UKH

One week later, here are updates to the plots from last Friday.  Methods are described in the old thread, linked below.

This week's disappointment - the ONS did not include an estimate of new cases/day in their weekly update today due to a lab processing error.

Notable points in the plots:

• Look at those cases drop - what a shame the ONS data is missing from infections
• My IFR analysis plot has changed on the left as I've included more ONS data for earlier weeks, which changes the trend line for ONS data in August (it was previously diving off towards zero which the new data bounds to be higher).
• The fraction of cases caught by test and trace appears to be improving - news to be highly welcomed
• The "jitter" in the cases data from the weekend effect is reducing.  This is a natural consequence of reduced case numbers but the jitter is also reducing in the "normalised residuals" plot which suggest it's improving more than just that would suggest; this is good news as it means a lower latency in the testing pipeline I think - not unexpected as this got worse as cases rose a few months back.

https://www.ukhillwalking.com/forums/off_belay/friday_night_covid_plotting-728...

• Cases and hospitalisations are well tipped over in to decay
• The halving time for cases is about 10 days - this is pretty good I think
• Some days deaths will tip over in to decay on my plot and others it won't.  It's coming soon I think.

Both the IFR and CFR plots show a crossing of the various lines for different lags.  That's a clear sign that things are tipping over from growth to decay, more discussion of that in [1].

[1] https://www.ukhillwalking.com/forums/off_belay/covid_-_death_rates_-_a_very_si...

No map plots this week as I used the time allotted to plotting to make a new plot, below.

Here I take the dashboard data by UTLA and age and do the deconvolution and lag-free smoothing as described in the last thread around plots 13-15.  I then rank all the UTLAs by their new tier designation and sub-rank them by their case rate on 17th November 2020.  Simple numeric thresholds (600 cases/day and 200 cases/day) would appear to correctly predict the tier level for about 92% of UTLAs.  Where it doesn't I have shown some "membership diagrams" linking UTLAs to the regional level at which the Tier is actually set - it looks to me like the outliers not predicted by the numeric threshold have been dragged up or down by other members of their region.

I am, to say the least, surprised that such a simple numeric threshold correctly predicts the tiers given the list of considerations in [1] that the government have used to decide the tiering.

I would be very happy if someone else was to do an independent version of this figure to test my methodology.

The most immediate concern I have in this figure is that there are a couple of UTLAs that look to clearly belong in Tier 3 that are being held in Tier 2 by their membership of the higher level blocks used for tiering.

Caution - There may be mistakes in the assigning of some UTLAs to Tier levels as I had to manually assign various UTLAs to regions in the government list, as I couldn't find a definitive mapping.  Please correct me if you see a mistake!

[1] https://www.gov.uk/government/speeches/returning-to-a-regional-tiered-appro...

These are not easy decisions, but they have been made according to the best clinical advice, and the criteria that we set out in the COVID-19 Winter Plan.

These are:

• case detection rates in all age groups
• case detection rates in the over 60s
• the rate at which cases are rising or falling
• positivity rate (the number of positive cases detected as a percentage of tests taken)
• pressure on the NHS

I mentioned a couple of errors on the other thread you posted this.

Have you done a similar analysis for cases rates per population. Eg a boundary of about 200/100k for Tier2/3? I'd have expected that to be closer? Funny if not.

Interesting that Cheshire west is high on that chart in Tier 2. My mum lives there so I've been following it fairly closely. They are under 200/100k now and dropping steeply so I think the allocation is fair.

I think they must have made a difficult decision over London as some areas at the Eastern end are definitely in Tier 3 territory. Not only do many of them fall at the top end of the Tier 2 group on your graph, but a few of them are still going up in the current data rather than falling. I guess the Govt just felt they had to keep London together because it's all so interconnected. Must be a sizeable risk.

Edit, being cynical, perhaps a high London hospital capacity is their excuse to keep London outside Tier 3 and simultaneously a way for them to claim they are really looking at a variety of metrics and not only infections..?

Thanks.  I've put the updated plot below.

It's hard to be closer than this, as this gets about 93% of the tier classifications right.  When I find a UTLA population dataset I will do as you suggest.  

The allocation doesn't appear to consider the trajectory of cases or of R; that's great for some places but may be an oversight for others.

Interconnection is a key - some low prevalence parts of areas near high prevalence areas are asking for lower tiers; without very hard travel boundaries that feels unwise to me. 

I'm very cynical about this right now; this is an interesting take - at a high enough level it all averages out over hospital occupancy and so the fine geographic details aren't so important...?  Down a rabbit hole.

So much positive language Wintertree!  I can't speak to the statistics but the mood music is nice to hear.  

Not looking forward to your Xmas updates : (

But:

The weather is still pretty mild so seasonal effects not showing through yet and we are just about to allow a massive amount of mixing at Christmas.

My guess is they will do what they have always done before and chuck the gains from lockdown away by easing up while the infection rate is still quite high.

Re ULTA populations. I'm not sure if this is useful and something you could paste in to a file then set up a script to get what you need, if not I think it would be a lot of manual effort.

This Wikipedia page has a table of districts by population in England. I'm 90% sure this is the same thing as LTLAs.

https://en.m.wikipedia.org/wiki/List_of_English_districts_by_population

This page provides a downloadable lookup table from LTLA to UTLA.

https://geoportal.statistics.gov.uk/datasets/lower-tier-local-authority-to-...

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

Looks like, the decision for tiers,  was based on the 19th November (not 17th?) and cases per 100,000. But it seems to support the theory it was just based on a single number, 225 for tier 2/3?

Thanks for that link - very useful.

I can imagine that absolute case levels and prevalences are almost linearly related - as I said on the other thread in the pub and should have qualified in my post here.   I assume UTLAs are somewhat sized to a certain population.  Still, I'm surprised that the absolute numbers correspond so closely.   

As you say, their figure on page 5 looks like a simple numeric threshold on cases/100 k, contrary to the five principles claimed on page 8.

Then again, page 8 claims:

Areas have then been allocated using the following principles. This includes the principle that if an area is not showing an improvement in trajectory of key metrics it remains in Tier 3:

But the "following principles" are a matrix criteria for remaining in a tier or for moving between tiers.  They do not give an initial tier.  This could be inferred from the "remain" criteria, but then the only indicators used to locate in this matrix are "prevalence" (the first indicator) and "trajectory", with no link given between "trajectory" and the 4 other indicators...  If "trajectory" was used to produce the initial tiers it had absolutely no effect over prevalence it seems.

Trajectory as a consequence of the Tier restrictions is not known until we move from Tiers to lockdown so I wonder if they have simply categorised each region according to prevalence (the first indicator) and will then use the other 4 indicators to determine "trajectory" as part of the reviews of tier levels. 

Yes, I noted that London, South East & East England were the only gross area groupings that were rising; all other areas were falling (on the BBC page).

[edit] looks like it's got worse today; more areas added to the +ve change in numbers.

Flip-book between these images to get a feel for it.  The per-UTLA data is quite noisy and I do some filtering to the case numbers; this can be a bit twitchy near the leading edge of the data (0 days).

Concerning to see a couple of London borougs rising and due to go in to T2.  Luton was rising with yesterday's release but is just dropping very slowly with today's along with 5 or so of the other high incidence T2 UTLAs.

If this trend doesn't reverse soon, the December 16th Tier review date seems a long way away...

I like that way of presenting the change with time and tier, it could be useful to spot any patterns occurring due to the different tiers in a few weeks.

Telford and Southend are both in Tier 2.

Update to my post on the other thread on Thursday-

"I've just added another column to my spreadsheet - rise/decline in live cases as a percentage of the live case count.  This is the only way to measure the success of Lockdown I as it eliminates the change in testing methodology.  Lockdown I resulted in a peak decline of ~5.5%, it's currently 2.2% decline with an increasing rise in rate of decline.  My calculated live cases are ~266k presently, about the same as a month ago. If live cases decline at the 5.5% of lockdown I (unlikely, given weaker restrictions and worse weather) we hit ~175k live cases on the 3rd December, the same number as 14th October.  If live cases decline at the current 2.2% we hit ~222k live cases on the 3rd, the same number as 21st October."

Current live cases yesterday ~226k, same as 21st October. Live case decline rate still consistently rising and up to 4.5%. This potentially results in hitting ~187k live cases on the 3rd, the same as 15th October.

Yes, the decrease is generally jaw dropping.  This is great news, and with rates dropping in the old T2, they should hopefully continue to drop in the new T2 and drop faster in the new T3.

How nice to have several different posters converging on optimistic takes on the data.

I updated my "Tiers" plot - it now shows the case plots for regions floating above their level on Nov 17th (which roughly corresponds to what the tiers were determined on.)   Doncaster is only above that level on my deconvolved method which is quite twitchy, but Medway and Newham look to be higher beyond the twitchy leading edge of the data.  There's a geographic cluster of London boroughs I've annotated as "East of Ealing" which look quite worrying given that they're about to be released in to T2.

Re: my "19:35 Sat" and "15:38 Mon" plots - these are misleading/wrong.  I was unintentionally reading prevalence data not case numbers from the gov.uk download.  (I'd like to blame the ambiguous description of the data but really I should blame myself for not cross checking the numbers with other data sources a in more detail). This means the data plotted is more or less a prevalence not a case number as the y-axis claims.  I say "more or less" as the data is separated in age bins, so when I merged the age bins the final value was biassed because the age bins are not all equally populous - the main effect seems to have been biasing measures higher when there were care home outbreaks etc and leading to a scale error in the Y-axis.

