CLARENDON LIBRARY OF L O G I C AND P H I L O S O P H Y

General Editor: L. Jonathan Cohen

THE CEMENT OF THE

UNIVERSE
The Clarendon Library of Logic and Philosophy brings together books, by
new as well as established authors, that combine originality of theme with
rigour of statement. Its aim is to encourage new research of a professional
standard into problems that are of current or perennial interest.
General Editor. L. Jonathan Cohen, The Queen's College, Oxford.

Also published in this series

Philosophy without Ambiguity by Jay David Atlas

Quality and Concept by George Bealer
Logic on the Track of Social Change by
David Braybrooke, Bryson Brown, and Peter K. Schotch
Psychological Models and Neural Mechanisms by Austen Clark
Sensory Qualities by Austen Clark
The Diversity of Moral Thinking by Neil Cooper
The Logic of Aspect: An Axiomatic Approach by Anthony Galton
Ontological Economy by Dale Gottlieb
Experiences: An Inquiry into some Ambiguities by J. M. Hinton
The Fortunes of Inquiry by N. Jardine
Metaphor: Its Cognitive Force and Linguistic Structure by
Eva Feder Kittay
The Cement of the Universe: A Study of Causation by J. L. Mackie
A Model of the Universe: Space-Time, Probability, and Decision by
Storrs McCall
The Nature of Necessity by Alvin Plantinga
Divine Commands and Moral Requirements by P. L. Quinn
Pluralism: Against the Demand for Consensus by Nicholas Rescher
Rationality: A Philosophical Inquiry into the Nature and the Rationale of
Reason by Nicholas Rescher
The Coherence of Theism by Richard Swinburne
Anti-Realism and Logic: Truth as Eternal by Neil Tennant
Ignorance: A Case for Scepticism by Peter Unger
The Scientific Image by Bas C. Van Fraassen
Chance and Structure: An Essay on the Logical Foundations of
Probability by John M. Vickers
Slippery Slope Arguments by Douglas Walton
What Is Existence? by C. J. F. Williams
Works and Worlds of Art by Nicholas Wolterstorff
 THE
CEMENT OF THE
UNIVERSE
A Study of Causation

J. L. MAGKIE

OXFORD
AT THE CLARENDON PRESS
 This book has been printed digitally and produced in a standard specification
in order to ensure its continuing availability

OXFORD
UNIVERSITY PRESS
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University's objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Bangkok Buenos Aires Cape Town Chennai
Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata
Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi
Sao Paulo Shanghai Singapore Taipei Tokyo Toronto

Oxford is a registered trade mark of Oxford University Press

in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© Oxford University Press 1980
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
Reprinted 2002
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose this same condition on any acqiiirer
ISBN 0-19-824642-0
. . . and as these [ Resemblance, Contiguity, and
Causation ] are the only ties of our thoughts, they
are really to us the cement of the universe, and
all the operations of the mind must, in a great
measure, depend on them.
HUME, Abstract
This page intentionally left blank
 Preface

THIS work, like my previous Truth, Probability, and Paradox, was

completed during the tenure of a Radcliffe Fellowship for the
years 1971-2 and 1972-3, and I have to thank the Radcliffe
Trustees for this award, the Master and Fellows of University
College for allowing me to accept it, and Mr. Peter Singer for
taking over, as Radcliffe Lecturer, nearly all my undergraduate
teaching.
Several parts of the book are based on previously published
work, and I have to thank the editors and publishers in question
for permission to use this. Chapter 3 includes material from an
article, 'Causes and Conditions', published in the American Philo-
sophical Quarterly, ii (1965), 245-64, which is used by permission
of the publishers, Basil Blackwell and Mott Ltd. Chapter 7 is
a considerably revised version of an article, 'The Direction of
Causation', published in the Philosophical Review, Ixxv (1966),
441-66, which is incorporated by permission of the editors. The
Appendix is similarly a revised version of an encyclopedia
article for which I have been asked to use the following
form of acknowledgement: ' "Mill's Methods of Induction" by
J. L. Mackie. Reprinted with permission of the Publisher
from the ENCYCLOPEDIA OF PHILOSO PHY edited by Paul Edwards.
Volume 5, pages 324-332. Copyright © 1967 by Crowell Collier
and Macmillan, Inc.'
Several other parts are based on unpublished papers which
I have read and discussed at various times and places. My
views about causation have developed in a haphazard way over
many years, and I have profited from many arguments and
suggestions which I cannot now trace to their original source.
I can only express my gratitude and offer my apologies gener-
ally to those whose ideas I have unconsciously plagiarized.
I would also like to thank Mrs. E. Hinkes for typing the book.
J. L. M.
February 1973
 Preface to the
Paperback Edition

