Here’s my response upon seeing these…

What on Earth happened during 2022 to cause that sudden surge in deaths and crude mortality? The word is once again WERGH, and this takes us into a decent-sized rabbit hole. There’s no way I can derive a sensible 5-year baseline, and so seasonal excess death has to be shelved as an idea for the time being. This also wades deep into John Dee’s Almanac territory and I feel I ought to take up further investigation over in that publication stream. It’s never bloomin’ straightforward!

…and so here we are over in the Almanac all ready to chew this one over!

Back To The Future

Back when the government was locking us all down, destroying the economy, and causing immeasurable suffering, pain and misery with its raft of highly dubious WEF-inspired policies, I used to eagerly await the weekly deaths registration data from the ONS to get a handle on what was happening (or not happening, as the case may be). It’s been a while since I populated my master spreadsheet of death, so here I am again brushing the dust-off some 2 years later, and heading over to this well-worn webpage.

In order to know where we’re going, we need to know where we’ve been. In this regard I’ve collated ONS weekly all cause death registrations since 1982. Together with ONS mid-year estimates for population, this will enable me to track crude mortality and derive those 5-year historic baselines of which I am not at all fond. Ideally, I would be swimming in the very latest date of death (occurrence) data instead of date of registration, but these valuable datasets are subject to hefty fees and the inevitable ONS processing delays following a successful FOI request. That being said, and as luck would have it, I have daily date of occurrence all cause death data by quinary age band for England & Wales from January 1970 right up to July 2023, and will be wheeling this wonderful resource out shortly.

But first things first… herewith ONS weekly registrations for all cause deaths since 1982:

Bizarrely, the 2020 pandemic doesn’t look that spectacular when put in a historical context; and this is before we’ve accounted for the rise in the population! It’s hard to judge what is happening because of this, so here’s the crude mortality for the same period:

We now observe a steady decline in crude mortality from 1982 to 2012, or thereabouts, at which point the curve appears to flatten out with no further gains. What may come as a surprise to some readers is to learn that the 2020 pandemic wasn’t the nastiest period in this 45-year history. The gold medal for mortality must go to 2001 w1 with a rate of 39.44 deaths/100k population; and the silver medal goes to 1999 w1 with a rate of 38.73 deaths/100k population. The 2020 pandemic manages to achieve bronze with a rate of 37.61 deaths/100k population, and is closely followed by 1989 w51 with a rate of 37.19 deaths/100k population.

Though rather lovely, these slides don’t help us understand what has been going on recently, so here they are again but zoomed-in to cover the last 10 years or so:

The pandemic turns out to be peculiarly well-defined. Perhaps too well defined, if you get my meaning. However, in many ways we cannot see the wood for the trees, so let me turn these raw data plots into error bar plots that grab weekly values and turn them into annual means; not just for the pseudo-summer season but the entire year:

Please bear in mind that at the time of writing 2026 is still in progress, so its mean may be a tad wonky!

In terms of the mean weekly count of all cause registered deaths we now see a gradual progression from 2007 – 2019, with a sudden and somewhat expected leap during the pandemic year of 2020. What is not expected is the continued elevated count for years 2021 through 2026.

If we turn to the all cause crude mortality plot, we get a similar picture, but being one of the sudden impact of the pandemic followed by a gradual recovery to a steady state similar to that observed for 2007 – 2019. This makes more sense but, because of this decline, we may deduce something odd is happening to estimates of the mid-year population. I’ve never totally trusted population estimates from the ONS since these can introduce a wide margin of error to what are essentially certified counts.

At this point I think I really need to go back and try to understand something more about seasonal death that is feeling a tad unusual. I’ll need to crayon 8 slides in total, and for that I’ll need a big mug of Grumpy Mule, and since I’m all out I need to pop to the shops.

Kettle On!

We know that age groups flip in and out of dominance over time; with the flu striking the elderly, and then the very young, and then those who have to work for a living. So… I’d like to step back and try and get a handle on the big picture – what age group has dominated the most over this 103-year study period? Once I’ve got that nailed I’ll start turning the handle on that whopping great list of temperature variables (60 juicy and well-seasoned independents to be precise). But, therein lies a problem worthy of a fresh section…

A Fresh Section

An astute reader has pointed out that, by examining so many variables, I’ll eventually find something that looks like a temperature effect. This is indeed so, and is a murky issue researchers will bury from public view. If my list of temperature variables stretched to 100-long (and it easily could), and if I’m using the ubiquitous 95% level of confidence to guide variable selection (p<=0.05), then we should expect 5 of that 100 to pop-up as being statistically significant when it’s nowt but the laws of chance playing with our minds. But the situation is worserer than this, and I shall reach for a hunk of probability theory and my crayons in order to explain.

