An update on John Campbell's analysis of Canadian mortality data and jiggery-pokery by statistical authorities
John Campbell asked me if I could explain what the confidence intervals meant in these Canadian excess deaths charts:
I sent him the response below
These are what are called classical statistical confidence intervals. I actually wrote a substack article about this very thing a while back:
The key thing is that the ‘expected number of deaths’ (the bold blue line in the graph) is the number of deaths that we would expect to see in each of these months if nothing ‘new’ was happening like a pandemic or a vaccine programme. But the problem is that this number is, of course, unknown and so is estimated based on some number of previous ‘normal years’. The uncertainly about the number is expressed in terms of a 95% confidence interval (x,y) where x is the lower bound and y is the upper bound. In lay terms (although this is not strictly correct - see below) this means we can be 95% sure that the true (but unknown) number of deaths we would see in that month (if nothing new was happening) is between the lower value x and the upper value y.
In the graph only the upper 95% value (this is the “upper 95% prediction value for expected number of deaths”) is shown as a red dotted line.
That’s because- if we are interested in excess deaths – then if what we observe is any number BIGGER than that – then we would regard it is a very unlikely that nothing new had happened – i.e. something new is likely causing this especially high number of deaths.
Now what probably confused you most about the graph is that it does not actually plot the observed number of deaths. Instead it plots a black dotted line (what it calls the “lower 95% prediction interval for adjusted number of deaths”. That’s because there may be uncertainty about the ‘true’ number of deaths observed (the fact that they call it ‘adjusted’ suggests they also think they need to adjust the actual observed number to take account of things like population changes, but strictly speaking those adjustments should have been accounted for in the estimate of expected number of deaths). But, whatever the assumptions are, the “lower 95% prediction interval for adjusted number of deaths” means that the ‘true’ observed number of deaths is very unlikely to be anything lower than this.
So, in summary, the actual number of observed deaths is ‘at least’ the black dotted line. So whenever this line is above the blue line it is probable that there were excess deaths and whenever this line is above the red dotted line it is almost certain there are excess deaths.
So it is likely (in the 2022 graph) there have been excess deaths every week except 2 April. In week of 22 Jan it is likely there were at least 1751 excess deaths (the difference between minimum observed and expected) while it is almost certain there were at least 1438 (the difference between minimum observed and maximum expected).
As a Bayesian I believe that the classical approach to confidence intervals is terrible and would always use Bayesian confidence intervals instead.
That’s because the formal meaning of the classical confidence interval is actually incredibly complex and very different to what most people think it means, whereas the Bayesian confidence interval means exactly what people intuitively think it should mean namely (in this case)
“There is a 95% probability that if things were as normal then the number of deaths in month M would lie between x (the lower bound) and y (the upper bound).”
Here is John’s subsequent video where he quotes my summary comments:
It turns out that the latest Canada data on excess deaths is actually much more serious that even suggested in John’s video. This website (with thanks to David Dickson) provides continually updated data and exposes multiple problems with the ‘official’ Canada data including the fact that the two main provinces are missing data (Ontario and Quebec). The excess deaths for the years 2020, 2021, and 2022 based on the 2010-2019 10-year average is especially revealing:
We believe that using a 10-year pre-covid (i.e., pre-2020) period is the best way to determine excess deaths, assuming stability and homogeneity in the population and in disease profiles. Many of the excess death figures you see for 2021, 2022 and 2023 from around the world are based on the previous 5 years only; moreover, while most (correctly) exclude the unusual covid year of 2020, it seems to have become standard to include the years 2021 and 2022 which (because of the impact of lockdowns and the vaccines as well as any continuing covid) were certainly not ‘normal’ years in any sense. Thus, for example, for its 2022 figures the ONS in the UK uses the years 2016, 2017, 2018, 2019, 2021 for its ‘baseline’ and for 2023 it uses the years 2017, 2018, 2019, 2021, 2022. We believe this is extremely duplicitous, since the high excess numbers in 2021 result in artificially suppressing the excess death figures in 2022, and the high excess numbers in both 2021 and 2022 result in even greater artificial suppression of the excess death figures in 2023.
We see the same in Australia where they estimate 2022 excess deaths using 2017-2019 and 2021 but do not include 2020. But in Australia 2020 was actually a normal year, despite the fact that the Austrialian Bureau of Statistics stated “2020 has been excluded as it did not resemble a typical mortality year.” So, by including a year that is higher than expected and excluding a normal year, the excess is manipulated to look smaller. See Arkmedic’s substack for details:
Even with these tricks to downplay the current excess death figures some people are noticing that there is a major problem, as this Daily Mirror article shows:
But of course, if you ignore (as the mainstream media does) the possibility that the vaccine may be a contributory factor, then it's all a mystery as Prof Coleman in that article suggests. He can't understand why excess deaths are higher when they should be lower after pandemic. But he highlighted two key reasons for excess deaths spike: "Britain’s getting older, and gaining a larger average Body Mass Index."
Of course, it could be people not taking their statins. Honestly.
Update: Here is David Dickson’s updated analysis of UK excess deaths using the 10-year 2010-2019 average:
I have thought that - that if the excess deaths continue over years, they need to exclude all the years with excess death from the averages, as otherwise the excess deaths will become normalized in those averages, and no longer appear as excesses in the data.
I just saw my cardiologist. Shortly after my booster vax (March 2022) I developed a cardiac arrhythmia (PVC’s, premature ventricular contractions). I need to lose weight (true, but not a cause of PVC’s), take my statins (again, nothing to do with PVC’s), and take a beta-blocker (which I already tried but discontinued because it had no effect). Of course, he totally rejected my position that the vaccine precipitated the PVC’s. Couldn’t happen, he said. I asked if he knew of a study that had investigated that. He did not. So, after a year of treatment I still don’t have a diagnosis and we’re just treating the symptom. BTW, I’m a retired pharmacist with a bit of knowledge regarding drugs and diseases.