A Novel Solution to the Problem of Old Evidence A Novel Solution to the Problem of Old Evidence Author(s): Jan Sprenger Source: Philosophy of Science , Vol. 82, No. 3 (July 2015), pp. 383-401 Published by: The University of Chicago Press on behalf of the Philosophy of Science Association Stable URL: https://www.jstor.org/stable/10.1086/681767 JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at https://about.jstor.org/terms The University of Chicago Press and Philosophy of Science Association are collaborating with JSTOR to digitize, preserve and extend access to Philosophy of Science This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms https://www.jstor.org/stable/10.1086/681767 A Novel Solution to the Problem of Old Evidence Jan Sprenger*y One of the most troubling and persistent challenges for Bayesian Confirmation Theory is the Problem of Old Evidence ðPOEÞ. The problem arises for anyone who models sci- entific reasoning by means of Bayesian Conditionalization. This article addresses the problem as follows: First, I clarify the nature and varieties of the POE and analyze vari- ous solution proposals in the literature. Second, I present a novel solution that combines previous attempts while making weaker and more plausible assumptions. Third and last, I summarize my findings and put them into the context of the general debate about POE and Bayesian reasoning. 1. Introduction: The Problem of Old Evidence. One of the most troubling and persistent challenges for Bayesian Confirmation Theory ðBCTÞ is the Problem of Old Evidence ðPOEÞ: A phenomenon E remains unexplained by the available scientific theories. At some point, a theory T is discovered that accounts for E. Then, E is “old evidence”: at the time when T is developed, the scientist is already certain or close to certain that the phenomenon E is real. Nevertheless, E apparently confirms T—at least if T was invented on in- dependent grounds. After all, it resolves a well-known and persistent tension between theory and observation. A famous case of old evidence in science is the Mercury perihelion anom- aly ðGlymour 1980; Earman 1992Þ. For a long time, the shift of the Mercury Received June 2014; revised November 2014. *To contact the author, please write to: Tilburg Center for Logic, Ethics and Philosophy of Science, Tilburg University, PO Box 90153, 5000 LE Tilburg, Netherlands; e-mail: j.sprenge @uvt.nl. yThe author wishes to thank the Netherlands Organisation for Scientific Research ðNWOÞ fo support of his research through Vidi grant 276-20-023. Peter Brössel, Clark Glymour, Bran den Fitelson, Stephan Hartmann, Jonah Schupbach, participants of the Tilburg Center fo Logic, Ethics and Philosophy of Science research seminars, and audiences in Dubrovnik and London improved the article with their helpful feedback. Philosophy of Science, 82 (July 2015) pp. 383–401. 0031-8248/2015/8203-0003$10.00 Copyright 2015 by the Philosophy of Science Association. All rights reserved. 383 This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms r r - r perihelion could not be explained by Newtonian mechanics or any other rep- utable physical theory. Then, Einstein realized that his General Theory of Rel- ativity ðGTRÞ could explain the perihelion shift. This discovery conferred a 384 JAN SPRENGER substantial degree of confirmation on GTR, much more than some pieces of novel evidence. Also in other scientific disciplines, newly introduced theories are com- monly assessed with respect to their success at accounting for observational anomalies. Think, for example, of the assessment of global climate models against a track record of historical data or of economic theories that try to explain anomalies in decision making under uncertainty ðe.g., the Allais or Ellsberg paradoxesÞ. All this is hard to capture in the Bayesian framework, where confirmation is expressed as an increase in an agent’s subjective degree of belief. On the Bayesian account, E confirms T if and only if the posterior degree of belief in T, p0ðTÞ, exceeds the prior degree of belief in T, pðTÞ. These two prob- abilities are related by means of conditioning on the evidence E and Bayes’s Theorem ðe.g., Howson and Urbach 2006Þ: p0ðTÞ:5 pðT j EÞ 5 pðTÞ pðE j TÞ pðEÞ : ð1Þ Here and in the sequel, reference to an accepted body of background as- sumptions K in the credence function pð·Þ is omitted for the sake of sim- plicity. Let us now apply the Bayesian calculus to POE. When E is old evidence and already known to the scientist, the prior degree of belief in E is max- imal: pðEÞ 5 1. But with that assumption, it follows that the posterior prob- ability of T cannot be greater than the prior probability: p0ðTÞ 5 pðT j EÞ 5 pðTÞ � pðE j TÞ ≤ pðTÞ. Hence, E does not confirm T. The very idea of confirmation by oldevidence, or equivalently, confirmation by accounting for well-known observational anomalies, seems impossible to describe in the Bayesian belief kinematics. This article investigates the variety of the problems that POE poses for BCT, reviews existing approaches, and finally proposes a novel solution. Section 2 distinguishes the dynamic and the static version of POE and briefly comments on attempts to solve the static problem. Section 3 analyzes the solutions of the dynamic version proposed by Garber ð1983Þ, Jeffrey ð1983Þ, Niiniluoto ð1983Þ, and Earman ð1992Þ. On these accounts, confirmation oc- curs through conditionalizing on the proposition T ⊢ E. Section 4 presents my own improvement on these attempts. Finally, section 5 puts my findings into the context of the general debate about POE and BCT. Throughout the article, I work in the framework of Bayesian epistemology ðBovens and Hartmann 2003; Hájek and Hartmann 2010; Hartmann and Sprenger 2010Þ and Bayesian philosophy of science ðHowson and Urbach 2006; Sprenger 2015Þ. I am using Bayesian Nets for representing depen- This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms dencies between propositions of interest—a technique that has recently proved its merits for modeling complex confirmation phenomena ðe.g., Dawid, Hart- mann, and Sprenger 2015Þ. THE PROBLEM OF OLD EVIDENCE 385 2. The Varieties of the POE. Part of the controversy about the POE con- cerns the question of what the problem really consists in. Eells ð1985, 1990Þ has introduced a helpful conceptual distinction between different problems of old evidence: 1. The Problem of Old New Evidence: E is learned after T is formulated, but even after updating our degrees of belief on E, we still say that E confirms T although by now, p0ðEÞ 5 1. ðAs before, p0 denotes the posterior probability distribution.