Some switched-on readers may ask if 6 of those 100 variables could easily pop-up as being statistically significant even though this is also a fluke; and the answer is yes! The same goes for 7 variables popping-up; and even 8, though the odds of this happening are less and less each time. We can crunch some probabilities for different numbers of variables popping-up using a dash of binomial theory and plot these out as a graph of the likelihood of each result happening, given a starting point of n=100 variables from which to choose, and given a ‘success’ threshold of p=0.05 (95% confidence). Here we are:

That there curve is pretty darn spread out! It peaks at 4 or 5 variables popping-up, as we may expect, these being the most likely outcomes due to chance whenever we run a modelling session with 100 variables for testing at p=0.05. But you can also see that just 3 might pop-up, as well as 6; and 2 might pop-up as well as 7 or even 8; albeit with a smaller probability of these particular outcomes happening. Now that’s getting a bit hairy!

A bold reader may ask, “what is the probability of 8 or more variables popping-up as being statistically significant, but is actually a fluke due to chance?” The answer to this requires I transform the probability function curve into a cumulative probability function curve, thus:

We now see that the cumulative probability for 8 or more variables popping up is 0.128 (or 12.8% in plain percentages). This is not just hairy but scary, since there’s a fighting chance that may happen!

The upshot of all this is that we could declare total nonsense without knowing it. So… here’s the thing… I’ve been upfront about this and will proceed cautiously, but out there, in the cut-throat world of grant funding and peer review, no researcher in their right mind will reveal their secret fumbling in the pantry day-after-day with variable-upon-variable until some combination crops up they can use to support their hypothesis. And when this eventually happens you can bet your bottom dollar that they’ll concoct a convincing argument as to why they chose that particular set of variables, and make it look like very sensible a priori reasoning. It is a harsh fact of life for the public to learn that intrepid researchers will bend the truth, and will happily do so, for they are engaged in their noble cause. And don’t expect the method and appendices you read in their well-groomed papers to be the methodology they honestly and truly stuck with. Consultant statisticians like me get to see how grimy things have gotten before research work is licked into passable shape (but not before we’ve signed a NDA keeping us mum). Despite the white lab coats, science can be a dirty business! Ranting is now done for the day, so let me turn the handle on some numbers…

Dominance

The idea behind this section is to look at the topline national figures for (rescaled) influenza mortality for all age groups across England & Wales over the period 1920 – 2022, and ask a neural network (multilayer perceptron - MLP) to go figure which age subgroups best describe what has been happening over this time. In order to achieve this I asked the MLP module to split the data file of 103 annual records into a 50:50 training to test ratio. Herewith the key outputs:

I’m happy with that.

Though perhaps not that discernible to the inexperienced eye, the difference between predicted and observed mortality increases with increasing mortality. Look at the jiggling about the black X=Y line and you’ll see there’s more of it as we move toward the right. This tells us that rather deadly national outbreaks are harder to predict using combinations of age subgroups, and so we may deduce they are more complex affairs.

Another clue to this is given in the lower table. The best predictor of national mortality is the oldest age group of 75+ years, as we may expect. This is followed by the 65 to 74-year group, as we may also expect. What we may not expect is that the 15 to 24-year group is the third best predictor of influenza mortality. This is interesting and I suspect this changes over time, with some years striking down our young. Though this is not exactly new news, what it usefully does is objectively indicate that it is the 75+ year group I should focus on if I want to consider the likely effects of heatwaves. This I shall now do…

Big Pot Cooking

What I’m going to do now is run MLP again, but submit influenza mortality for the 75+ year group as the dependent variable, and submit that giant raft of 60 temperature variables that I listed in part 3 as independent predictors. Here’s the gubbins:

I’m not so thrilled with this because of the skew in the observed vs. predicted graph. The MLP model is clearly (and badly) under-predicting years when mortality in the oldest age group is soaring. This tells us that temperature, and change in temperature, isn’t really to blame for deadly years amongst the vulnerable, otherwise that cloud of purple points would be hugging the X=Y line. A handy null result right there, then!

If we now turn to the hit parade of candidate independent variables, then top of the charts is the mean maximum daily temperature attained during the autumn months of September, October and November. Closely linked to this is the standard deviation of the mean daily temperature for autumn, which is no doubt highly correlated. And then we’ve got the mean daily differential temperature for the summer months in the bronze medal position. This is fascinating, and suggests one of the roots of influenza might lie in temperature fluctuations in the run up to the influenza season – though we must keep that under-predicting model in mind. I guess I better look at these three variables in more detail just to check we aren’t being led astray.

