UKH

https://www.bbc.co.uk/news/52282844

I'm looking at the London Weekly Deaths in London.

It is supposed to be illustrating 'Excess Deaths' for each week.  So the number of actual deaths recorded is shown as a blue bar, and the 5 year average a grey bar. However, the grey bar stays perfectly horizontal; it doesn't seem to vary, even though of course the average number of deaths each week varies quite a lot, steadily declining each spring.

Is my understanding correct; and if so, should we be paranoid that the ONS is presenting data in such a way is to obfuscate  what is actually going on? Why not show the 'average deaths' as a single jagged line?

In reply to Rob Exile Ward:

I think you'll find that's a BBC graphic, not an ONS one. It's the BBC failing to show seasonal trends in the 5 year average.

In reply to Rob Exile Ward:

The average deaths is shown as a box that would contain the line rather than a line to make the graph simple and easy to read.  

The ups and downs of the average deaths isn't particularly relevant to what is being shown.  A box shows the variation.   

As you can see from the graph the average deaths doesn't vary much compared to the excess deaths.  

In reply to Rob Exile Ward:

Think you'll find the grey is a range of deaths. Was surprised it was so consistent myself so took a look at the ONS data ( https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriage... ) and interestingly it does seem to be very linear.

In reply to Jon Read:

Well it says 'Source: ONS' though it IS ambiguous about whether it's the graphic or just the data that is being credited. It seems a crass mistake to make IF my understanding is correct.

In reply to Rob Exile Ward:

Personally I'd read that as the data used to draw the graph came from the ONS, but it is ambiguous.  

In reply to PhilWS:

You're quite correct, I'm surprised, looking at the data it looks like peak deaths are in January and maybe earlier, outside the scope. So I assume the lower and upper limits of the grey bar are for the entire period, which isn't quite what the text describes! So it is odd.

In reply to Rob Exile Ward:

The box represents the upper and lower limits of deaths over the entire period. It would be better if they had used two dotted lines.

In reply to Removed User:

I think dotted lines would be clearer too.  But I guess someone thought it looked better as a box.  

In reply to Rob Exile Ward:

I'm going to save this graph to use as an example when I get back to work. 

In reply to Rob Exile Ward:

The problem is that Coronavirus isn’t seasonal. However, neither are average deaths over 5 years. So showing seasonal averages would be pointless. 
 

The text does say “historical upper and lower limits”. 

In reply to DancingOnRock:

Yes I get that. Upper and lower limits of what? What historical period? Why is it even shown like that when the text above describes in perfectly good detail how the average is calculated?

It happens that average deaths ARE cyclical, and there is quite a range between, say, Jan and May (though not so great between end of March and end of May, granted). However if cyclical averages are going down as the year progresses - which they are - and total deaths are staying the same or increasing - which they may be - then the pandemic is getting worse.

I think the choice of that wide horizontal bar obscures this, and it is an 'unusual' way of displaying data. Why wouldn't you just compare excess deaths in each time period with average deaths in the same period?  Otherwise you are comparing apples - deaths in each selected week- and oranges - the range of deaths averaged over a several month period. It's just wrong.

In reply to Rob Exile Ward:

If you draw the wiggly line it would fit inside the box.  That’s what it means by upper and lower limits.  It won’t go any higher than the top of the box, or lower than the bottom.  

Given how little the gap between upper and lower, compared to the bars the person who drew the graph felt that it doesn’t add anything to draw the wiggly line.  I agree.  

If you want to see it with the wiggly line the source of the data is linked, it won’t take long, but I really don’t think it’s relevant.  

In reply to Rob Exile Ward:

Is this ONS graph more accurate? (pdf)

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

In reply to Rob Exile Ward:

Not the graph but this paragraph is shockingly misleading and helpful to the Tory nonsense:

"But it will also be vital to establish how many deaths may be happening because of the restrictions, if people are not getting the treatment or support they need for other health conditions."

It isn't the restrictions which are causing excess deaths from other factors like cancer/heart disease, they are indirectly caused by the virus.   If there were no lock down restrictions the number of hospital admissions for Covid infections would be far higher and there would be even more disruption to treatment of other conditions.  Removing lockdown will increase, not decrease, excess deaths from other conditions.

