Urban75 Home About Offline BrixtonBuzz Contact

Coronavirus in the UK - news, lockdown and discussion

Todays numbers make the recent plateau look slightly more like a slow increase than a plateau, but that was sort of expected hence the caveats I added to my waffle when speaking of plateaus in recent days.

So long as people talk about that daily data, it is somewhat inevitable that I will feel like posting my graph several times a week, recently most usually to put large numbers in context it seems.

View attachment 240414
Which is very useful.
Thanks for what you do.
 
I think it's probably best not to get to hung up on the daily numbers just because it's bad for you. The broader picture remains one of cases going down in parts of the country, and slowing or plateauing elsewhere, so things appear to be going in the right direction - maybe the higher number of deaths in the daily numbers today is an outlier, or some late reporting, or whatever. Maybe they're an indication of a bigger problem, but it's likely not to be.
I'd agree that the daily figures certainly don't lift the spirits, but they are published by Gov and to completely overlook those, frankly desperate, numbers seems almost disrespectful to all of those affected by the deaths.

That said, I'm fully aware that I'm able to do the easy task of posting up the daily toll in the knowledge that elbows will do the much harder task of putting them in context.
 
Daily figures don't give any kind of precision - you have to wait days or weeks for that. But, looked at as part of an ongoing pattern they do give a clue of where we might be now rather than a week or two ago. Of course it's better to look back with hindsight to get an accurate picture of where we were, but with a rapidly changing situation I like to have an idea (without certainty) where we are. Also, I'm impatient to know what's going on.

If you want to be pedantic, all top line national figures are largely meaningless as they disguise significant variations in the regional and demographic pictures.
The deaths reported each day often vary, literally, by a factor of 2 or 3 from one day to the next.
 
I know most of my focus is on daily deaths per date of death rather than by day of reporting, but I think if the 7 day rolling average of daily announced deaths by date of announcement is used instead, a reasonably useful and not too misleading picture is also generated.

Cheers for the thanks by the way. If I went the extra mile to produce graphics that were not open to misinterpretation, then my daily deaths graph should really have an area near the end that is colour coded to indicate that numbers over that date range shouldn't be taken seriously because the data is still catching up in large ways for those dates. But I just had an initial attempt at this and my choice of colour leads to a quite disgusting result. But I'm sticking it here anyway just for the benefit of anyone who might not have picked up on that aspect of my graphs yet.


Screenshot 2020-11-25 at 18.42.43.png
 
I'd agree that the daily figures certainly don't lift the spirits, but they are published by Gov and to completely overlook those, frankly desperate, numbers seems almost disrespectful to all of those affected by the deaths.
I've more or less ignored the daily figures since April, and really don't understand how that could seem disrespectful. It's data that only really makes sense or is any use in context. Posted to make a blunt political point or similar - as many are doing this afternoon - is closer to disrespectful IMO.
 
I've more or less ignored the daily figures since April, and really don't understand how that could seem disrespectful. It's data that only really makes sense or is any use in context. Posted to make a blunt political point or similar - as many are doing this afternoon - is closer to disrespectful IMO.
Really don't think it's making an overtly political point to express anger or profound disappointment/sadness that the death of 696 of our compatriots has been announced today. Although the fact that other political decisions might have saved a number of them from dying is a salient point.
 
I guess if there are fewer cases the case fatality will lessen as there can be less pressure in the hospitals and therefore better care given to each patient, probably more can be admitted to hospital at an earlier stage of the illness in the first place too.
 
I had to use daily deaths a lot for the first wave because the testing system barely even captured the tip of the iceberg and there was an important period where I had access to approximately no hospital data at all. Far more measures of how things are going are available this time, too many for me to make a serious attempt to cover them every day, I just pick and choose, often based on what other people are talking about.

Sometimes it has been necessary to make blunt political points using deaths data because otherwise the chance of having the right measures at the right time and getting across the seriousness of the situation and the extent of governmental failure was reduced.

The daily figures, no matter of what quality or quality of interpretation, have also formed part of a thin strand that ended up in many peoples daily lockdown routine earlier in the year, eg via the daily press briefings, and in a situation like this I can see why people still find it tempting to follow along.

I've got daily estimates for things like people in England admitted to hospital with Covid-19 who live in care homes, and a crude estimate of daily hospital-acquired covid infections. So far I havent much felt like adding them to the constant stream of info.
 
