Data & analysis thread, started 1 December

[NoGoodPunsLeft]

New thread!

Link to previous:

Data and analysis thread, started 12 November www.mumsnet.com/Talk/coronavirus/4077794-data-and-analysis-thread-started-12-november

@LJC1234 and @JacobReesMogadishu it’s worrying isn’t it.

@MRex

There is another thing about London, it got hit hard the first time. So most people seem to perceive the restrictions of Tier 2/3 as being protective rather than punishment. Whereas in some parts of the UK (as so frequently evidenced on mumsnet) it's clearly seen as punishment.

Generally rule-breaking starts the chain of cases increasing. Once cases in the community reach a certain level at MSOA level though, there has to be more spread, and quickly, between households, workers, university / FE/ older secondary schools - all those who can't isolate and are around so many contacts that they must meet someone with the virus. Less compliant areas get hit harder, but it spreads from specific MSOAs first. So I think it would surely be possible to devise a simple formula now that has:
weekly rolling case volume avg residents per household % working outside the home * (11 - deprivation score) = N,
Where any number above say 2000 needs urgent action. I'm amazed nobody seems to have tried to model it to identify the risk number yet.

Very interesting

I have been pondering why areas in London are so varied. Or moreover why some are so low. Although they may have gone up since I last looked.

I think any equation like that is going to be deemed a racist algorithm. It's essentially a recipe for locking up ethnically diverse inner city areas in perpetuity.

That would be really interesting to try @Mrex. Though I wonder if the relationships would really be strong enough to predict where the next big outbreak will be? For instance, Hastings is a pretty deprived town, and yet has seen very low rates throughout. And some wealthier areas (eg York) saw big student outbreaks. I'm sure you could model some of the variance based on deprivation, % in manual occupations (best proxy you'd get data on for working outside the home, though not perfect), overcrowding, population density, etc. But what would you do with it? Target areas that ought to have had bigger outbreaks than they have, as they're high risk of going next maybe? But I'm not sure any restrictions you imposed on that basis would be politically acceptable. Businesses in already deprived areas would be hard hit, and feel unfairly treated if case rates were still low.

@JellyBabiesSaveLives

Although you take the rules of your own tier with you, the only major difference between tier 2 and tier 3 is that tier 3 people aren’t supposed to meet other households in private gardens. The rest of the rules apply to businesses and organisations.

Maybe they need to make the areas really really big, with borders in very rural areas.

Pubs and restaurants offer indoor dining in Tier 2 but only takeaway in Tier 3. I suspect this makes a significant difference.

Ooh some new data coming to the dashboard soon - a heat map of cases by age and area

twitter.com/pouriaaa/status/1338221444875116546?s=21

[quote PrayingandHoping]@JacobReesMogadishu
"Yes, I know a lot of rural MPs asked for it and Boris said no it would be too confusing. In which case surely it's too confusing for londoners as well? Unless he thinks we're all thicko country bumpkins. 🤷‍♀️

And yes Gove wants it area by area. So muswell Hill could be a different tier to hampstead for example (just picking 2 places I know are close)"

When the MPs complained last time they were told that it would be under discussion for the review this week. It was a no last time.

So if it does change this time, it won't purely be because of London pressure, v much outside London pressure too[/quote]
Fair enough. But if they change it for London they will have to change it for the rest of the country. So I will expect my county rather than one blanket tier 3 level to be split into the 8 different council sections and evaluated on their own merits and if needed be in different tiers.

Movement between areas is more simple and fluid in London than in many other areas though- due to the fact that they are geographically closer and the tube makes transport easier. It is simple here in South Devon, I live in walking distance, right on the border, of South Hams and Torbay and close to Teignbridge. I know children in the South Hams who go to school in Torbay and Teignbridge, and Torbay children at school in South Hams and Teignbridge, and Teignbridge children at school in Torbay and South Hams. One reason for lots of changing areas is the grammars in Torbay attracting children from out of the area and then Torbay families deciding to go outside as they don't agree with the two tier system. I know kids from Exeter who are going to Torbay grammars. Lots of mobility with jobs as well and personally I shop in all three districts, Totnes close in South Hams and Paignton in Torbay and slightly further away Newton Abbot in Teignbridge because it has shops Totnes and Paignton don't have.

