EXPLANATION, CAUSATION AND DEDUCTION

EXPLANATION, CAUSATION AND DEDUCTION

THE UNIVERSITY OF WESTERN ONTARIO SERIES IN PHILOSOPHY OF SCIENCE A SERIES OF BOOKS IN PHILOSOPHY OF SCIENCE, METHODOLOGY, EPISTEMOLOGY, LOGIC, HISTORY OF SCIENCE, AND RELATED FIELDS Managing Editor ROBERT E. BUTTS Dept. of Philosophy, University of Western Ontario, Canada Editorial Board JEFFREY BUB, University of Western Ontario L. JONATHAN COHEN, Queen's College, Oxford WILLIAM DEMOPOULOS, University of Western Ontario WILLIAM HARPER, University of Western Ontario JAAKKO HINTIKKA, Florida State University, Tallahassee CLIFFORD A. HOOKER, University of Newcastle HENRY E. KYBURG, JR., University of Rochester AUSONIO MARRAS, University of Western Ontario JURGEN MITTELSTRASS, Universitiit Konstanz JOHN M. NICHOLAS, University of Western Ontario GLENN A. PEARCE, University of Western Ontario BAS C. VAN FRAASSEN,Princeton University VOLUME 26

FRED WILSON Philosophy Department, University of Toronto, Canada EXPLANATION, CAUSATION AND DEDUCTION D. REIDEL PUBLISHING COMPANY A MEMBER OF THE KLUWER ACADEMIC PUBLISHERS GROUP DORDRECHT/BOSTON/LANCASTER

Libraty of Congress Cataloging in Publication Data Wilson, Fred, 1937- Explanation, causation, and deduction. (The University of Western Ontario series in philosophy of science; v.26) Bibliography: p. Includes indexes. 1. Explanation (Philosophy) 2. Logic. 3. Causation. I. Title. II. Series. BD237.W54 1985 162 85-10797 ISBN-J3: 978-94-010-8818-3 e-isbn-j3: 978-94-009-5249-2 DOl: 10.1007/978-94-009-5249-2 Published by D. Reidel Publishing Company, P.O. Box 17, 3300 AA Dordrecht, Holland Sold and distributed in the U.S.A. and Canada by Kluwer Academic Publishers, 190 Old Derby Street, Hingham, MA 02043, U.S.A. In all other countries, sold and distributed by Kluwer Academic Publishers Group P.O. Box 322, 3300 AH Dordrecht, Holland D. Reidel Publishing Company is a member of the Kluwer Group. All Rights Reserved 1985 by D. Reidel Publishing Company Softcover reprint of the hardcover 1st edition 1985 No part of the material protected by this copyright notice may be reproduced or utilized in any foim or by any means, electronic or mechanical, including photocopying, recording or by any information storage and retrieval system, without written permission from the copyright owner

To Gustav Bergmann

TABLE OF CONTENTS CHAPTER 1 / THE DEDUCTIVE MODEL OF EXPLANATION: A STATEMENT 1.1. Explanation and Deduction 1.2. The Humean Account of Laws 1.3. The Evidential Worth of Law-Assertions 1.4. That Some Explanations Are Better than Others 1.5. That Technical Rules of Computation Are Laws ix 2 10 23 43 66 CHAPTER 2 / THE REASONABILITY OF THE DEDUCTIVE MODEL 73 2.1. Why Ought the Deductive Model Be Accepted? 75 2.2. Are There Reasoned Predictions Which Are Not Explanations? 86 2.3. Is Correlation Less Explanatory than Causation? 106 2.4. Is Causation Inseparable from Action? 118 2.5. Are There Explanations Without Predictions? 136 2.6. Explanation and Judgment 159 CHAPTER 3 / EXPLANATIONS AND EXPLAININGS 3.1. Explanations in the Context of Communication 3.2. Formalist Criticisms of the Deductive Model 3.3. Explanations and Explanatory Content 3.4. Narrative and Integrating Explanations 3.5. Are Laws Evidence for, or Part of, Explanations? 3.6. Can We Know Causes Without Knowing Laws? CONCLUSION NOTES Notes to Chapter 1 Notes to Chapter 2 173 173 187 231 251 263 290 343 345 345 351 vii

viii T ABLE OF CONTENTS Notes to Chapter 3 353 Notes to Conclusion 365 BIBLIOGRAPHY 367 NAME INDEX 375 SUBJECT INDEX 377