A Reduction

I shall start by being saucy and ignoring my own advice on using cross-sectional methods on a reduction made from the big pot boil-up: herewith a Pearson bivariate matrix:

Well, there’s a tasteless table for you!

Of the three top predictors chosen by MLP to predict influenza mortality in the elderly, then only the very best performing predictor manages to return any kind of a result in terms of overall correlation. Even so, this is a pretty thin result (r = -0.218, p=0.027), and I’m going to bet it fails the ARIMA acid test.

Hang on, one ARIMA with fries coming up…

Yep, there we go.

The term fails to reach statistical significance in an ARIMA(0,1,1) model structure, with p=0.270. This tells me we are looking at cross-sectional artefact once again, and this time it has crept in through use of a neural network that I tried to use to render that big list of veggies down.

I’m all cooked-out, so shall now withdraw with a slice of Mrs Dee’s buckwheat flour Victoria sponge layered with Mr Dee’s homemade gooseberry jam, and a lavish amount of buttercream…

Kettle On!

Yes indeed, I’ve got nice-looking and rather interesting results on the table, but these can easily be misleading. I’ve already mentioned a potential confounding factor of warmer winters tending to be recent winters, and earlier in this series we saw mortality for influenza in sharp decline over the years. Put the two together and you’ve got bias bigger than my brunch today, and it sure was a big brunch!

Time series modelling needs to come to the rescue, and that means I’ll be reaching for my ARIMA spanner once again. If I get another positive result supporting the hot winter = less …

Read more

Mean Daily Temperature Range

More boring tables for you…

These were generated by an ARIMA(0,1,1) model that did a mi…

Read more

Over in my sister publication of John Dee’s Climate Normal you’ll find me tackling the Central England Temperature Record (HadCET) – being the longest running temperature record in the world – and over in the Notes bit of Substack you’ll find me firing-off thought pieces that hopefully will get people questioning how the Met Office go about the business of selling man-made global warming (a.k.a. climate change) on behalf of the globalists. Here’s one such weeny note:

Whenever the Met Office announc…

Read more

There is absolutely nothing on show here, though we probably guessed this back when button-pushing for part 5 resulted in a soggy bottom. Next up, then, are those cross-correlation plots (CCF) that have a habit of turning our hypotheses upside down:

Read more

At this point I can go in search of the source of what might well be an artefact hiding down in the data, or I can broaden the scope of the study to include the potential effect of the Earth’s magnetic field on human health. Now, before we all reach for the tin foil again, I want us to do some bedtime reading, starting with this sample of papers:

Azcarate et al. Influence of geomagnetic activity and atmospheric pressure on human arterial pressure during the solar cycle 24. Advances in Space Research (2016)

Babayev and Allahverdiyeva. Effect of geomagnetic activity variations on the physiological and psychological state of functionally healthy humans. Advances in Space Research (2007)

Babayev et al. (2012) Potential effects of so…

Read more

To get the ball rolling we have to define ‘extreme mortality’ and I’ve gone for the easy option of declaring this to be a rescaled influenza mortality for England & Wales exceeding 5 deaths per 100k population for the period 1965 – 2022. This simple definition yields 7 extreme years, and if I re-jig that penultimate plot of part 3, it looks like this:

This makes it pretty clear that there’s no obvious and immediate link between cosmic ray counts, solar activity and influenza mortality (if we ignore the likelihood of lagged effects, that is). If cosmic rays were a problem, then those whopper extreme mortality blobs would cluster top left; if solar activity was a problem then they’d congregate bottom right; but tha…

Read more

Read more

Read more

One Dozen Papers

I think, that with this racy topic, I better start out by presenting the dozen papers I’ve come across to date, so you can get a feel for where this idea has come from and how it has developed:

1: Iagodinskiĭ VN, Aleksandrov IuV. O tsiklichnosti épidemicheskogo protsessa pri grippe v sviazi s periodichnost’iu solnechnoĭ deiatel’nosti [On the cyclic nature of the epidemiologic process of influenza in relation to the periodicity of sola…

Read more

For one thing, all those little extra analyses that consecutive Clinical Directors and Departmental Managers wanted at the last possible minute before their many meetings are not included, as are the myriads of weeny little jobs that surgeons wanted me to undertake whilst standing looking over my shoulder and thrusting a USB in my face. Then there were the ‘suits’ working for NHS England whose shared hot-desks were down the end of the corridor; they’d approach me with, “John, if I buy you lunch could you do this analysis for me?”

