death toll down 20% (I think). I know we shouldn’t over analyse on such a small sample etc, but sod it, I am ! Got to be the vaccine/immunity kicking in. I’ve got everything crossed.
The data looks increasingly compelling.
The "Linear scale" plot below shows deaths for the older age bins (numbers too small to be meaningful in younger ones, and the 50_59 bin is going that way).
Ignoring 50_59 (a lot of noise), the exponential decay rate is consistently more aggressive (for negative) with increasing age. This means decay is faster in older ages in a way that accounts for absolute numbers of people
The "Log scale" plot shows the same thing, but with a log-y axis and will all age bins normalised to the same peak. Now you don't have to look at the measured rate constants to see that deaths for 80+ (the first to be vaccinated) are plummeting.
The update to D1.x also shows this - plotting the exponential rate constants for cases (X) and deaths (Y) vs age. The older curves are further down and left on the page in general, showing faster decay of both with age.
The other worrying trend in this plot - which it looks likely to feature heavily in Saturday's update - is the drift towards the right of the rate for cases. They look to be on an unavoidable course for growth; the data on this plot is heavily filtered to make it look comprehensible; the unfiltered rate constant plots suggest a couple of age bins around 40 have just returned to growth. I'll hold off showing them now as I prefer to compare like for like with the day of the week the plots are made from, given the weekend effects in the data.
This is all as a result of changes "locked in" before the unlocking roadmap was laid out which may have more behavioural effects on growth rates. So, it looks like we really are in a race against time with vaccination.
> Hope we can get the vaccine supply sorted: Van Tam admitted supply shortage.
Greater Manc is a good area to analyse: 2.8 mill population and varied demographics and the odd affluent area and higher than average Covid. I read cases is Stockport (affluent) had risen recently so that’s the 4th (out of 10) LA to have seen an increase in the last few weeks. The whole area is slowing down compared to the country average and has been for a while. Two of the areas that slowed down have speeded up again (only a bit). I wonder if we are seeing things start to reach an equilibrium with immunity balancing out the virus spreading ? As wintertree says, a race against time with the vaccine (given easing of restrictions). Whilst we are dealing with a totally negative situation, it is fascinating.
Forget to add: your weather theory is quite possible, but the recent bad spell was just a very cold snap in GM - we had more wintery weather earlier on.
Also, I’ve heard quite a few reports of teenagers and adults having boozey parties - folk are at their whits end and are going ‘f*ck it’. (no judgement attached - I’m done with that !)
I am slightly less gloomy about cases having seen today's data release. It looks at this stage like the data for this Monday (ie 23rd, two days ago) is likely to end up showing a bigger fall again compared to last Monday's, which would suggest we are just seeing a bumpy slowing of the rate of fall rather than a levelling or turning up towards a rise. To be able to say this though, I'm putting a lot of faith in the data being far more consistent than it used to be in terms of reporting lag - until recently I couldn't have made anything of two day old data at all, and this might still be too early. We'll know for sure in another couple of days so by next Saturday night the data going in to Wintertree's graphs will give a good picture of the situation I think.
I have the Covid19 symptom study app on my phone and check the case rate in Edinburgh every day. For a while it looked as if we'd reached an equilibrium position with cases oscillating around, neither rising nor falling. However things are declining now although the rate is slowing as one would expect. I wouldn't expect things to get worse again though unless some new mutation reared its ugly head.
I just checked on testing rates and no. of tests carried out. Something funny has happened in the last week in the the %age of positive tests has increased while the number of tests carried out has dropped by about 15%.
It's not clear to me if that's data by specimen date or reporting date; as the 23rd's specimen date data is still highly provisional I assume it's by reporting data.
> Tuesday's numbers of new cases are significantly lower than Monday's which breaks the cycle of periodicity we've been seeing. You'd expect the number of new cases to be higher not lower.
I agree that there's a change in the pattern there, but I find it hard to interpret the meaning of it.