The purpose of this essay is to defend the deductive-nomological model of explanation against a number of criticisms that have been made of it. It has traditionally been thought that scientific explanations were causal and that scientific explanations involved deduction from laws. In recent years, however, this three-fold identity has been challenged: there are, it is argued, causal explanations that are not scientific, scientific explanations that are not deductive, deductions from laws that are neither causal explanations nor scientific explanations, and causal explanations that involve no deductions from laws. The aim of the present essay is to defend the traditional identities, and to show that the more recent attempts at invalidating them fail in their object. More specifically, this essay argues that a Humean version of the deductive-nomological model of explanation can be defended as (1) the correct account of scientific explanation of individual facts and processes, and as (2) the correct account of causal explanations of individual facts and processes. The deductive-nomological model holds that to explain an event E, say that a is G, one must find some initial conditions C, say that a is F, and a law or theory T such that T and C jointly entail E, and both are essential to the deduction. Thus, if the law T is 'All Fare G', or, iii symbols, '(x) (Fx :J Gx)" then the argument, assuming its premisses are true, and that the general premiss is a law, T: (x) (Fx :J Gx) C:Fa E: Ga will be an explanation of a's being G, and equally, if a's being F is observed prior to its being observed to be G, it will be a prediction of a's being G. Thus, to defend the deductive-nomological model of explanation is also to defend the symmetry thesis, that every explanation is a potential prediction an that every prediction - that is, every reasoned prediction, every prediction based on laws - is an explanation. The criticisms we shall be examining proceed largely by way of counterexample. One group attacks the model by way of the symmetry thesis, ix

x purporting to discover explanations which are not predictions and (reasoned) predictions which are not explanations. Another group attacks the model on formalist grounds, giving examples which satisfy the requirement of law-deduction but which do not seem to be explanations, or showing how, formally, any event seems to be able to explain any other event. A third group attacks the model on grounds that it fails to do justice to the pragmatics of explanation. A fourth group attacks the model on grounds that there are explanations, and, specifically, causal explanations, which do not involve laws or deduction from laws. Interestingly enough, as we shall see, the last three groups are not unrelated. For a matter-of-fact generalization 'All Fare G' to serve as a premiss in an explanatory argument, it must be a law. A necessary condition for a matter-of-fact generalization to be a law, is that it be true. This is not sufficient for a generalization to be a law, however, since there are true matter-of-fact generalizations, e.g., "all the coins now in my pocket are copper", which are, as one says, accidental generalities, i.e., not laws. In order to discuss the criticisms indicated above, it is necessary for this essay to take some stand on the issue of what constitutes a law. The position adopted is that of Hume (Treatise of Human Nature) and Mill (System of Logic) that what distinguishes those generalizations that count as laws and those that do not is contextual or pragmatic. Lawfulness is not a syntactic or semantic feature of generalizations but rather is, as Rescher (Scientific Explanation) puts it, a matter of imputation. A generalization is accepted as a law, or, as I shall say, is law-asserted just in case that we are prepared to use it for purposes of explanation, prediction, and the support of contrary-tofact conditionals. And one is justified in law-asserting a generalization just in case that the evidence for its truth has been collected in accordance with the rules of the scientific method. This is the position proposed by Humeand defended by Mill, and, more recently, Bergmann (Philosophy of Science), Nagel (The Structure of Science), and Rescher. The present essay does not undertake a full-scale defence of this Humean account of laws. It has, of course, come under attack recently, both by those, e.g., Armstrong (Universals and Scientific Realism), who speak of primitive necessary connections among properties - a theme that goes back to Aristotle - and by those, e.g., Lewis (Counterfactuals), who propose to analyze lawfulness in terms of possible worlds. But as a single essay cannot do everything, many of the points raised by these critics are not dealt with; the aim is not so much to look at objections to the Humean account of laws as to look at objections to the deductive-nomological model of explanation. For

