Discussion about this post

User's avatar
Sian's avatar

" if they intuit that their personal risk of murder is ~zero, they’ll respond against anything that worsens their odds, no matter how slightly. "

I think we've just come onto the reason why people react to mass shootings the way they do. It's disruptive to their sense of risk. They know they're generally immune to 'normal' murder, but the mass shooter, rare as it is, disrupts the algebra. This experience is inherently disconcerting.

Expand full comment
Eidein's avatar

> But that assumes the risk is evenly distributed, and of course it is not. Statistics change based on how far you zoom in. This is an example of Simpson’s paradox. An easy way to demonstrate it is to zoom out as far as possible. The worldwide average homicide rate is roughly 6.1 murders per 100,000 people. But that doesn’t tell you anything about your own personal murder risk. You can keep going and make it even more useless — zoom out further to the whole universe, whose murder rate asymptotically approaches zero. Equally useless for telling you whether you should walk outside alone at night.

This is one of my Big Ideas, and I'm sure I've shared it here before.

ALL statistics about "The United States" are meaningless, because they are statistical aggregates over a highly heterogenous population that do not generalize to any individuals in that population

So, to make up some illustrative numbers, you might hear that the murder rate in the US is 10/100k. So you might think, I have a one in ten thousand chance, per year, of being murdered.

Except, inevitably, the way they got that 10/100k murder rate was by averaging over a bunch of sleepy quiet areas where the murder rate is essentially 0, and then a bunch of violent inner city neighbourhoods where the murder rate is 200/100k. The average is technically correct, but meaningless if your goal is to evaluate your own personal risk of being murdered.

This is true of _**ALL**_ stats in the US, on all subjects, always. You have to drill down and be way more specific and targetted in your statistical analysis before the output becomes useful

Expand full comment
4 more comments...

No posts