Prior Probabilities: an Information-theoretic Approach

Abstract. General theoretical principles that enable the derivation of prior probabilities are of interest both in practical data analysis and, more broadly, in the foundations of probability theory. In this paper, it is shown that the general rule for the assignment of priors proposed by Jeffreys can be obtained from, and is logically equivalent to, an intuitively reasonable information-theoretical invariance principle Some of the implications for the priors proposed by Hartigan [1], Jaynes [2], and Skilling [3], are also discussed.