View Single Post
Old 10-17-2018, 12:55 PM
Damuri Ajashi is offline
Join Date: Jul 2006
Posts: 20,317
Originally Posted by iiandyiiii View Post
This isn't how statistics work. Like it or not, there's a lot of science to statistics, and things like risk ratio and confidence level are based in good science. That's how we can extrapolate from polling and margins of error to the feelings of the whole country. It's no more "statistical noise" than a poll of ~800 people is. It's by no means a perfect and conclusive measure, but it's a piece of data that provides useful information.
It provides about as much useful "data" as the one shooting of an Asian boy does on police inclination to shoot Asian boys. The ONE statistician that they used for this backed away from the "study"

So I understand something about statistics. Do you?

What pro-publica did was engage in P-mining or data dredging.

If you pick through a data set looking for a particular relationship long enough, you will find a statistically significant relationship somewhere. Statistical significance just tells us that there is a 95% chance that a result is not random. It doesn't tell you what might cause the non-randomness. That is what a study does. N The Pro-Publica "study" just throws the number out there with no context other than "look, cops killed a disproportionate of black people" with the implication that it is racism.

You ever play roulette and sometimes you see a long line of red or black numbers. If you just took that clump of red numbers you would might be able to reach the conclusion that red is more likely to be rolled than a black number. But its just noise.

Maybe this is true. I'd certainly be interested in looking further into it.
Why? It's statistical noise