I'd like to be more conscious about my Bayesian-type updates of my beliefs based on general accounts of what people say. So far, I've started using a rule-of-thumb that somebody telling me something is so is worth approximately 1 decibel of belief (1/3rd of a bit)
You don't really believe that. When someone introduces themselves to you under normal circumstances, your probability distribution about what their name is concentrates immensely in a few seconds. See this paper by Robin Hanson for this point and related discussion.
I think The Invisible Gorilla has some statistics, and some googling shows there are quite a few papers. Would be interesting to see with what kind of correlations you come up with.
Interesting and important question, although I tend to agree with AlexSchell that the claim made about taking eyewitness testimony seriously is unrealistic.
Not sure why someone downvoted it...
After reading and taking classes in judgement i really am curious why we put any stock in eyewitness testimony. The only thing i can thing of is that it might be the only type of evidence in a trial but should that be suffienct to convict someone?
This mentions some of the limitations of eyewitness testimony; does anybody here have any references giving any hard numbers about how reliable eyewitness accounts are, under any given circumstances?
I'd like to be more conscious about my Bayesian-type updates of my beliefs based on general accounts of what people say. So far, I've started using a rule-of-thumb that somebody telling me something is so is worth approximately 1 decibel of belief (1/3rd of a bit); evidence, but about the weakest evidence possible, nulled by any opposing accounts, and countered by any more substansive evidence.
If possible, I'd like to know exactly how reliable such testimony tends to be in one particular set of circumstances - time since the thing being reported, level of emotional involvement, etc - to use as a baseline, and at least roughly how strongly such factors change that. (I'll actually be very surprised if this particular set of data currently exists in ready form - but I'll be satisfied if I can get even order-of-magnitude approximations, so that I know whether or not the rules-of-thumb I end up using are at least within plausible spitting distance.)