I've put an updated plot below using actual daily case rates; these no longer correlate so well with the Tiering assignments as the later were based on prevalence not actual case rates - it was ruminating on that that got me digging through the data and checking my understanding.

The big picture isn't changed much - generally the same UTLAs that were rising are still rising, and those that were level are still level.   

I'm off to fund my Dunce cap now.

This isn't cynical, because the main justification for restrictions all along has been to prevent hospitals being overwhelmed. Therefore, if hospital capacity is large in a particular area, it would be expected to be in a lower tier than somewhere with similar infections but less hospital capacity.

That makes sense, but then the effect of having a lot of “T3 level prevalence” areas in London under T2 is twofold - especially when some have seen cases growing under lockdown :

1. Those areas will see cases rise - or rise faster - under T2.  With an exponential mechanic it’s only a matter of time (*) before cases in those areas overwhelm the healthcare capacity in the surroundings 
2. Those areas of high - and under T2, likely rising - prevalence will bleed through in to the surrounding areas creating more cases there.

(*) less time I think than before the vaccination roll out could break the exponential rise.  Ergo, the sooner these areas are brought under control, the better.

Edit: it’s always possible that targeted local interventions - local public health teams contact tracing, more local enforcement, asymptomatic LFT screening etc - can stop the growth whilst under T2.  The plan for Christmas also requires low prevalance when the temporary relaxation of rules kicks in...

I think this is what you need?

https://www.nomisweb.co.uk/query/construct/summary.asp?reset=yes&mode=c...

I've managed to download the inevitable xls with data for April 2020 that looks reasonable based on a couple of checks.

This week's updates.  A few brief comments:

• Plot 1 - look at those curves drop.  Cases dropping to a slower exponential decay?  This lockdown lite has been surprisingly effective.  I hope to read studies of this eventually; but I take it as a positive for for all the control measures now in place compared to the first time around.  
• Plot 2 - having clear peaks in both lets us estimate the IFR by defining the lag as the space between the.  Dividing one peak value in to the other gives about 0.8%.  All the talk from various people about how this wave was going to be less lethal isn't really playing out much.  This drops if you use the REACT survey for estimates of infections.  I'd like to see an informed dig in to why that and ONS differ so much.
• Plot 3 - a bit redundant given the comments on plot 2.  Noticeable that all curves appear to be trending upwards though;  worrying but perhaps related to demographic shifts?
• Plot 4 - not sure this really says much.  I was kind of hoping things would be getting better as pressure eases on test and trace.  That it's not suggests the problem lies largely in core capability and/or public engagement not sheer workload pressures.  Perhaps. 
• Plot 5 - Jitter in Cases - this is really dropping with the datapoint lying much closer to the trend line.  This is good because I think the reporting lag corresponds to lag on entry to test-and-trace.

All measures are now clearly in to decay it seems including deaths.  All have halving times that at a guess are headed towards ~ 15 days - that is cases will halve every 15 days etc.  Of course, soon enough we move from lockdown 2 to the new tiers so they might get bigger (meaning it takes longer for cases etc to halve).  I'm hopeful that - at a national level (more in another post below...) they will remain halving times and not doubling times.

The CFR estimates are as inscrutable as ever, encoding the varying efficiency of the testing system at catching infections.  It's notable that the 12-day lag line is almost horizontal as they cross due to the peak in cases and deaths; this suggest to me that 12-days is a reasonable lag to pick for an accurate readout of CFR.

This now shows a normalised case number per day per 100,000 people thanks to the ONS mid-year population survey from 2019 data.

This plot has its highest number yet of places exceeding their case level from the time the tiering was decided.

As ever, the right hand side of the case curves - which is what's plotted - are twitchy to future days of data release, but the same places are cropping up in general, along with some new ones.  A lot of it looks to be London boroughs to me, and many more aren't dropping much.  Remember - the period for which I have data and which is plotted is all from before the lockdown was released, and these areas are being released in to Tier 2.  

Thew two red dots just under their tiering positions are Blackburn with Darwen and Kent.  Medway continues to soar although it looks like it may be levelling off.  3 Kent MPs voted against the tiering in the commons vote.

Thanks. I see from the ONS data that cases are falling in school children as well, so my concerns about Christmas have been eased a bit. Even so, given the easing in restrictions, I think it looks like around 1% of people could be positive at Christmas. So if you have a bubble of 10 the chance of at least 1 being positive is more than 10%. 

The Blackburn case is worrying given how much that area has had problems throughout. Looking at their data in a bit more detail it looks like their cases might have bottomed out a week ago and started to rise again. Lockdown fatigue? Worrying.

Good job keeping all this up. I think it'll get very interesting again I'm a couple of weeks as the tiering effects start to be seen.

That’s really interesting. Chances of London going into T3 after Xmas? Highest R rate at the moment - 0.9 to 1.1 is not much but that will increase post lockdown.

The big question is will T3 keep a lid on things. Initial signs are encouraging. Secondary question is what impact Xmas will have. I’m a bit more optimistic than I was a month ago but I still think we’d do well to get through winter without another lockdown.

Number of hospital admissions hasn’t reduced all that much and neither has the number of patients in hospital. We find ourselves in a precarious position and with limited headroom. Infections have fallen but only back to where we were at the start of October. A month of lockdown-lite to undo a month of growth, give or take. Not a great return on investment, so to speak.   

Wales is a perfect demonstration of how to lose control very quickly, again. I’m glad BoJo has been more sensible (never thought I’d say that!). I’m a bit concerned about the measures being prematurely weakened when they come up for review in early Feb but that’s a while away yet. 

Yes; interesting times.  I’ve a couple more plots in mind to do in the tiering.  I think the review is going to be immensely politicised.

It’s hard to see half of London avoiding T3 at this point; it all depends on how much worse things get under T2 and it’ll be another week or two before that’s really clear.  

I think we’re going to start seeing a big fall in national level hospital admissions, it’s tipped over and will follow cases.  Still almost no sign of influenza in the PHE data which is promising, but I wonder if that could spike from Christmas?

Re: Johnson; starting with his lockdown 2.0 announcement I’ve felt that he either understands the problem really well (cf March 2020) or is now being briefed by someone who does; either way he is developing and this is good I think.  Perhaps he’s learnt that not all opinions are equal when it comes to matters of science vs politics.

Just browsing some of the Covid data and I noticed that the specimen-date infections have been rising for the last 3-4 days in some regions, especially the East and London. These are the days not yet captured in the 7-day averages as there may still be more cases to add. That means the only way to see what LAs are rising from the dashboard is to look at the graphs for each one individually, so I haven't done it.

Bit worrying as I'm sure it's too early to see the effects of relaxing lockdown yet. Do you have any insight from your deconvolved data?

Just did a bit more looking around and it seems there are quite a few places in the general south and east that have turned upwards again in the last few days, which will show in the 7 day averages over the next couple of days. I wonder if this is due to people starting to relax once they knew lockdown was about to end..?

Looks like the government is going to be in a pickle if it keeps London out of Tier 3 in the next review but leaves the major northern urban areas in it....

There’s now “raw” data in the download and I’m using that for the most recent plots.  When I did last night’s update a bunch more London boroughs had passed their level from when tiering was set, ie cases were rising.  I’ll update it again tomorrow; I want to add markers for all areas where R is positive.

In reply to Blunderbuss:

Yup.

Thing is, is there much point having some areas of London T3 and others T2? The T3 people will just go to T2 to do whatever they can’t do in T3 (pubs etc)... London is a big place of course but I think T3 across the board would make a lot more sense.

As you say, the review on the 16th will be subject to various political pressures.

On my way to the Depot today I took a slight diversion to have a look at one of the streets in the shopping area in central Birmingham. Very busy, though not as rammed as it usually is due to the Xmas market, which was cancelled a while back. Got the hell out of there and went to the Depot, which was quieter than at any time I’ve seen since August. Not sure if people are staying away or going shopping...

It is a big place, but it has tube trains that shrink it massively.  You are right, an area and places in 'commuting range' need to be the same.

The situation in London has rather broken the formatting on my plot, which assumed only a few areas would be rising above the level of cases from when the data was apparently used to determine Tiering.

The plot now looks like a dogs dinner.  The inset shows the worst affected places.  This is now using the by-specimin-date data without a rolling average from the downloads so it's more current.  The data is however very noisy (weekend effect etc, and much smaller numbers than national level data) so it's filtered with a 21 point 3rd order SG filter.  This means that plots for the most recent week or so of data can change quite a lot with future data releases.   The benefit of the SG is that it doesn't introduce any lag.  I've added a plot without it below.  I present and compare both because there's always a concern that the filter is exaggerating the effects of noise leading to a false rise.  It looks to me like it's actually moderating the decay and growth over the raw data for most areas. 

I'm a bit agog at the way this is suddenly going.  This plot is for the day before lockdown was released to Tiers and the review date is 15 days from the point this plot represents, and 9 days from now.

I'm on the fence here.  

As you say, people will move between neighbouring parts of London.  To me that's reason to throw the T3 net wide, rather than to delay T3 - but I do not have the same priorities as the government.  The main driver for the government is avoiding healthcare overload, and that can be shared between neighbouring parts of London - at least for a few weeks [1].

On the other hand, the risk for individuals is higher in the higher rate areas and many people will use their local Tier information rater than numbers and graphs to judge their local risk, and could well get a false sense of comfort for being in T2, which can only accelerate the growth to the point T3 has to happen.

