Thread by @roberto_blanko

#Covid19 myths Ep. 2: Thread

(1)
Self-proclaimed "experts" are spreading this type of chart where I added lovely remarks.
From this chart you can calculate e.g. monthly Excess Deaths. Here for Sweden.

But does it allow you to calculate the Excess Deaths correctly?
Spoiler: NO

(2)
In general, Excess Deaths are the difference between the observed deaths to an average death count in a defined time period.

So, you need 2 values:
#1: The observed value. This value is correct on the chart
#2 The average or baseline. This value is wrong

(3)
Authorities and real experts would calculate the monthly average of deaths by averaging the value of the respective month of the previous 3 years.
But in this chart a rolling average of the previous 36 months is used.

To illustrate this difference here is an example:

(4)
For April 2020 the correct way to calculate the baseline is the average of Apr 17, Apr 18 and Apr 19. It really is that simple.

The chart however uses a rolling average of the previous 36 months: Apr 18, May 18, Jun 18,…,Nov 19, Dec 19, Jan 20,…

(5)
The rolling average method is unsuitable for analyzing parameters with an uneven distribution over the year, such as monthly temperatures or deaths.

This table shows monthly deaths per million in SWE from 2004 to 2019:

(6)
As you can see deaths decrease over time AND more people die during winter, independent of the severity of the flu.
So, the usage of a rolling average over 36 months which is used in the graph leads to a baseline which is too low for winter and too high for summer months.

(7)
Let’s do a quick thought experiment to highlight the mistake of the chart’s average calculation:

Let’s say for the last 3 Christmases I got my kids 120 presents each year in December, because Christmas is in December (I know: breaking news!)

(8)
We take Christmas because it’s an event which is distributed unevenly over the year and 120 Christmas presents because I'm a great dad and it allows an easy calculation.

Based on the experiences of the past 3 years, let’s compare the methods by answering 2 simple questions.

(9)
1st question: How many Christmas presents will my kids get in October 2020?

Educated guess: 0, because on average they got 0 Christmas presents for the last 3 October months (Oct 17, Oct 18, Oct 19)

Using the chart’s logic: 10, because they got 360 over the last 36 months

(10)
2nd question: How many Christmas presents will my kids get in December?

Educated guess: 120, because on average they got 120 for the last 3 December months

According to the chart’s logic: 10, because they got 360 Christmas presents over the last 36 months

(11)
Summary: The graph’s average calculation is utter NONSENSE

If you see this graphic from now on, always remember the logical fallacy here:

According to the graph’s logic, I am a bad father, because this year I won’t give my kids any Christmas presents in October

(12)
General information:
Excess mortality means that more people have died within a certain time period than expected.

This method is used e.g. to determine the severity of an influenza season by calculating the Excess deaths.

Here is the statement by the creator of the chart that the 36 months rolling average is used:

https://twitter.com/FrankfurtZack/status/1305498636437131264?s=19

(13)
Improvement

The discussed chart mixes 2 effects:
#1 Overall Mortality decreases over time
#2 Peaks in mortality occur over time
By mixing effects the graph has no significance and is visually misleading.

A much better data display could look like this:

(14)
Improvement Pt. 2

To better display Excess deaths I chose an approach where the monthly distribution of each year is displayed with additional information about the peak values.

The death peak in April 20 during the #Covid pandemic is an absolute exception