Jouni writes that SAS has now a Bayesian module. I agree with Jouni that “The Bayesian probability reflects a person’s subjective beliefs” is not really the kind of phrase you expect to hear from a modern practicing Bayesian methodologist. This definition would immediately invalidate the use of Bayesian methods in any field of science, I’d think.” Well, maybe not in sociology . . .
>>"The Bayesian probability reflects a person's subjective beliefs"
Isn't that a tautology too, after all many of a person's belief's are learned and therefore somewhat Bayesian in their processes. Running some Bayesian probability on a computer is just augmented externally whatver goes on in our heads.
Actually you could say the same for all statisics, its just less isomorphically coherent or something.
Meika,
I think the point is that when presenting scientific results, many people prefer these results to be objective rather than biased by your own beliefs. I think this is one of the main arguments people use against bayesian methods and why many people, I believe, look at how sensitive their results are to their prior (prior beliefs).
All science is subjective in the sense that we have to make choices. Certainly statistics is subjective in the choice of information to include, the choice of model, method, or algorithm to use, and so forth. In this context, I do not see Bayesian statistics as more subjective than other issues.
On the specific point of subjectivity of Bayesian probability distributions, see the examples in Chapter 1 of Bayesian Data Analysis with assignment of probability distributions (for football scores and record linkage) based on empirical data.