[1] Wish I'd found this back in March.  🎵 You can run but you can't hide from exponential growth 🎵  -  youtube.com/watch?v=bghbxemp4kQ&

I think the argument about whether different parts of London should be separated will become academic within a few days. Most areas of the capital now look to be rising, not only the east. There are a couple of obvious outbreak areas just to the south west (Kingston, Runnymede/Woking) that may well spread inwards. I forecast that by the time of the government's next review (16th?) it will be obvious the whole capital needs to be tier 3. I only hope they are already not far gone enough that tier 3 becomes inadequate. It looks like things are moving fairly fast in some places.

By the way, if you wanted to tidy up your plot, would it be easy to adapt the colour scheme? You don't really need the red/yellow for tiers 3 and 2 as they are separated left/right anyway. You could colour code by region to see themes in where was 'above the line' and then remove all the individual labels/arrows to tidy it up. Eg just have one colour for London, one for the remainder of the SE, one for each of the other UK regions. You would still have the labels on the X axis.

Parts of northern Kent are going nuts....like they've not been in lockdown (lite).

Yes, the colours were for correspondence with another plot, but I'm not using it much, so I can separate the tiers and use colour for regions or for trajectories - good suggestion.  I'm going to look at some spider-lines for several regions in T2.  Perhaps I'll colour the grey lines by the Tier level, seeing as that's what they correspond to.

It's a bit of an effort to cram as much information as possible in to a single plot without totally overloading it.  Showing the recent trajectory frees up the need for a crappy inset plot opening up some space...  (Less if Medway wasn't stretching the y-axis so...)

In the mean time, a quick variant suggested by a colleague - dots fade in intensity back in time from the leading edge to the tiering levels.  This helps visualise the different behaviours.

I thin the next step is to make a stacked line plot for the cases/100k in each of the UTLAs in each of (T2, T3) x (Falling, Rising).  I also really need to get my map view back, and persuade the site owners to let me attach a .MP4 movie of it...

Adding this I found some new bugs in this plot - the horizontal and circle data markers were separately ranked and didn't coincide with the same places.   Sorry!

I thrown the code out (it's been evolving and getting worse for a month) and made a new clean version.  I've added the region colouring you suggested and now only annotate rising UTLAs not in a coloured region which restores some clarity.

It's the same places as usual poking above the lines.

I've added some cumulative plots for the case rates broken down by region.    This is determined done by comparing case counts on Nov 17th and Dec 2nd.  Worryingly everywhere seems to be having an up-tick over the last few days, although it's notable that the number of tests being run at a national level hasn't plummeted with the number of positive results as you'd expect for a symptom driven testing regime so it's possible that part of the rising cases is an improvement to what fraction of cases testing and test-and-trace are catching.  Just a hunch really...  But if testing was previously supply limited rather than demand driven, it would make some sense.  We'll have to wait a few weeks to compare the current period to the ONS data.

More plots for (some) T3 areas

You've added most of the Liverpool region in your "Falling- London T2" plot.

I think your main graph is quite powerful now. I liked the fading effect too. Maybe you should send the output to someone in sage once a week!

Quickly looking at the latest data, I think the chances of us not going into lockdown #3 early in the new year are less than zero; especially after the xmas free for all.

On the bright side (not really, it's all terrible), Italy are soon going to re-overtake us in the number of deaths - it looks like their wave 2 is going to be worse than the initial wave. What are they doing that's even worse than how the UK is dealing with it?

Thanks.  The fading effect is making a comeback into the new, simpler code.  When I’ve got that and a time lapse movie of the maps together I’m going to try circulating it to a wider audience than UKC.

In reply to Michael Hood:

I don’t understand why the lockdown seems to have failed part way through, and with so many areas having two weeks of likely growth between the most recent data and re-tiering my hopes are fading fast for going in to the xmas feee for all with a low enough prevalance to limit the fallout.  The pre-lockdown T3 had been highly effective as had the start of lockdown it seems.    What’s changed?  Weather?  It seems to synchronous for a shift in compliance.

ZOE data isn’t showing the rebound that’s appearing in the pillar 1+2 data, just a slackening of the exponential decay.  This is odd as ZOE should lead P1/P2 by a few days.  So perhaps as I mused above, the fraction of infections being detected by P1/P2 as cases is rising, meaning that the rise in cases (in places with a small rise) doesn’t necessarily mean a rise in infections there.  A counterpoint to this is that some areas are rising and some are falling so there’s clear heterogeneity making it unlikely to be a systemic bias.  I also trust the ZOE data a bit less than I used to (which was already below ONS and React in my book).  The ONS update due late this week will be critical to understand what’s really going on.  The hospitalisation data looks like it might be moving from exponential decay to a plateau which suggests at the very least cases have levelled off.  If they’re genuinely rising everywhere that’ll show in hospitalisations soon enough.

I wonder if the shift was down to a societal “vaccine euphoria” weakening?  A failure of messaging if so as it’s clear that the Pfizer vaccine changes little for many months, potentially not until the late spring and now that date is at risk with the supply drama.

My live case data indicates a peaking in the rate of decline at 4.2% about 2-3 days ago - it's lower at weekends, but today's volume isn't really any higher than the weekend. Peak drop in live case numbers was ~10.7k on the 30th November, today it was ~2.8k. Current live case count is ~199k, the same as 16th October i.e. 2 weeks away from a lockdown if we unlocked to where we were in terms of restrictions at that time. Although most of the country is tier 3 now, cases are now much more widespread I suspect.  I suspect there's some premature slackening of behaviour due to the vaccine announcements and some 'sod it, I'm going to enjoy xmas' behaviour too.

I just did a random sample of the latest case data graphs on the dashboard for a bunch of local authority areas outside of London and the South East up to about 5th December, which although not included in the averages yet is close enough to trustworthy that you can use it when eyeballing the graphs. I wrote it all up then ukc lost my draft post My conclusion is there are still as many areas falling as rising right up to 5th Dec, but most are broadly flat. So I think your paragraph above is overly pessimistic. To be honest I think the cumulative graphs are very difficult to see the behaviours of individual areas in. Perhaps surprisingly, the behaviour of the pandemic has remained strongly regional even through lockdown.  The main problems are London, the rest of the South East (especially places bordering London all the way around it, but most obviously Kent and Essex) and the areas captured in East of England. In those areas cases are rising in most LAs. In most LAs elsewhere, the lockdown kept r≤1 throughout, with a few bumps here and there. Why is it so regional? I'm not sure - perhaps it is purely behavioural? Have people in areas who had got used to low case rates pre lockdown come out and relaxed too much as soon as the lockdown replacement system was announced?

Edit to say that the overall England total on the dashboard up to about 5th December is just about flattened off, which suggests outside the south east there is still more fall than rise. There isn't an uptick at the end in this total. I think your filtering must be doing something a bit funny in the cumulatives. Is it carrying through trends in the rate of change that occurred as falls were slackening and turning them into rises as the measured rate of change passes around zero?

Tutor group really like it. However, some of them are now confident we (Cumbria) are heading for Tier 1 on 16th. Hopefully it won't change their behaviour. 

What this plot doesn’t show is hospital occupancy which will still be elevated in Cumbria compared to the current T1 regions.  So whilst things are moving in a good direction, there isn’t the slack in the system to let up on most controls as if it goes wrong, by the time action can be taken healthcare capacity could be exhausted.

I wonder what on earth is going on there? From one of wintertree's earlier plots it looks like the problem is predominantly working age.

Is there a correlation with the trainline(s) into London, is it just insecure workers forced to keep commuting or maybe a cultural problem at Southeastern? Are they workers on the rushed brexit border? Is it just what happens where you have high prevalence of this disease, the R increases because the environment becomes contaminated or the social barriers somehow break down? Is it undiagnosed school or institutional outbreaks creating a highly connected reservoir? Something else? It's just really odd.

jk

Rotten luck - I've lost a couple of corking rants that way.

Yes; they're a bit of a disappointment.  That's mitigated by separating them based on tier and overall behaviour in the last few weeks, but it's still not great.  They are I think useful to see if anywhere is particularly driving the trend or not.  I'll try balancing the curves about the y-axis and a different sorting order before abandoning the idea.

That's always the worry with the leading edge of the data.  I think I trust it tough, the filter is responding well to what the actual data is doing, and is not mis-interpolating a rise. Below is a stacked plot with and without the filtering.  But - is it appropriate to correctly smooth this rise, or will future data change how it is interpreted?  There last two weekend lows were similar sizes, but the Monday spike is much larger this time around.  It could be that the backlog was cleaned quicker this week and it's going to drop more, or it could be that things are rising.  We need a couple more days to know.  My plot of the England data also looks like it may be rebounding from a plateau.

Monday/Tuesday is the worst time of the week to try and judge the immediate trend - Friday I think is the best.    Last week's data did not have the usual fall-off from the Monday spike which could mean changes to the latency in the system or it could mean that cases were rising against the fall-off from the spike (Which I think is weekend mail being collected and processed...?)

Understand that properly and I suppose you understand which levers to pull to fix this.  

What they are good at is mesoscale understanding of the leading edge.  The noise from the "weekend effect" on UTLA data makes it very hard to follow, and England level data is too coarse.  

Here are some falling regions (over Nov 17 to Dec 5th) with and without a filter.   I think the filter is "correct" but is misleading in that the rise its showing is real but will disappear as the week pans out and more data feeds in, due to the weekend effect.  It's still worrying that the decay has just about bottomed out before lockdown ended.    

I think the lesson here is to be extra cautious interpreting the data on a Monday/Tuesday.

I think you're right.  Although the fall does look to be bottoming out - during lockdown - which is almost as pessimistic.