Þ 2. The Problem of Old Evidence: E is known before T is formulated. 2aÞ The Problem of Old Old Evidence: Even after formulating T and discovering that T accounts for E, E seems to confirm T. 2bÞ The Problem of New Old Evidence: Why does E confirm Tat the moment where T is discovered or when it is discovered that T accounts for E? Items 1 and 2a describe the static ðEells: “ahistorical”Þ aspect of POE: be- lief changes induced by the discovery of T, or the fact that T accounts for E, have already taken place. Still, we would like to say that E is evidentially relevant for T: when faced with a decision between T and a competitor T 0, E is a good reason for preferring T over T 0. Item 2b, however, captures the dynamic ðEells: “historical”Þ aspect of the problem: it refers to the moment in time when T and its relation to E are discovered. Here, the challenge for the Bayesian is to describe how the discovery of a new theory T and its explanatory successes raises our confidence in T. The Bayesian usually approaches both problems by different strategies. The standard take on the dynamic problem consists in allowing for the learning of logical truths. In classical examples, such as explaining the Mer- cury perihelion shift, the newly invented theory ðhere: GTRÞ was initially not known to entail the old evidence. It took Einstein some time to find out that T entailed E ðBrush 1989; Earman 1992Þ. Learning this deductive relation- ship undoubtedly increased Einstein’s confidence in T since such a strong consilience with the phenomena could not be expected beforehand. However, this belief change is hard to model in BCT. A Bayesian rea- soner is assumed to be logically omniscient, and the logical fact T ⊢ E should always have been known to her. Hence, the proposition T ⊢ E can- not be learned by a Bayesian: it is already part of her background beliefs. To solve this problem, several philosophers have relaxed the assumption of logical omniscience and enriched the algebra of statements about which we have degrees of belief. New atomic sentences of the form T ⊢ E are This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms added ðGarber 1983; Jeffrey 1983; Niiniluoto 1983Þ, such that BCT can ac- count for our cognitive limitations in deductive reasoning. Then, it can be shown that under suitable assumptions, conditioning on T ⊢ E confirms T. 386 JAN SPRENGER I comment on these efforts in the next section. The response to the static problem is different. Eells ð1990, 209Þ pro- poses that “E is ðactualÞ evidence for T . . . if, at some point in the past, the event of its confirming T took place.” On that definition, the solution of the static problem in 1 and 2a would reduce to the solution of the dynamic problem ðcf. Christensen 1999, 444Þ: when we can show that at the time of the formulation of T, E confirmed T, then E would also be evidence for T afterward ðEells 1990, 210Þ. However, most takes on the dynamic problem do not try to show that E confirms T at the relevant moment in time: rather, this work is done by the proposition T ⊢ E. This strategy also fails to ac- count for changes over time in our assessment of the strength of the evi- dence that E provides for T. Therefore, Colin Howson ð1984, 1985, 1991Þ developed a more princi- pled take on the static problem. He gives up the Bayesian explication of confirmation as positive probabilistic relevance relative to actual degrees of belief. Rather, he suggests to subtract the old evidence E from the agent’s background knowledge K and to evaluate the confirmation relation with respect to the counterfactual credence function pK =fEg: “The Bayesian as- sesses the contemporary support E gives H by how much the agent would change his odds on H were he now to come to know E. . . . In other words, the theory is explicitly a theory of dispositional properties of the agent’s belief-structure” ðHowson 1984, 246Þ. For instance, if T is a hypothesis about the bias of a coin and E the outcome of some series of coin tosses, then eliminating E from the background knowledge allows us to describe how E raises the probability of T since we would have definite and non- trivial values for pK =fEgðE j TÞ and pK =fEgðE j :TÞ ðHowson 1991, 551–52Þ. As the above quote indicates, Howson thinks that BCT is essentially a counterfactual or dispositional theory. I happen to agree with him on this point, but there are a couple of technical problems with Howson’s specific choice of a counterfactual credence function. In particular, E may be en- tangled with other propositions that are part of K. As Chihara ð1987Þ notes, just removing E from the set of background assumptions K will not work in general if K is supposed to be a deductively closed set. If we ignore this feature, we would sacrifice a main advantage of Howson’s counterfactual ap- proach: that it need not give up the elegance of the standard Bayesian ratio- nality assumptions ðe.g., epistemic closureÞ in order to account for the POE. Alternatively, Howson may choose to evaluate the confirmation relation with respect to the agent’s credence function at some point in the past ðe.g., just before she learned EÞ. But the agent’s belief in E may have gradually grown over time, and no such time point may exist. Moreover, without This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms knowledge of E, T may even not have been formulated ðcf. Glymour 