IN this edition a serious misprint has been corrected on page

164, where sixteen words were omitted after line 24; in order to
accommodate these words small changes, which do not alter the
sense, have been made on this page, minor misprints have been
corrected on pages 51 (line n), 52 (line 10), 99 (line 8 up),
224 (note 28), 255 (line 12), 263 (line 15 up), 303 (line 12),
and 328 (under 'Inus condition' and 'Laws'). Additional
notes have been inserted referring to pages 80, 143, 190, 214,
and 238, being marked by asterisks and printed on p. 322.
There is a short Additional Bibliography on page 326, Other-
wise the text is unchanged.
Some reviewers and other readers have found my argument
hard to follow, mainly because, instead of first stating my theory
of causation and then defending it, I work towards it from
several different starting points and by way of attempts to solve
a number of problems, partly criticizing and partly taking over
the views of a good many writers. My position is in some respects
intermediate between those of the best known rival schools of
thought about causation—it has, understandably, been attacked
both for being too Humean and for not being Humean enough
—and I could hardly define it intelligibly or explain why I hold
it except by tracing the struggles through which I have reached
it. Nevertheless, it might have been helpful if I had begun with a
rough outline and conspectus of my final view, and I shall now
try to make good that deficiency. This may be of some use as an
introduction, but it will itself be more fully understood if
readers also come back to it after having read the book; it may
then guard against misinterpretations by indicating how the
different parts of my theory fit together.
My treatment is based on distinctions between three kinds of
 PREFACE TO THE PAPERBACK E D I T I O N ix
analysis, factual, conceptual, and epistemic. It is one thing to
ask what causation is *in the objects', as a feature of a world that
is wholly objective and independent of our thoughts, another to
ask what concept (or concepts) of causation we have, and yet
another to ask what causation is in the objects so far as we know
it and how we know what we do about it. Verificationism,
which has been (openly or covertly) influential in much recent
philosophy, has tended to blur these distinctions, suggesting
that our concepts and what our statements mean, and also
what we can intelligibly claim to be the case, or even speculate
about, are tied to what is somehow verified or at least verifiable.
My contrary insistence on these distinctions reflects a funda-
mentally realist view.
In the objects, causation can be seen as a relation between
concrete events. But more precisely it is a relation between
instantiated features: there will be some conjunction of features
in the cause-event (including the 'circumstances' of what we
initially identify as the cause) which are all and only those
causally relevant to the set of features that we take as the effect.
(Once a specific effect has been chosen, it is a wholly objective
matter what features are causally relevant to it.) But what
constitutes this causal relevance which connects the cause-
features with the effect-features? There are regularities of
succession, typically complex (what is regularly related to
some effect-feature will be a disjunction of conjunctions of
positive and negative causal factors), and these regularities may
be quantitative, involving functional dependences as well as
simple presences and absences of items on either the cause or the
effect side, and again may be statistical or probabilistic rather
than deterministic. But regularities, of any sort, do not exhaust
the matter. There is also a direction of causation, an asymmetry
between cause and effect which is not simply temporal order,
though it is related to this. I say something below about our
knowledge of this asymmetry and our concept of it; but the ob-
jective asymmetry is, I admit, obscure and elusive. Again, in
an individual cause-effect sequence, over and above the fact
that it instantiates some regularity, there is more than one sort
of continuity of process, in particular, as well as spatio-temporal
continuity, there is what I call qualitative continuity: some-
thing carries over from cause to effect, causal relations are
x P R E F A C E TO THE PAPERBACK EDITION
forms of partial persistence as well as of change. Thus, contrary
to the Humean view, there is more in objective causation than
regularity, but contrary to the traditional rationalist view, this
extra element is nothing like logical necessity. Since this
objective causation significantly resembles both the continua-
tion of self-maintaining processes and the survival of things
without change (or without easily observable change), these
two kinds of persistence might well be included in an extended
concept of causation in the objects.
Causation as we know it commonly falls far short of this.
The above-mentioned regularities are usually very incompletely
known. Yet I stress that such 'gappy' regularities can be empiri-
cally discovered or confirmed (particularly by methods of
eliminative induction) and can also be used. Causal relations are
not knowable a priori, and I also reject all transcendental
arguments. But I suggest that there is some ground for a pre-