Possibly, but Allerdale, Eden and Copeland had very low rates at the start of lock down. The rates in Carlisle and South Lakes were affected by two school outbreaks that resulted in 300 positive tests. If this is significant, and there was not significant spread to the general population, it might mean there is hospital capacity at the moment? Cumbria does however have an elderly population so they might want a greater safety margin.

What are we to make of this? Peston suggesting massive revisions have been made to the ONS data. Did we know that?

https://www.itv.com/news/2020-12-08/covid-did-boris-johnson-order-englands-...

Revisions have not been made to the ONS data.  

The estimates the ONS make using their data evolve every week as more data comes in.  They have done since the start and they continue to do so.

The data is what it is - the results of the random sampling surveys conducted each week.

The estimates are what they are - the most statistically likely value behind what is being sampled, as informed by the data.  Add more data with more weeks > estimates change including those for previous weeks.

The value Peston is hi lighting comes at the end of one time series - it is the very most recent point.  It was the best estimate available at the time.  Estimates at the end of an evolving time series are very twitchy - I qualify this on a lot of my posts.  They will change as more data comes in in the future, further constraining estimates of what becomes the past.

This is not revision of data  but estimates changing to reflect new data.  This is absolutely standard although not always so well explained for a general audience.  I'd expect Peston to know better though.

Disclaimer - this is my assumption about what's going on.  Perhaps there was a revision other than the addition of new data.  I do my own analysis of the weekly pilot infection survey outputs rather than whatever the process is in the Excel sheet and that shows a steep rise - as did P1+P2 testing data, REACT, ZOE, hospital admissions and deaths...

As it stands, the whole point is irrelevant as the reason for lockdown was not spitballing over true infection counts but clear and direct evidence (not estimates) that hospital admissions were rising to the point we'd run out of care facilities nationwide in less than two weeks from the lockdown date if action wasn't taken.  As it happens, with the benefit of data that didn't exist when we went in to lockdown, it looks like T3 in the north was reversing growth in cases (it takes a while to show through) and escalating most of the country to T3 may have been sufficient.  But it is prudent to make the best possible choice with the data available at the time. 

Peston says this:

No other course of action seemed sensible, given that the ONS survey on October 30 showed the incidence of coronavirus in the community in England had surged from 4.3 per 10,000 people on October 3 to 9.52 on October 17, the latest date for data then available.

My recollection of the decision to lockdown was that the briefing was focused on data sources from P1/P2 and hospitalisations, so Peston seems to be shoehorning this in.

The ONS data was not, and cannot be, the primary tool for policy making. It is too slow and not sufficiently geographically segregated. However they can use it together with the React study and Zoe to get an idea of how representative the more up to date pillar1/2 testing data is. It is this that was used to make lockdown decisions (together with hospital forecasts) and Tier 1/2/3 decisions.

  Is this some technical statistical definition of the meaning of "revise data"?

  In common parlance if a number changes, on the basis of more information becoming available or for another reason, then the number has been "revised". Hence preliminary economic estimates are adjusted and known as "revised data" when more complete data becomes available.

  Anyway, that's by the by,thanks for your explanation

Data is factual.  

Estimations are statistical inferences made from applying some model to the factual data.

Existing data hasn't changed.  

The earlier statistical inference is not revised - it stands as a reflection of the data at the time and is still there for Peston to badly screenshot into his article. 

More recent statistical inferences for that period using additional data from more recent time periods give different values.

I think it's fair to say that a more recent statistical inference revises the estimates over previous inferences. 

I would say they should be known as "revised estimates" for the sake of clarity.   Calling them "revised data" is misleading because they are not factual, they are best estimates inferred from factual data.  Hopefully the factual data is augmented with more factual data and not revised - that is none of the pre-existing factual data going in to the model changed.  Perhaps it does with economics, it shouldn't do with these ONS surveys. 

Many of the problems in this world around data and statistics arise when one person's output estimates become another person's input data without then achieving a shared understanding over the estimates including measures of uncertainty and regions of the estimates likely to be revised by future data.

  So would you say that no data except in hard sciences (and then not always) should actually be termed "data" because it is not "fact".  Most "data" in ,for example economics and probably epidemiology, is actually just an estimate of the objective "data" ie. the facts, (based on the representative actual data that can be measured)  .

Exactly.  It's a calibration check on P1/P2 data and very useful for estimating the IFR,  but it's always a couple of weeks too late for policy making.

I would say No -perhaps one of the hardcore scientists will be along to disagree with me.  In science, a measurement quantified by its uncertainty is factual information.  It's hard data.  I took a micrometer, I measured it, what I wrote down is factual.  Hopefully the uncertainty I have quantified is accurate then my factual reading is compatible with the underlying reality.

The factual data the ONS have (to oversimplify grossly) are the number of swabs they sent out, the number that were returned and the fraction of those that tested +ve.

By "facts" I don't mean that the data is full measurement of the underlying unknown, but that it is factual or measurable quantity taking from somehow sampling the underling unknown reality.

These factual measurements are used to inform statistical estimates about the underling unknown reality.  The factual measurements are used with some scientific method(s) to reach a conclusion that "We think the underlying reality is X ± Y" where Y is some uncertainty.   This forms a best estimate of the underlying reality . As more measurements come in, that estimate can change.  

The model here (ONS estimates) is not a factual measurable of the underlying unknown realty. It's the result of adjusting some model (with pre-concieved tenants) until it best fits the factual measurements taking in to account their uncertainties.  This process relies on the estimates of uncertainty being accurate.   

And that process of fitting a model to the data is very twitchy to the leading edge of the data as it develops over time.  

Another way of looking at it is that the 'data' is the input to the analysis process, and the 'estimate' is the output.

So, I maintain data was not revised.  More data was added and this revised estimates made from the data.

  That is a long winded way of saying “yes”

  How would term a final GDP number?

A long winded way of saying no. 

I think I've explained very clearly the difference between measurements and estimates, and how one is data and one is inference from data.

Irrelevant to a discussion on ONS estimates of infection rates from random sampling data.

Yes, which what I was describing.

  But very relevant to clarifying the distinction  between data and estimates, especially in the context of my previous post.

So if you understand the difference between data and inference/estimation from data, you'll hopefully accept why I disagree strongly that ONS data has been "revised", rather that the inference/estimation has changed as additional data was brought in with the passing of time.  Data was not revised, inference/estimation changed.  

Not really, because we're taking about medical/scientific data and inference, so the language of those fields is rather more relevant than that of economics.   I don't know how academic economists use terminology but if it's like their use of maths it will be rather slapdash and arbitrary, and in a way that bakes in far more assumptions than they're aware off...

  Your analysis and charting is very useful and  interesting. Like a lot of scientists and other experts you are caught (whether you acknowledge it or not) by the dilemma of whether you stick rigidly to the specific terminology and jargon of your field as used by your colleagues and which you require your students to understand, or whether, in order to make your work more accessible to a wider audience you are prepared to use commonly understood (if strictly speaking incorrect) terminology. As a journalist Peston (who I happen to think is a bit of a dick) understands that in order to reach his audience he has to use terminology as it is widely understood and usually without great loss of meaning to the general audience (as opposed to the expert).

PS. On your definition, which I assume is correct one for statistical usage, the final GDP number is not "data". It is a final "estimate". So, like I suggested, most economic data is not not actually data. I suspect the same is true pf epidemiology. But it's the terminolgy that millins of people use and understand for practical purposes.

I don't think so.  I am well aware that most people will refer to a spreadsheet of numbers as "data" or "statistics" regardless of what they are.  What I am doing is explaining why I think Peston is wrong to refer to data as having been revised.

As I said upthread:

Many of the problems in this world around data and statistics arise when one person's output estimates become another person's input data without then achieving a shared understanding over the estimates including measures of uncertainty and regions of the estimates likely to be revised by future data.​​​​​​

This recognises that different groups of people have different views of what constitutes "data" vs interpretation etc.

Peston's article is pretty crap - he doesn't cite the specific files he's showing, and he carefully crops the screenshots (photographs of a monitor FFS) to omit the certainty bounds, so we can't see if the difference between the two estimates is even significant or not...  (It probably is given the typical bounds of the ONS datasets but I'm not diving through a whole bunch of pages to find the files he doesn't correctly reference in order t find out...)

Which is why it's reprehensible in the extreme of a journalist to make such a conflation of terms.  Given the shoehorning in of a link to lockdown - despite this data not really pertaining to that decision - it almost looks deliberate...

Lol

At a very basic level, the official stats show the average number of cases for the UK flattening out at c.15k / day from 26 November. Presumably that's the base line for lockdown light. That seems consistent with c.25k / day in the last ONS survey - 'real' numbers have been consistently roughly double the level of reported cases (give or take). I suspect that's too many for contact tracing to keep a lid on things. As soon as restrictions are relaxed, case numbers and R are likely to creep up... I imagine we'll be seeing the impact of that in reported cases around this time next week.

Interesting that some areas but not others are creeping up in your graphs. It's not just London and the SE. I guess there's an interaction with previous infection and tier levels. An element of naturally acquired herd immunity in previous high infection areas? I doubt the impact is that great though. Behavioural impact of local messaging around the situation being really bad previously? Ripple effect of previously higher tier levels? Not sure there's a single explanation as apparently similar areas are showing different trends. It's a complex picture.

Re Italy - possibly due to less government support? Much as people complain about the UK government not spending enough on supporting X Y Z sectors / groups, the level of government support here has been extraordinary and as a result the (official) unemployment rate is significantly lower than it could have been. Currently 4.5-5% compared to 8% following the 2008 crisis which had a fraction of the impact on GDP compared to Covid. Forecast to rise to 7.5% next year unfortunately but that's still pretty good going considering we've got a c. 11% reduction in GDP this year. Of course more could be done and/or the support could be better targeted but fundamentally the level of support has been a lot higher compared to a lot of countries. Why? Because we could afford it from an economic / fiscal point of view.  