1980, 87–93Þ. All in all, the credence function p that is supposed to ground the THE PROBLEM OF OLD EVIDENCE 387 K =fEg confirmation relation between T and E is rather indefinite even in cases where everybody agrees that E confirms T. As a consequence, we can hardly determine whether the degree of confirmation conferred on T by E is strong, weak, or nil. Another solution proposal for the static problem is based on the choice of a specific confirmation measure ðFitelson 1999, 2014Þ. Christensen ð1999Þ contends that for the measure s*ðT; EÞ 5 pðT j EÞ 2 pðT j :EÞ, the POE does not arise. For instance, if T entails E, then:E also entails ¬T, which implies pðT j :EÞ 5 0 and s*ðT; EÞ 5 pðT j EÞ > 0. According to s*, old evidence E can substantially confirm theory T, whereas the degree of con- firmation is 0 for measures that compare the prior and posterior probability of T, such as dðT; EÞ 5 pðT j EÞ 2 pðTÞ or rðT; EÞ 5 logð pðT j EÞ=pðTÞÞ. Christensen’s move has its merits for cases where pðEÞ is close to, but not entirely equal to, 1 ðalthough it is questionable whether s* is a good ex- plicatum for the degree of confirmation; see Eells and Fitelson 2000Þ. But in the classical POE where pðEÞ 5 1, pðT | :EÞ may not have a clear-cut definition since pðT j :EÞ 5 pðT ∧: EÞ=pð:EÞ involves a division by 0. And if pðT |:EÞ has to be evaluated relative to a counterfactual credence function, not much may have been gained with respect to Howson’s pro- posal. The static POE thus proves to be a tough challenge for Bayesian confir- mation theorists. Are prospects any better for the dynamic problem? 3. The GJN Approach to POE. The first models of the dynamic POE have been developed by Daniel Garber, Richard Jeffrey, and Ilkka Niiniluoto in a group of papers that all appeared in 1983. Henceforth, we refer to the family of their solution proposals as GJN solutions. In order to properly compare my own solution proposal to the state of the art, and to assess its innovative value, I will briefly recap the achievements of the GJN models and elabo- rate their limitations and weaknesses. The GJN models take into account that a scientist is typically not aware of all possible theories and their relations to the evidence, thus parting with the assumption of logical omniscience that characterizes the ideal Bayesian reasoner. Consequently, the relevant piece of evidence is not E itself but the learning of a specific relation between theory and evidence, namely, that T implies E or accounts for E. The notational convention to write this prop- osition as ⌈T ⊢ E⌉ conceals that we do not necessarily deal with a strict logical deduction—also explanatory relationships may fall under the scope of this model ðGarber 1983, 103; Eells 1990, 212Þ. Such cases count, after all, as confirmatory arguments for T in ordinary scientific reasoning. How- This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms ever, thinking of deduction facilitates notation and can be used as a guide for intuition and the development of an adequate Bayesian model. What the GJN models aim to show is that conditionalizing on the prop- 388 JAN SPRENGER osition T ⊢ E increases the posterior probability of T. Eells ð1990, 211Þ dis- tinguishes three steps in this endeavor: first, parting with the logical om- niscience assumption and developing a formal framework for imperfect Bayesian reasoning; second, describing which kind of relation obtains be- tween T and E; and third, showing that learning this relation increases the probability of T. While the GJN models neglect the second step, probably in due anticipation of the diversity of logical and explanatory relations in sci- ence, they are quite explicit on step 1 and step 3. Garber’s ð1983Þ model focuses on step 1 and on learning logical truths in a Bayesian framework. After all, no reasoner is ever logically omniscient. Learning logical/mathematical truths can be quite insightful and lead to great progress in science. The famous, incredibly complex proof of Fermat’s Last Theorem may be a good example ðsee Wiles ½1995� for a short versionÞ. Garber therefore enriches the underlying language L in a way that T ⊢ E is one of the atomic propositions of the extended language L0. Garber also demands that the agent recognize some elementary relations in which the proposition T ⊢ E stands to other elements of L0: pðE j T; T ⊢ EÞ 5 1 pðT; E; T ⊢ EÞ 5 pðT; T ⊢ EÞ: ð2Þ These constraints express the closure of degree of belief under modus po- nens. Or alternatively, if an agent takes T and T ⊢ E for granted, then she maximally believes E. Even in a boundedly rational picture of Bayesian rea- soning, such as Garber’s, knowledge of such elementary inference schemes sounds eminently sensible. Garber then proves the following theorem: there is at least one probability function on L0 such that every nontrivial atomic sentence of the form T ⊢ E gets a value strictly between 0 and 1. Thus, one can coherently have a genuinely uncertain attitude about all propositions in the logical universe, including tautologies. Finally Garber shows that for any atomic L0-sentenceof the form ⌈T ⊢ E⌉, there are infinitely many probability functions such that pðEÞ 5 1 and pðT j T ⊢ EÞ > pðTÞ. A similar point is made by Niiniluoto ð1983Þ, although with less formal rigor and elaboration. While Garber’s efforts are admirable, they only address the first step of solving the dynamic POE: he provides an existence proof for a solution to the POE but no set of conditions that guide our judgments on when learn- ing T ⊢ E confirms T. This lacuna is closed by Richard Jeffrey ð1983Þ, who published his solution in the same volume in which Garber’s paper ap- peared. Jeffrey considers the proposition T ⊢ E as an object of subjective un- certainty. We follow Earman’s ð1992Þ presentation of Jeffrey’s ð1983, 150– 51Þ solution, which contains the following assumptions: This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms a. pðEÞ 5 1. b. pðTÞ; pðT ⊢ EÞ; pðT ⊢:EÞ ∈ ð0; 1Þ. g. pð½T ⊢ E� ∧ ½T ⊢:E�Þ 5 0. THE PROBLEM OF OLD EVIDENCE 389 d. pðT j ½T ⊢ E� ∨ ½T ⊢:E�Þ ≥ pðTÞ. h. pðT; :E; T ⊢:EÞ 5 pðT; T ⊢:EÞ. From these assumptions, Jeffrey derives pðT j T ⊢ EÞ > pðTÞ, which amounts, given the constraint pðEÞ 5 1, to a solution of the dynamic POE. The strength of Jeffrey’s solution crucially depends on how well we can motivate condition d. The other conditions are plausible: a is just the stan- dard presumption that at the time where confirmation takes place, E is al- ready known to the agent; b demands that we not be certain about the truth of T or T ⊢ ±E beforehand, in line with the typical description of POE; and g requires that T do not entail E and :E at the same time. In particular, T has to be consistent. Finally, h is a modus ponens condition similar to equa- tion ð2Þ: the joint degree of belief in T, :E, and T ⊢ :E is equal to the joint degree of belief in T and T ⊢ :E, demanding that the agent recognize the implications that the latter two propositions have on :E. Hence, d really carries the burden of Jeffrey’s argument. This condition has some odd technical consequences, as pointed out by Earman ð1992, 127Þ. Assume, for instance, that pðT ⊢ EÞ 5 pðT ⊢:EÞ, which may be a plausi- ble representation of our momentary ignorance regarding the implications of T for ±E. Then it follows that pðT j T ⊢ EÞ ≥ 2pðTÞ, which implies that the prior degree of belief pðTÞ must have been be smaller than .5. In other words, d cannot be satisfied for theories that are rather probable a priori. This restriction is ad hoc and quite troubling since it severely limits the scope of Jeffrey’s solution: why should probable theories not be confirmed by old evidence? To this I would like to add that in the absence of specific motivation, it is very surprising that the posterior probability of T should be at least twice as large as the prior probability. The real problem with d is, however, not technical but philosophical. Jeffrey ð1983, 148–49Þ supports d by mentioning that Newton was, on for- mulating his theory of gravitation G, convinced that it would bear on the phenomena he was interested in, namely, the mechanism governing the tides. Although Newton did not know whether G would entail the phenomena as- sociated to the tides or be inconsistent with them, he used his knowledge that G would bear on the tides as an argument for endorsing it and for tempo- rarily accepting it as a working hypothesis. To my mind, this reconstruction conflates an evidential virtue of a theory with a methodological one. We are well advised to cherish theories of which we know that they make precise predictions on an interesting subject matter, even if we do not yet know what these predictions look like in detail. This is basically a Popperian rationale for scientific inquiry: go for theories that have high empirical content and that make precise predictions and develop This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms them further. They are the ones that will finally help us to solve deep sci- entific problems. It is plausible that Newton, on deciding to further pursue his theory of gravitation, followed this methodological rule when discov- 390 JAN SPRENGER ering that it would have some implications for the tides phenomena. But following this rule is very different from arguing that the informativity and empiricalcontentofatheoryincreasesitsplausibility.Actually,Popperð1959/ 2002, 268–69Þ thought the other way round: theories with high empirical content rule out more states of the world and will have low ðlogicalÞ prob- ability. This is just because they take, in the virtue of making many predic- tions, a higher risk of being falsified. Indeed, it is hard to understand why in- creasing the ðunconfirmed and unrefutedÞ empirical content of T provides an argument that T is more likely to be true. The reply that p describes a sub- jective rather than a logical probability function will not help much: even if p is a purely subjective function, it remains opaque why increasing the class of potential falsifiers of T will increase its plausibility. Jeffrey’s condition d is therefore ill-grounded and at the very least too controversial to act as a premise in a solution of the POE. Earman ð1992, 128–29Þ considers two alternative derivations of pðT j T ⊢ EÞ > pðTÞ where different assumptions carry the burden of the argument. One of them is the inequality f. pðT j T ⊢ EÞ > pðT j ½T ⊬E� ∧ ½T ⊬ :E�Þ, but it is questionable whether this suffices to circumvent the above objec- tions. What Earman demands here is very close to what is supposed to be shown: that learning T ⊢ E is more favorable to T than learning that T gives no definite prediction for the occurrence of E or ¬E. In the light of the above arguments against d and in the absence of independent arguments in favor of f, this condition just seems to beg the question. The second alternative derivation of pðT j T ⊢ EÞ > pðTÞ is proposed by Jeffrey ð1983, 149Þ himself and relies on the equality w. pð½T ⊢ E� ∨ ½T ⊢:E�Þ 5 1. However, as Earman admits himself, this condition is too strong: it amounts to demanding that on formulating T, the scientist was certain that it either implied E or ¬E. In practice, such relationships are rather discovered grad- ually. As Earman continues, discussing the case of GTR: “the historical ev- idence goes against this supposition: . . . Einstein’s published paper on the perihelion anomaly contained an incomplete explanation, since, as he him- self noted, he had no proof that the solution of the field equations . . . was the unique solution for the relevant set of boundary conditions” ð1992, 129Þ. Taking stock, we conclude that Garber, Jeffrey, Niiniluoto, and Earman make This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms interesting proposals for solving the dynamic POE but that their solutions are either incomplete ðGarber, NiiniluotoÞ or based on highly problematic assumptions ðEarman, JeffreyÞ. I now present a novel solution proposal that THE PROBLEM OF OLD EVIDENCE 391 also aims at the dynamic problem but that makes use of a slightly different conceptualization. 