xi this reason, while we adopt the Humean account of lawfulness, its defence is only sketched in, just enough for our purposes with no full-scale discussion attempted. In adopting the Humean account of laws, we introduce not merely the objective conditions of validity and true premisses for an argument to count as an explanation but also the subjective condition of the general premiss being supported by scientific evidence for an argument to be acceptable as an explanation. It is subjective because the evidence available varies from time to time and from person to person, so that a generalization may not be lawassertible at one time or for one person but law-assertible at another time or for another person. It therefore varies from subject to subject; and, unlike validity and truth, is not objective, independent of the subject. The thesis we are defending, then, is that law-deduction is necessary and sufficient for scientific explanations, bearing in mind that what renders a generalization a law is not a matter of either syntax or semantics nor, indeed, of truth alone, but also evidential context. Included in this thesis are the subtheses that there are no scientific explanations which are not potentially predictions, and no (law-based) predictions which are not explanations. Furthermore, also included is the subthesis that all causal explanations, insofar as they are genuine explanations, are scientific and therefore both deductive from laws and potentially predictions. Several gambits will be used in replying to the critics of the deductivenomological model as we attempt to show how their proposed counterexamples really do fit with that model. Thus. one of the immediate consequences of the Humean-contextualist account of laws is that an argument may be acceptable as explanatory in one evidential context but not in another. We shall make use of this point in several of our replies to critics of the deductive-nomological model of explanation. Another of the immediate consequences of the Humean account of laws is that, since we observe only a sample and not the population, and since a generalization is about a population and not the sample, we never have conclusive evidence that a generalization we assert or law-assert is true. It may well be, then, that a generalization the law-assertion of which is supported by available data is in fact false. Thus, an argument may well be subjectively acceptable as an explanation but still not an explanation if the objective condition of true premisses is not fulfilled We shall also make use of this point in replying to critics of the deductive-nomological model of explanation.

xii Deductive-nomological explanations interest us because they provide knowledge that permits prediction and control. That is, we have certain pragmatic interests in obtaining such knowledge. We also are sometimes interested in such explanations for their own sake; i.e., to speak with Veblen, we are interested in them out of idle curiosity. Now, one explanation may better satisfy these cognitive interests - pragmatic interests and idle curiosity - than does another. This permits us to distinguish better and worse among explanations according to the extent to which they satisfy the cognitive interests that motivate us to obtain explanations. We shall make extensive use of this point when we reply to critics of the deductive-nomological model. In particular, we shall discover that many of them fail to distinguish, what we shall insist are distinct, an argument being a non-explanation from an argument being a weak explanation. Those who claim to find predictions which are not explanations are often guilty of this confusion. It should be noted, by the way, that while our cognitive interests, that is, our subjective values, determine a scale of better and worse among explanatins, once that scale is give then whether or not an explanation falls at a higher or lower place on this scale is an objective matter. If we have two explanations available, and one is better than another on the scale of worth defined by our cognitive interests, then, other things being equal, it would be wrong to use the weaker, less desirable, explanation as an act of explaining something to someone. It is important to distinguish (I) and explanation, i.e., an argument, (2) the acceptability of an argument as an explanation, and (3) the use of an acceptable explanation in an act of explaining. In particular, an argument may be acceptable as an explanation but for all that not appropriate for use in an act of explaining. Critics of the deductive-nomological model often confuse an argument's being inappropriate for use in an act of explaining with its being a non-explanation. Some critics of the model tend to rely upon an assumption that any deductive-nomological explanation can have a law-premiss no more complicated in logical form than 'All F and G'. This is often the case with those who claim to have found explanations which are not predictions. These criticisms can be shown to be invalid once it is noted that there are laws in such explanations, and that these laws do not violate the symmetry thesis, but that the logical form of the explanatory argument is more complex than that of a syllogism in Barbara. The same point also serves to rebut criticisms based on the idea that some explanations somehow rely upon 'intuitive judgment' rather than scientific reason to determine their acceptability; and is relevant, too, to