11 days later and the BBC are breaking the news to Londoners that maybe they need to stick to the rules if they want to avoid T3.  As far as I'm concerned it's too late for the majority of London to avoid T3 - both in terms of infection control and politically if some northern areas with lower prevalence but still high hospital occupancy are to be kept in T3.

https://www.bbc.co.uk/news/uk-england-london-55228119

Todays plot below with stacked detail plots for London (no errant Liverpool regions on them now...) for UTLAs that have risen and fallen since Nov 17th.    Grouped together, the falling regions have more recently levelled off, and this is still data from when they were released from lockdown to T2.

Edit: The Y-axes on the stacked plots are messed up as I've not adjusted them to start at zero on the new balanced shape.  A fix for another day...

Indeed and I’ve seen almost no discussion of this complexity in the media.  I don’t know if SAGE are digging in to it, I’ve not been reading their minutes lately. 

I'm sticking with what I said in lockdown 2. A more uniform approach (maybe with the few Tier 1 exceptions) would have been way better for public health (and hence the economy) and for the public to understand. They let London and some of it's suburb's off the hook for reasons I still can't understand and treated some nothern metropolitan areas a bit harshly and lied about Liverpool's change being due to mass testing. Now things are a complete mess again. The Kent tory MPs look especially deluded. All to be expected of course.. standard Boris SNAFU. If we get no deal chaos added onto this I fear for late winter.

People need to be looking more at Independent SAGE, as SAGE information has an inevitable slight government bias.

https://www.independentsage.org/

These rises were well underway before lockdown was released though.  What changed?  Something physical affecting transmission, colder weather?  Or people making the mental leap that “it can’t be so bad if we’re being released into T2 soon”?  Or widespread misunderstanding that the Pfizer vaccine was sufficient and ready to make this all go away?

The problem with public understanding strikes me as less about the measures and more about the messaging - excepting some parts of London that clearly weren’t ready for T2 and where putting them in to T2 sends a very wrong and damaging message.

I agree although as a Kentish person I know has pointed out, Kent has two very different sides, one of which is low prevalence and going to scream loudly.  An argument for more local measures perhaps.

Re: Independent SAGE; it is very good that they are so clearly qualifying what the LFT results mean, it is unbelievable that it falls on them to do this.

It did have but not any more. All LAs or high and or rising. The lowest (Tunbridge and Sevenoaks); will not be low for long, looking at the latest data. Dover is catching Thanet. There is definitely no single LA in Kent that I would place outside Tier 3 now even if they were all judged individually.

At the time we had that discussion about the Tiers on 27/11 I could support the decision they made about London, but things have moved fast since then and it was already clear by the time tiering came in on 02/12 that East London really needed to be Tier 3. It's now clear the whole of London and Essex do as well. If we wait until 16/12 we could be talking about the whole south east quarter of the country below a line from Portsmouth to Norwich via Northampton, perhaps excluding a few bits of rural Norfolk and Sussex.

The big problem with this Tiering system is already the same as one of the biggest problems of the last one - timeliness. It takes up to 2 weeks before a change is obvious in the data. If politicians acted immediately at that point we might still be ok. But politicians incentives are all wrong because fixing the problem means taking difficult actions before their constituents think the problem exists. In practice no council or MP wants to take accountability so instead the vast majority campaign against restrictions and wait until the shit has really hit the fan and it registers on the agenda of the health secretary or the prime minister. By the time they do anything it is already too late. We have got there now - Hancock was talking about London, Kent and Essex on the radio yesterday - but even he does not have the political capital/influence/whatever to put London in Tier 3. It will have to be Boris and that means waiting until the 16th.

It's satisfying but very annoying that ukc seem to 2 weeks ahead of the decision makers in working out what is going on.

https://www.wunderground.com/history/monthly/gb/southend-on-sea/EGMC/date/2...

There's a cool start and end to Nov' with a warmer breezy spell in the middle. Precipitation data is missing but the breezy spell seems to have been predominantly Atlantic winds so likely damp. There's no really marked change mid month. London City's weather looked very similar but it's on an urban heat island.

It's the pronounced difference in trajectory between superficially similar seeming areas that puzzles me. We're all exposed to the same national level messaging and I'm not aware of much at the local level (highstreet and roadside signs, the odd leaflet).

jk

I do think there is a big difference in perceived risk and therefore behaviours at local levels. When Liverpool region went in to extra restrictions at the end of October there was a big change in the number of people visible on the streets, in number of people wearing masks, in number of people taking care to social distance, etc in my area of Sefton.  It got to people because they could see the rates here were amongst the highest in the country and we were being singled out for extra measures (well, alongside the NE.) That has probably carried through to an extent. Other areas that went into Tier 3 before lockdown are mostly continuing to see good trends, although many have flattened.

Whether the opposite is true of London and Kent, who knows. I do wonder whether many people in areas who thought they were way down the list pre lockdown are just in denial about the fact the risks are now high and/or have taken the view that if lockdown wasn't enough to keep their rates low then they are no longer willing to comply with whatever follows. Anecdotally, my uncle in law (Thanet based) says the problem is that people in Medway are all idiots. I'm sure it's not that simple

Another example is Hull where the infection rate was low in the data pre lockdown but then shot up soon after lockdown began (in practice the incr.ease had occurred pre lockdown but this wasn't obvious to a casual observer.) There were local news stories with the council asking fairly desperately for more help from government to stop people dying because they couldn't see a way out of the situation. Presumably that had an effect on behaviours, because ever since the rates in Hull have been dropping fast, and continue to drop now.

Edit t say, that Hull example, along with Liverpool, is a good counterpoint to the observation I made in my previous post about local politicians and their incentives.

Been a whilst since I lived in Kent(and there has been some rejuvenation attempts since) but the medway towns definitely had a rather special reputation.

There are the various stories about superspreaders. If they are right then a few of them by chance in a particular region could probably significantly shift the patterns.

On other anecdotal info. Normally I have been shopping fairly late at night but due to various planning issues and being damned hungry went out at lunch to an out of town supermarket. It seemed pretty much the same as this time last year with all the OAPs bimbling around slowly shopping with the only difference being a mix of masks covering with various efficiency).

Interesting you say that, I'd have been a little sceptical until yesterday but I did notice the marked difference between Wetherby high street (tier 3, nearly everyone masked outdoors) and York (tier 2, almost nobody masked outdoors). The demographic is a bit different, more older folk out in Wetherby at lunchtime than in suburban York at a weekend but the difference was very marked and I hadn't seen it before.

It's amazing really that the situation nationally (or at least within tiers*) is so finely balanced that almost imperceptible ingrained attitudinal differences may now be the difference between epidemic growth and decline locally.

Hull had puzzled me for a while. As the rest of the huge northern ribbon development got progressively worse Hull held out despite having many of the same socioeconomic issues as Liverpool then bang, there's covid everywhere over a couple of weeks, not synchronised with the unis going back, not clustered around the campus as it was here in York, it was city wide (and beyond) then as you say, straight back into decline. It must just be worsening weather and fatigue/complacency coming together in a toxic mix I suppose but I'm still intrigued how they held out so far into October.

I suspect there has to be a correlation between places people stay, where families are clustered and places with stubbornly high covid hospitalisation. The frequent inter-generational mixing that brings (whatever the law has to say about it) is more dangerous than more age stratified rule breaking and it simply can't occur to the same degree in destination cities where people are removed from their families. I wonder if Hull isn't one of the places people stay to the same degree the northern mill and mining towns maybe are? I'm guessing a bit, I don't know Britain that well.

I think the quality of local leadership and their understanding of the situation differs enormously. Understanding good leadership now puts you in a stronger position on polling day has to help. I wonder if the timing of local elections is playing into this, recently elected leaders with longer left can take longer views? It'd be a subtle noisy effect but i wonder if it's real and might be unpicked. That said, it's probably just (not) understanding the problem, the health vs economy idea has never really been properly put to pasture and is still being pressed hard particularly on the right.

jk

When it comes to understanding the difference, one thing to think about is a dependance of R on prevalence.  Imagine a model where R is constant for low prevalence but has one or more step increases with prevalence.  These could be things like local test-and-trace exhaustion, loss of control and widespread infection in an institutional setting (secondary school, care home, hospital) and so on.  This is not a fixed dependance but one with statistical probabilities - it takes bad luck and high prevalence.

If all regions start with low prevalence, some will surge ahead when bad luck takes hold, others will plod on with a slower exponential rise.

If all regions start with high prevalence, it doesn't really matter as they're likely to be under strict measures eliminating much of the bad luck, and the effect(s) will be present everywhere.

If all regions start with medium prevalence and reasonable control measures, I think you could get a bifurcation of trajectories - if somewhere gets no bad luck, prevalence keep dropping to well below the critical levels where R could increase, ruling the effect out.  If they get back luck before prevalence drops below those levels, the "super-exponential" effect of prevalence increasing R kicks in.  As prevalence rises, the control measures (especially for T2 regions...) are no longer enough and it just continues on its rise.

The role of super-spreaders means that small number statistics still play a role even at the level of 100ks of people.  We saw this with loss of control in the NW and NE heralded by super spreading from specific individuals or events.

I'm very drawn to this as a model for the bifurcation of trajectories under statistical noise.  Of course, now we've switched to tiers it'll all change again.

Yup.  You have to look ahead with the lags in policy change>infection rates>cases>hospitalisations>policy change being so long.  