> It did occur to me that anomaly we've been seeing could have been weather related.
From my mid-week plots it looks like there's a "shoulder" of level cases after the period of cold weather; this can be seen in the log plot you linked by drawing two lines of parallel gradients and applying wishful thinking (annotated screenshot below). The same sort of thing shows in my mid-week plot; I'm not sharing it as it ends with a weekend which is deceptive; we'll see how it develops when Monday's data is fed in, I might post it tomorrow.
Re: your comment on the change in test numbers - Pillar 2 testing was notably down and following a different weekly pattern during the snow week - see "Testing by Pillar" on . Positivity would rise if the snow was more of a deterrent to people who had less reason to believe themselves infected from more moderate symptoms etc, perhaps. Rank speculation. I'm not sharp enough to find a meaningful way of including this data in the plots I do other than as a standalone plot; but I'm tentatively encouraged by the next weekly peak around the 22nd - I still think the "specimen date" is the data the specimen enters the lab system, so if the snow caused postal delays this may yet all square itself off.
I think I'm sticking with my early assessment from Monday's discussion with Wicamoi of a mix of effects between weather and a partial slackening of decay rates. A lot of confusion here though.
In reply to Si dH:
> To be able to say this though, I'm putting a lot of faith in the data being far more consistent than it used to be in terms of reporting lag - until recently I couldn't have made anything of two day old data at all, and this might still be too early
In reply to Mick Taylor:
> Forget to add: your weather theory is quite possible, but the recent bad spell was just a very cold snap in GM - we had more wintery weather earlier on.
My take on it is disruption of people going for tests, and disruption of the delivery of postal test kits to households, and then from households to test labs. This causes a falsely large drop, followed by a rising signal - all of which looks very confusing until more time has passed... Snow out of area for GM can affect this if it messes with the postal side of the sample kit distribution (no idea how distributed the origins are) and the sample processing which is centralised to a few sites.
Looking at the log plots (eg as I scribbled above) I’m coming round to the idea that it’s directly temperature related. Looking at the change over 2 weeks in the demographics (plot below, single day rather than the usuals 7-day average), the biggest change was working age adults.
More windows being closed in workplaces, busses and households? If so, that’s really quite promising in terms of seasonality and the coming months.
Hope you're right, but we've been here before. Remember the puzzling rise in cases during the second English lockdown? Late November I think. I was looking for an environmental explanation for that at the time and it kind of correlated with a cold dry spell in Scotland and down through England as far as London. Then a better explanation emerged....
Indeed, but there was a very clear geographic signal that time, we’d watched the failure of lockdown spread in to London from “East of Ealing”. This time there’s no smoking gun in either the demographics or geographics, and the effect is quite correlated across the nations, with a tightly correlated onset.
Its got to be behavioural or weather (direct or indirect). I was thinking indirect with sample processing latency but now I’m leaning towards a direct component as well, basically on ventilation and open windows as well. Anything behavioural is probably masked by all this.
Your point about possible change in testing regimes is important. I'm not aware of any significant changes but there may be some cumulative effect from enhanced local testing to week out the SA variant and testing being required for travellers returning from overseas. The other thing which will throw the numbers out will be schools testing. In fact with the number being 'relatively' low now, the schools testing could have a significant impact. Imagine testing about 10m kids - even if you find 0.1% who are asymptomatic (possibly more), that would still be 10k extra cases, though if that's spread over a week it won't have as much of an impact I suppose.
It is indeed a race against time to get people vaccinated but for now I'm not massively concerned. It all depends on what impact the return of the schools will have.
You are right, the schools testing is bound to bring the numbers up a bit even with no underlying change in rates.
I followed the results of the surge testing in Southport and it's effect on local case rates. Cases rates spiked to probably double where they started in the local area after the surge testing began (it went black on the dashboard map for a few days) and then gradually reduced to the point they are back in line now with other local areas, even while testing was still high. The surge testing was a large% increase in testing in the local area.
