Coherence as an ideal of rationality

Probabilistic coherence is not an absolute requirement of rationality; nevertheless, it is an ideal of rationality with substantive normative import. An idealized rational agent who avoided making implicit logical errors in forming his preferences would be coherent. In response to the challenge, recently made by epistemologists such as Foley and Plantinga, that appeals to ideal rationality render probabilism either irrelevant or implausible, I argue that idealized requirements can be normatively relevant even when the ideals are unattainable, so long as they define a structure that links imperfect and perfect rationality in a way that enables us to make sense of the notion of better approximations to the ideal. I then analyze the notion of approximation to the ideal of coherence by developing a generalized theory of belief functions that allows for incoherence, and showing how such belief functions can be ordered with regard to greater or lesser coherence.Many people influenced the present version of this essay. Ban van Frassen, Richard Jeffrey, David Lewis, Mike Thau, and Alan Hájek provided extensive and invaluable written comments on the entire essay. Mark van Roojen provided helpful comments on Sections 3 and 4. Mike Than and John Barker provided essential aid when I was formulating the proofs in section 6. Finally, I am grateful for valuable discussion of the essay with Ned Hall, Fiona Cowie, Jim Woodward, David Hilbert, and Frank Arntzenius.