For today's lunch break.  The "Weekend effect" in the cases data is massive with a big lull in cases by specimen date on a Saturday and Sunday followed by a compensating spike on a Monday and to a lesser effect on a Tuesday.

This causes the data to bob up and down like a yoyo, and any attempt to measure the exponential rates over less than several cycles is highly biassed by where the lulls fall - if the last date point is a Sunday the rate prediction is biassed down; if it's a Monday it's biassed up.

Neither the moving average nor the Savitsky Golay filter I've been using deal great with this; any convolutional filter such as these is meant for use with data where the noise, the variance in the data, is random - but here if the "noise" is low on a Sunday it's always high on a Monday.  

What I've done instead is to make some code that iteratively redistributes excess cases from a Monday and a Tuesday to the weekend immediately before.  It repeatedly fits a linear form to data in a window ±3 days from (sat, sun, mon) and then alternately redistributes 60% of Monday excess or 20% of Tuesday date to the weekend.  This is done in steps informed by the excess ratios estimated from comparing raw data with a trendline, with the iterative approach hopefully giving more stability and accuracy, as the linear fit becomes less biassed by the weekend effect as it goes.  It's a bit belts and braces but that's a 5 minute job.  Perhaps a Bayesian person out there can do a proper model driven version...

The results look pretty good to me - there's a remaining split between later week days and the weekend but the scale of the variation is reduced by ~2.1x.

I then fit an exponential to the last 7 days in the raw data points and the "de-weekended" datapoint.  The doubling time of this exponential can be measured to give an idea of how bad things are, and it can be extrapolated forwards to show that visually.  This is not a prediction, just a way of understanding what's happening right now.

For all regions, the de-weekending gives a longer (better) doubling time for data as of Today.  Here are the values.  My interpretation would be that all regions are crossing over from a decay in cases to a growth, and that Wales is furthest along that process.   This hasn't eliminated all the bias from the data, and it works better in England than the other regions - so the "real" doubling times may be a big longer for Scotland and Northern Ireland.  For Wales, the trend is sufficiently large to overwhelm the weekend biassing effect I think.  

07.3 days - Wales
14.4 days - Northern Ireland
22.5 days - England
46.0 days - Scotland

So, this is the most up-to-date estimate I can make a situation just developing in the data.   

This is data up to Dec 4th, which is just 2 days after England was released from lockdown.  So, I would expect the doubling time to decrease substantially as the transmission rate goes up with the shift to T2 restrictions down south, and as their absolute numbers take over driving the national level figure against a backdrop of falls in most T3 regions. 

The patten within England is much more complicated on a regional and even more local scale, as discussed extensively up-thread.

In reply to geode:

As a new poster to these forums, would you like to share your views on the Covid situation with us all?

My post above gives the methods used in some detail.  I take the data, try and fix the weekend sampling holes in it, and then measure the exponential rate on the most recent week of data.  I put some time in to fixing the weekend sampling holes as they otherwise make the situation look even worse on some days (there is some discussion on this with Si dH upthread) and better on others.

Having only recently joined us you may have missed the accusations levelled at me by some past pop-up posters about cherry picking data to push my "agenda"; I do wonder how they'd react to me going to some effort which results in a less pessimistic outlook.   

In terms of "projections" - they are mathematical projections of the most recent 7 days of data.  I am not projecting that this is what will actually happen.  It's a way of understanding the immediate situation.

My suspicion is that in England things are going to get worse than these projection as all the infections embodied in the cases were caused before the end of lockdown, and as we are currently in a region where the doubling time is weird sort of average of the having time of falling cases in some regions and the doubling time of rising cases in others, and as the centroid of the cases shift towards the growth regions, they will drive the doubling time more.

Everything about these plots could change in the next week - we seem to be at a tipping point, and projecting exponential processes are very sensitive to noise around those.

In reply to geode:

The residual effect of the weekend in that 7-days is probably still pushing doubling times down a bit (as I noted in the post) but I also think we're in a transition from decay to growth and have't seen the "final" doubling time yet.  Then, throw in release from lockdown to T2/T3 and the way I see it, if we are to be able to release restrictions (as promised) across England for Christmas, the only chance is to impose T3 on a lot of the South East and London immediately, and to pour resources onto the ground in the highest growth regions to stem that through mass asymptotic testing, boots-on-the-ground contact tracing and heightened enforcement.  With a pessimistic hat on there's about one week left to take action that could save Christmas for London and the South East.  

Perhaps the data will develop more optimistically in the next week or so.  If we wait to find out, and it doesn't...  Well, then prevalence at a national level is right back to "break healthcare" levels just around Christmas.

Tbh I still think it's too early to forecast the growth rate as you have from the last couple of days of data. The latest weekend data we have is still lower than the weekend before, the lowest Monday data is still lower than the Monday before that. The first day on which specimen-date infections were higher than the same day in the previous week was Wednesday 3rd, and you only have data up to one day after that. Those two days of data could easily just be a noise effect. I think you are correct that we are transitioning to net growth and we can see the behaviour of the second differential in cases as it has approached zero, but I'm not convinced it's valid to assume it continues to rise from that point, or at what rate. You may be right. But it still looks to me like the rate in much of the country may just reach a roughly flat equilibrium, or it may rise slowly but not on an exponential, or it may behave as you predict. (Edit, I guess we'll find out in only a couple of days if the apparent rise continues.)

I do agree with you about the prospects in the South East and Wales and therefore, if no action were taken, I would agree a return to exponential growth was inevitable once prevalence in those areas became dominant, but that's still a few weeks off I think.

You could well be right for England and Scotland.  For Wales the trend is well established by now and it's starting to look established for Northern Ireland.

Yes; the noise is very significant to any attempt to measure the instantaneous doubling times right now - as I said to geode "Everything about these plots could change in the next week - we seem to be at a tipping point, and projecting exponential processes are very sensitive to noise around those.".   I think I should have better qualified that my 12:34 post. 

I'm sure it isn't valid - as I said up thread this is an extension of what the leading edge of the data is doing, and is not a prediction for what is going to actually happen.  It doesn't take in to account the stepping down from lockdown either.

In terms of the actual doubling time values being suggested for England and Scotland - 22.5 days and 46.0 days - these are very slow exponential rises that for the week or so aren't really that different from a plateau in respect to the noise.  So, we probably have to give it another week to know with some certainty.  Still, it's making me nervous re: England.

Equilibriums are rarely held for long in this sort of situation; and that's extrapolating from the lockdown-era behaviour to the post-lockdown one; not that T3 is soo different.  

Did you mean London?  How soon it takes over driving England level data really depends on if other areas keep falling or not.  Wales is looking worrying exponential over the last week; I've not been following the news there but something doesn't look good.

This was the best effort I could make to project ahead from the most recent 7 days of data - that's not saying that it's a very good method or appropriate enough.  But at a turning point there's no other way of doing it using more data.  We'll find out in another week if it was total garbage or not.

I just looked at the most recent few days of incomplete data up to 7th Dec and it does confirm we are in growth again ie the data on 3/4th are not just noise. I went to edit my previous post but was too slow.

Re: the South East (incl London) taking over the influence on national growth rate, I would expect that might take a little while on the assumption that very few areas elsewhere keep falling post lockdown, since most seem to have already flattened off. Some will probably start rising again too (or already are doing, slowly) but possibly at a different rate.

Fair comment re 22.5 days being a slow doubling time - it just looks bad on your graph being over a fairly long timescale. Appreciate the effort you put in!

This is where we need a Bayesian approach using statistics on the noise and on what the reporting-lag kernel looks like for each day of the week.  That could make a better prediction using the incomplete edge of the data.  I think it would need way more time than I can allot to a hobby project, as I'd be starting at page 1 of the Dummy's Guide to Bayesian Statistics.  

Yes; having looked at the data if I'd had a much shorter number for England I wouldn't have posted it, as I think that would have been quite irresponsible given the issues you raise of the the method, but 22.5 days isn't so far from from a short term plateau and as you say, there's growth in the provisional data for the most recent weekend, which likely means more growth in the actuals.

Making measurements from the leading edge of data about living things is a core professional interest for me, so it's quite useful to work outside of my immediate concerns, it helps to stop tunnel vision forming over methods etc.

Edit: Today’s data release just landed.  Not looking good.

I agree with you that the trend in England is a worry. However, I suspect your exponential projection for Scotland is based on noise rather than signal. As you point out the positive trend is very weak, and had you taken virtually any other start date a simple linear trend would probably have been negative. What's more there is no logic for the starting date you chose, because there have been no changes in lockdown measures in Scotland for weeks, so there is no particular reason to expect a recent change to the longer term declining trend. Tomorrow is the relevant date in Scotland for the start of a change, because that will see the first (albeit slight) loosening of measures for much of the population.

On your de-weekending, I'm no statistician, but the residuals seem to suggest that the weekend estimates are still too low, and that Tuesday and Wednesday are still too high. Of course there may well be a non-uniform distribution of the true number of cases across the days of the week, but even so maybe there's scope for improving the de-weekending with a couple of tweaks to your code, even without a Bayesian approach?

Finally I'd like to offer you my sincere thanks for all your many and varied efforts regarding Covid on this forum - you've been nothing short of magnificent.

All good points.  I'm fitting an exponential to the data and it comes back with a 46-day doubling time - that's very close to a plateau behaviour.  I've taken starting dates 10, 9, 8 and 7 days in to the past using today's data release for Scotland - outputs below(which still stops on the same date as yesterday's).