In reply to tom_in_edinburgh:

It is possible for the lockdown to be causing deaths *and* for it to be a net life saver vs the alternative.

lockdown will also be averting deaths (eg reduced car accidents).
 

to the OP: it’s a bad graph. This is what they should have done https://twitter.com/victimofmaths/status/1265204619535417344

In reply to victim of mathematics:

So what is the mechanism for the lockdown causing deaths?

Potentially a lockdown could cause deaths from hunger but that isn't happening here.  There are more people dying of cancer/stroke/heart attack than would have happended before Coronavirus but it is incorrect to attribute those deaths to the lockdown.   It isn't the lockdown which has caused hospitals to be full of coronavirus patients and have to restrict normal services, it is the virus, and the effect would be worse without the lockedown.

In reply to tom_in_edinburgh:

Domestic abuse?

I think we’re probably closer in our positions than you think, but it’s a little dangerous to boldly state what is attributable to lockdown vs the virus. If people couldn’t get treatment because services were overwhelmed with COVID patients, yes that’s clear enough, but what about people reluctant to access treatment because they *perceive* services that they should be staying at home even if services could cope. You could argue that was the fault of government messaging/lockdown.

Basically it’s messy and being overly reductive is a bad idea (IMO).

In reply to Rob Exile Ward:

If you can bear to read an article in the telegraph you will see why the 5 year rolling average is not always the best tool for comparison. It's from May so figures given are for the time until then but they make a strong case. Pandemic or not we are underprepared for any statistical anomalies. 

https://www.telegraph.co.uk/politics/2020/05/14/cant-gauge-impact-covid-19-...

However I guess the BBC would rather simplify the situation than give such a lengthy explanation. A bit like the charts or Brass Eye.

When the arrow goes up it flashes. 

In reply to JohnBson:

I'm interested in their argument, but I'm not registering with the Telegraph just to read it. Can you summarise?

Worth noting that many sources as well as the ONS (e.g. National Records of Scotland, the Economist, the Financial Times) are using excess deaths above 5-year weekly average (mean or median) (or a longer average). 

In reply to marsbar:

I don't think that's quite right?  Isn't it more likely to be the 95% limits within the 5 year period?

Edit: because Jan 10 is outside the box!

In reply to Bilberry: where's the 95 coming from - and any special rationale for the P95 rather than the P99 as an example?  

Personally I think the BBC graph is ok - it's showing the min/max over a 'normal' pandemic free 5 year period.  I don't think you'd gain a lot by going to a month varying number, especially as March and april are months where you are going to get warm and cold years - are you going to adjust again for weather.   You'll miss the wood for the trees-.  The problem is the initial assumption the grey bar is an average as opposed to normal upper/lower bounds

In reply to Rob Exile Ward:

It’ll be the average max over 5 years and years average min over 5 years. 
 

COVID isn’t seasonal so I suspect they’re trying to show is how it compares to deaths from Flu. The max and min are shown so you can still use the extremes to compare.(Ironically deaths from flu appear to have been down this year due to social distancing)
 

It’s definitely decreasing because we are now not seeing excess deaths in the 15-64 age range in week 21 but we did in week 20. 

In reply to wbo2:

 

It’ll be the rolling average to reduce the effect of spikes from bad epidemic years. Or do you mean ‘COVID pandemic free’?

In reply to DancingOnRock: does it say that?  I read it as just the min and max from the previous 5 years to establish normal bounds.  The 5 years referred to in the text is not necessarily related. They're still useful to see.  

In reply to wbo2:

No. But the ONS graphs usually use 5 year average. 
 

https://www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriage...

In reply to Rob Exile Ward:

You'll be pleased to see it shows the line as well this week. 

In reply to marsbar:

Which explains that it represents upper and lower limits and therefore makes perfect sense.
If I remember correctly, that explanation wasn't present on the original one?

In reply to marsbar:

Do you have a link, sorry I can't find it.

In reply to Rob Exile Ward:

It's an update to the article that you originally linked to:
https://www.bbc.co.uk/news/52282844

In reply to FactorXXX:

I may be being thick but I still can't see it.

In reply to Rob Exile Ward:

The first graph down has a grey horizontal box on the Y Axis which represents The Historical Upper and Lower Limits.
If a section of the histogram falls within it, then the death rate is as per normal.
The only thing missing is over what time period is counted as historical.  Believe it's 5 years?

In reply to Rob Exile Ward:

Sorry  https://www.bbc.co.uk/news/uk-51768274 

Quite a way down

In reply to marsbar:

Thanks. That I now understand.