We have buses and trains as well although I have to confess we don't run to a tube. I'm sure there are places all round the country that are similar so I don't think London is unusual.

@JacobReesMogadishu 100% it needs to be nationwide!

@MRex
Interesting theory.
It could be partially underpinned by Richmond. Although adherence to rules seemingly are low many people are out and about in shops, parks, even in restaurants and cafes. But on the other hand most of R. is mid/higher class so they can afford to wfh or isolate if need be. Hence numbers are not growing as fast as non-adherence to the rules would suggest.
The problem is that I can spin this in many other ways to "prove" totally different points as well. :(

But let's do the maths
170 (top of my head) 2.5 (apprx. from datarich.gov) 80% working outside the home * 341 = N,

N=11,594,000

Are you sure you wanted to multiple these all together? % with score number for example?

% working outside the home- based on www.nomisweb.co.uk/reports/lmp/la/1946157276/report.aspx#tabempunemp
and going by the fact that managerial/office is the most likely to be able to wfh

add. info: 91% white

Could someone from a totally diff are do this as well? (who can find the numbers easily because they know the area...)

@NuttyinNotts
Why is something that is describing an are racist? The algorithm doesn't care about purple, it will just assign a number to it....
Also, if, let's say, based on this algorithm we see that Tottenham borough needs to be moved to T3 - how is that racist?

@MRex
but it spreads from specific MSOAs first.
I had this inkling feeling too so looked at the gov dashboard map view and slid the timescale to check, but within Richmond I didn't see an obvious pattern.

Oh, sorry I didn't mean to literally multiply them straight like that - it could be 10*case volume and only deprivation or only % working out of the home... Just that it feels like there should be some kind of better prediction of where problematic spread is starting rather than just when the NHS will hit strain.

@NuttyinNotts - The only point is reducing spread. Through the summer none of London would have been locked down by any metrics because cases were low, still now most inner city areas wouldn't be caught, it's some deprived areas but only because they are afflicted with cases. Identifying the next MSOA with problematic spread would bring focus to testing, instead of letting it affect more and more areas.

@TheSunIsStillShining - look at Malden Manor for a recent example, then you can also follow it back to the Wimbledon outbreak.

@MRex

Oh, sorry I didn't mean to literally multiply them straight like that - it could be 10*case volume and only deprivation or only % working out of the home... Just that it feels like there should be some kind of better prediction of where problematic spread is starting rather than just when the NHS will hit strain.

@NuttyinNotts - The only point is reducing spread. Through the summer none of London would have been locked down by any metrics because cases were low, still now most inner city areas wouldn't be caught, it's some deprived areas but only because they are afflicted with cases. Identifying the next MSOA with problematic spread would bring focus to testing, instead of letting it affect more and more areas.

I'm a math genius as it is quite obvious :) Apart from feeling off I didn't actually think one minutia about how the formula should work. I assumed you did the thinking :)

As far as I can remember all forecasting algorithms were guarded as some big secrets and no unis wanted to share. I remember prof Friston being asked to go into detail but declined.
Maybe they made it public since?

The trouble I have with the forecasting algorithms is that they are at country and at most borough level. But a spread of 5 cases in each MSOA across the borough seems to have a smaller effect than a sudden hit of 30 cases in one MSOA; the first tends to reduce nicely over a few weeks, the second always seems to rise AND have increases in neighbouring MSOAs (that may not be in the same borough). Forecasting by borough would put the first as more problematic because there are more total cases, but the second looks to be the problem in real life; and none of these algorithms seem to be using location data to highlight risks between boroughs. I'm now just dithering around looking like I'm describing how exponential growth might be an issue as though it's news, so probably I should have quit while I was behind, or at least should try explaining myself again another time.