xiii criticisms based on the idea that there are explanations which are causal but which do not involve laws, or law-deduction. Finally, the distinction between better and worse among explanations enables one to distinguish laws into various kinds, e.g. process laws, causal laws, correlations. Criticis of the deductive-nomological model are often guilty of confusing the concept of explanation with one of the special cases, very often that of causal explanation. However, while a law that gives a correlation may not yields a causal explanation, and may be weaker, or worse as an explanation, than a causal explanation, it does not follow, as too many wrongly infer, that correlations have no explanatory power, that arguments with correlations as their law-premisses are non-explanatory. The Humean version of the deductive-nomological model that we propose to defend is laid out in Chapter 1. This chapter is essentially exposition, rather than defence. It begins by stating briefly the thesis that explanation of individual events is by deduction from generalities and initial conditions. However, not all generalities yield explanations: for the latter, the general premiss must be a law. The thesis we are defending is that the Humean account of laws suffices. On this account, the characteristic of being lawlike is contextual: a generality is lawlike just in case that we are prepared to use it to support the assertion of counterfactual conditionals. It is then argued that some generalities are more worthy of being treated as lawlike than others; a generality is worthy of being treated as lawlike just in case that evidence for it has been gathered, or at least advanced, in terms of the norms of scientific research. This involves a crucial distinction between the objective and the subjective (or contextual) conditions that a deduction must satisfy before it can count as, or reasonably be used as, an explanation. Among the relevant contextual elements are matters of evidence. It is not the point of this essay to analyze the logic of confirmation. (Some aspects of this problem - or set of problems - are discussed in my Reasons and Revolutions, currently being prepared for publication.) But clearly, the topic is important enough for what we are about that the issues cannot be wholly avoided, and so a brief discussion of these matters is included. It is suggested that the basic patterns of confirmation are those described by Hume and John Stuart Mill. But, using some ideas of Peirce, it is also suggested that these patterns are compatible with ideas more recently presented by Kuhn. However, no full discussion of these points is attempted. Mill's account of the Principle of Causation leads directly to the idea of what Bergmann has called "process knowledge". It has been suggested by

xiv Bergmann that this notion provides an ideal or standard of excellence for the evaluation of explanations of individual facts and processes. This notion of better and worse explanations is developed in Chapter 1, and will play an important role in what follows. Again, the thrust of the discussion at this point is expository; defence of the notion will be given in Chapter 2. John Stuart Mill's views on confirmation have often been condemned as foolish. Chapter 1 attempts to show that Stuart Mill's position on the justification of the Principle of Causality and, more generally, on the confirmation of laws about laws is a position that is at least plausible, sufficiently plausible, anyway, to permit us to rely upon them for our limited purpose. That purpose is the defence of the Humean version of the deductivenomological model of explanation. This defence begins in Chapter 2. This chapter begins by outlining the positive reasons for accepting the model as one that -- relative to certain cognitive interests - ought to be adopted. It then proceeds to discuss various ways in which the deductive-nomological model has been challenged. It argues that, contrary to some, there are no reasoned predictions (Le., predictions based on deductions from laws) that are not explanations. It argues that, contrary to some, correlations provide explanations just as do causal relations. It argues that, contrary to some, the concept of causation is independent of that of action. It argues that, contrary to some, there do not exist explanations that are not also predictions. Throughout these arguments the distinction between process laws and imperfect knowledge that falls short of this ideal plays a crucial role: in general the argument is that the objections to the deductive-nomological model that we discuss appear successful only if this distinction is ignored. The chapter closes with a criticism of the views of those who hold that 'judgment' plays a role in explanation that is not allowed for by those who defend the deductive-nomological model. Chapter 3 develops the contrast between explanations and acts of explaining. In this context, it examines and rejects a number of recent formalist criticisms of the deductive-nomological model. The chapter then goes on to discuss whether laws are part of, rather than merely presupposed by, certain sorts of scientific explanation, and whether causes can be known, e.g., by 'judgment', without our also knowing laws. It is argued that laws are indeed part of scientific explanations, and that in knowing a cause one ipso facto knows a law. Generally speaking, aspects of the deductive-nomological model are developed only to the extent that they are needed for replying to critics. For example, the defence of the criteria for better and worse scientific