The deaths reported each day often vary, literally, by a factor of 2 or 3 from one day to the next.
Yes they do. And the figures announced on Sunday and Monday are always artificially low, the figures on Tuesday and to a lesser extent Wednesday are artificially high. But if you follow the figures and take all this into account they do give an indication of what's going on. The UK Covid dashboard gives figures by date reported and sample date/date of death if you want to compare. It's not as meaningless as you make out, if not as accurate as looking back a week or two later, taking into account community surveys as well.
 
Really don't think it's making an overtly political point to express anger or profound disappointment/sadness that the death of 696 of our compatriots has been announced today. Although the fact that other political decisions might have saved a number of them from dying is a salient point.
I was talking to a guy earlier who - cause he pays attention mostly to the headline deaths figures - is convinced that the current lockdown restrictions are pointless and aren't doing anything. He was miserable and hopeless, and it took a while (and a couple of elbows graphs, so thanks for that...) to show him that things actually are (most probably) starting to improve.

It's been a really difficult few months round here - there's been various additional special measures since September - I'm seeing lot of this kind of hopelessness about, and it's a bit frustrating and sad, as it's at least partly influenced and informed by this kind of information, shared without context, with a frowny face or a swear about Boris Johnson - there's also a political partisan aspect to it too, with one incomplete story shared by the lockdown sceptics, denialists & tories, and another incomplete story being shared by the other side. We're pretty privileged here to be able to have a detailed and in depth conversation about this stuff without it dissolving into rancour. Very often.
 
Yes they do. And the figures announced on Sunday and Monday are always artificially low, the figures on Tuesday and to a lesser extent Wednesday are artificially high.
The figures on certain days of the week tend to be artificially high or low. But this is also unreliable.

For example you could compare two recent sundays (15th and 22nd) and see a massive rise from 168 deaths to 398 deaths.
Then you could compare the two mondays directly after (16th and 23rd) and see a moderate fall from 213 deaths to 206 deaths.

Even comparing with the same day of the week, a week ago, is totally unreliable, as a one-off excercise. The numbers are only any use seen in quite a wide context. And if you have to go and look at them in that wider context to make any sense of them, what's the point of posting up the numbers on their own? The risk of people getting a completely wrong impression of what's going on is massively greater than any benefit of just seeing the bare numbers.

As others have said, if you're going to post a single figure at least present a rolling average (which is also something that updates every day).
 
Rolling average is the way to go. There is plenty of variation in the data due to all sorts of reasons. The general trajectory of the rolling average is enough to get a flavour of change over time imho. Going into greater detail just makes you study variations and to me at least isn't that isn't very interesting.
 
Rolling average is the way to go. There is plenty of variation in the data due to all sorts of reasons. The general trajectory of the rolling average is enough to get a flavour of change over time imho. Going into greater detail just makes you study variations and to me at least isn't that isn't very interesting.

And, the different between deaths reported today compared to last Wednesday [79] shows the 7-day rolling average has increased by just over 11 a day.
 
But the rolling average isn’t that helpful for other analysis. The daily figures are what are used in averages. If I want to know what the average is for a week, or a month, or three days, I need the daily data. Publishing averages on their own aren’t that useful for others to use.
 
Although daily data doesn't really work if the figures are coming in a few days later. Weekly figures (if they are correct) would seem to give the best for analysis.
 
Although daily data doesn't really work if the figures are coming in a few days later. Weekly figures (if they are correct) would seem to give the best for analysis.
But they are the quickest measure we have. In the midst of a pandemic, I don’t think it would be acceptable if they just released weekly data.
 
Although daily data doesn't really work if the figures are coming in a few days later. Weekly figures (if they are correct) would seem to give the best for analysis.

Weekly stuff has the advantage of simplifying the picture and making changes over time very obvious. It doesn't really help with lag, and often lags even further behind than daily data, because typically we have to wait a whole week before getting a release that is then out of date (or still incomplete) before its even published. I do use various weekly things for past analysis but its not that timely.
 
But they are the quickest measure we have. In the midst of a pandemic, I don’t think it would be acceptable if they just released weekly data.

Yes true, but if you're trying to find trends then I think weekly is what you'd need. And if you want to find changes in trends - i.e. see what has led to the changes in trends - you'd work from the weekly figures again.
 
Weekly stuff has the advantage of simplifying the picture and making changes over time very obvious. It doesn't really help with lag, and often lags even further behind than daily data, because typically we have to wait a whole week before getting a release that is then out of date before its even published. I do use various weekly things for past analysis but its not that timely.

Yes agreed, but again the daily is actually inaccurate because the figures take a few days to come in. They're valuable but not if you want to see trends. Yes we have to wait a week but then that weekly figure is actually correct because all the figures have come in?

And yes totally - for past analysis - but that's what you need if you want to see what changes in policy have led to changes in the data.

