UK Personal Independence Payments & Accumulated Vaccine Doses (part 2)

A statistical rummage through the data held on UK GOV servers that reveals a strong correlation between COVID-19 vaccination dosing and cleared new claims for personal Independence Payments

In part 1 of this series I ranted on about flippant use of the maxim ‘correlation does not equal causation’ by expertistas™ and lay persons alike to bury findings that ought to be triggering a great deal of investigation by regulatory bodies and others. We’re floundering in a situation akin to running an emergency ambulance service with no 999/111 facility, no SW radio, no cellular transmissions, no phone line and no pigeon loft. Supping soma and arguing how the mRNA vaccines cannot possibly cause harm whilst people continue to perish is pretty much the stance taken by die-hard expertistas™, whom I am now calling VRDs (vaccine research deniers).

Ranting all done we saw a remarkable rise in new claims for personal independence payments (PIP) since July 2021 that correlated strongly with accumulated total doses administered to the nation of England. I brandished a spanner that goes by the moniker of autoregression using Prais-Winsten estimation to mop up serial correlation (and avoid being labelled a naughty boy) and we ended-up looking at a pretty convincing relationship to the tune of 247 new claims per million doses of elixir.

Today, Today And Today

Today I want to follow through with another slide that should raise an eyebrow, this being a dual time series plot of monthly new cleared claims with accumulated total administered doses for England for the period April 2015 – October 2022:

There is that same striking correspondence that was first unearthed over in the US by Phinance Technologies. I’ve seen a lot of correlations over my 39 years in the numbers business and this sure smells like it is causal, but perhaps we better listen to an expertista™ whose comforting salve of hyper-experience and/or grant funding and/or shares and/or ego is going to soothe our troubled minds.

Alternatively, we can look at a very simple error bar plot of mean monthly new cleared claims by data period and let the evidence speak for itself:

We observe little difference in claim rates between the pre-pandemic and pandemic periods (p=0.496; Kruskal-Wallis independent samples non-parametric test of significance), and a dirty great difference between pandemic and post-vaccine periods (p=0.004; Kruskal-Wallis independent samples non-parametric test of significance).

Whilst we can’t argue against that we can argue over what is causing the sudden and unprecedented surge in new cleared claims for PIP. Are these vaccine damaged people or are they emotionally crippled in some way or other (or both)? Why didn’t claims rise during the first three most terrible waves of COVID compounded by lockdown angst and NHS service breakdown?

I shall suggest that it would be a jolly good idea to find out, but the chances are that we’re going to be badgered into sipping soma with the expertistas™ who, being on top of the mountain, obviously know better than everyone else.

Rhubarb!

There’s nothing inherently wrong with analysing cumulative data providing you know what you are doing when it comes to assumptions of independence of measure. The primary benefit of looking at cumulative total doses administered is that it serves to act as a low pass filter so we may separate the signal from noise. Observation, both anecdotal and clinical, reveals that adverse reactions to the vaccine range from minutes to several months down the line. This huge spread of reaction is compounded by the spread of rollout, uptake and dosing regime, all this resulting in one ginormous and amorphous mass of delayed reactions that are well-spread over time like manure over my rhubarb. Think of the vaccine as leaving a ‘legacy’, with cumulative dosing being the best proxy measure of that legacy.

Ping Pong

The situation is not unlike the problem faced by sonar: a discrete impulse (ping) is sent out from a vessel and we listen intently for the return echo (pong). If the reflective terrain is complex that pong will be stonkingly complex, and we’ll need some fancy algorithms to detect a baddie sub hiding next to a wreck. The trouble with vaccination data at the level of the population is that our ping isn’t discrete, but a great wad of impulses. Thus, in analytical terms, a ‘vax wad’ is sent out and a ‘harm smear’ is returned. If we integrate one or more of those signals by using cumulative data we stand a better chance of baddie sub detection.

Cross Correlation

What I’m going to do now is try my hand at sending out a ping and listening for a pong across the time domain, and I am going to do this using a technique called cross correlation, which is widely used in engineering.

I am going to take the month-on-month change in total doses administered and correlate this with the month-on-month change in new claims. I am then going to slide the differenced series for dosing counts along by one month and check the correlation at a lag of one month. If I repeat this sliding trick for lags up to ±8 months we arrive at the following slide:

See that large red bar sticking up beyond the upper 95% confidence limit (dashed line) at a lag of +3 months? Well, that is telling us that an increase in the month-on-month changes in total administered doses is followed three months later by a rise in month-on-month changes in new cleared claims. That bar is our ‘pong’ indicating where the baddie sub lies. Thus, despite using rather messy raw wad data, we’ve gone and detected what looks like a delayed impact of vaccination on PIP claims.

Once again the evidence mounts, but we can always choose to sip the soma of the expertistas™ if that makes us feel better.

Kettle On!