-7: 46.3 days - Doubling
-8: 50.2 days - Doubling
-9: 99.2 days - Doubling
-10: 114.6 days - Falling

So, the trend towards more recent data is towards faster growth - consistent with tipping over from decay to growth.   It could all be in the noise - the more I stare at the dots on the graph the more it just becomes a blur...  I've put the Scotland graph below.    You can see the weekend sampling drop is much more smeared out over the weekdays for Scotland.  This indicates to me that for this particular data (with 5 datapoints after the weekend), a measure of the doubling time will be biassed to be shorter (worse) by even the residual effects after my belts-and-braces redistribution.  As you say - the details are lost to the noise.  Big picture though - cases aren't falling there any more.  As R itself tends to have both velocity and to some degree acceleration, with policy acing somewhere between R and its velocity perhaps, this isn't a good development at all.

All good points.  Counterpoint - the change from decay to rise in England appears to have happened in the middle of the lockdown period, so changes to the behaviour don't have to correlate to lockdown changes right now.   For Scotland, there is a logical reason to pick values in the range -9 to -7 days - this is when the trend line bottoms out.  This could indicate that whatever factor reversed the effect of lockdown in England is also at work in Scotland.  7 days is pretty arbitrary for Wales however - but given the clear trend there it's also a much less critical parameter.  Pushing the fit back a few days gives a doubling time of closer to 10 days.

I'd like to give some bounds of certainty on the exponential values, but the noise is so shonky, correlated and un-gaussian (technically large number poissonian) that the basic techniques I know for propagating uncertainties through function fitting aren't valid.

I think you're right, but the residuals aren't very similar from one week to another so there's not a very sound basis to use for this, and my gut feeling is that a belts and braces approach here starts to get a bit too dependant on free parameters (i.e. the potential for accidental or deliberate bias).  I'm not sure how an actual statistician would view my efforts to date over Monday and Tuesday...  

Thanks!

General comment. Just watching Hancock’s briefing- announced mass testing of school children in London. The government didn’t  have the courage to do the right thing - stick SE England into tier 3 when it was clear cases were increasing - so they want to quickly get numbers down by the quick and easy hit of school transmission. 
My daughter, just back from London, just commented  ‘it’s bad in London coz it’s been a free for all for the past few months.’  Sticking London/SE England into tier 3 for a week or so before the 5 days of Xmas is a bit late and would go down like the proverbial in a swimming pool with lots of folk. 
edit.....

Chris Whitby: ‘it’s not inevitable that things will get considerably worse (over Xmas).’

Mick Taylor: ‘Chris, you are a space cadette who, like many of your colleagues, often appear to have no understanding of how people actually behave.’

R will increase as prevalence increases due to this calculation https://en.wikipedia.org/wiki/Infection_rate even if nothing else changes - in a nutshell I describe it as if you don't get infected by person A one of persons B-G will get you instead.  The question is, how much of an effect is it?

That's the risk of catching the disease though; R is more akin to the risk of an infected individual transmitting it.  At low prevalence, the later should be invariant of prevalence. (At high prevalence, it decreases because you can't transmit it to someone who already has it).  

Some ways high prevalence might boost R I can think off:

• If someone is simultaneously exposed to more than one infectious person and if viral load affects the probability of getting infected in a more than linear way (schools, hospitals, dense accommodation) 
• Exhaustion of PPE in healthcare or care homes
• Mistakes with infection control due to staff exhaustion in healthcare. Edit: not just mistakes but exhaustion of other resources - the recent thread from one of the forum’s paramedics is exactly the (disgraceful) sort of thing that raises the number of people someone will infect when resources are stretched and certain decisions are made.
• ???

It’s good that they’re taking extra measures at least.  I’m not convinced quarantining school kids will do it - the demographics make it look to me like it starts in working aged adults and spreads down in age from there.  But if they PCR confirm any kids found +ve in the mass testing, that puts them and their households into quarantine which may restrict other downstream contacts in the household from the household’s index case.

Oh dear oh dear.  Our local public health person on the council is strongly pushing a message basically saying “Well you can meet grandparents at Christmas, but really please please don’t”.

"By far" the fastest rise in infection rates in those areas was among 11-18-year-olds, the health secretary said.

Does anyone have any figures to give an idea of what "by far" means

And all good counterpoints as expected. Perhaps I'm being too sensitive here - in my work I am forever battling those trying to see positive trends in a time series over too short a time scale given the stochastic nature of the data. In those cases, the ones seeking the positive trend just want to exploit it for reckless personal gain and to general detriment. While your moral position is quite the opposite, it seems to me that you too are seeing a positive trend over a short period which could just be noise. I suppose I would caution you about the reputational cost of being too negative (given those who will seek to undermine you for their own personal agendas). I dare say you'll know how to deal with them in any case.

Thanks for your additional analysis on start date and doubling times - seems I was wrong, there were at least three other start dates that would give a positive trend! But a projection that flips between positive and negative with the addition of just three data points (and data points that are known to be somewhat shoogly at that) doesn't seem like a very strong case for making any kind of projection to me.

On the start date being justified by the plateau - I've always though this is circular logic. The plateau could just be noise - we know there's plenty of it around.

But having said all that you could of course be right. Maybe climate effects on viral transfer are stronger than lockdown measures, maybe lockdown fatigue has set in with the northern darkness. Who knows? I hope you will update us next week with good news!

Edited just to make clear that this post refers only to Scotland - the bad news elsewhere is more obviously true.

I hope I've qualified this suitably well that it's clear I'm not predicting based on these plots, and I share all your skepticism over looking for structure in very noisy data.  

It's a refreshing change to have the outright attacks on my character and integrity and to choose to ignore them, rather than dealing with academic peer review in some fields where you get people driven by all sorts of agendas that you have no idea about...

Critically for Scotland though, estimating a doubling time of 44.6 days is hardly running through a crowded theatre shouting "Fire" so I dare say my reputation will survive.  It is a very good point that you raise however, thank you.

It could just be noise - and it's happened before in the Covid data.  But there are two pieces of context not in the plots - the similarity with timing to the effects in parts of England and the provisional data (not shown) which already has higher counts for the most recent weekend than the last weekend in this analysis.  So, my weak and un-compelling analysis (the least worst I can do with so little data), provisional data and geographic context all come out pessimistic.  It is what it is, and we’ll know more this time next week.

So true.  This pandemic could be used to illustrate a whole book on how data is used to mislead people.  Which is why I've taken the trendline off the plot below that includes provisional data for Scotland to show the rise in weekend cases week-on-week.  Fit a trendline to this and you get a nice graph you could use to misrepresent cases as falling dramatically.  

I can update these tomorrow night but then there's not much point returning to it until next Wednesday or so, as the tail end of the mostly non-provisional data is about to be salted by weekend counts that can't be recovered until the Monday data is out of the provisional zone.  Let's indeed hope it's a happy news update.

Again good points, and you were indeed very clear that you were not making a prediction. That the apparent upswing in cases is simultaneous in Scotland and England is interesting, but slightly perplexing. However, I've just been having a look at some data on test positivity rate in Scotland. It's very variable throughout the country, but as a national average there appears also to be a small upswing in test positivity at the same time as the upswing in cases - more support for your depressing projection. And tomorrow more than half the population of Scotland is due to move from level 4 to level 3.

To be fair that's basically what Whitty has been saying too. Last briefing he even made some dark comment about "if you want them to still be around next Christmas." He has been very keen to influence people to be sensible. He was also very negative sounding about reviewing the restrictions for London next week, without being able to be expliciy about it - writing ; wall.

The most notable thing about this briefing was what Rob has pointed out - Hancock claiming that the rise in London was driven by 11-18s and students and that in older adults the infection rates are "flat." I'd be very surprised if that was true. 

Thanks; I should really force myself to watch the briefings.

Yes.  I'm not convinced.  I seem to recall Hancock proclaiming early on that there was no evidence of transmission within schools.  He needs to be carful with all his competing comments that apparently suit the moment - he might undermine the teacher's support for full in-person teaching if he keeps it up...

I can’t find cases by age. However, they will easily be able to get data on ‘children tested positive’ because that’s when they close that class bubble.  Manchester  Evening News used to publish a list with ‘isolating bubbles’.

The cynic in me thinks they do not want to put that area into tier 3 so they are going for the school method in the hope it will peg case numbers down. 

I'm not convinced either. This was in a local paper a couple of days ago

Nearly half of new cases in Essex's worst-hit area can be tracked back to schools, new data suggests.

Those between the ages of 10 and 18 are being told to stringently follow social distancing guidelines ahead of tomorrow's (Wednesday, December 2) lockdown ease.

From 12.01am tomorrow, the entirety of Essex will revert back to a Tier 2 lockdown, banning indoor household mixing and limiting outdoor gatherings to six people.

However, experts are warning teenagers from Essex's Covid hotspots - especially Basildon, Essex - to follow the rules to the letter to try and curb a worrying spike in cases.

Basildon currently has the highest rate of infections in Essex, and new data has shown that there has been a clear rise in case in the 10-18-years-old age range.

The data suggests that 40 per cent of new cases in Basildon can be tracked to schools.

However, it is important to note that those that have tested positive, and included in school figures, may have been infected outside of education environments.

but I think there are only 8 secondary schools in Basildon? You could have inserted 85% for Carlisle a couple of weeks ago, 6 schools, but that didn't spread to the wider community.  

edit mick I think you a probably correct, if it hundreds of cases in a few schools. Hancock might have a point if it is thousands of cases in hundreds of schools.

Government is desperate not to put London into Tier 3 imo and is spinning any old reason for them not to....theres a lot of cynicism in West Yorkshire and probably elsewhere in the north. 