4. A Novel Solution of POE. The troubles with both Howson’s approach to the static problem and the GJN approach to the dynamic problem seem to be technical at first sight. Howson has trouble with the counterfactual credence function pK =fEg, and the GJN-style solutions by Jeffrey and Earman are based on implausible assumptions. How are we supposed to take it from there? The aim of this section consists in solving the dynamic problem and in showing that if E is already known, learning R:5T ⊢ E raises the subjective probability of T. To achieve this goal, we combine ideas from Howson with the machinery of the GJN models. Specifically, we assume a couple of con- ditions on when R confirms T relative to E. The first condition has the fol- lowing qualitative form: ð1Þ If ¬T is already known, then learning R 5 T ⊢ E leaves the prob- ability of E unchanged. To motivate this constraint, consider the following fictitious scenario. Con- sider the hypothesis that Steven is going on a ski trip in the Alps. We already know that Steven had to cancel the trip because his employer did not give him leave. Now, Steven tells us: “Oh, by the way, if I were to go on this ski trip ðTÞ, I would soon buy a new winter jacket ðEÞ.” Does this utterance raise our credence that Steven will buy a new winter jacket soon? Plausibly not. Remember that we already know that the ski trip is canceled when learning T ⊢ E. Does it lower it? Plausibly not. Steven’s statement neither under- mines nor supports those reasons for buying a winter jacket that are inde- pendent of his ðnotÞ going on a ski trip ðe.g., the winter temperatures in his home placeÞ. More generally, learning that certain events are predicted by a refuted hypothesis is just irrelevant to our assessment of the plausibility of these events. Nostradamus’s astrological theory is in all probability wrong; therefore, on learning the content of his predictions, we should neither raise nor lower our credence in the events that his theory predicted to happen. Formally, ð1Þ can be written as pðE j :T; RÞ 5 pðE j :TÞ or pðE j :T; RÞ 5 pðE j :T; :RÞ. However, one may worry about the credence function involved. If E is old evidence, then pðE | ·Þ 5 1, regardless of which prop- osition stands to the right of the vertical dash. In this reading, ð1Þ would be utterly trivial. Therefore, we have, like Howson, to define a counterfactual credence function p and to give it an interpretation. Instead of just elimi- nating E from the background knowledge, p should represent the degrees of This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms belief of a scientist who has a sound understanding of theoretical principles and their impact on observational data but who is an imperfect logical rea- soner and lacks full knowledge of the observational history ðcf. Earman 392 JAN SPRENGER 1992, 134Þ. Such probabilistic assessments are typically made by scientists who assess evidence and review journal articles: How probable would the actual evidence E be if T were true? How probable would E be if T were false? When T and :T are two definite statistical hypotheses, such judg- ments are immediately given by the corresponding sampling distribution. Also in more general contexts, such judgments may be straightforward, or at least a matter of consensus in the scientific community. This proposal resembles Howson’s proposal for introducing counterfac- tual reasoning into the POE, but it is not bound to a particular credence func- tion. I say more about this in section 5. Let us now proceed by formulating constraints on p that allow us to solve the dynamic POE. The first one char- acterizes the elementary inferential relations between E, T, and R: ½1� pðE j T; RÞ 5 1. If T is true and T entails E, then E can be regarded as a certainty. In this scenario, R codifies a strict deductive relation between T and E; later, we re- lax this condition and also consider weaker ðe.g., explanatoryÞ dependencies. The second condition just spells out ð1Þ in terms of the p-function: ½2� pðE j :T; RÞ 5 pðE j :T; :RÞ > 0. In other words, if :T is already known, then learning R or :R does not change the probability of E, as argued above. Moreover, we should not be certain that E can only occur if T is true. Finally, we have the following inequality: ½3� pðE j T; :RÞ < 1 2 pðR j :TÞ pðR j :TÞ � pðR j TÞ 1 2 pðR j TÞ : This condition demands that the value of pðE j :R; TÞ be smaller than the threshold on the right-hand side.1 When the ðdubiousÞ Jeffrey condition d is satisfied and R and T are positively relevant or neutral to each other, ½3� is 1. In this formulation of condition ½3�, it is implicit that 1 > pðR j TÞ; pðR j :TÞ > 0 since, otherwise, either the expression would be undefined ðdivison by 0Þ or pðE j :R; TÞ would have to be smaller than 0, which is impossible. We will use this fact in the proofs. This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms trivially satisfied since in that case, pðR j TÞ ≥ pðR j :TÞ, implying that the right-hand side of ½3� is greater or equal to 1. But also if R and T are neg- atively relevant to each other, ½3� is plausibly satisfied. To see this, note that THE PROBLEM OF OLD EVIDENCE 393 ½3� can be written in the form pðE j :R; TÞ < OðR j :TÞ=OðR j TÞ, where O represents the betting odds for a proposition that correspond to a par- ticular degree of belief. When the mutual negative impact of R and T is not too strong and the two betting odds are quite close to each other, ½3� will be satisfied as long as pðE j :R; TÞ is not too close to 1. And given that T is assumed to be true, but that by ¬R it does not fully account for E, E should not be a matter of course for a rational Bayesian agent. These three conditions are jointly sufficient to prove the following result ðproof in the appendixÞ: Theorem 1. Let T, R, and E be three elements of an algebra A with asso- ciated probability measure p, and let the