You sure? 90% to 30% isn’t really an orders of magnitude difference. Of course, taking that into account, “beyond a reasonable doubt” should be something like “at least 99.9% probability of guilt” (or whatever incidence of innocent people you’re willing to throw in prison / execute), in which case a shift from 90% to 30% would be a shift from “not guilty” to “slightly more not guilty”.
I don’t understand what you’re asking if I’m sure about. I’m saying that given the premise that brazil84 accepts the forensic argument against Knox and Sollecito’s guilt, it does not make sense to update his probability by so little.
He already stated that given that he accepts the argument, he would update to 30% confidence, which conveniently frees us from the need to consider the probability outside the argument.
It’s possible that he really meant “given that I accept the argument at low confidence, as opposed to not at all, as I currently do,” but that would make his statement awfully misleading, and brazil84 has demonstrated too little familiarity with Bayesian reasoning for me to read his statements with that much charity.
He should mean “given that I assign the argument a comparable credence to other things I would state that I accept”—which shouldn’t be 1. I think this probably shouldn’t be a big enough difference to matter (I missed the “given...” in the context, in my response), but it would be interesting to update the numbers to verify.
You sure? 90% to 30% isn’t really an orders of magnitude difference. Of course, taking that into account, “beyond a reasonable doubt” should be something like “at least 99.9% probability of guilt” (or whatever incidence of innocent people you’re willing to throw in prison / execute), in which case a shift from 90% to 30% would be a shift from “not guilty” to “slightly more not guilty”.
These should really be different standards.
I don’t understand what you’re asking if I’m sure about. I’m saying that given the premise that brazil84 accepts the forensic argument against Knox and Sollecito’s guilt, it does not make sense to update his probability by so little.
Aha—I’d misread you as saying that the update was too big, rather than too little.
He could accept it at less than full confidence.
He already stated that given that he accepts the argument, he would update to 30% confidence, which conveniently frees us from the need to consider the probability outside the argument.
It’s possible that he really meant “given that I accept the argument at low confidence, as opposed to not at all, as I currently do,” but that would make his statement awfully misleading, and brazil84 has demonstrated too little familiarity with Bayesian reasoning for me to read his statements with that much charity.
He should mean “given that I assign the argument a comparable credence to other things I would state that I accept”—which shouldn’t be 1. I think this probably shouldn’t be a big enough difference to matter (I missed the “given...” in the context, in my response), but it would be interesting to update the numbers to verify.