For the whole of Sefton (of which the surge testing area is a relatively small part, maybe ROM 10% population?) you can see how the testing rates, positivity and cases evolved below (surge testing began on the 3rd or 4th of February I think, there were also record rates going for LFT locally that week after the discovery of the SA variant was announced.) On the UTLA level the effect on cases was only a bump for a few days. More obviously, you can see that although cases went up in the first week of the testing, positivity very obviously came down (compare it to the positivity during the peak of testing rates in early January when infection rates in Sefton were very high.) So perhaps we should be looking at that alongside absolute rates once schools start back.
> Examination of local data might provide some insight. I guess SW England is rarely as cold and as reluctant to ventilate as NE Scotland.
We'll, most of the places currently showing week on week rises do seem to be broadly on the eastern side of the country, with a couple of exceptions. That was where the weather came from. Locally speaking it didn't get as cold here (nr Liverpool) and there was no clear change in the cases trend.
So you might be right but it's very difficult to be sure.
> Examination of local data might provide some insight. I guess SW England is rarely as cold and as reluctant to ventilate as NE Scotland.
As Si dH says, it looks eastern. I haven’t looked in to regional data within Scotland yet, but in England it looks like the eastern regions may have blipped the hardest. We’ll know soon enough.
If this turns out to be as simple as people closing more windows due to the could, it’s a really useful finding. We already know indoor air turnover is important. First thing I’d do is remove all ventilator windows from busses, especially upstairs on school run double deckers.... It also gives me high hopes for this week given the glorious sunshine...
> My take on it is disruption of people going for tests, and disruption of the delivery of postal test kits to households, and then from households to test labs. This causes a falsely large drop, followed by a rising signal - all of which looks very confusing until more time has passed... Snow out of area for GM can affect this if it messes with the postal side of the sample kit distribution (no idea how distributed the origins are) and the sample processing which is centralised to a few sites.
It may never fully resolve itself even if the weather caused no behavioural change impacting spread (unlikely). If I understand right the probability of detection varies with time from symptoms to swabbing then changing that latency not only delays the reporting of results (which will work itself out over time) but it also alters the probability of detection (which won't drop out completely when all delayed tests are processed). I suspect that effect would be subtle compared to everything else likely going on.
> I am slightly less gloomy about cases having seen today's data release. It looks at this stage like the data for this Monday (ie 23rd, two days ago) is likely to end up showing a bigger fall again compared to last Monday's, which would suggest we are just seeing a bumpy slowing of the rate of fall rather than a levelling or turning up towards a rise. To be able to say this though, I'm putting a lot of faith in the data being far more consistent than it used to be in terms of reporting lag - until recently I couldn't have made anything of two day old data at all, and this might still be too early. We'll know for sure in another couple of days so by next Saturday night the data going in to Wintertree's graphs will give a good picture of the situation I think.
Still looking good with today's data - I think the case rates at the beginning of this week will show a notable fall vs those last week and put us closer to the previous trend. I have just had a look at a bunch of case graphs for the UTLAs in Wintertree's watch list and the majority are making improvements (turning back downwards).
Edit, graph below is today's update for England. It is looking like cases for Monday 22nd will only just scrape over 10000, vs 12500 on the 15th.
I also just noticed something that might be of interest from a data collection point of view. Today's change in reported cases for the 15th is negative (-16), meaning cases must have been reallocated to a different date or maybe somehow determined to be false. I didn't realise they were retrospectively doing this for cases, and I don't know how. Previous I thought they could only ever go up. There are quite a few other days with negative changes too although generally in single figures.
(Wt - also note there is a warning that today's hospital data update on the dashboard has been delayed by up to a week due to a technical issue. Apparently it's still available via the NHS website but might put off some of your charts this week.)
I think I saw some downwards revisions on cases when I was measuring the lag; certainly it’s common with deaths - I think they were often a case of a date being reassigned with -ve and +ve historical revisions having some balance.
