Excess Deaths - England & Wales

ONS cooking the books again!

Thank You.

Just a quick update. First, thank you for the overwhelming support to crowdfund the useful mortality data. I’ll probably set up a GoFundMe or something to make it easy for everyone who has pledged. Any other suggestions?

Excess Deaths 2020 to 2022, England & Wales

Secondly, I reread one of my related old articles and your recent support compelled me to revisit it. As it happens, the ONS has since added 2022 data, so, FWIW, just in time for tomorrow’s debate on excess deaths, here is my re-analysis of their latest annual data.

Actual vs Expected (excess) deaths, England & Wales, 1995 to 2022, based on registration-date data from the ONS

So, there you have it… In spite of the best efforts of the ONS to make their ubiquitously shoddy mortality analysis even more corrupted by fiddling the baseline, dead people don’t lie.

There are more people dying “unexpectedly” in 2022 than there were in 2020 - the year of the COVID! Nice try, ONS, but you’ve been found out yet again.

Inevitably, I’m not the only one to report this:

Too many deaths are to be expected

And to compound the issue, even if there was no excess in 2021 and 2022, this would still be a negative indicator due to the lack of pull-forward effect (you can only die once). So, in fact, both 2021 and 2022 are even worse than they appear (which is already bad enough).

There were quite a few other insights, but given the limitations set out below, I’m going to refrain from reporting them until we get the better data.

Methodology

Detailed methodology of the robust and parsimonious model here but note:

• All distributions required two regimes, indicating a new mortality “population” emerging since around 2000 to 2010. H/T to Dr Clare Craig - most likely due to immigration but cannot yet definitively confirm.

Here’s a couple of charts to demonstrate the quality of the model. You can see them all in the linked spreadsheet below.

Accommodating the entire mortality distribution of a single birth cohort with immigration confounding with just two, continuous, two-parameter models is a clear indication of its robustness.

Limitations

All of the limitations stem from the poor quality of the underlying data:

1. Registration date - the data is by registration date rather than occurrence. As obvious as it should seem, this puts any information from the data (which is no more than is contained in the occurrence-date data since both sets are missing the same deaths!) in the wrong place. What the ONS considers to be a benefit in not having to revisit and amend prior reports as a result of new information being received that might contradict its conclusions, again, I’m kind of thinking that’s the whole point of using the most complete data that is available, no?

2. Annual data - not only does the annual data lack sufficient granularity to help us identify the potential causes of the excess mortality since 2020, in aggregating by calendar year, the ONS rather inconveniently cuts through two mortality years, which run mid-calendar year in the northern hemisphere. Doh.

3. Incomplete data - In spite of being more than 15 months post year-end, it seems the ONS is incapable of providing data for this series for 2023. This is all the more strange given how readily even allegedly inferior national statistics agencies all around the world were able to report COVID deaths within days of them happening. Weird.

4. Confounding - as confident as I am that, empirically, all the mortality distributions for people aged 55+ in England & wales are bimodal, I can only speculate as to the reason why (immigration?). So, while numerically, the results are satisfying, we’d still need a rational explanation to accept their legitimacy.

Fortunately, all of these limitations are dispensed with once we get the daily, occurrence dataset from the ONS. Just a shame we still have to wait another 12 weeks for it as I can analyse it in a mere few hours once finally received.

Link to spreadsheet which contains 90 independent Gompertz analyses - 45 male, 45 female for birth cohorts 1917 through 1962.