following three conditions be sat- isfied: ½1� pðE j T; RÞ 5 1. ½2� pðE j :T; RÞ 5 pðE j :T; :RÞ > 0. ½3� pðE j T; :RÞ < 1 2 pðR j :TÞ pðR j :TÞ � pðR j TÞ 1 2 pðR j TÞ : Then, R confirms T relative to old evidence E; that is, pðT j E; RÞ > pðT j EÞ. In other words, learning that T entails E also confirms T. This result solves the POE on the basis of a conceptualization that com- bines elements from the GJN and Howson’s counterfactual strategy. We use the main idea from the GJN models—the confirming event is the dis- covery that T accounts for/explains E—but we spell out the confirmation relation relative to counterfactual credences, as Howson envisions. However, in many cases of scientific reasoning, condition ½1�; that is, pðE j T; RÞ 5 1, may be too strong. It may apply well to the Mercury peri- helion shift, which is deductively implied by GTR, but it may fail to cover cases where T accounts for E in a less rigorous manner ðsee also Earman 1992, 121; Fitelson 2014Þ. If we allow for a weaker interpretation of R ðe.g., as providing some explanatory mechanismÞ, then we are faced with the possibility that even if we are certain that T is true, and that T explains E, the conditional degree of belief in E may not be a certainty. And pðE j TÞ < 1 could even make sense if the relationships between T and E are deductive: This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms the proof of T ⊢ E could so complex that the involved scientists have some doubts about its soundness and refrain from assigning it maximal degree of belief. Again, Fermat’s Last Theorem may be a plausible intuition pump. 394 JAN SPRENGER For covering this case, I prove a second theorem that covers the case of pðE j T; RÞ 5 1 2 ε for some small ɛ > 0 ðproof in the appendixÞ. Theorem 2. Let T, R, and E be three elements of an algebra A with asso- ciated probability measure p, and let the following three conditions be sat- isfied: ½10� pðE j T; RÞ 5 1 2 ε for some 0 < ε < 1. ½20� pðE j :T; RÞ 5 pðE j :T; :RÞ > 0. ½30� pðE j T; :RÞ < ð1 2 εÞ 1 2 pðR j :TÞ pðR j :TÞ � pðR j TÞ 1 2 pðR j TÞ : Then, R confirms T relative to old evidence E; that is, pðT j E; RÞ > pðT j EÞ. In other words, learning that T accounts for E also confirms T. The motivations and justifications for the above assumptions are the same as in theorem 1. Condition ½10� just accounts for lack of full certainty about the old evidence, and ½20� is identical to ½2�. Moreover, condition ½3� of the- orem 1 can, with the same line of reasoning, be extended to condition ½30� in theorem 2. Condition ½30� sharpens ½3� by a factor of 1 − ɛ but leaves the qualitative argument for ½3� intact. As long as pðE j T; :RÞ and pðE j T; RÞ decrease by roughly the same margin, the result of theorem 1 transfers to theorem 2. Thus, we can extend the novel solution of POE to the case of residual uncertainty about the old evidence E—a case that is highly relevant for case studies in the history of science. If we compare this solution of the POE to Jeffrey’s and Earman’s proposals, we note that our assumptions ½1�, ½2�, and ½3� are silent on whether Jeffrey’s d—or Earman’s ϕ and w, for that matter— is true or false.2 Hence, we can discard Jeffrey’s dubious assumption d that increasing empirical content makes a theory more plausible, without jeop- ardizing our own results. 2. A Mathematica notebook file that checks the relevant probability models is appended in a zip file online as proof of this claim. In the model checking, I used the PrSAT decision procedure for probabilistic models invented by Branden Fitelson ð2008Þ. This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms I have thus provided a solution of the dynamic POE that makes less de- manding assumptions than Jeffrey and Earman and that achieves stronger results than Garber and Niiniluoto. The final section discusses the reper- THE PROBLEM OF OLD EVIDENCE 395 cussions of these results on the general debate about POE and BCT. 5. Discussion. This article has analyzed some Bayesian attempts to solve the Problem of Old Evidence ðPOEÞ, and it has proposed a new solution. I started with a distinction between the static and the dynamic aspect of the problem. Then I presented my criticism of the Garber-Jeffrey-Niiniluoto ðGJNÞ approach and my own solution proposal to the dynamic POE. Like Howson, I rely on counterfactual credences, but unlike him, I use them for developing a more plausible and robust solution to the dynamic POE. Since I have already defended the specific assumptions of theorems 1 and 2, this last section is devoted to placing my proposal in the literature and to a gen- eral synopsis of the relation between POE and Bayesian Confirmation The- ory ðBCTÞ. Let us first reconsider the relation between the static and the dynamic POE. The static problem concerns, in a nutshell, the question whether E provides evidence for T, and the dynamic problem concerns the question whether discovering T ⊢ E confirms T. The words “evidence” and “con- firmation” are often used synonymously, but they have a different emphasis, namely, explanatory power for the data versus an increase in degree of belief. This is mirrored in the literature in the foundations of statistics ðe.g., Berger and Wolpert 1984; Royall 1997; Lele 2004Þ and also in Bayesian measures of evidential support. Some of them, like the log-likelihood ratio lðT; EÞ 5 pðE j TÞ=pðE j :TÞ, express the difference in explanatory power of T and ¬T for the evidence, whereas others, such as the difference measure dðT; EÞ 5 pðT j EÞ 2 pðTÞ, focus on the increase in degree of belief in T. If we are interested in the dynamic POE—modeling our increase in degree of be- lief in T—then by assumption, old evidence E cannot confirm T. This is why many philosophers have shifted their attention to studying how learning the proposition R :5 T ⊢ E affects our degree of belief in T. In the static POE, however, we compare how expected E is under the competing hypotheses. In scientific practice, such assessments are usually