sumption of some element of persistence. I also hold that
the traditional problem of induction can be solved (though the
solution is not given in this book), and hence that there is
empirical support for the assumption that there are some
regularities to be found, an assumption which is used in
applications of the eliminative methods. The continuity or
persistence which, in my view, replaces the rationalists' neces-
sity is a thoroughly empirical feature, in no way mysterious, but
observable or discoverable in principle, though in many cases
it is not immediately perceivable. Our knowledge of causal
asymmetry is based partly on our experience of bringing about
one thing by way of another. But causal asymmetry in itself
is not constituted by this experience, and even our knowledge
of it does not rest wholly on this experience: there is also an
observable asymmetry with respect to what I call the dispersal
of order, which yields a non-reversible direction of explanation.
Whereas regularities of one sort or another play a large part
in causation both in the objects and as we know it, our concept
of causation is not primarily a concept of the instantiation of
regularities: we locate causation in singular cause-effect
sequences, and assume that there is some distinctively causal
relation, some influence or operation, in each such sequence
taken on its own. But neither is it a concept of anything like
logical necessity or even a priori knowability or intelligibility
 PREFACE TO THE PAPERBACK E D I T I O N xi
of sequences. The basic notion is best captured by various
contrary-to-fact conditionals. What we call a cause typically is,
and is recognized as being, only a partial cause; it is what makes
the difference in relation to some assumed background or causal
field. And where we take A to be thus a partial cause of B, we
can say that if A had not occurred, B would not; a cause is
taken to be in this counterfactual sense necessary in the circum-
stances for B, though sometimes also sufficient in the circum-
stances as well, or perhaps only sufficient in the circumstances
and not necessary: we have alternative counterfactual concepts
of causation. But these counterfactual conditional relationships
do not exhaust our concept of causation, for this concept
includes also the notion of the asymmetry between cause and
effect; conceptually, that asymmetry seems to be based on a
contrast between what is fixed and what is (at any time) not
fixed: effects become fixed only by way of their causes. Our
concept of causation also includes some presumption of a
'causal mechanism', and hence of the continuities, both spatio-
temporal and qualitative, that I have ascribed to causation in
the objects.
Since I distinguish the analysis of the concept of causation
from the factual analysis of causation in the objects, I am not
committed to giving a fully determinate or tidy conceptual
analysis; rather I insist that our concept is in several ways a bit
indeterminate: 'cause' can mean slightly different things on
different occasions, and about some problematic cases, for
example of over-determination, we may be unsure what to say.
But it is still a fairly unitary concept: we do not have one concept
for physical causation and another for human actions and
interactions (as someone might be forced to say who took our
concept of physical causation to be that of regular succession);
we can and do assert similar counterfactual necessity (and at
times sufficiency) about fields of all different sorts.
Having thus distinguished our concept of causation from
causation as it is and even as it fa known, I must also explain
how these different items are related to one another. The
counterfactual conditionals which are central in our concept
are not, I hold, capable of being simply true, of describing liter-
ally a fully objective reality. But they are suggested by the
resemblances and differences that we find when we compare
xii PREFACE TO THE PAPERBACK E D I T I O N
observed sequences with one another. They are also sustained
by statements about regularities of sequence, even incompletely
known regularities; while these regularities are themselves con-
firmed (by way of the eliminative methods of induction) by
those same combinations of resemblance and difference. A
world in which events take place in accordance with such laws
of working as I ascribe to causation in the objects, even
though these laws are very incompletely known, will be one in
which the ways of thinking that constitute our causal concepts
are both reasonable and useful. Also, both our experience of
action and manipulation and the observable asymmetry with
respect to the dispersal of order, which are central in our dis-
covery of the direction of causation, suggest the contrast be-
tween fixity and non-fixity which is at the heart of our concept
of this asymmetry, and which may also belong to the objective
world.
I should also explain how the various chapters contribute to
this rather complex account.
Chapter i is a critical study of Hume's account of causation,
arguing that while his negative theses are largely correct,
Hume radically misinterprets the ordinary notion of causal
efficacy or necessity which he then proceeds to explain away.
Chapter 2 is devoted to the analysis of this ordinary concept in
