ONS Vaccination Deaths Analysis (part 3)

Another poke at the vital dataset that should be error-free but is not

We’re on the fifth day of the new ONS file of files with a revision for outright bloopers already released, but with a bunch of issues remaining that is bigger than my full English breakfast (and I like a big breakfast). I heartily recommend the following reading/viewing:

Norman Fenton and John Campbell discuss latest ONS deaths by vaccination status data (49mins)

Deaths among the ghost population by Dr Clare Craig

the new UK ONS data is out and it's worse than before by elo gato malo

The latest ONS data on deaths by covid vaccination status by Professors Neil & Fenton

England Deaths by Vaccination Status by Joel Smalley

ONS Data: 25% Excess Mortality Among the Boosted is Obscured by Undercounting of the Unvaccinated by Igor Chudov

There are also some splendid threads on Twitter from @ClareCraigPath, @profnfenton, @josh99, @OS51388957, @USMortality and the inimitable @Jikkyleaks.

Big Time Mess

I think it fair to say that the Office for Nobbled Statistics (ONS) has messed-up big time with arguably the most important dataset they’re going to churn out this year. One commentator declared that the dataset has “more holes than Swiss cheese” and another asked why a pathologist (Dr Clare Craig) can spot multiple errors in a file that took a team of trained officers 7 months to prepare. How is it even possible to get males and females mixed up, and what’s with the missing entries?

By now some of you will have gathered that the latest release omits Jan, Feb and Mar 2021 these being critical early months in which the first vaccine-induced deaths would have been recorded. ONS are also ignoring all teenagers and children under the age of 18, so we don’t get to analyse those tragic deaths of youngsters that have been making headlines.

On the technical side they’ve resorted to using ASMR (age standardised mortality rate) in breakdowns by age group that don’t need standardisation, and they decided to run with the overly coarse bin of 18 – 39 years. There are issues with ASMR that I might cover in a later article; suffice it to say that this doesn’t always do what it is supposed to do and can obfuscate a great deal.

Neither is the cohort used representative of the nation of England as a whole, it being a subset of people whose data records have been successfully and unambiguously linked to the ONS 2021 census. That is to say they left a whole bunch of people out simply because they couldn’t link them via their NHS number (not everybody has a NHS number) – Dr Clare Craig seriously digs into this in her fabulous substack, ‘Deaths Among The Ghost Population’. A consequence of this is that unvaccinated souls are under-represented in the database and this distorts derived ASMRs. As if that wasn’t sufficient they’ve also ignored people who died very quickly after receiving their injection. I wonder why?

Can all this be attributed to sheer sloppiness and error?

I think not, but let me tell you why…

An Aside

As a former UK government PSO/G7 scientist and section leader for a policy area my work sometimes pushed me in the direction of the ONS, and so I would attend high level meetings at their offices, chewing over the numerical fat with PSO/PEO/G7 types from their end. Planning the 2001 census was one such area of work, along with development of work streams and outputs across a variety of topics. So when I say the quality of the datafile released to the public five days ago matches that expected from a rookie SO/EO then, as a former gov-bod, I know what I’m talking about. There’s no way a datafile on such a hot topic would be released to the public without the scrutiny of the G7 in charge, who would be clearing release with appropriate Assistant Secretaries. Absolutely nothing would be left to chance, and especially so with a datafile this explosive. I can only conclude that we are witnessing deliberate acts of obfuscation that will invariably be covered by the Official Secrets Act (even though I left government service years ago the OSA still binds me). The clever part of this is that public-facing officers at ONS would not necessarily be aware of all that is going on, this typically being on a strict need-to-know basis.

Yes, But What About A Slide?

After that indulgent rant I better get my crayons out! A question folk may have at this point is how much of a subset is the ONS cohort? A related question is how representative is it? The latter is the superior question imho because if the ONS cohort (a.k.a. sample) is genuinely representative then we don’t need to worry about them counting every single head. With that in mind I delved into the file and yanked out the person years figures for vaxxed and unvaxxed souls for each month. I then nicked Dr Craig’s natty formula for conversion of these into individual person equivalents by multiplying by 365 and dividing by the number of days in the month.

NIMS Numbers

What we need now is something to compare this head count with and I opted for the National flu and COVID-19 surveillance report: 23 February 2023 (week 8) in which we find figure 67; and a splendid figure it is too:

Data for this figure can be found in the accompanying ODS file; and what splendid data they are too, with numbers of vaccinated and unvaccinated souls presented on a weekly basis by age band, all freshly squeezed from NIMS. Now NIMS runs quinary bands from 75 – 79y down to 20 – 24y, thereafter becoming 18 – 19y, 16 – 17y, 12 – 15y and 5 – 11y; now that’s what I call sensible! In order to flip these into comparable counts I summed everything from 18 years to beyond 80 years.

Missing In Action

With all this done I faced a spreadsheet with some very interesting numbers that told a fascinating story over time. In terms of proportions I discovered the ONS cohort varied from 91.6% of the NIMS cohort down to just 78.2% of the NIMS cohort, with a 24-month mean of 89.6%. If we view this difference as ‘missing’ folk then this stretches from 8.4% all the way up to 21.%, with a 24-month mean of 10.4%. That’s a fair old percentage of folk to have excluded, and especially if exclusion leads to significant bias.

But that’s enough wittering, here are those slides I promised:

I think we can see now why the ONS cohort should really be called a sample and a biased one at that (please read the reference materials provided at the outset). That great leap from just under 40 million individuals to just under 46 million individuals between March and April 2021 is going to have an effect, and that alone may well skew derivation of ASMR depending on who was included and how the ONS went about this.

---

Not a lot to say about this pair of slides, being pretty darn self explanatory.

---

This is my favourite slide of the day (apologies to other data warriors if they’ve already covered this). We now see in terms of raw head counts just what sort of individual was excluded from the ONS cohort over time compared to the altogether more robust NIMS cohort. In the first few months both series play see-saw, with the first big period of bias alarm popping up May 2021 – October 2021 in which vaccinated folk were going missing in comparison to unvaccinated. This bias swings into reverse from November 2021 through to October 2022 when we observe consistent under-sampling of the unvaccinated. These swings are bound to mess up derivation of ASMRs to the point where I consider them useless.

I reckon that’s enough for today – my stomach is telling me lunchtime is coming up fast and there’s an aroma wafting up from the kitchen that suggests to me that Mrs Dee has baked a tray of flapjack by way of surprise!

Kettle On!