Speaking of dashboard delays, as well as the cases latency really dropping, the dashboard is being updated very close to 4 pm every day, until recently it could be any time up to midnight or so. It’s really good news that someone has done a successful job of tightening it all up.
Re: the data - its looking good isn’t it. The week of much reduced growth will raises cases quite a bit when schools reopen but they’ll still hopefully be sub 6k/day and falling on March 8th.
> Sub 6k in less than 2 weeks seems optimistic - it’s falling but definitely plateauing.
I’m pretty hopeful we’ll get there - it’s almost two weeks from the leading edge of the lagged data, and I think the plateau was a temporary phenomenon that’s now in the past, with a “normal” exponential decay rate returned. We’ll see what all the plots show when I run them tomorrow night.
> Hopefully as we get into the[...] kids at school sort of age brackets
That's going to need additional regulatory approval for the vaccines. I've not seen it discussed in the UK press; Arstechnica ran a story on it from an American view a couple of weeks back . I've not followed the juvenile trials much; I think AZ started trialling in children quite recently, so they're unlikely to be complete before the autumn.
> This is updated every Wednesday and it might be interesting to play with if anyone has more free time than me.
Someone has decided to flag revised values in that spreadsheet by putting an 'r' in the cell in front of the number. As it's a Friday I'm not going to rant about this... I can probably twist my arm as far as doing some simple exponential rate constant vs traffic level plots - but how meaningful it is I don't know as both are downstream of obvious causal factors such as lockdowns and schools being turned on/off. Thanks for the link.
What is interesting at a very quick glance is the upwards creep in traffic levels immediately after a serous lockdown.
> Someone has decided to flag revised values in that spreadsheet by putting an 'r' in the cell in front of the number.
> I can probably twist my arm as far as doing some simple exponential rate constant vs traffic level plots - but how meaningful it is I don't know as both are downstream of obvious causal factors such as lockdowns and schools being turned on/off. Thanks for the link.
> What is interesting at a very quick glance is the upwards creep in traffic levels immediately after a serous lockdown.
For me, the most interesting thing is probably the potential to compare some sort of actual numbers with people's subjective views on lockdown compliance. I hear a lot of stuff like "everyone gave up on lockdown after Barnard Castle" or "this doesn't feel like a lockdown, traffic is just like normal", but how much is that actually reflected at least in the traffic levels? Is anything actually happening, and if so, is it wholesale abandonment or a slight kick-up in the level?
>> That weekend is when the deep powder snow was drifting over roads up North and when East Anglia has actual proper snow. If I'm write, the drop from red to blue will be biggest mid-week as blue is "artificially" raised on the left, and red is "artificially" lowered on the right.
> Well done for making a (conditional) prediction - let's see. I thought about the snow when you posed the question "but where from?" ... then got more interested in the signal for your asking.
Updated plot below. The drops from red to blue were indeed smaller on the following weekend (on the log plot, where size of drop is week-on-week fractional drop, and so accounts for the weekend weirdness).
The first datapoint of the new week (Monday, yellow) has a larger drop.
So, I think it probably was some combination of samples being delayed by weather and increased transmission from the inclement weather, perhaps overlaid on a background of a gradually slackening decay rate. It more for less looks like we lost a week of exponential decay to the cold.
I was very happy to see the decrease grow again with this yellow Monday bar.
Sadly Leicester isn't even the highest urban area in the East Midlands. The concern I guess is the vaccine reluctance might be higher than typical because of the high proportion of the population that are BAME.
We met our first ever ant-vaxer yesterday on our daily walk. A very chatty 86 year old catholic who was recovering from cancer and who won't be getting vaccinated as it had foetal cells in it. When challenged he said our views were just lies. Then we got the Bill Gates rant!
At least the 86 year old isn’t going to be a superspreader due to limited social contacts. He may well be going to the GP and to hospital, so that’s not great but he won’t be going to work and generally won’t be getting out and about very much other than going for a local walk. So far, the senior population has been very accepting of the vaccines overall (even with the vaccine hesitancy among some groups, overall uptake is over 95%). The simple reason is that they know they are at great risk if they catch Covid.