terms of counterfactual conditionals along with a preliminary
treatment of the asymmetry which this concept involves; it
offers a psychological explanation of our counterfactual think-
ing, much as Hume offered a psychological explanation of the
different sort of necessity—a priori knowability—which he
wrongly took to be what, in our concepts, distinguishes causal
from non-causal sequences. Chapter 3 studies the complex form
of causal regularities and relates this to our uses of them and to
our methods of discovering them; these methods are studied in
further detail and defended against some stock objections in
the Appendix on Eliminative Methods of Induction. Chapter 4
contains mainly negative criticism both of Kant and of more
recent writers, Strawson and Bennett, who have developed some
of his ideas. Its conclusion is that no sort of transcendentalism
is needed as a supplement to Hume or as a replacement of some
of his doctrines, though it is conceded that causation is a very
pervasive and structurally important feature of the world as we
 PREFACE TO THE PAPERBACK E D I T I O N xiii
know it, and one on which much of our knowledge of the world
depends—for example, it plays a vital part in the determination
of the metric of the time dimension. Chapter 5 defends the
conceptual analysis already given by comparing it with the
account given by Hart and Honore (and by applying it to
some of their legal examples), and also with the accounts of
Ducasse and of some thinkers who appeal to a commonsense
knowledge of particular kinds of causing. It shows that we do
not need to recognize a radical plurality of causal concepts,
but also that there is more in causing than can be simply
perceived; though causing can indeed be observed, the observa-
tion includes an important element of interpretation and relies
greatly on knowledge drawn from the observation of other
cases.
Chapter 6 rebuts Russell's thesis that the concepts of causa-
tion and causal law are obsolete, and shows that the relation of
functional dependence, which he would substitute for it, is
better understood as an addition to and extension of the
'neolithic' concept. It adds that once this extension has been
made, we might well extend the concept in two further ways,
to include the continuation of a uniform process a.nd the
persistence of an unchanging thing as further varieties of
causation.
Chapter 7 deals with the direction of causation, mainly as a
problem in conceptual analysis. More criticism has been direc-
ted against this chapter, I believe, than against all the others
together, and I admit that the topic is highly controversial and
that the positive account suggested at the end of the chapter is
not altogether satisfactory. But most of the chapter is given to
the establishing of a number of negative theses: the soundness
of these has been acknowledged, at least implicitly, for they
have hardly been challenged. Thus it is shown that the concept
of causal asymmetry is not simply that of temporal succession,
but equally that the direction of time cannot be reduced to the
direction of causation (pages 161-6), that it is not that of one-
way determination, as Russell suggested (pages 166-7), an(^
that though causal direction is related to 'effectiveness', to the
possibility of using a cause to bring about an effect, and our
knowledge of the asymmetry depends partly on this sort of
experience, our concept is of an objective asymmetry on which
xiv P R E F A C E TO THE PAPERBACK E D I T I O N
effectiveness somehow depends, and is not the anthropocentric
concept of effectiveness itself (pages 168-73). My discussion
(pages 173-83) of backward (time-reversed) causation has
often been misunderstood. I do not suggest, and I do not believe,
that backward causation occurs, but only that it is conceptually
possible, and I use this conceptual possibility and an examina-
tion of what might be evidence for backward causation merely
to throw light on the concept of causal direction. This exami-
nation strongly suggests that this concept is connected with a
contrast between fixity and non-fixity. It is also shown that en-
tropy increase does not itself constitute the asymmetry of either
time or causation. But a notion of the dispersal of order, based
on examples introduced particularly by Popper, is shown to be
related to a direction of explanation which must also connect
with the contrast between fixity and non-fixity, and this
dispersal of order is shown to be involved in the stock examples
of the increase of entropy (pages 183-9). These criticisms and
preliminary discussions point clearly towards a conceptual
analysis based on the notion of fixity; but I concede that the
analysis actually offered on pages 190 will not do. I have
returned to this problem in a later article on 'Mind, Brain, and
Causation' (see the Additional Bibliography). Whether the
conceptual analysis based on fixity can be read also as factual
analysis is briefly discussed on pages 191-2, where it is also
considered what objective asymmetries will be left if it cannot.
The theme of Chapter 8 is the supposed necessity in causal
connections and the distinction (in concept, knowledge, and
reality) between causal laws and accidental universal proposi-
tions. After distinguishing many different jobs that 'necessity'
