"Bayesian statistics" is regarded as an advanced topic, which it kind of is, because it's usually taught only to undergraduate mathematicians. To a lot of people, it's the ceiling of statistical thinking. But I think that's a failure of our education system-- it should be the floor.

This makes me want to pull my face down while screaming from frustration, because so much critical thinking is improved by this.

The kicker is that it's a really simple subject that we can teach to children. In the same stroke we teach "the difference between median and mode", we can start introducing conditional knowledge. The average temperature on Earth might be 60F, but if you know it's midnight in December in Vermont, maybe you'd estimate it's lower.

We express this as P(A|B) (the probability of event A given we know B) and E(A|B) (the estimated value of A given we know B). So, one might express "the expected temperature outside given it's midnight in December in Vermont" as E(temperature outside | it's midnight in December in Vermont).

The BIG thing is that so much of the racist ideology that abuses statistics just falls apart with the most elementary application of conditional (Bayesian) thinking.

There's this classic racist statistic that far-righties love, "Black people make up 13% of the population but commit 50% of violent crimes".

They try to draw racist conclusions from that, but we can dismiss it for a number of reasons (they're arrested for 50% of crimes, it's a statistic from the FBI, this falls away when accounting for socioeconomic factors, etc).

Assuming you're familiar with those arguments, I'd like to add some more:

• This gives us P(race | crime) if you trust the FBI (lol), but if you wanted to enforce demographic-based policing you'd want P(crime | race). And you simply can't reverse that. You can't generally find P(A|B) from P(B|A).
  • Btw, we do have P(crime | sex), and it would tell us we should pre-emptively police men. Somehow, the far-right miss the conclusion that women should necessarily have more rights than men.
  • Also, we already have race-based policing. It's very common, and in excess of what any statistics might suggest is proportional.

This next part follows a pattern of "proof by contradiction", which means assuming the statement you want to disprove, and showing how it contradicts itself. This means assuming a racist statistic.

So, even if we trusted the FBI's P(race | crime) as fully capturing reality, and even if we could infer P(crime | race) from that, and even if we had P(crime | black) > P(crime | not black and not white) > P(crime | white), it all falls apart:

• P(criminal | black) would still be a very low number! Most people are not committing crimes! Even people who have committed crimes are not constantly committing crimes.
• You still have other knowledge to condition yourself on. Race would still likely be one of the lowest information-theoretically useful factors to consider. (Information theory is something we can teach to kids btw.)
• To enforce this, this ultimately requires an overarching Government with a very epistemically robust and efficient information gathering system. Which:
  • Is something we can readily dismiss, lol, and
  • See the above point, that race would not be the most information theoretically useful factor, and
  • Something something "Minority Report" something something "torment nexus"

And again, Bayesian thinking is just the floor. The reason racism has never been empirically or statistically or theoretically justified is because it doesn't hold up to facts and logic.

thank u for coming to my ted talk