My concerns is more about vaccine hesitancy / anti vaxxers among the younger, working population, ie broadly the under 65s and particularly the under 50s, who may feel like they are relatively too young and immortal. These people will be going to work and socialising, so they will be the ones picking up and spreading infection.
I hope that by the time those age groupes are offered vaccines at least some of those who are vaccine hesitant will consider the facts and change their minds. Number of people who have died from Covid: over 100,000, based on 4m recorded cases, so at most 20m actual cases (assuming only about half the cases get picked up and that pretty much all of the 4m was in the second wave). Number of people who have died from having the vaccine: nil, based on 19m vaccinations to date in the UK alone. I get that Bill Gates is working hard to suppress news of vaccine related deaths but I think even he would struggle to suppress 100,000 deaths in the UK alone!
The over 60s rate is nearly three times higher the national average and the mayor says ‘it’s very worrying.’ Thats an understatement. He mentions economic deprivation, densely packed housing and the problems self isolating. The UK has many similar areas, if not worse, yet their cases aren’t rising. He’s also writing to Nick Hancock. They will know were the gaps in vaccination are, just come out with it. Time so stop messing around. We are constantly being told (correctly in my view) that local people know their own communities and their problems and are best at working out solutions (with support from the experts). Time to look closer to home for the solution as there has to be a link with low vaccine take up within their BME communities. I really do think we are fast approaching a time when Covid wards in some areas will have a very , very high % of people from BME communities.
It's possible to read too much in to the data quoted about Leicester. The article says:
"Nationally the Covid-19 rate in over 60s is 83.8 cases per 100,000 people but here it is 225.4 cases per 100,000."
It could also have said: "Nationally the Covid-19 rate across all ages is 114 cases per 100,000 people but here it is 250 cases per 100,000."
So yes the difference is worse in over 60s, but not much. No one in their 60s has had the vaccine long enough to build immunity yet anyway (unless they got the vaccine early) and allowing for lag from initial infection to case detection, it's highly likely the cases we are seeing now are from a time when very few people in their early 70s had any immunity either. So looking at the over 60s data and drawing conclusions about vaccine effectiveness is a bit out on a limb. It's possible this is down to vaccine take-up, but I wouldn't call it a strong indicator.
Leicester's rate is now starting to drop again like most other places too.
Not questioning vaccine effectiveness, but when one of the UKs most multicultural cities has such a high rate of cases in the older population, and all the evidence suggests the older BME communities have low vaccine take up, strikes me there is, most likely, a clear link.
Ed: Also worth noting it was a Leicester hospital that reported very low take up of the vaccine in BME staff the other week. Not a good example from people who should know better.
> Hopefully Wintertree also has good news to report this evening.
He has - it's been clear things were back on track for 48 hours. Be good to see the detail though.
(You can now get a less delayed look at progress with case rates if you use the dashboard by looking at the grey bars on the charts as well as the blue. The grey bars are the most recent 5 days and are considered 'incomplete' because they might still increase due to a delay in processing tests. Historically lots of tests took several days to be processed and reported so this cutoff was important. They aren't included in any of the data averages, the data on the interactive map or some of the data downloads the dashboard provides. WT cuts his own data at the same point. However, over the last month I've been tracking how much these things change and in general you can now trust the last two grey bars (ie 3 days ago and 4 days ago) to be pretty accurate. The data from 2 days ago is also useful but can be expected to increase another 20% or so.)
News Cairngorms recreated in computer game Minecraft
Scotland's largest National Park has been rendered on Minecraft. The popular building game was used to create a representation of the topography of the entire Cairngorms, as part of an educational project to encourage young people to learn about...
Press Release Berghaus to promote diversity, inclusion and mental health with new ambassadors
Product News DOLOMITE Velocissima
Digital Feature An Oral History of the First Winter Ascent of K2