has been expected to do, I argue that whereas it is often
thought that the ability of causal laws to entail or sustain
counterfactual conditionals is what marks them off from
accidental universals and shows them to have some meaning
stronger than that of merely factual universality, this is a
mistake: merely factual universals, if supported directly or
indirectly by inductive evidence, will sustain counterfactuals
(pages 193-204). So far, then, I am supporting the Humean
rejection of any necessity stronger than factual universality.
But after this problem has been completely resolved, there is
still some distinction between causal laws and merely accidental
 PREFACE TO THE PAPERBACK EDITION xv
universals, and this is not adequately accounted for in the stan-
dard Humean way by noting that causal laws fit together into
a body of scientific theory (pages 204-8). My positive solution of
this problem contains two elements. I show how we can distin-
guish the basic laws of working of some system both from pure
collocation statements and from derived laws which are mixed
products of the basic laws and the pure collocations; but this
distinction is not made by the application of any simple criteria,
but requires a thorough understanding of the system in question.
I then suggest that the basic laws of working are forms of
persistence, that qualitative continuity is to be found (though it
is often not obvious) in causal processes. This notion underlies
the demand for 'causal mechanisms', and is an empirical
counterpart of the necessity postulated by Kneale and other
rationalists. I also claim (against, e.g., Nagel) that there is
some a priori expectability about such continuity or persistence
(pages 208-30). In these respects I go beyond Humean views.
Chapter 9 discusses the form of statistical or probabilistic laws,
and shows that these may be of three different kinds, only one
of which introduces a basic indeterminism. Chapter 10 discusses
the ontology of causation. After criticizing the use often made
of a Fregean argument to show that causal statements cannot be
extensional (pages 250-6), I note that both concrete events and
facts are, and can coherently be, recognized as causes and effects,
and so distinguish 'producing causes' from 'explanatory causes'
(pages 257-60). The latter name has, unfortunately, proved to
be misleading: since explanation is relative to interests and
previous beliefs and expectations, it has been thought by some
readers that explanatory causation is similarly relative or
subjective. But this is not what I intend. I develop the notion of
fact causes or explanatory causes by distinguishing causally
relevant features and formulating the ideal of a minimally
complete causal account which would mention all and only the
relevant features. This relevance of cause-features to effect-feat-
ures is a thoroughly objective matter, owing nothing to interests
or previous beliefs or expectations; explanatory causation, thus
developed, is the form which causation in the objects is most
accurately seen as taking, while our ordinary recognition of
concrete event causes is seen as a legitimate concession to the
incompleteness of our knowledge (pages 260-9).
xvi P R E F A C E TO THE PAPERBACK E D I T I O N
Chapter 11 begins with a summary of the previous ten
chapters and an outline of the view of efficient causation
developed in them, as an introduction to the question whether
teleology, final causation, and purposive behaviour can be
fully explained in terms of efficient causation, or whether some
kind of irreducible teleology should be recognized in addition
to efficient causes. The 'logical connection argument1 which has
been thought to show that the relation between will or desire
and action cannot be causal (in anything like a Humean sense)
is rebutted. While several uses of distinctively ideological
descriptions of processes and behaviour are admitted and
explained, it is argued that no objective processes are in
themselves irreducibly teleological. We have teleological
concepts, and teleology belongs to some processes as we know
them, but not to the processes in the objects. In reality, efficient
causation only is the cement of the universe.
This account of causation does not, of course, stand entirely
on its own, but is connected with things that I have said else-
where. It rests partly on logical and epistemological views which
are not defended here; for the logical ones I have argued
particularly in Truth, Probability, and Paradox, for the epistemo-
logical ones particularly in Problems from Locke but also in *A
Defence of Induction'. The project of defending an essentially
empiricist view of causation which, however, explains elements
in our thinking which are initially at variance with such a view
is also connected with my main meta-ethical thesis in Ethics:
Inventing Right and Wrong, The most radical challenge to my
denial of the objectivity of values is the argument that it
presupposes an empiricism which is unable to cope with many
other features of the world, including causation, that it would
nevertheless be intolerable to deny. This challenge is (in part)
rebutted by showing that the empiricism in question can,
after all, give an adequate account of causation and our know-
ledge of it.
August 1919 J. L. M.
 Contents

