John Goldin sends in this, from an interview with Alan Dershowitz:
Q: The lawyerly obligation to not change your mind, to defend a position right or wrong–do you find that it seeps over into the rest of your life?
A: No, it doesn’t because I’m a professor first, and as a professor I’m always changing my mind. I mean, my students go crazy in my class because I’m the most orthodox Bayesian in the world. [Bayesian probability theory is a way of modeling how the human mind reasons about the world. It assumes that people have prior beliefs about the probability of a given hypothesis and also beliefs about the probability that the hypothesis, if true, would generate the evidence they see. Taken together, these beliefs determine how people update their faith in a hypothesis in light of new evidence.] I do everything based on Bayes analysis, and Bayes analysis is always based on shifting probabilities and constantly changing and being adaptive and fluid.
Although, who am I to argue with a guy who got Jeremy Irons off on a murder charge?
It's ironic that Dershowitz calls himself a Bayesian when, during O.J. Simpson's trial, he's made the staggering claim that since less than 1 in 1000 women who are abused by their husbands are subsequently killed by them, the fact that Simpson had abused his wife is irrelevant in terms of raising the probability that he has also killed her. See here: http://expertvoices.nsdl.org/cornell-info204/2010…
You're thinking of Ron Silver. People confuse them all the time.
Nice cheap shot, Andrew. So now there's a 'reductio ad Dershowitz'?
Rick: Hey, why is that a cheap shot? Dershowitz saying he does everything based on Bayes analysis is like me saying I do all my statistics based on strict constitutionalism.
Gelman states:
Rick: Hey, why is that a cheap shot? Dershowitz saying he does everything based on Bayes analysis is like me saying I do all my statistics based on strict constitutionalism.
Dershowitz can distinguish between having a coherent set of subjective probabilities and announcing them to the world. People need to be modeled as optimizing agents with goals and motivations. In Dershowitz's case, getting an obviously guilty client off is his goal, and he will offer any semi-credible (or incredible) line of BS to further his goal. Doesn't mean his subjective probability of the guilt of his client isn't darn near close to one. So while Dershowitz probably does take in new information and integrate it (presumably as close to Bayes theorem as he can handle), it doesn't mean he'll let it from carrying out his job (the updated probabilities will affect how he carries it out, however).