The system presented by the author in The Logical Foundations of Statistical Inference suffered from certain technical difficulties, and from a major practical difficulty; it was hard to be sure, in discussing examples and applications, when you had got hold of the right reference class. The present paper, concerned mainly with the characterization of randomness, resolves the technical difficulties and provides a well structured framework for the choice of a reference class. The definition of randomness that leads to this framework (...) is simplified and clarified in a number of respects. It resolves certain puzzles raised by S. Spielman and W. Harper in their contributions to Profiles: Henry E. Kyburg, Jr. and Isaac Levi 1982). (shrink)

First, a comment on a pessimistic note: Salmon says we can't be sure there is any such thing as inductive inference: in demanding that some explanations have the form of correct inductive inferences, “we may be laying down a requirement which cannot be fulfilled.” To doubt that we can fulfill that requirement is to doubt that we can formalize inductive logic. It may be true, but why begin the fight by throwing in the sponge? It is also true that there (...) are difficulties involved in formalizing acceptance rules for inductive conclusions, but these difficulties may be overcome. It is false to claim that we ‘cannot claim to know exactly’ what these acceptance rules are. Only two years ago I did just that. It was true that my claim was shown to be erroneous a year later, but that fate can befall almost any worthwhile claim. (shrink)

Quantities are naturally viewed as functions, whose arguments may be construed as situations, events, objects, etc. We explore the question of the range of these functions: should it be construed as the real numbers (or some subset thereof)? This is Carnap's view. It has attractive features, specifically, what Carnap views as ontological economy. Or should the range of a quantity be a set of magnitudes? This may have been Helmholtz's view, and it, too, has attractive features. It reveals the close (...) connection between measurement and natural law, it makes dimensional analysis intelligible, and explains the concern of scientists and engineers with units in equations. It leaves the philosophical problem of the relation between the structure of magnitudes and the structure of the reals. What explains it? And is it always the same? We will argue that on the whole, construing the values of quantities as magnitudes has some advantages, and that (as Helmholtz seems to suggest in "Numbering and Measuring from an Epistemological Viewpoint") the relation between magnitudes and real numbers can be based on foundational similarities of structure. (shrink)

Bishop Butler, [Butler, 1736], said that probability was the very guide of life. But what interpretations of probability can serve this function? It isn’t hard to see that empirical (frequency) views won’t do, and many recent writers-for example John Earman, who has said that Bayesianism is “the only game in town”-have been persuaded by various dutch book arguments that only subjective probability will perform the function required. We will defend the thesis that probability construed in this way offers very little (...) guidance, dutch book arguments notwithstanding. We will sketch a way out of the impasse. (shrink)

In, Levi has laid out the issues involving chances, frequencies, and direct inference with admirable precision. Nevertheless, puzzles remain. The chief puzzle to which I wish to draw attention is this: Under certain circumstances, we can combine knowledge of chances and knowledge of frequencies to yield new knowledge of chances. If Petersen is “drawn at random” from among Swedes, and we also know that the proportion of Protestants among Swedes is 0.9, then we can say that the chance that Petersen (...) is a Protestant is 0.9. But if we apply this principle generally, we are led to generally trivial results: direct inference yields probabilities constrained only to lie in a broad interval like [0, 1]. On the other hand, if we can't always combine knowledge of chances and knowledge of frequencies to get new knowledge of chances, or if this knowledge can be overridden by other considerations, how do we know when we can usefully apply direct inference? (shrink)

Charles Morgan has argued that nonmonotonic logic is ``impossible''. We show here that those arguments are mistaken, and that Morgan's preferred alternative, the representation of nonmonotonic reasoning by ``presuppositions'' fails to provide a framework in which nonmonotonic reasoning can be constructively criticised. We argue that an inductive logic, based on probabilistic acceptance, offers more than Morgan's approach through presuppositions.

Since Ramsey, much discussion of the relation between probability and belief has taken for granted that there are degrees of belief, i.e., that there is a real-valued function, B, that characterizes the degree of belief that an agent has in each statement of his language. It is then supposed that B is a probability. It is then often supposed that as the agent accumulates evidence, this function should be updated by conditioning: BE(·) should be B(·E)/B(E). Probability is also important in (...) classical statistics, where it is generally supposed that probabilities are frequencies, and that inference proceeds by controlling error and not by conditioning. I will focus on the tension between these two approaches to probability, and in the main part of the paper show where and when Bayesian conditioning conflicts with error based statistics and how to resolve these conflicts. (shrink)

Kyburg proposes the following test for the simplicity of a theory: the complexity of a theory is measured by the number of quantifiers that occur in the set of statements comprising the theory. (staff).

Quantitative laws are more typical of science than are generalizations involving observational predicates, yet much discussion of scientific inference takes the confirmation of a universal generalization by its instances to be typical and paradigmatic. The important difference is that measurement necessarily involves error. It is argued that because of error laws can no more be refuted by observation than they can be verified by observation. Without much background knowledge, tests of a law mainly provide evidence for the distribution of errors (...) of measurement of the quantities involved. With more background knowledge, the data may contribute either to our knowledge of the error distributions, or to the grounds we have for accepting or rejecting the law. With enough background knowledge, data may verify as well as refute laws. (shrink)

There are a number of reasons for being interested in uncertainty, and there are also a number of uncertainty formalisms. These formalisms are not unrelated. It is argued that they can all be reflected as special cases of the approach of taking probabilities to be determined by sets of probability functions defined on an algebra of statements. Thus, interval probabilities should be construed as maximum and minimum probabilities within a set of distributions, Glenn Shafer's belief functions should be construed as (...) lower probabilities, etc. Updating probabilities introduces new considerations, and it is shown that the representation of belief as a set of probabilities conflicts in this regard with the updating procedures advocated by Shafer. The attempt to make subjectivistic probability plausible as a doctrine of rational belief by making it more flowery — i.e., by adding new dimensions — does not succeed. But, if one is going to represent beliefs by sets of distributions, those sets of distributions might as well be based in statistical knowledge, as they are in epistemological or evidential probability. (shrink)

The dominant argument for the introduction of propensities or chances as an interpretation of probability depends on the difficulty of accounting for single case probabilities. We argue that in almost all cases, the``single case'' application of probability can be accounted for otherwise. ``Propensities'' are needed only intheoretical contexts, and even there applications of probability need only depend on propensities indirectly.

We examine the notion of conditionals and the role of conditionals in inductive logics and arguments. We identify three mistakes commonly made in the study of, or motivation for, non-classical logics. A nonmonotonic consequence relation based on evidential probability is formulated. With respect to this acceptance relation some rules of inference of System P are unsound, and we propose refinements that hold in our framework.