1) “This means that probability values that are 10 times higher are 10times common.” Shouldn’t it be probabilities that are 10 times lower are 10 times more common?
2) In the section on speculative bias, you say that “grounded and speculative threats are identical in all ways, except that the speculative threats are much more uncertain”. Shouldn’t the frequencies of grounded actual (blue) and speculative actual (yellow) look the same then?
Point 2 is something that confused me at first as well. The reason they are different is that we are looking at the performance of the top apparent threat. If we were perfectly good estimators, this would be the same as the top actual threat, but we aren’t: the threat of the top pick is generally going to be lower than the top actual threat due to uncertainty and the curse.
For the grounded estimator, the process of ranking threats gives useful information, and it means that the top threat picked is much higher than you would get from picking at random. Whereas in the speculative case, we are much closer to just picking at random, and thats reflected in the yellow curve which looks a lot like the power law sampling we are drawing from.
Questions for clarification:
1) “This means that probability values that are 10 times higher are 10 times common.” Shouldn’t it be probabilities that are 10 times lower are 10 times more common?
2) In the section on speculative bias, you say that “grounded and speculative threats are identical in all ways, except that the speculative threats are much more uncertain”. Shouldn’t the frequencies of grounded actual (blue) and speculative actual (yellow) look the same then?
Point 1 was a typo, thanks for pointing it out!
Point 2 is something that confused me at first as well. The reason they are different is that we are looking at the performance of the top apparent threat. If we were perfectly good estimators, this would be the same as the top actual threat, but we aren’t: the threat of the top pick is generally going to be lower than the top actual threat due to uncertainty and the curse.
For the grounded estimator, the process of ranking threats gives useful information, and it means that the top threat picked is much higher than you would get from picking at random. Whereas in the speculative case, we are much closer to just picking at random, and thats reflected in the yellow curve which looks a lot like the power law sampling we are drawing from.