One person I'd like to see uncanceled is Amy Cuddy. I think there's a there there with power posing that can be teased out with more subtle experimental design. We've overcorrected on reproducibility crisis and forgotten that absence of evidence is not evidence of absence.
Conversation
Replying to
absence of evidence is evidence of absence, though
lesswrong.com/posts/mnS2WYLC
oyhus.no/AbsenceOfEvide
1
2
Yeah, this is a narrow bayesian argument that simply does not apply outside conditions that are sort of ergodic in the behavior space. The general principle is more broadly applicable. You need to establish exhaustiveness of coverage to claim otherwise.
1
1
The basic analysis is Taleb's black swan argument, and more sophisticated ones are possible. A general problem with bayesian epistemology is that you cannot know if the distribution you observe to form a prior is in fact typical, and if it is, must necessarily remain so for ever
1
3
But if you go meta, isn't the fat tails black swan situation the more rare case?
If x has no evidence wouldn't it be more accurate to update a bit towards absence in the general case?
1
1
You can run a meta distribution of distribution types...
1
The problem isn’t purely a statistical one. Once an exception to an assertion is found, a possible world where it is the default becomes plausible as well. You can breed black swans to be dominant. If power posing works under some conditions, the norm could spread to dominance.
“Absence of evidence != evidence of absence” is a broader epistemological heuristic than the statistical one. Think infinitude of primes, counterexamples to seemingly universal truths etc. It is a category error to use narrow statistical analysis to refute the broader principle.
1


