Coronavirus - Data Analysis Thread

[Elysium]

A number of people have suggested we should have a separate thread for analysis of the data around the pandemic. The idea being that this should be factual and free from arguments, conspiracy theories and mask wars.

I can't see it lasting, but I am willing to give it a go.

To kick us off. Two charts to put 2020 all cause deaths into perspective:

1. A heat map of ONS all cause deaths for 2010-2019, adjusted for ONS population estimates with 2019 as the benchmark. The 2020 is actual:



2. The same data presented as a graph



The Spring outbreak is clearly much more significant than the autumn one.

Edited by Elysium on Sunday 20th December 14:53

[Elysium]

This graph divides all cause deaths into those 'due to' COVID and those where COVID is not noted as a cause of death.

Data is from ONS and population adjusted to the 2019 population. The dotted lines are the 5 year minimum, average and maximum and the green line is the point at which 2020 becomes 'statistically significant', which is 2 x standard deviations from the average:



This shows statistically significant non-COVID excess deaths in the spring outbreak, but in the Autumn non-COVID deaths are below the 5 year minimum.

[Elysium]

I also wanted to share this simple graph, which shows COVID deaths and Excess deaths (variance from the 5 year average) throughout 2020:



Spring excess deaths are approximately 5 times higher than Autumn, which I think provides some context for the relative scale of each 'wave'.

Edited by Elysium on Sunday 20th December 15:05

[Elysium]

This graph shows the 7 day average of cases for London and Wales. I have identified the time periods where the Welsh Firebreak and English Lockdown could have had an impact on transmission.

Data here: https://coronavirus.data.gov.uk/details/cases

Both show a dip during tighter restrictions with a rapid rebound. London has rebounded harder, but was growing faster prior to restrictions. The dip at the end is reporting lag:

Edited by Elysium on Sunday 20th December 15:28

[Elysium]

This graph helps illustrate the potential hidden scale of the pandemic.

Cases, admissions and deaths from here: https://coronavirus.data.gov.uk/

The infection curve is calculated from daily deaths assuming a constant IFR of 0.6% and a time from infection to death of 21 days:



I believe this shows the vast scale of infections that was probable in the spring, but that we were completely unable to detect. And also that, despite massive increases in testing and case numbers we may still be missing a great many cases in the autumn.

Admissions and are deaths barely visible at this scale.

Edited by Elysium on Monday 21st December 23:09

[Elysium]

oyster said:

Elysium said:

This graph helps illustrate the hidden scale of the pandemic.

Cases, admissions and deaths from here: https://coronavirus.data.gov.uk/

The infection curve is calculated from daily deaths assuming a constant IFR of 0.6% and a time from infection to death of 21 days:



This shows the vast scale of infections that was likely in the spring, but that we were completely unable to detect. It also shows that, despite massive increases in testing and case numbers we must still be missing a great many cases in the autumn.

Admissions and are deaths barely visible at this scale.

What’s the sum total of the area under the infections graph?
We know the sum under the cases graph is circa 2m.

Extrapolated out, that graph suggests herd immunity isn’t so far away.

The Govt site currently reports 67,401 deaths

So the area under the curve will be 67,401/.006 = 11.23 million, which is 19% of the 2019 England and Wales population.

[Elysium]

Some other interesting data sites:

1. A COVID 'tier' visualiser, which is now somewhat overtaken by events.

https://coviddatashare.s3-eu-west-1.amazonaws.com/...

2. This website lets you display a number of data sets by area:

https://victimofmaths.shinyapps.io/COVID_LA_Plots/

3. Swedens live numbers:

https://c19.se/

4. A great site to compare latest info by country:

https://datagraver.com/corona/#/

[Elysium]

I have no idea how to share the data set behind these graphs, but it is all taken directly from here:

https://coronavirus.data.gov.uk/

My cases per 100k graph compares positive cases by specimen date to tests processed two days later.

1. Tests and Cases. Although cases have been increasing, we have also been testing more, topping 440k tests processed in a day for the first time. This suggests to me that we have reversed the lockdown 'dip', so what happens next will be interesting:



2. Cases per 100k tests, admissions and deaths. Cases and admissions now showing an uptick. Deaths will follow:







3. Overlay of the above and corresponding growth rates:

[Elysium]

Well spotted. This is essentially the same curve as my 'cases per 100k tests' graph.

Unfortunately its only for England at the moment.

[Elysium]

Otispunkmeyer said:

Has anyone seen any US thanks giving data? Just seen anecdotes about how the prognosis was more more more death and despair if people meet up for Thanksgiving. But they did and the data suggests otherwise...

US and EU27 fatalities per million from the datagraver website:

[Elysium]

Try this one if you prefer:

https://ourworldindata.org/coronavirus-data-explor...

