by Bayesian ‘old evidence’ problems Why is the subjective Bayesian supposed to have an old evidence problem? The allegation … goes like this: If probability is a measure of degree of belief, then if an agent already knows that e has occurred, the agent must assign P(e) the value 1. Hence P(e|H) is assigned a value of 1. But this means no Bayesian support accrues from e. For if P(e) = P(e|H) = 1, then P(H|e) = P(H). The Bayesian condition for support is not met … How do subjective Bayesians respond to the charge that they have an old evidence problem? The standard subjective Bayesian response is … “The Bayesian interprets P(e|H) as how likely you think e would be were h to be false” … But many people — Bayesians included — are not too clear about how this
Topics:
Lars Pålsson Syll considers the following as important: Theory of Science & Methodology
This could be interesting, too:
Lars Pålsson Syll writes Randomization and causal claims
Lars Pålsson Syll writes Race and sex as causes
Lars Pålsson Syll writes Randomization — a philosophical device gone astray
Lars Pålsson Syll writes Keynes on the importance of ‘causal spread’
Bayesian ‘old evidence’ problems
Why is the subjective Bayesian supposed to have an old evidence problem?
The allegation … goes like this: If probability is a measure of degree of belief, then if an agent already knows that e has occurred, the agent must assign P(e) the value 1. Hence P(e|H) is assigned a value of 1. But this means no Bayesian support accrues from e. For if P(e) = P(e|H) = 1, then P(H|e) = P(H). The Bayesian condition for support is not met …
How do subjective Bayesians respond to the charge that they have an old evidence problem? The standard subjective Bayesian response is …
“The Bayesian interprets P(e|H) as how likely you think e would be were h to be false” …
But many people — Bayesians included — are not too clear about how this “would be” probability is supposed to work.
Yes indeed — how is such a “would be” probability to be interpreted? The only feasible solution is arguably to restrict the Bayesian calculus to problems where well-specified nomological machines are operating. Throwing a die or pulling balls from an urn is fine, but then the Bayesian calculus would of course not have much to say about science …