xv explanations of individual fact is merely sketched. Again, since our concern is with explanations of individual facts and events, another sort of deductive explanation, that of laws by theories, receives very little attention. Both these topics -- criteria for better and worse, and the nature of theoretical explanation - receive a more detailed discussion in my forthcoming Reasons and Revolutions. In this connection it is perhaps worth noting a proposal concerning explanations advanced by a number of recent authors such as Michael Friedman (,Explanation and Scientific Understanding', 1974) and Philip Kitcher (,Explanatory Unification', 1981). While these authors disagree about what constitutes unification (cf. Kitcher, 'Explanation, Conjunction and Unification', 1976), they do agree that "to explain is to unify" - Kitcher cites E. M. Forster's injunction: "Only connect" - and argue that models developed in accordance with this injunction ought to replace the now discredited covering law or deductive-nomological model. Now it seems obvious that in a sense I do not agree with this position, since the whole thrust of the present essay is to argue that the deductivenomological model has in fact not been discredited. That being so, there is no need to replace it. On the other hand, the injunction is not without merit: there is indeed a real point to the idea that explanation consists in unification. But of course, the deductive-nomological model already acknowledges this point. This model asserts that to explain an event is to subsume it and another, independent event, under a law that connects the two. It is part of the very idea of the deductive-nomological model that explanation of individual events consists in unifying those events by means of laws. There are, of course, degrees of unifying power. The connection between events that some laws establish is tighter than the connection established by other laws. What we argue below is that, so far as concerns individual events, the ideal of explanation is process knowledge. When one explains an event in terms of process knowledge, that event is located as part of the ongoing development of a system over time, a development for which the process knowledge enables one, on the basis of a knowledge of the present state of the system, to predict the state of the system at all future times and at all past times. Process knowledge fills in all details of a process. Thus, when an event is explained by means of process knowledge it is gaplessly connected to all other events in an ongoing process. In the case of individual events, then, the injunction "Only connect" is fulfilled to the greatest degree possible when process knowledge is attained.

xvi Kitcher, Friedman, and others who adopt their position are thus mistaken in their view that defending the deductive-nomological model is somehow incompatible with defending the thesis that explanation consists in unification. The notion that I'to explain is to unify" is, however, one that has its greatest impact at the level not of explanation of individual facts but at the level of theoretical explanation, the explanation of laws. Our cognitive interest in scope, as well as our interest in theories that can guide research into hitherto undiscovered laws, both lead us to seek theories that unify bodies of law. Such laws about laws (to use Mill's phrase) explain the laws they unify, and, insofar as they apply to areas where the specific laws are not yet known, they predict, and guide researchers to the discovery of, new laws. It is hard to over-estimate the importance of such unifying laws. Yet a thorough discussion of the issues would be out of place in t~e present essay, which has as its main focus of concern the explanation of individual facts rather that the explanation of laws. To this extent, then, a pur sua 1 of these aspects of the issues raised by Kitcher and Friedman is out of place. On the other hand, the issue cannot entirely be avoided, if only because the laws about laws, the unifying laws that explain other laws, can themselves at times be used to explain individual facts. The discussion of Mill in Chapter I gives an account of how certain sorts of generic over-hypotheses can unify more specific hypotheses. According to Kitcher ('Explanatory Unification', p. 519), "The problem of explanation is to specify which set of arguments we ought to accept for explanatory purposes given that we hold certain sentences to be true". The constraints on such arguments are contraints on the laws that appear in these explanatory arguments; and these constraints are themselves factual, that is, lawful. These conditions "jointly imposed by the presence of nonlogical expressions in the pattern and by the filling instructions" as well as "conditions on logical structure" (ibid., p. 518) are precisely the constraints imposed by the generic laws about laws that Mill discussed. The view that we sketch, then, of how theoretical explanation proceeds by unification is close indeed to that of Kitcher. Kitcher does not develop his position at any length. He does not attend to the two specific forms of theoretical unification that have played so great a role in the actual history of science, to wit, unification by means of a composition law, and unification by reducing one area (the macro-area) to another (the micro-area). Nor does Kitcher develop the idea that unifying theories are indispensable tools for guiding research. For a detailed analysis of a generic law and its role in concept formation, one might look at my essay 'Is Operationism Unjust to Temperature?'; and for their role in research