Introduction 1
1. Hume's Account of Causation 3
2. The Concept of Causation—Conditional Analyses 29
3. Causal Regularities 59
4. Kant and Transcendentalism 88
5. Common Sense and the Law 117
6. Functional Laws and Concomitant Variation 143
7. The Direction of Causation 160
8. The Necessity of Causes 193
9. Statistical Laws 231
10. Extensionality—Two Kinds of Cause 248
11. Teleology 270
Appendix. Eliminative Methods of Induction 297
Additional Notes 322
Bibliography 323
Additional Bibliography 326

Index 327
This page intentionally left blank
 Introduction

THIS book has as its theme an attempt to discover or elucidate

the nature of causation. It is, of course, part of the business of
the various sciences to discover particular causal relationships
and causal laws; but it is part of the business of philosophy to
determine what causal relationships in general are, what it is
for one thing to cause another,, or what it is for nature to obey
causal laws. As I understand it, this is an ontological question, a
question about how the world goes on. In Hume's phrase,the cen-
tral problem is that of causation 'in the objects'. In approaching
this ontological question it will, indeed be necessary to examine
the meaning, and the logic, of several kinds of causal statements,
the uses of what we can classify as causal language; to engage,
in other words, in conceptual analysis. But questions about the
analysis of concepts or meanings are distinct from questions
about what is there, about what goes on. Of course, we may, or
rather must, accept the use of causal language as a rough guide
to what we are to take as causal relationships, as indicating—
though not as authoritatively marking out—our field of study.
But it is always possible that our causal statements should, in
their standard, regular, and central uses, carry meanings and
implications which the facts do not bear out, that our ways of
speaking and reasoning about the situations or sequences which
we recognize as causal should (explicitly or implicitly) assert of
them something which is not true. It is equally possible that
some or even all of the situations and sequences that we recog-
nize as causal should have features which we have not fully
recognized, or perhaps which we have not discovered at all,
and which therefore are not built into the meaning of our causal
language. Conceptual and linguistic analysis in the causal area,
then, is a guide to our main topic and an introduction to it;
but it is not itself our main topic, and with regard to that topic
its authority is far from absolute.
Also subsidiary to our main topic, but even more important
 a INTRODUCTION
as an introduction to it, is the epistemological question how we
can and do acquire causal knowledge, how we learn about causal
relationships, test, refute, establish, or confirm causal claims
and hypotheses. I would reject the verificationist theory that
the meanings of causal statements, among others, are directly
determined by the ways in which they are, or even can be,
verified or tested, though of course it must be conceded that
there are indirect connections between what a statement means
and the ways in which it can be checked. Equally I would
reject any phenomenalist or subjective idealist theory that what
is there is constituted or determined by how things appear to
us (or to me). But the possibilities of rinding things out may set
limits, perhaps quite narrow limits, to what we have any right
to assert to be there. An ontological claim which lacked
epistemological support would be nothing but an airy specula-
tion, and, while an outright denial of it would not be in order,
to spend any great time on it would be as much out of place in
philosophy as in science.
When I say that the theme is the nature of causation, I am
taking the word 'causation* in a very broad sense. What goes on
may be governed by laws of functional dependence rather than,
or as well as, by connections between distinguishable cause-
events and result-events. There may be continuing processes
which it would be not at all natural to describe in the ordinary
terminology of cause and effect. There may be statistical laws,
probabilistic laws of working in contrast with strict causal, that
is, deterministic laws. There may be human actions and re-
actions which do not fit into some preferred framework of
mechanical or Humean causation. There may be some room or
even some need for ideological description of some or all of
what goes on. None of these topics is excluded a priori by a
narrow interpretation of the causation whose nature is the sub-
ject of this inquiry, and in fact some attention will be paid to
each of them. The causation that I want to know more about is
a very general feature or cluster of features of the way the
world works: it is not merely, as Hume says, to ust but also in
fact, the cement of the universe,1
1
An Abstract of a Treatise of Human Nature, p. 32.
 I

