A widely endorsed thesis in the philosophy of science holds that if evidence for a hypothesis was not known when the hypothesis was proposed, then that evidence confirms the hypothesis more strongly than would otherwise be the case. The thesis has been thought to be inconsistent with Bayesian confirmation theory, but the arguments offered for that view are fallacious. This paper shows how the special value of prediction can in fact be given Bayesian explanation. The explanation involves consideration of the (...) reliability of the method by which the hypothesis was discovered, and thus reveals an intimate connection between the 'logic of discovery' and confirmation theory. (shrink)

The incredible achievements of modern scientific theories lead most of us to embrace scientific realism: the view that our best theories offer us at least roughly accurate descriptions of otherwise inaccessible parts of the world like genes, atoms, and the big bang. In Exceeding Our Grasp, Stanford argues that careful attention to the history of scientific investigation invites a challenge to this view that is not well represented in contemporary debates about the nature of the scientific enterprise. The historical record (...) of scientific inquiry, Stanford suggests, is characterized by what he calls the problem of unconceived alternatives. Past scientists have routinely failed even to conceive of alternatives to their own theories and lines of theoretical investigation, alternatives that were both well-confirmed by the evidence available at the time and sufficiently serious as to be ultimately accepted by later scientific communities. Stanford supports this claim with a detailed investigation of the mid-to-late 19th century theories of inheritance and generation proposed in turn by Charles Darwin, Francis Galton, and August Weismann. He goes on to argue that this historical pattern strongly suggests that there are equally well-confirmed and scientifically serious alternatives to our own best theories that remain currently unconceived. Moreover, this challenge is more serious than those rooted in either the so-called pessimistic induction or the underdetermination of theories by evidence, in part because existing realist responses to these latter challenges offer no relief from the problem of unconceived alternatives itself. Stanford concludes by investigating what positive account of the spectacularly successful edifice of modern theoretical science remains open to us if we accept that our best scientific theories are powerful conceptual tools for accomplishing our practical goals, but abandon the view that the descriptions of the world around us that they offer are therefore even probably or approximately true. (shrink)

an observation to formulate a theory, it is no surprise that the resulting theory accurately captures that observation. However, when the theory makes a novel prediction—when it predicts an observation that was not used in its formulation—this seems to provide more substantial confirmation of the theory. This paper presents a new approach to the vexed problem of understanding the epistemic difference between prediction and accommodation. In fact, there are several problems that need to be disentangled; in all of them, the (...) key is the concept of overfitting. We float the hypothesis that accommodation is a defective methodology only when the methods used to accommodate the data fail to guard against the risk of overfitting. We connect our analysis with the proposals that other philosophers have made. We also discuss its bearing on the conflict between instrumentalism and scientific realism. Introduction Predictivisms—a taxonomy Observations Formulating the problem What might Annie be doing wrong? Solutions Observations explained Mayo on severe tests The miracle argument and scientific realism Concluding comments. (shrink)

Evidence for a hypothesis typically confirms the hypothesis more if the evidence was predicted than if it was accommodated. Or so I argued in previous papers, where I also developed an analysis of why this should be so. But this was all a mistake if Howson and Franklin (1991) are to be believed. In this paper, I show why they are not to be believed. I also identify a grain of truth that may have been dimly grasped by those Bayesians (...) who deny the confirmatory value of prediction. (shrink)

Evidence that confirms a scientific hypothesis is said to be ‘novel’ if it is not discovered until after the hypothesis isconstructed. The philosophical issues surrounding novel confirmation have been well summarized by Campbell and Vinci [1983]. They write that philosophers of science generally agree that when observational evidence supports a theory, the confirmation is much stronger when the evidence is ‘novel’... There are, nevertheless, reasons to be skeptical of this tradition... The notion of novel confirmation is beset with a theoretical (...) puzzle about how the degree of confirmation can change without any change in the evidence, hypothesis, or auxiliary assumptions... There have not yet appeared any obviously satisfactory solutions to these problems Much of the literature on novel confirmation relies on the Bayesian analysis of conditional probabilities. Let H represent a hypothesis, E an event that confirms the hypothesis, and B some relevant background information. Denote by Pr the conditional probability of x given y. There are various plausible measures of the degree of support that E lends to H. Among these are: Degree of support = Pr Degree of support = Pr – Pr Degree of support = Pr – Pr Degree of support = Pr – Pt In the work cited, Campbell and Vinci offer a somewhat more involved Bayesian interpretation. Formula is discussed by Gardner [1981] who points out that under this formulation there can be no role for novelty. 'The function Pr contains no third slot in which to insert a temporal relation between the invention of H and the inventor's learning E. Obviously, then, this relation could not possibly affect E's support of H.' Formula, on the other hand, suggests a role for novelty. Bayes's Theorem allows us to rewrite the formula as Degree of support = Pr x [Pr – 1] We can use 1/Pr as a measure of the novelty of E. Then shows that the degree of support increases with novelty of E. In the paper already cited, Campbell and Vinci discuss shortcomings of this analysis. Formula expresses an alternative offered by Howson [1984]. In that formula, Pr represents the probability of H assuming that only B – {E} is known. This allows for the possibility of non-novel facts generating support for hypotheses. Niiniluoto [1984] argues for a variant along the lines of, in which we account for the possibility that the theorist was unaware that his hypothesis entails E. One problem in deciding among these approaches is that the choice of a definition for the degree of support appears arbitrary. What kind of argument could justify the choice of one definition over another? It is our position that there can be no basis for addressing this question in the absence of an explicit model of the process by which hypotheses are generated. Only in the presence of such a model can the various conditional probabilities be given meaningful interpretations. We provide such models in Sections 1, 2 and 4. The simple model of Section 1, incorporating strong assumptions, yields the conclusion that novelty is irrelevant. When these assumptions are relaxed in the later sections, novelty becomes relevant for a variety of reasons. It is at least potentially the case that scientists have more information about their own abilities than is publicly available, and this information might influence their decisions about whether even to attempt novel prediction. If this is so, then it should be incorporated into the model of hypothesis generation. This requires an explicit discussion of how scientists respond to incentives and how the incentives themselves evolve, which in turn takes us into the realm of economic theory. We have addressed these issues in another paper, written for an audience of economists. The results of this research are summarized in Section 3. (shrink)