Looking at this, up to Dec 3rd, there is no evidence to say schools are an issue. 

https://www.essex.gov.uk/local-outbreak-control-plan/data-cases-in-essex-by...

I very much doubt if anything different was happening in schools last week.

Well of course the schools will be off during the Xmas break so the most significant cause of spread will be eliminated so we can confidently continue in tier 2 knowing that these measures will be sufficient to control the spread of infection.

You heard it here first 😁 (from someone who lives in the north)

Here's the latest England UTLA status plot, a stacked plot of rising London UTLAs and a couple of degogrpahic breakdowns for those rising UTLAs.

I'm happy with the last plot - it's the raw data put through my de-weekending algorithm.  It doesn't look awfully jagged like the raw data, and it doesn't look horribly over-filtered like the results of a moving average or an SG filter sufficient to remove the weekend effects do.  I think this is a good way of viewing the data.  

The image plots are absolute case numbers not prevalences.  It looks like there has been a rise in school ages in the last week of the data.  This appears to rise in parallel with the age range 30-55   This is in contrast to the earlier rise in November which seemed centred around ages 25-35.  My interpretation would be that there's been a demographic shift from younger workers to older workers and their school children.  Prevalence has been high in working aged adults before without being high in the schools; so I'm not sure schools are the core of the problem - although more activities are moving indoors with the cold/rain etc. What worries me is the apparent recent rise in the ages 40-55; if that continues to develop upwards in age and on with time, it's very bad news for hospitalisation rates.  The bimodal nature of the current demographic distribution means any simplifying analyses need careful thought - mean age of infection is dropping currently for example, and the rates > 65 are not changing much so if either of those are used as barometers of trouble, they're not firing of warning signs.  But the full plot is absolute setting my warning signs off.

Thanks

"What worries me is the apparent recent rise in the ages 40-55; if that continues to develop upwards in age and on with time, it's very bad news for hospitalisation rates. "

How many Christmas bubbles are going to contain half a dozen school kids, four parents in their 40's and grandparents in 70's? On that thought, is it going to be a week or so before the US can see the effect of Thanksgiving?

Good insight, thanks.

Unfortunately, Whitty doesnt seem to have the balls to contradict his masters when they talk nonsense. I guess he thinks its preferable for him to stay in a role and try to influence (rather than get sacked for saying "I'm sorry, PM, but you're talking fantastical bollocks again"), but, as we know, this government wont listen to experts.

De-weekending, what a great word!

Couple of points on those graphs, which look very scary. I agree that cases will go up again but is it reasonable that they would do so at a faster rate than they did back in October? I'm looking at the English data in particular. After all, we now have heavier tiering. Admittedly it's colder / darker but I don't suppose Dec vs Oct is going to be that much different. Of course Xmas is a wildcard.

The other point is, as you say, we are at a tipping point so any projections will be massively affected by noise / small adjustments. Once we're actually on the upward curve again, the future should become clearer.

Your knowledge of stats is vastly superior to mine (I can't remember much of my maths A level...) so all I can do is look at the high level picture. Which isn't great... As you say, your graphs are ever evolving but the trend is clear. Covid isn't going away any time soon. 

Odds on another lockdown light for 3-4 weeks starting in early Jan? 50-50 I reckon.

Agreed. I’ve seen a lot of films that could do with that treatment.

On the other hand, if he does, how long is he going to last in his position?  In his position he can at least tell the PM things behind closed doors.  Rock and a hard place.  Government relationships with scientific advisors have been getting worse for two decades.  It's close to rock bottom now I think.   This government seems to consider scientifically based predictions as just another equally valid opinion.  Previous ones at least knew they were following politics, not science...

I reckon odds of  1/2. The government will be heavily influenced by what goes on in SE England. Hospitalisation here will already be on the rise and Xmas will push them up even more. LightLockDown from mid January.

On a brighter note, I predict that by mid Feb, many vulnerable people will have been vaccinated, Oxford vaccine will have been in use, better weather - no more lockdowns and easing of tiers. I think a sense of normality will arrive quicker than many people think. 

It's a standard requirement of civil servants to not get involved with public political arguments with the government. The only way he can contradict Boris directly on anything important is to resign immediately afterwards. I never cease to be amazed how many people don't realise this and see the CMO and CSA as independent.

It almost looks worth running a series of fits only through matching days of the week (or the average of small clusters Sat+Sun, Mon+Tue, Others). Introduces a bit of lag but it'd be interesting to see how the extrapolated tails wag about relative to each other over time.

jk

I wonder if the increase in 10-20 age bracket is really increased prevalence or detection, we must be well into the pre-xmas screening of students. 18-20 would capture first year students, likely in halls and likely the student group at most risk of catching covid (facilities shared among 10+ people unlike 2nd/3rd year house-shares). We'd expect extensive routine testing in this period to be finding mild/asymptomatic cases in this group. As ever it's a real shame that group isn't split at 18.

The matching blip 25-30 years above does make it look like it might be school age families but if that's the case, why aren't we finding more asymptomatic students.

jk

BBC says detections in student populations are very low. 81% of people coming into Guernsey last weekend (mostly students) were tested and no cases were found which tallies up too.

That's what I meant by this:

"I guess he thinks its preferable for him to stay in a role and try to influence"

I see it as a subtly different angle - does he think it is better for us the public for him to stay in role and try to influence the cabinet vs the alternative?  I can't imagine that staying in role is particularly good for him vs basically doing anything else.  Fishing, for example.  I think I'd rather take up fishing than be this government's CSO or CMO.  As much as I'd like him to give us, the public, a frank download of his views this would probably resist in someone less capable and far less wiling to talk bluntly behind closed doors coming in to role.

Edit: Although that could be exactly what you mean - when I read then re-read your message I interpreted "preferable for him" in two very different ways.

Yes; I generally limit the plotting outputs now to a Friday to be consistent.  I'll update them this Friday then park them until next Friday.  A lot can change in a week though...

As Toerag said, these are coming back with incredibly low positivity.  After the mistakes from the start of term, the undergrads have been demonised and rigidly instructed quite a lot - and it's working.   

If I was UK King of Statistics I would haul every agency in to my court and tell them nobody is going home until they agree on a unified set of age brackets for reporting.  There would be a string hint that 18 is a notable age...

Perhaps it is reasonable - as this rise started during lockdown 2.0 - after a really quick drop earlier in the lockdown (in part driven by old-T3 restrictions up North before the lockdown), and many areas were released from the apparently ineffective (for them) lockdown in to new-T2.

I'm beyond guessing now.  We'll see how Christmas goes.  With all the brouhaha made love the Pfizer vaccine another "lockdown" (lite) is going to be a very hard sell for the government.  

That's what I meant; preferable for the nation, not for him personally. I'd have put a comma after 'him', if I'd meant the latter...

Well that's good news and bad. Means the bump in prevalence <20's is very likely real and great news that the students aren't taking the covid. Presumably Guernsey is using PCR screening at the border?

jk

Can we call it a trend yet given the numbers today.

I think so.  There is a correspondingly lagged rise in hospital admissions, and the fall in deaths is running out of steam as we'd expect if it was heading towards a more lagged rise.  The ONS did not release a daily infection count in today's weekly update.  ZOE is showing a fall but then again it's just ZOE and is increasingly falling from my good books.  The MRC nowcast [1] out this evening has overall infections rising and is forecasting rising deaths.

[1] https://www.mrc-bsu.cam.ac.uk/nowcasting-and-forecasting-11th-december-2020...

The total number of patients in hospital on the government dashboard is rising - and there isn't much headroom left.

Any idea how much is being driven by london?

I haven’t got regional grouping in my plots. If you go  into the MRC link above they do regions; I’d say the growth is - very crudely - 50/50 between London and the rest of The South.  East Midlands is holding at a high level.

Just noticed on the dashboard today the numbers in English hospitals over the last few weeks have been revised upwards by over 1k....

It's straightforward to see regional turning points from the dashboard cases data only about 2 days behind, as you can see how many cases are being added each day and you know the incomplete days can only go up, not down.

London, South East and East of England (mainly Lincolnshire) are in bad shape, going up sharply.  The driving areas of the south east outside London are Kent and Essex. East Mids is starting to rise slowly.  South West as a whole is now flat, along with all the other regions except Yorkshire/Humber, which is still in decline.

Obviously the flat regions all of have areas of slight up and down in them. 

The overall picture gets worse with every day's new data at the moment. I'm willing the North West to stay flat until Christmas but not that hopeful.

The trends over time in the MRC model don't reflect the trends in cases data seen through the lockdown period. Do you know if they have any explanation for that?

I just read what they say on the website about their method and data sources. It sounds like an interesting method but I can only assume their frequency of anti body testing on a regional basis is too low for them to pick up trends with time in a meaningful way, or that their modelling method is messing too much with the data. The differences from the evidential testing data (way beyond any error you would assign - basic trends rather than magnitudes) are large enough that I can't really put any trust in the output, which is a shame as it sounds like an interesting way of doing things.

Yes; it's an oddball fusion of different data sources.  I think it's more accurate for the past than for the present, as then deaths data feed in to it.  Likewise the antibody testing will tune for the time-varying fraction of infections detected as cases, but that only starts to work for more than a few weeks in to the past.  

I think you’re a good month ahead with your mid Feb prediction. I suspect vaccine rollout will take a while for various reasons. We could really do with another sunny, warm spring though and not just for climbing!

No deal Brexit meets Covid lockdown.  That'll be great.  I really hope they've been thinking about how to feed and provide toilets for all the stuck lorry drivers they are forecasting without them getting into situations where Covid could spread.