Hume's Account of Causation

THE most significant and influential single contribution to the

theory of causation is that which Hume developed in Book I,
Part III, of the Treatise, summarized in the Abstract, and re-
stated in the first Enquiry. It seems appropriate to begin by
examining and criticizing it, so that we can take over from it
whatever seems to be defensible but develop an improved
account by correcting its errors and deficiencies.
Hume's conclusion seems to be summed up in the two defini-
tions of a cause as 'an object precedent and contiguous to an-
other, and where all the objects resembling the former are
placed in a like relation of priority and contiguity to those
objects that resemble the latter', and again as 'an object pre-
cedent and contiguous to another, and so united with it in the
imagination, that the idea of the one determines the mind to
form the idea of the other, and the impression of the one to
form a more lively idea of the other'.
These definitions, repeated with minor changes in the
Enquiry, identify causation first with regular succession and then
with succession together with a psychological association. They
at once give rise to well-known and obvious difficulties and
objections. How are the two definitions even to be reconciled
with one another? They are clearly not equivalent, nor even
coextensive.1 There may well be regular successions such that
ideas of the successive items provoke no associations, and again
successions which are not regular but are such that the ideas of
the successive items are suitably associated. Even if we confine
ourselves to the first definition there are still many problems.
How close or how remote are the resemblances to be? Is
1
Cf. J. A. Robinson, 'Hume's Two Definitions of "Cause"', Philosophical
Quarterly, xii (1962), reprinted in Hume, ed. V. C. Ghappell.
4 HUME'S ACCOUNT
contiguity required or not (it is not included in the first definition
in the Enquiry, and reasons for not insisting on it are mentioned
later in the Treatise}?2 If there is a unique, never repeated,
sequence of events, does it count as causal or not by this defini-
tion? Would we not want to say that such a unique sequence
might either be, or not be, causal, but that this question would
not be settled either way merely by its uniqueness? And in
general, would we not want to say that there can be sequences
which are regular, but in which the earlier item does not cause
the later, either where the two items are collateral effects of a
common cause, or where the regularity of their succession is
just accidental? More dubiously, are there not causal sequences
which nevertheless are not regular: e.g. if tossing a coin were
an indeterministic process, so that my tossing this coin just as I
did this time might equally well have led to its falling heads or
tails, still, since it in fact fell heads, did not my tossing it cause
its falling heads? In any case, there are complications about
necessary causes, sufficient causes, necessary and sufficient
causes, combinations of causal factors, counteracting causes, a
plurality of alternative causes, causal over-determination, and
so on which are entirely neglected if we speak just of regular
succession, about 'all the objects resembling the former . . .'.
Can even precedence in time be insisted on? Are there not
causes which are simultaneous with their effects, and might
there not conceivably be causes which succeed their effects?
Even the first definition, then, does not seem to agree at all
well with what we ordinarily require for causation. It might be
clarified and made precise and perhaps qualified or extended,
and so turned into something more satisfactory. This would be
a fairly lengthy task. I shall undertake this task later,3 but to
pursue it now would lead us away from Hume's distinctive con-
tribution to the subject. In fact, we may suspect that while the
two definitions in some way represent the conclusion of Hume's
discussion, the way we have just been looking at them somehow
misses the point, that what Hume took himself to have estab-
lished was something to which the difficulties I have been
hastily surveying have little relevance. Perhaps a definition of
'cause', in the sense in which a definition is shown to be defec-
tive by such points as those mentioned, is not what Hume was
1 3
Book I, Pt. IV, Sect. 5. In Chapter 3.
 HUME'S ACCOUNT 5
primarily concerned to give. This suspicion is strengthened by
the reflection that in the Abstract, where Hume selects for special
attention this 'one simple argument' and claims to have 'care-
fully traced [it] from the beginning to the end'4—thereby in-
dicating that he regarded the discussion of causation as his
greatest achievement in the first two Books of the Treatise—he
does not give the two definitions: the conclusion is stated (on
pages 22 and 23) without them.
Indeed, the main topic of all the passages to which I have
referred is 'the nature of our reasonings concerning matter of
fact'.5 This is made explicit in the Enquiry and in the Abstract
