Read on Twitter
How to mislead with charts in one easy step.
(this chart was cited as support that there is a relation between economic performance and COVID incidence)
(this chart was cited as support that there is a relation between economic performance and COVID incidence)
Step 1/ Use a presentational format borrowed from a different field that is completely inappropriate for the message (and somewhat arbitrary) but looks like it supports your argument
The "correct" way of showing a relationship is by ... showing a relationship. But of course if you draw a trend line it shows precisely ... nothing
so instead use some technique borrowed from AI and separate those into cluster, and if we are lucky that looks like we are onto something
cut if you do a cluster analysis on that data the only reason why you'd do it that way is because you have something to show; without an agenda you'd have at least those
1/ the horrible four FR/UK/IT/ES
2/ the fab two (SK, ID)
3/ Belgium
4/ Everyone else
1/ the horrible four FR/UK/IT/ES
2/ the fab two (SK, ID)
3/ Belgium
4/ Everyone else
we can argue whether we split the Americas and Sweden into their own cluster, but for the initial purpose it does not matter: the initial interpretation is solely driven by 1/ and 2/ which is 6 countries in total; for everyone else this relationship does not exist
Now 1/ is also the countries who locked down hard but late, so this might explain why they lost a lot of buck without bang; 2/ is South Korea essentially; Indonesia might or might not have correct numbers
it is funny by the way -- if you put what I assume to be the same data on a log scale it looks somewhat more linear and connected
(and guys, really -- log is not ln; log is log10)