Traditional analyses of the curve fitting problem maintain that the data do not indicate what form the fitted curve should take. Rather, this issue is said to be settled by prior probabilities, by simplicity, or by a background theory. In this paper, we describe a result due to Akaike [1973], which shows how the data can underwrite an inference concerning the curve's form based on an estimate of how predictively accurate it will be. We argue that this approach throws light (...) on the theoretical virtues of parsimoniousness, unification, and non ad hocness, on the dispute about Bayesianism, and on empiricism and scientific realism. * Both of us gratefully acknowledge support from the Graduate School at the University of Wisconsin-Madison, and NSF grant DIR-8822278 (M.F.) and NSF grant SBE-9212294 (E.S.). Special thanks go to A. W. F. Edwards.William Harper. Martin Leckey. Brian Skyrms, and especially Peter Turney for helpful comments on an earlier draft. (shrink)

This article investigates the implications of string theory for the conception of scientific theory confirmation. The classical understanding of theory confirmation is based on the assumption that scientific theory building is underdetermined by the available empirical data. Several arguments are presented, which suggest a devaluation of this ‘principle of scientific underdetermination’ in the context of string theory. An altered conception of scientific progress emerges that is not based on the notion of theory succession.

This text provides a critique of the subjective Bayesian view of statistical inference, and proposes the author's own error-statistical approach as an alternative framework for the epistemology of experiment. It seeks to address the needs of researchers who work with statistical analysis.

has offered a lovely example to motivate the intuition that a successful prediction has a kind of confirmatory significance that an accommodation lacks. This paper scrutinizes Maher's example. It argues that once the example is tweaked, the intuitive difference there between prediction and accommodation disappears. This suggests that the apparent superiority of prediction to accommodation is actually a side effect of an important difference between the hypotheses that tend to arise in each case.

Maher (1988, 1990) has recently argued that the way a hypothesis is generated can affect its confirmation by the available evidence, and that Bayesian confirmation theory can explain this. In particular, he argues that evidence known at the time a theory was proposed does not confirm the theory as much as it would had that evidence been discovered after the theory was proposed. We examine Maher's arguments for this "predictivist" position and conclude that they do not, in fact, support his (...) view. We also cast doubt on the assumptions of Maher's alleged Bayesian proofs. (shrink)

It is argued that the evolution of fundamental microphysics throughout the twentieth century is characterised by two interrelated developments. On the one hand, the experimental signatures which confirm theoretical statements are moving towards the fringes of the phenomenal world and, at the same time, leave increasingly wide spaces for entirely theoretical reasoning with little or no empirical interference. On the other hand, assessments of limitations to scientific underdetermination gain importance within the theoretical process.

An enduring question in the philosophy of science is the question of whether a scientific theory deserves more credit for its successful predictions than it does for accommodating data that was already known when the theory was developed. In The Paradox of Predictivism, Eric Barnes argues that the successful prediction of evidence testifies to the general credibility of the predictor in a way that evidence does not when the evidence is used in the process of endorsing the theory. He illustrates (...) his argument with an important episode from nineteenth-century chemistry, Mendeleev's Periodic Law and its successful predictions of the existence of various elements. The consequences of this account of predictivism for the realist/anti-realist debate are considerable, and strengthen the status of the 'no miracle' argument for scientific realism. Barnes's important and original contribution to the debate will interest a wide range of readers in philosophy of science. (shrink)

An enduring question in the philosophy of science is the question of whether a scientific theory deserves more credit for its successful predictions than it does for accommodating data that was already known when the theory was developed. In The Paradox of Predictivism, Eric Barnes argues that the successful prediction of evidence testifies to the general credibility of the predictor in a way that evidence does not when the evidence is used in the process of endorsing the theory. He illustrates (...) his argument with an important episode from nineteenth-century chemistry, Mendeleev's Periodic Law and its successful predictions of the existence of various elements. The consequences of this account of predictivism for the realist/anti-realist debate are considerable, and strengthen the status of the 'no miracle' argument for scientific realism. Barnes's important and original contribution to the debate will interest a wide range of readers in philosophy of science. (shrink)