Scientific arguments often attribute the credit or blame for particular experimental or observational results to fragments of theories rather than to the whole of them. The thesis of Theory and Evidence (Glymour 1980) is that there are structural criteria that account for at least part of the distribution of credit and blame; the book also attempts a tentative formal characterization of one structural criterion. David Christensen (1983) has produced a series of counterexamples which show that the formal characterization is untenable. The question remains whether the failure is due to an oversight in the formal theory or to the falsity of the very idea that there are structural criteria for evidential relevance. Since the counterexamples are uniformly eliminated by a single formal principle, I incline to the first answer.

This paper attempts to define Exploratory Data Analysis (EDA) more precisely than usual, and to produce the beginnings of a philosophy of this topical and somewhat novel branch of statistics.

A data set is, roughly speaking, a collection of k-tuples for some k. In both descriptive statistics and in EDA, these k-tuples, or functions of them, are represented in a manner matched to human and computer abilities with a view to finding patterns that are not “kinkera”. A kinkus is a pattern that has a negligible probability of being even partly potentially explicable. A potentially explicable pattern is one for which there probably exists a hypothesis of adequate “explicativity”, which is another technical probabilistic concept. A pattern can be judged to be probably potentially explicable even if we cannot find an explanation. The theory of probability understood here is one of partially ordered (interval-valued), subjective (personal) probabilities. Among other topics relevant to a philosophy of EDA are the “reduction” of data; Francis Bacon's philosophy of science; the automatic formulation of hypotheses; successive deepening of hypotheses; neurophysiology; and rationality of type II.

When natural selection theory was presented, much active philosophical debate, in which Darwin himself participated, centered on its hypothetical nature, its explanatory power, and Darwin's methodology. Upon first examination, Darwin's support of his theory seems to consist of a set of claims pertaining to various aspects of explanatory success. I analyze the support of his method and theory given in the Origin of Species and private correspondence, and conclude that an interpretation focusing on the explanatory strengths of natural selection theory accurately reflects neither Darwin's own self-consciously held views, nor the nature of his support. Darwin's methodological and philosophical arguments were at once consistently empiricist and more sophisticated than such interpretations credit to him.

Theories are conceptualized as predictors of phenomena using computable functions acting on prior world information. Formally, the concept of bounded prior world recursive function is defined and used as a candidate for a potential theory. An artificial world of fact is then constructed for which there exist two inconsistent best theories, in that they cannot be improved upon, and these theories are maximally inconsistent in that every best theory is a compromise. It is argued that in such a world the scientific objective would be to find these maximally inconsistent theories. The construction is motivated as an attempt to illustrate the status of theory in the social sciences and in economics, in particular.

In (1981), 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?

The question is raised, what is going on when there is a tendency for something to happen, but it does not happen. A conspicuous case is that in which there is an equilibrium between opposing forces. One answer, given, for example, by Anscombe and Geach, Harré, and Bhaskar, is that such tendencies are in a sense real, though not manifest. The other answer, given, for example, by Braithwaite and Mackie, is that statements of tendencies are conditional statements about what is manifest but ones in which some antecedents are not specified. The first of these answers is criticized and the second defended.

Glymour's “bootstrap” account of confirmation is designed to provide an analysis of evidential relevance, which has been a serious problem for hypothetico-deductivism. As set out in Theory and Evidence, however, the “bootstrap” condition allows confirmation in clear cases of evidential irrelevance. The difficulties with Glymour's account seem to be due to a basic feature which it shares with hypothetico-deductive accounts, and which may explain why neither can give a satisfactory analysis of evidential relevance.

This paper discusses “Russell's hypothesis” that the world sprang into existence five minutes ago. The three most widely accepted “solutions” to the Russell's hypothesis problem are shown to be unsatisfactory.

Two main points of interest are involved in the paper's discussion. First, I show all the widely accepted “solutions” to be unacceptable by using the same device—alternatives to Russell's hypothesis. The device, which has never previously been applied to this problem, is a familiar one in discussions of the problem of induction. (Parallels between these two problems are sketched at the paper's end.) Second, the discussion introduces the new concept of crucial consequences for hypotheses that are otherwise different in only trivial linguistic ways.

What is the relation of epistemology, understood as the study of the evaluation of knowledge claims, and empirical psychology, understood as the study of the causal generation of a person's beliefs? Quine maintains that the relation is one of “mutual containment”.

To ground judgments of credal probability on knowledge of chance via direct inference, one should appeal to the information about chances available relative to the most specific description known to be true of the trial event.

Thus, to obtain a judgment of credal probability concerning the hypothesis that coin a landed heads at t given that it is known that at t it is known that a was tossed by Levi in 728 Philosophy Hall, the pertinent knowledge of chances concerns the chances of coin a landing heads on a toss of coin a which is (i) a toss at t, (ii) a toss in 728 Philosophy Hall and (iii) a toss by Levi. In many contexts, appeal is made to knowledge of the chance of coin a landing heads on a toss. This appeal, when legitimate, takes for granted that the additional information about the specific tossing listed in (i), (ii) and (iii) is stochastically irrelevant to landing heads and, hence, may be ignored. Such an assumption is a substantive presupposition about chances and not an a priori, self evident or pragmatic principle of methodology.

Larry Wright's analysis of teleological description of goal-directed behavior, though ingenious and insightful, errs in the following ways: it incorrectly claims that intentional human action exhibits consequence-etiology, making it impossible, contrary to his claim, for reference to consequence-etiology to be metaphorically transmitted to teleological descriptions of nonhuman behavior; it does not remove the threat of reverse causation for nonhuman behavior; it assumes in the face of contrary evidence that reference to purpose drops out in metaphorical extension; and it cannot account for unrequired behavior directed toward impossible goals.

Philosophy of science may indeed perform a valuable normative function through historical analyses of characteristic errors, as Bechtel (1982) argues. But considerations of the two types of errors elected for this role, and harder looks at some of the historical examples prosecuted for these errors, disclose few helpful hints beyond the wisdom provided in Chapter I of freshman science texts coupled to a clear view of the future. To avoid churlishly spurning at first glance this offer of assistance, however, let us look in turn at the two candidates with their supporting examples.

In his book, The Justification of Scientific Change, Carl Kordig claims that the radical meaning variance view of Feyerabend and others becomes ensnared in a self-referential paradox. The accusation fails because it rests upon a confusion of that view with its “counterparts” in other linguistic/theoretical contexts.

Rosenberg (1983), in his comments on our article (Sober and Lewontin 1982) concerning the units of selection controversy, has matters precisely backwards. He chides us for conflating a causal theory with a mathematical model, whereas this is precisely the error we argued is at the core of the position defended by Williams (1966) and Dawkins (1976). Their idea—genic selectionism—confuses the computational adequacy of a mathematical model with a substantive causal theory. The main point of our paper was simply this: the fact that evolution by natural selection can be “represented” in terms of selection coefficients that attach to single genes does not imply that selection is always selection for or against the possession of those single genes.