xvii two other essays 'Kuhn and Goodman: Revolutionary vs. Conservative Science', and 'A Note on Hempel on the Logic of Reduction', are relevant. Some aspects of the latter issue are also dealt with in Section 3.3, below. The whole complex of issues is discussed in some detail in another forthcoming essay of mine, Empiricism and Darwin's Science. Another topic that is certainly worth discussing is that of statistical explanations; but this too is almost completely ignored. This in spite of the fact that it is of considerable relevance to what we are about. I can only plead by way of excuse that one cannot deal at once with all the objections to the deductive-nomological model. The objections I have considered all fit together, so it is they that I have discussed, rather than challenges to the deductive-nomological model based on so-called 'statistical explanations'. If the reader wishes some very brief indication of my views on the latter, he might glance at my review of N. Rescher's Scientific Explanation. A number of critics of the deductive-nomological model who shall be discussed have been heavily influenced by the doctrine of so-called 'Oxford philosophy' that 'meaning is use'. I have critically evaluated the latter doctrine in a monograph, Meaning Is Use, currently being prepared for publication. Certain parts of the present work supplement and elaborate points about scientific explanation made more briefly in this latter monograph. It is my belief that many objections to the deductive-nomological model depend upon an inadequate analysis of certain examples of scientific theories and of certain historical developments of science. For that reason, I have ventured detailed analyses where appropriate. In particular, an extended analysis of geometrical optics occurs in Chapter 3. Much of this is straightforward physics, or history of science, yet I am convinced that only by presenting these straightforward things in a deliberately philosophical context can many objections to the deductive-nomological model be adequately met. It is also my view that it is sometimes important to know not only that a conclusion follows but also how it follows. Sometimes a knowledge of the logic of the deduction is important to understanding its philosophical implications. This is particularly so with respect to some points of Davidson and of Kim that are discussed in Chapter 3. I have therefore tried to present such deductions in sufficient detail to make clear what is going on. The result is no doubt more symbolic logic than is comfortable for many. My only excuse is that experience with students and colleagues over the years has convinced me that clear understanding can be attained only by a patient workingthrough of the details.

xviii I would also like to thank the following authors for permission to quote. from the works cited: S. Bromberger: 'Why-Questions'. First appeared in Mind and Cosmos: Essays in Contemporary Science and Philosophy, Robert Colodny, editor. Copyright by the University of Pittsburgh Press, 1966. Used by permission. A. W. Collins: 'Explanation and Causality'. Mind 7S (1966). Used by permission of the editor of Mind. D. Davidson: 'Causal Relations'. First published in Journal of Philosophy 64 (1967). Reprinted in D. Davidson: Actions and Events, Oxford University Press, 1980. Quoted from the latter by permission of the Oxford University Press. S. Toulmin: Philosophy of Science. Hutchison, 1953. Used by permission. G. H. von Wright: 'On the Logic and Epistemology of the Causal Relation'. First appeared in Logic, Methodology and Philosophy of Science IV, P. Suppes et al., editors. North-Holland Publishing Co., 1973. Used by permission. Portions of the following publications have been incorporated into the present essay: 'Hurne and Ducasse on Causal Inference from a Single Experiment', Philosophical Studies 3S (1979), pp. 305-9. Copyright 1979, D. Reidel Publishing Co. 'Goudge's Contribution to Philosophy of Science', in L. W. Sumner, J. G. Slater, and F. Wilson, (eds.), Pragmatism and Purpose: Essays in Honour of T. A. Goudge, University of Toronto Press, Toronto, 1980. Copyright 1980, University of Toronto Press. 'Mill on the Operation of Discovering and Proving General Propositions', Mill News Letter 17 (1982), pp. 1-14. Copyright 1982, University of Toronto Press. Editors and publishers are thanked for permission to include this material. A number of key ideas in this essay were developed at a Working Conference on Causality held at Dalhousie University, July-August 1973. All the participants should be thanked for the stimulation they provided. The organizer, David Braybrooke, deserves special thanks. So, too does the Canada Council which partially funded the conference. The draft of the book was completed while I was on leave from the University of Toronto in 1979-80, made possible by a leave fellowship granted by the Social Sciences and Humanities Research Council of Canada.