@MRex

The trouble I have with the forecasting algorithms is that they are at country and at most borough level. But a spread of 5 cases in each MSOA across the borough seems to have a smaller effect than a sudden hit of 30 cases in one MSOA; the first tends to reduce nicely over a few weeks, the second always seems to rise AND have increases in neighbouring MSOAs (that may not be in the same borough). Forecasting by borough would put the first as more problematic because there are more total cases, but the second looks to be the problem in real life; and none of these algorithms seem to be using location data to highlight risks between boroughs. I'm now just dithering around looking like I'm describing how exponential growth might be an issue as though it's news, so probably I should have quit while I was behind, or at least should try explaining myself again another time.

If you had to take an educated guess where do you think we will be a month from now as a country as a whole, on the rise or decline

And I wont hold you to it :) Just interested

@MRex

I think I get what you are saying (or not, we'll see).
Chain and cluster are different. Cluster=group where infection is contained. Eg get's tested on day 1 of symptoms but has been almost nowhere in 7 days.
Chain = what it says on the tin

If you want to do prediction on that level than it's a minimum of the combination of the following data:

• UID for the infected person
• PErson data (age, gender, race, occupation, comorbidities at least)
• how infectious was the person (if day-1 before testing and if viral load=very high then infectious score=6)
• geo tags of movement
• movement type
• vehicles used for movement

- Vehicle score should factor in all vehicle details: no of ppl on vehicle, ventilation (assign a number on a scale based on irl validity - eg double deckers don't have openable windows=9, small bus fully opened windows=2,etc... ), driver confinment (if plexi, then infection possibility=low; if proper mas then infection possibility =low,...etc)

• events (this is basically all the places that that person went to)
• event score

this is made up by with similar logic to vehicle. School type A=19; School type B=3, wedding=55,...

• number of ppl interacted with filtered by certain criteria
• number of items touched (eg a store assistant will leave more viruses by just doing their job than me going in and touching almost only the things I actually buy. A checkout person touches everything, but briefly, so passing big viral load is unlikely, but passing to hundreds=likely)
• self assigned score on introvert/extrovert scale
• mask wearing by patient
• mask wearing by others
• time (as in what happened/happens when) - to be used for when creating chains

now thinking through what it would show and how. And how would that be useful.

1. some of this data should be available through TTR, Google, schools, tfl, venues, NHS
2. Feed in multiple known patients that are actually known to be super spreaders as well
3. Feed in x months of data
4. create a visual chain of infection with potential infection points (if unknown) and if unknown % of likeliness.
5. Train the model: if super spreaders and known chains have been

fed to the model than confirm those. Then run the model again. Results: it should come up with clusters potential chains. These can be then analysed to see spreader patterns.

This can then be used to run scenarios where if parameters change the cluster patterns and chains would also change.

omg, sorry, got carried away :)
does anyone know how o get research funding? I really would like to do this now :) I know a couple of DS guys who'd love the challenge....
oh wait...I don't have any contacts to MIx... oh crap....

DH has done a lot of work on longevity and deprevation. He has some uk maps he showed me recently of correlations by regional areas. Overlay the maps of deprevation, high co morbidities, low life span, and latterly Covid, and they match exactly.

Um wouldn't you need to factor in GDPR / public outcry / people not being honest about contacts etc ??

Can I ask just a data access issue if anyone has insight.

When an LA / PHE says cases are doubling every four days is this very localised?

There was a MSOA vulnerability index encompassing a wide range of measures produced earlier in the year, with a full csv download option

www.arcgis.com/home/item.html?id=519df61254b349c982b7eef1ba153470

Quite an alarming change in covid picture in UK over the last fortnight. Hard to know to what extent government will respond given the promised upcoming 5-day window over Christmas.

Number of councils in England with rising cases week-over-week now stands at 196. Very different picture than a fortnight ago when just 20 councils had rising cases week-over-week. Number of councils in England: 315.

Full breakdown for each council on www.covidmessenger.com

Interesting Little Owl. I'm in Leeds and although the 7 day figures continues to fall (137/100000 today), I've noticed from your emails that the figures are definitely flattening, with higher figures over the last couple of days.