I agree it is also bayesian evidence for that! My current guess is it was more in the other direction, as in general I think there are more people breaking rules for bad reasons than for good reasons, but I’m not that confident, and would be interested in hearing from someone who disagreed about this (in specific or in general) and why.
I think that there isn’t just one bag of people breaking rules, and some number of the marbles in that bag are “for good reasons” and some number “for bad reasons.” I think there are clusters, and types, and certain kinds of rule-breaking are predictive of other kinds of rule-breaking.
I think that me not wearing shoes at university is evidence that I might also disdain sports, but not evidence that I might steal.
I think that trying to think in terms of “for bad reasons” and “for good reasons” as two flavors in one bucket is likely to lead one to make wrong updates. Like, the model is oversimplified and causes fearful swerves.
(In my usage, which may or may not be standard, if something feels like a Bayesian update for each of two different mutually exclusive directions, you sort of cancel out the overlap and then only refer to the net remainder as the thing for which you have Bayesian evidence. Like, if it independently seems like a 3 update to the west, and a 5 update to the east, when you consider each separately, you say “a Bayesian update of 2 to the east” or similar.)
I agree it is also bayesian evidence for that! My current guess is it was more in the other direction, as in general I think there are more people breaking rules for bad reasons than for good reasons, but I’m not that confident, and would be interested in hearing from someone who disagreed about this (in specific or in general) and why.
I think that there isn’t just one bag of people breaking rules, and some number of the marbles in that bag are “for good reasons” and some number “for bad reasons.” I think there are clusters, and types, and certain kinds of rule-breaking are predictive of other kinds of rule-breaking.
I think that me not wearing shoes at university is evidence that I might also disdain sports, but not evidence that I might steal.
I think that trying to think in terms of “for bad reasons” and “for good reasons” as two flavors in one bucket is likely to lead one to make wrong updates. Like, the model is oversimplified and causes fearful swerves.
(In my usage, which may or may not be standard, if something feels like a Bayesian update for each of two different mutually exclusive directions, you sort of cancel out the overlap and then only refer to the net remainder as the thing for which you have Bayesian evidence. Like, if it independently seems like a 3 update to the west, and a 5 update to the east, when you consider each separately, you say “a Bayesian update of 2 to the east” or similar.)