Then there was the Director of the Primary Care Trust who’d phone and say, “John, are you free right now?” - can’t say no to the purse holder! And then, pretty much every other morning at 8am, was a queue of pale-faced senior house officers waiting outside my office, who were all going to get bawled at terribly by their consultant at 8:30am for not handing in their clinical audit project.

The key for early morning arr…

Read more

I’ve added a locally estimated scatterplot smoothing function (LOESS) to guide the eye and we can see that this trundles along reasonably horizontally until the year 1999 or thereabouts, after which it starts to climb. That climb over the period 2000 – 2019 means increased variability in mortality, and ChatGPT reckons this can be taken as evidence of an ineffective vaccine campaign (and possibly vaccine harm); for vaccines, as with any other intervention, will serve to reduce the variability of clinical outcomes when genuinely beneficial. One way to think of this is that instead of one bunch of people surviving and another bunch d…

Read more

Well, I’ve now tried that suggestion and the situation is still as clear as mud!

Instead of divergence post-2000 we discovered convergence, and we figured that this could be due to the older age group surviving better in the post rollout period; but that it could also be due to the younger age group faring less well! And there’s a nasty plot twist I ought to mention… this faring better or worse may not be a real-world biological thing; it may simply be administrative artefact as coding fashions co…

Read more

Not necessarily.

All that I have done is use a particular statistical technique on a particular set of data and come away with a single negative result. What we can safely say after reading part 20 is that any benefit that is alleged to accrue from the national vaccination programme for those aged 65 and over hasn’t jumped on the table, song a bawdy song and slapped us square in the face. In short, there’s no big and bold effect that many would have expected when it comes to preventing death. There are many possible reasons why this is so.

One possibility is that genuine benefit might be a thing but it might offer too small an effect to be detected within the noise of a national dataset prone to coding whimsy. I have to say that weedy benefits arising from any medical intervention do not exactly inspire me, but it remains a crumbly fact that the world is a complex plac…

Read more

Let us first have a look at the boring tabular output of the model runs, and then we’ll look at some colourful slides:

Read more

What I forgot to do (after a substantial luncheon that got me all drowsy) was filter the dataset to start in 1920 and not 1901 in order to avoid the crazy pandemic of 1918/19. What dropped out is worth serious cogitation and more than a few questions need to be asked!

Below is a simple linear regression of causal influenza mortality (deaths per 100k population) for the 65 – 74 y age group for the period 1901 – 2022. The y-axis is logarithmic and I’ve fitted an exponential model. But here’s the thing: if I hadn’t labelled the global p…

Read more

A highly pertinent question, to boot, for it was in 2000 that the UK government decided on a comprehensive national rollout for the flu vaccine for those aged 65 years and over. The big black fly buzzing in the soup is the WHO ‘rule 3’ coding update of 2001 that effectively wrecked continuity in the death certification time series. The trouble is I’m a curious sort and wondered what residuals would be churned out from models thrashing through the period 1920 – 2022, ideally with some sort of indicator variable to mop up the policy changes after 2001.

Exponential smoothing models are rather simple univariate affairs and cannot cope with the notion of additional indicator variables so I decided to rummage in the pantry for an alternative. I didn’t want to reach for my ARIMA spanner and fancied some fancy regression or other. To help me decide I started out by plotting four scatterplots of the four dependent var…

Read more

Einstein got there first when he said this in Geometrie und Erfahrung:

Die Mathematik genießt vor allen anderen Wissenschaften aus einem Grunde ein besonderes Ansehen; ihre Sätze sind absolut sicher und unbestreitbar, während die aller andern Wissenschaften bis zu einem gewissen Grad umstritten und stets in Gefahr sind, durch neu entdeckte Tatsachen umgestoßen zu werden. Trotzdem brauchte der auf einem anderen Gebiete Forschende den Mathematiker noch nicht zu beneiden, wenn sich seine Sätze nicht auf Gegenstände der Wirklichkeit, sondern nur auf…

Read more

By way of appetiser here’s how Brave Search A.I. summarises the flu jab situation:

The seasonal influenza vaccination programme in the UK was introduced in England during the late 1960s to protect individuals in clinical risk groups who were at higher risk of influenza-associated morbidity and mortality. Over time, the programme has been progressively expanded to include additional groups. The age threshold for vaccination was extended to include all individuals aged 65 years and over in the year 2000, pregnant women in…

Read more