counterfactual, as How- son rightfully remarks. That is, we standardly interpret pðE | TÞand pðE | :TÞ as principled statements about the predictive import of ±T on E, without referring to our complete observational record. Such judgments are part and parcel of scientific reasoning, for example, in statistical inference, where the- ories T, T 0, and so on, impose definite probability distributions on the obser- vations, and our credences pðE | TÞ, pðE | T 0Þ, and so on, follow suit ðsee also Ramsey 1926; Sprenger 2010Þ. So I believe that a resolution of the static problem should be not technical but conceptual—by spelling out why central aspects of scientific reason- This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms ing are suppositional and counterfactual. If we managed to substantiate this claim, the static POE would vanish because the relevant concept of evi- dence would differ from an increase in degrees of belief. This would not 396 JAN SPRENGER imply a farewell to Bayesian reasoning since subjective degrees of belief are required for weighing hypotheses within complex models, integrating out nuisance parameters, and so on. However, making such a counterfac- tual theory of scientific reasoning explicit and articulating the role of BCT in this context goes beyond the scope of this article. I hope to address the challenge in a future paper by means of a detailed analysis of scientific reasoning patterns. It is also worth mentioning that our treatment of the POE allows for a distinction between theories that have been constructed to explain the old evidence E and those that explain E surprisingly ðlike Einstein’s GTRÞ. In the first case, we would not speak about proper confirmation. Indeed, if we accommodate the parameter values of a general theory T such that it explains the old evidence E, then R is actually a certainty, conditional on E: pðR | EÞ 5 1. This is because T has been designed to explain E. As a con- sequence, pðT j E; RÞ 5 pðT j EÞ, and R fails to confirm T. Whereas in the case of a surprising discovery of an explanatory relation between T and E, pðR | EÞ < 1. The degree of confirmation that R confers on T then gets bigger, the more surprising R is—a property that aligns well with our in- tuitive judgment that surprising explanations have special cognitive value. Finally, a general but popular critique of Bayesian approaches to the POE is inspired by the view that the POE poses a principled and insoluble prob- lem for BCT. For instance, Glymour writes at the end of his discussion of the POE: “our judgment of the relevance of evidence to theory depends on the perception of a structural connection between the two, and . . . degree of be- lief is, at best, epiphenomenal. In the determination of the bearing of evi- dence on theory there seem to be mechanisms and strategems that have no apparent connections with degrees of belief ” ð1980, 92–93Þ. What Glymour argues here is not so much that a specific formal aspect of the Bayesian apparatus ðe.g., logical omniscienceÞ prevents it from solving the POE but that these shortcomings are a symptom of a more general in- adequacy of BCT: the inability to capture structural relations between evi- dence and theory. This criticism should not be misunderstood as claiming that confirmation has to be conceived of as an objective relation that is in- dependent of contextual knowledge or contingent background assumptions. Rather, it suggests that solutions to the dynamic POE mistake an increase in degree of belief for a structural relation between T and E. But what makes E relevant for T is not the increase in degree of belief pðT j EÞ > pðTÞ but the entailment relation between T and E—hence, Glymour’s verdict that BCT gives “epiphenomenal” results. To my mind, this criticism is too fundamental to be a source of concern: it does not only affect solutions to the POE, but it directly attacks the en- This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms tire Bayesian explication of confirmation as increase in degree of belief. However, BCT can point to a lot of success stories: explaining the confir- matory value of evidential diversity, mitigating the tacking by conjunction THE PROBLEM OF OLD EVIDENCE 397 paradoxes, resolving the raven paradox, and so on ðsee Crupi 2013 for a detailed reviewÞ. What I have shown in this article is that the confirmatory power of old evidence might be added to this list. Appendix Proofs Proof of Theorem 1. First, we define e1 5 pðE j T; RÞ e2 5 pðE j :T; RÞ e3 5 pðE j T; :RÞ e4 5 pðE j :T; :RÞ t 5 pðTÞ r 5 pðR j TÞ t 0 5 pðT j RÞ �r 5 pðR j :TÞ: By making use of ½1� ðe1 5 1Þ, ½2� ðe2 5 e4 > 0Þ, and the extension the- orem pðX j ZÞ 5 pðX j Y; ZÞpðY j ZÞ 1 pðX j :Y; ZÞpð:Y j ZÞ, we can quickly verify the identities pðE j TÞ 5 pðE j T; RÞpðR j TÞ 1 pðE j T; :RÞpð:R j TÞ 5 r 1 e3ð1 2 rÞ pðE j :TÞ 5 pðE j :T; RÞpðR j :TÞ 1 pðE j :T; :RÞpð:R j :TÞ 5 e2�r 1 e4ð1 2 �rÞ 5 e2 that will be useful later. Second, we note that by Bayes’s Theorem and as- sumption ½1�, pðT j E; RÞ 5 pðT j RÞ pðE j T; RÞ pðE j RÞ 5 1 1 pð:T j RÞ pðT j RÞ pðE j :T; RÞ pðE j T; RÞ � �21 5 1 1 1 2 t 0 t 0 � e2 � �21 : ðA1Þ This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms Third, we observe that by ½1�, ½2�, and the above identities for pðE j TÞ and pðE j :TÞ, 398 JAN SPRENGER pðT j EÞ 5 pðTÞ pðE j TÞ pðEÞ 5 1 1 pð:TÞ pðTÞ pðE j :TÞ pðE j TÞ � �21 5 1 1 1 2 t t e2 r 1 e3ð1 2 rÞ � �21 : ðA2Þ We also note by ½3� that e3 5 1 2 �rð Þ=�r½ � r= 1 2 rð Þ½ �. This allows us to derive r 1 e3ð1 2 rÞ < r 1 1 2 �r �r r 1 2 r ð1 2 rÞ 5 r � 1 1 1 2 �r �r � � 5 r �r ; ðA3Þ and in addition, 1 2 t 0 t 0 5 pð:T j RÞ pðT j RÞ 5 pðR j :TÞ pðR j TÞ � pð:TÞ pðTÞ 5 1 2 t t � �r r : ðA4Þ All this implies that pðT j E; RÞ pðT j EÞ 5 A3ð Þ; A4ð Þ 1 1 1 2 t 0 t 0 � e2 � �21 � 1 1 1 2 t t e2 r 1 e3ð1 2 rÞ � � > A5ð Þ 1 1 1 2 t 0 t 0 � e2 � �21 � 1 1 1 2 t t e2�r r � � 5 A6ð Þ 1 1 1 2 t 0 t 0 � e2 � �21 � 1 1 1 2 t 0 t 0 e2 � � 5 1; completing the proof. The second line has also made use of