[Elysium]

bigandclever said:

coanda said:

bigandclever said:

That’s not data, it’s a terrible, unintelligible chart.

Let's see your attempt at answering Otis' question then.

It’s unreadable. Maybe it was great on the poster’s screen, but by the time it gets to mine it’s illegible.

eg

You have clicked on it once to open it in thumbsnap. Click on it again to make it high res. It’s how the image hosting here works unfortunately. The link to the website that produced it is earlier in the thread.

[Elysium]

[redacted]

[Elysium]

Country comparisons are difficult, but I think we need to do more to understand why east asian countries are experiencing a very different pandemic to Europe and the USA:

[Elysium]

When comparing countries Sweden and UK normally come up because the timing of their epidemic curves is similar and they have different lockdown policies. People also argue that Sweden should be compared to neighbouring states.

This is from Euromomo, where you can choose countries to compare z-scores and zoom in on a particular timescale. In this case all of 2020:

https://www.euromomo.eu/graphs-and-maps/#



UK lockdown was wk 13, so if we assume 3 weeks from infection to death the impacts should be seen wk 16.

This comparison between Sweden and neighbouring states suggests they faced difficult levels of infection between wks 13 and 16.

[Elysium]

ONS data published today:

https://www.ons.gov.uk/peoplepopulationandcommunit...

This shows ONS all cause deaths with those 'due to' COVID highlighted in orange. The dotted lines are the average, max and min for 2015-2019 adjusted for the 2019 population. The solid green line is the limit of statistical significance or 2 x standard deviations from the average.

I think it is interesting that we had sigificant non-COVID excess death in the spring, but non-COVID deaths are now below the 5 year minimum. This requires further explanation.



This second graph uses the same data set, to show 'due to COVID' deaths and excess deaths. I think this gives a good idea of the relative scale of the spring and autumn outbreaks:

[Elysium]

survivalist said:

Otispunkmeyer said:

Elysium said:

This graph shows the 7 day average of cases for London and Wales. I have identified the time periods where the Welsh Firebreak and English Lockdown could have had an impact on transmission.

Data here: https://coronavirus.data.gov.uk/details/cases

Both show a dip during tighter restrictions with a rapid rebound. London has rebounded harder, but was growing faster prior to restrictions. The dip at the end is reporting lag:

Edited by Elysium on Sunday 20th December 15:28

Like I replied in the other thread. Looks like lockdown does have an effect. It is just short lived and doesn't effect the eventual outcome given what looks like a quick "reversion to the mean" once the restrictions stop.

So perhaps there is a use for them. But the expectations should be that it's not going to make the virus go away. It will come straight back again the moment you emerge. Obviously we cannot lockdown ad infinitum, but perhaps they can be used to buy some time for the services.

But then we are not seeing anything like as bad as it was in spring, so are they actually being genuinely useful?

Do they publish the number of tests that are carried out on a daily basis by region? If so it would be interested if this was mapped against this graph. Just wondering if, when locked down, fewer people would have a test? After all, if you're at home with no plans to go out for a couple of weeks, what's the point in getting tested anyway?

This shows a couple of dips in testing, but only the second of these coincides with the period where the second English lockdown would have an impact on cases:

[Elysium]

It’s effectively the ONS excess deaths data, but I think this is a great ‘Florence Nightingale’ graph shared by Carl Heneghan:



https://www.cebm.net/covid-19/covid-19-florence-ni...

[Elysium]

So far

Just seen that the Gov staging site now provides positivity data at local authority level for England:

https://coronavirus.data.gov.uk/details/testing

Hastings, which is caught up in the 'Southern Surge' is hitting 15%



However, Liverpool and Nottingham were 19.9% and 32.5% respectively when they peaked:





Interesting that positivity remained steady during Liverpools mass testing spike.

[Elysium]

WatchfulEye said:

Elysium said:

So far

Interesting that positivity remained steady during Liverpools mass testing spike.

Perhaps, I'm misunderstanding you. But this was one of the unexpected findings of the mass testing programme in Liverpool - it found surprisingly few additional cases.

As your graph shows, as the mass testing program ramped up, the percentage of positive results plummeted - because the large increase in testing was not accompanied by a large increase in positive results. Before mass testing started, there were 13.7k tests per week, with a positivity rate of 11.4% (1560 new cases per week). At peak of mass testing a week later, there were 46.9k tests per week with a positivity rate of 3.2% (1500 cases per week).

One potential explanation which was raised locally was that there was poor engagement with mass testing (which was optional) by people in the highest risk population groups. To somewhat oversimplify, the more precautions people were taking (and therefore the less likely they were to contract it), the more likely they were to agree to testing.

That is exactly what I was drawing attention to.

Interesting that we see something similar in the latest data for Havering:

Edited by Elysium on Thursday 24th December 14:43