Not trying to devalue your daily data charts at all, but if you want to see trends then you'd need to look at weekly data I think (as you say).
 
Yes agreed, but again the daily is actually inaccurate because the figures take a few days to come in. They're valuable but not if you want to see trends. Yes we have to wait a week but then that weekly figure is actually correct because all the figures have come in?

No I wouldn't say that. Not all deaths are registered within a week of the death, or otherwise entered into data collecting systems within that amount of time.

Take for example the weekly ONS number. The report published on November 24th includes deaths registered up to November 13th. It also includes other data that covers deaths by date of death rather than date of registration, again going up to November 13th, but the numbers given in that report for some days leading up to the 13th will have grown larger again by the time they publish the next version next Tuesday.

Personally I dont need stuff broken down into weekly figures in order to understand trends, I can look at wiggly daily lines and see a picture, but people vary in what sort of a picture they want and what sort of mental processing they perform when looking at graphs, so I dont have a fixed view of what the 'right' approach is.
 
Yep totally agree - but you need to use the data appropriately. Daily figures (allowing for delays in updating them) are good for headline. There are also differences in weekend/Monday figures, too, as they get updated? Weekly figures correct for these.

If you're trying to find out what changed that affected the results then you need accurate figures. Daily figures will be fine if they've been corrected - so daily figures are fine from a few days ago (as elbows has shown in his latest graph). Otherwise weekly figures will be more accurate for looking at trends.
 
Yes agreed, but again the daily is actually inaccurate because the figures take a few days to come in. They're valuable but not if you want to see trends. Yes we have to wait a week but then that weekly figure is actually correct because all the figures have come in?

And yes totally - for past analysis - but that's what you need if you want to see what changes in policy have led to changes in the data.

Not trying to devalue your daily data charts at all, but if you want to see trends then you'd need to look at weekly data I think (as you say).
The whole point of elbows charts is to illustrate the fact that the closer you are to the present date, the less confident you can be about what the trend is at that point.
 
No I wouldn't say that. Not all deaths are registered within a week of the death, or otherwise entered into data collecting systems within that amount of time.

Take for example the weekly ONS number. The report published on November 24th includes deaths registered up to November 13th. It also includes other data that covers deaths by date of death rather than date of registration, again going up to November 13th, but the numbers given in that report for some days leading up to the 13th will have grown larger again by the time they publish the next version next Tuesday.

Personally I dont need stuff broken down into weekly figures in order to understand trends, I can look at wiggly daily lines and see a picture, but people vary in what sort of a picture they want and what sort of mental processing they perform when looking at graphs, so I dont have a fixed view of what the 'right' approach is.

In that case, the monthly figures are the accurate ones? Again I'm talking about looking back to see what changes in policy started to affect the actual numbers of people infected (for example). For that you need stable, accurate figures.
 
In that case, the monthly figures are the accurate ones? Again I'm talking about looking back to see what changes in policy started to affect the actual numbers of people infected (for example). For that you need stable, accurate figures.

All of the data that is recorded by actual date of death etc is the best data eventually, whether its presented in the form of individual days, weeks or months. For all of it, to get the most accurate version you just have to be prepared to wait quite a long time, to know that the last week+ of data is incomplete to varying degrees. Stability in death figures takes time to develop, so averaged out daily reported deaths is an alternative measure to look at, as are hospital daily admissions and these days, with a larger testing system, daily positive cases is also some guide.

Theres nothing to stop me taking daily data and turning it into charts that show it as a weekly number instead. I could even do a rolling thing and produce a chart every day, but still only showing single numbers for each 7 day period. I'll probably try it as an exercise just to see how it turns out. It will be regional. I havent decided whether to do it for cases or deaths or something else. eg if the indie SAGE one posted above is of interest than I can do a version of that I suppose, but I usually avoid adjusting things to be shown as a rate per 100,000, partly because I havent checked what figure for different regions populations are used to achieve this.
 
What I'll probably do its wait for tomorrows data & tier announcements, see how they frame the decisions and data, and see if there are any obvious explanations or stories of interest that can be told with data in a different form.

Because the SAGE papers made clear that government should be making the new tier location decisions based not just on current rates of prevalence of the virus in those areas, but also on the rate of increase/decrease.
 
The Guadians take on what tiers places are likely to end up in contains no surprises that I can think of.


Nor are the complaints surprising. Although if the bloody Covid Recovery Group of MPs manages to maintain a fair chunk of its focus on making the test & trace system better then it is yet possible that that bunch of anti-lockdown numbskulls could manage to contribute pressure towards achieving something actually useful.
 
Back
Top Bottom