ONS Baseline Derivation For Excess Death: ignorance or corruption?

More jiggery-pokery from an organisation that ought to know better

I have been busy preparing datafiles for another stab at quantifying vaccine harm using ARIMA time series techniques, an effort that fizzled out like a damp squib in part 9 of my Vaccine & Death series (but which promises to build back better).

All depends on how we derive excess death, and the technique adopted by the Office for National Statistics (ONS) is to subtract prior 5-year means from observed counts of all cause death on a week-by-week basis. I’ve grumbled on about this approach before, pointing out that disease doesn’t come and go like clockwork, thus an early or late flu season will throw a spanner in the works when we try to compare this with a historically-derived baseline whose bugs had a different idea.

Grumbling aside this is an easy calculation to make and the concept is intuitively accessible for most people; that is, we all readily understand that by subtracting the historic seasonal pattern for death - the ‘baseline’ - we will end up with a surplus or deficit that will require explanation if alarming.

A Fly In The Ointment

A fly in the ointment is how to handle the derivation of excess death for weeks during 2021 given that the 5-year prior mean baseline would necessitate involving the pandemic year of 2020. The ONS neatly side-stepped this by ignoring 2020 and by using the period 2015-2019 for derivation of excess death for both 2020 and 2021. A crude approach, maybe, but a reasonable one in the circumstances.

Not leaving anything to chance I derived my own baseline series for 2020 and 2021 and this precisely matched figures published by ONS. I had assumed that the ONS would adopt the same strategy for 2022 to avoid the abnormal years of 2020 and 2021 but I was wrong: they’ve changed their method in such a way that we need to start calling them the Office for Nobbled Statistics.

The Office for Nobbled Statistics

In the derivation of excess deaths for 2022 the Office for Nobbled Statistics has decided to completely ignore the mayhem caused by the pandemic and use the 5-year baseline period of 2017-2021 as though nothing was ever out of the ordinary. A gob-smacker indeed but I kid you not! Herewith a screenshot of my modest spreadsheet revealing the trick that has been played:

In the column headed EWDeath5yrDEE you’ll find my calculation for the prior 5-year mean baseline using the data from EWDeaths (England & Wales all cause weekly deaths; n.b. the file is sorted by week number then by year to give me an easy time of it). Grab your trusty hand-held calculator and derive the average of EWDeaths running down from 2010 - 2014 and you should arrive at 11,838.8, this being my baseline for 2015/w1. The ONS provide a figure of 11,838 under the column headed EWDath5yrONS… and I’m not going to quibble about a difference of one!

Work your way down you’ll see pretty minor differences of 25 - 26 deaths along with evidence of the ONS using the period 2015-2019 twice to arrive at baseline estimates of 12,175 for 2020/w1 and 12,175 for 2021/w1. They didn’t repeat this method to arrive at 12,175 for 2022/w1 and, instead, they’ve flipped to using the mean for 2017-2021 to arrive at 13,298 baseline deaths; this producing a surplus of 1,098 deaths beyond my reckoning - a significant amount by any standards (+9%).

If I adopt the same trick I arrive at a baseline of 13,134.8 deaths, but this means I’ve ignored anything the pandemic threw at the nation along with the mayhem that ensued. By flipping the calculation in this manner the ONS are effectively claiming that 2020 and 2021 were nothing special and make for ideal baselines for deriving excess death during 2022. I have to keep slapping my face to ensure I’m not hallucinating and that they’re actually gone and done something this crass!

Why Does This Matter?

Because it greatly reduces estimates of excess death during 2022 - a most critical time during which many analysts are attempting to assess vaccination benefit/harm. If you want to hide a mounting death toll I cannot think of a better way of nobbling excess death unless you start conjuring numbers out of thin air.

The number of excess deaths that have been effectively ‘hidden’ as a direct result amounts to an estimated 7,475 deaths for the period 2022/w1 - 2022/w10. This is shocking and very likely the reason why my vaccine harm/benefit modelling has been turning in ambiguous results. I thus have to start all over from scratch in part 10, and ideally without punching people.

A Visual Treat

I always like a good graph to finish, so here is a time series plot revealing the discrepancy between the ONS baseline estimates and my own baseline estimates. You’ll see a superb match right up to early 2021 when the ONS slightly over-egged the mixture. The curves dance as one again right through to the 2022/w1 debacle.

Note also that the ONS estimate starts to rise above my reference curve just as the series fizzles out – all such rises will serve to lessen estimates of excess death for 2022.

Kettle On!