e2 > 0, as ensured by ½2�.3 QED 3. It might be objected that we have tacitly assumed that t 5 pðTÞ < 1. This is right insofar as if pðTÞ 5 1, then no confirmation can occur. However, pðTÞ < 1 is entailed by ½1�–½3�. If pðTÞ 5 1, then also pðE j RÞ 5 1, because R states that T implies E, and T is ðalmostÞ certain. However, pðE j RÞ 5 pðE j :T; RÞpð:T j RÞ 1 pðE j T; RÞpðT j RÞ This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms , Proof of Theorem 2. By means of performing the same steps as in the proof of theorem 1, we can easily verify the equalities THE PROBLEM OF OLD EVIDENCE 399 pðT j E; RÞ 5 1 1 pð:T j RÞ pðT j RÞ pðE j :T; RÞ pðE j T; RÞ � �21 5 1 1 1 2 t 0 t 0 � e2 1 2 ε � �21 pðT j EÞ 5 1 1 pð:TÞ pðTÞ pðE j :TÞ pðE j TÞ � �21 5 1 1 1 2 t t e2�r 1 e4ð1 2 �rÞ ð1 2 εÞr 1 e3ð1 2 rÞ � �21 5 1 1 1 2 t t e2 ð1 2 εÞr 1 e3ð1 2 rÞ � �21 ; where we have made use of ½10� and ½20� 5 ½2�. We also note that ½30� implies ð1 2 εÞr 1 e3ð1 2 rÞ < ð1 2 εÞr 1 ð1 2 εÞ 1 2 �r �r r 1 2 r ð1 2 rÞ 5 ð1 2 εÞ � r � 1 1 1 2 �r �r � � 5 ð1 2 εÞ r �r : This brings us to the final calculation: pðT j E; RÞ pðT j EÞ 5 1 1 1 2 tð Þ=t½ � e2= 1 2 εð Þr 1 e3ð1 2 rÞ½ �f g 1 1 1 2 t 0ð Þ=t 0½ � � e2= 1 2 εð Þ > 1 1 1 2 tð Þ=t½ � 1= 1 2 εð Þ½ � �r=rð Þe2 1 1 1 2 tð Þ=t½ � � �r=rð Þ � e2= 1 2 εð Þ 5 1; where we have, in the penultimate step, also applied equation ðA4Þ. This completes the proof. QED and from ½2� and the implicit condition in ½3� ðpðT j RÞ < 1Þ it follows that this ex- pression is smaller than 1. Hence, we have a contradiction, and pðTÞ must be smaller than 1, too. This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms REFERENCES Berger, J. O., and R. L. Wolpert. 1984. The Likelihood Principle. Hayward, CA: Institute of Math- ematical Statistics. 400 JAN SPRENGER Bovens, L., and S. Hartmann. 2003. Bayesian Epistemology. Oxford: Oxford University Press. Brush, S. G. 1989. “Prediction and Theory Evaluation: The Case of Light Bending.” Science Chih Chri Crup Daw Earm Eells —— Eells Fitel —— —— Garb Glym Háje Hart How —— —— How Jeffr Lele Niin Popp Ram Roy 246:1124–29. ara, C. 1987. “Some Unresolved Problems in Bayesian Confirmation Theory.” British Journal for the Philosophy of Science 38:551–60. stensen, D. 1999. “Measuring Confirmation.” Journal of Philosophy 96:437–61. i, V. 2013. “Confirmation.” In Stanford Encyclopedia of Philosophy, ed. E. Zalta. Stanford, CA: Stanford University. http://plato.stanford.edu/entries/confirmation/#BayConThe. id, R., S. Hartmann, and J. Sprenger. 2015. “The No Alternative Argument.” British Journal for the Philosophy of Science 66 ð1Þ: 213–34. an, J. 1992. Bayes or Bust? Cambridge, MA: MIT Press. , E. 1985. “Problems of Old Evidence.” Pacific Philosophical Quarterly 66:283–302. —. 1990. “Bayesian Problems of Old Evidence.” In Scientific Theories, ed. C. W. Savage, 205–23. Minnesota Studies in the Philosophy of Science 14. Minneapolis: University of Min- nesota Press. , E., and B. Fitelson. 2000. “Measuring Confirmation and Evidence.” Journal of Philosophy 97:663–72. son, B. 1999. “The Plurality of Bayesian Measures of Confirmation and the Problem of Measure Sensitivity.” Philosophy of Science 66:S362–S378. —. 2008. “A Decision Procedure for Probability Calculus with Applications.” Review of Sym- bolic Logic 1:111–25. —. 2014. “Earman on Old Evidence and Measures of Confirmation.” Unpublished manu- script, University of California, Berkeley. http://fitelson.org/research.htm. er, D. 1983. “Old Evidence and Logical Omniscience in Bayesian Confirmation Theory.” In Testing Scientific Theories, ed. J. Earman, 99–132. Minnesota Studies in the Philosophy of Science 10. Minneapolis: University of Minnesota Press. our, C. 1980. Theory and Evidence. Princeton, NJ: Princeton University Press. k, A., and S. Hartmann. 2010. “Bayesian Epistemology.” In A Companion to Epistemology, ed. J. Dancy et al., 93–106. Oxford: Blackwell. mann, S., and J. Sprenger. 2010. “Bayesian Epistemology.” In Routledge Companion to Epistemology, ed. S. Bernecker and D. Pritchard, 609–20. London: Routledge. son, C. 1984. “Bayesianism and Support by Novel Facts.” British Journal for the Philosophy of Science 35:245–51. —. 1985. “Some Recent Objections to the Bayesian Theory of Support.” British Journal for the Philosophy of Science 36:305–9. —. 1991. “The ‘Old Evidence’ Problem.” British Journal for the Philosophy of Science 42: 547–55. son, C., and P. Urbach. 2006. Scientific Reasoning: The Bayesian Approach. 3rd ed. La Salle, IL: Open Court. ey, R. 1983. “Bayesianism with a Human Face.” In Testing Scientific Theories, ed. J. Earman, 133–56. Minnesota Studies in the Philosophy of Science 10. Minneapolis: University of Min- nesota Press. , S. 2004. “Evidence Functions and the Optimality of the Law of Likelihood.” In The Nature of Scientific Evidence, ed. M. Taper and S. Lele, 191–216. Chicago: University of Chicago Press. iluoto, I. 1983. “Novel Facts and Bayesianism.” British Journal for the Philosophy of Science 34:375–79. er, K. R. 1959/2002. Logic of Scientific Discovery. London: Routledge. sey, F. P. 1926. “Truth and Probability.” In Philosophical Papers, ed. D. H. Mellor, 52–94. Cambridge: Cambridge University Press. all, R. 1997. Scientific Evidence: A Likelihood Paradigm. London: Chapman & Hall. This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms Sprenger, J. 2010. “Statistical Inference without Frequentist Justifications.” In EPSA Epistemol- ogy and Methodology of Science: Launch of the European Philosophy of Science Associa- tion, vol. 1, ed. M. Suárez et al., 289–97. Berlin: Springer. ———. 2015. “Confirmation and Induction.” In Oxford Handbook of the Philosophy of Science, Wile THE PROBLEM OF OLD EVIDENCE 401 ed. P. Humphreys. Oxford: Oxford University Press, forthcoming. s, A. 1995. “Modular Elliptic Curves and Fermat’s Last Theorem.” Annals of Mathematics 142:443–551. This content downloaded from �������������87.79.184.140 on Mon, 04 May 2020 12:30:16 UTC������������� All use subject to https://about.jstor.org/terms