Those stats by Chris Giles on the true number of excess deaths towards the end of the last thread were a bit sobering.
He should be here, not hanging round with the riff-raff on Twitter, then he might make more accurate posts...
The data mostly being referred to are the 'weekly registrations'. They are the normal numbers published by the ONS for years, BUT they aren't that good when we're trying to find out what is happening during viral growth and decay. He says
"Data is for registrations of deaths on average happening by April 13, when equivalent hospital number was 13,423"
This is not the right way to do this.
Each death has the following data:
- date of death
- venue of death (care home, hospital)
- location of usual residence (e.g., for someone who is hit by a car in Devon, helicoptered to hospital Bristol, dies there), but was on holiday from Hillingdon, this is 'Hillingdon'
- date of registration
- registry office - could be anywhere, but probably the place of death
- place of death (Bristol hospital here)
There are therefore two key dates:
- date of death
- date of registration, which is likely to be within a week of death but AVERAGES 6 months in cases where a post-mortem and inquest is conducted.
Now we know that not only is there a delay to registration of death, but that the LENGTH of this delay varies according to the characteristics of the death, specifically the location (London & SE slowest), and the type of death (doctor-certified or not).
www.ons.gov.uk/peoplepopulationandcommunity/birthsdeathsandmarriages/deaths/articles/impactofregistrationdelaysonmortalitystatisticsinenglandandwales/2018
In addition, we get delays of registrations over Christmas, New Year, Easter, etc. so that while there is likely an effect of people 'holding out' for Christmas and dying shortly after, there's also an administrative backlog resulting in outsized death figures in the New Year, which aren't 'true', but merely deaths pushed from one week to another.
So when Mr. Giles is referring to the 'average date of death', this is not a meaningful figure, in that 'a care home death in SE England' will be registered after 'a hospital death in SE England', and then 'a hospital death in NE England' will be faster still.
So what you will be doing is mixing, say hospital deaths in London from 13 April with care home deaths in London from 10 April with hospital deaths in NE England from 15 April and care home deaths in NE England from 12 April, and so on.
Which on a time-sensitive basis is not very useful at all.
It makes more sense to look at the 'Date of Death' (which are not in the main release but in the companion LA-based release), but here you have to project forward based on the characteristics of death, so for example where you have 1000 deaths in hospital in NE England recorded as dying 11-17 April and registered by 25 April, then you can estimate that the total number of deaths in hospital between 11-17 April in NE England will turn out to be a number, say, 10% larger than that, while the number of deaths in hospital in London might be 15% higher than those so far reported and in care homes in London 20% higher and so on.
Only if you have a meaningful time period, i.e. a calendar week of deaths, are your numbers meaningful. Talking about excess deaths in the context of an annual average of a numerical week over period 2015-2019 doesn't make much sense either when Easter will be messing with the data.
The headline '22,351' figure doesn't represent the number of people that died in the week 11-17 April