(p. 11). In the Treatise, Part III of Book I is entitled 'Of Know-
ledge and Probability'; knowledge is dispatched in Section I,
leaving the other fifteen sections for probability. In Section 11
he explains that he is using the word 'probability' in an un-
naturally broad sense6 to cover all 'arguments from causes or
effects'; in view, therefore, of his repeated claim that all infer-
ences concerning matter of fact, that is, all non-demonstrative
inferences, are founded on the relation of cause and effect, 'prob-
ability' equally covers all inferences concerning matter of fact.
Hume's broad purpose, then, is 'to enquire what is the nature of
that evidence which assures us of any real existence and matter
of fact, beyond the present testimony of our senses, or the records
of our memory',7 in other words to establish the character and
the justification, if any, of all non-demonstrative inferences
about empirical matters; causation comes into view because
Hume thinks that it is the foundation and immediate subject of
all such inferences. Very soon, of course, he focuses on necessity
or necessary connection both as what distinguishes causal
sequences from non-causal ones and as somehow intimately
linked with causal inference, so that his specific task is to give
an account of the idea of necessary connection and of whence
and how it is derived, but this is still only part of the wider
study of empirical inferences. And of course his serious con-
clusion is the largely negative one that these inferences are to
be ascribed to imagination, and custom rather than to reason,
that we do not discover any necessity in 'the objects' that could
+ Abstract, p. 5. s Enquiry, Sect. IV, Pt. II, Sclby-Bigge, p. 32.
* Borrowed, as he says in Enquiry (Selby-Bigge, p. 56 n,), from Locke.
7
Enquiry, Selby-Bigge, p. 26.
6 HUME'S ACCOUNT
serve as a basis and justification for these inferences, but that
instead our very idea of necessity arises from those inferences:
the supposed necessity is based on the inference, not the infer-
ence upon any perceived necessity.3 This negative thesis, this
reversal of what seems to be the true or natural order9 of things}
is what appears to Hume himself to be his great new discovery.
Since this is his main theme and purpose, it is not very sur-
prising if his first definition, his account of causation as we know
it in the objects, is imprecise and carelessly formulated. Pre-
cision about this was not required for the job in hand. He
might, indeed, have explained himself somewhat as follows:
Causation as we observe it in 'objects' of any kind—physical pro-
cesses, mental processes, the transition from willing to bodily move-
ment, or anywhere else—is something that we might roughly
describe as regular succession. Exactly what it is or may be, within
the bounds of this rough description, does not matter for the present
purpose. All that matters is (i) that it should be something that
could, in those cases in which we form our idea of causation, give
rise to a suitable association of ideas and hence, in accordance with
my psychological theory of belief, to belief in the effect when the
cause is observed or in the cause when the effect is, and (ii) that it
should not be anything in which there is an observable necessity (or
efficacy or agency or power or force or energy or productive quality)
or anything at all that could supply a rival explanation of our idea
of necessity, competing with the explanation given in terms of
association, belief, and projection.
With this understanding of Hume's purpose, we can now
study the arguments by which he reached and supported what
he acknowledged to be a paradoxical conclusion. But these
arguments are complicated, and it will be helpful, in dis-
entangling them, to use the device of a structure diagram.10 Hume's
thesis (a] that causation in the objects, so far as we know, is only
regular succession, is a corollary or further conclusion based
(partly) on his main conclusion (b] that necessity is in the mind,
not in objects, though we tend to project it onto the objects.
* Treatise, I, in, 6, Selby-Bigge, p. 88.
» Treatise, I, m, 14, Selby-Bigge, pp. 167-8.
10
Such a diagram is, I hope, self-explanatory. Propositions are represented by
letters, and arrows between letters, as in b ->• a, indicate that the author whose
argument is being analysed is saying that b gives some support to a. In constructing
the diagram I am not myself asserting that there are such logical or evidential
relationships.
 HUME'S ACCOUNT 7
This main conclusion is supported by the confluence of three
lines of thought. The first of these says that (c) no knowledge of
one 'object* on its own would tell us that it would cause a cer-
tain effect or that it was the effect of a certain cause, and hence
that (d) what might be variously described as causal knowledge
or causal beliefs or the making of inferences from observed
events to their causes or to their eifects—(the inference from the
impression to the idea'—arises purely from the experience of
constant conjunction. The second line of thought culminates
in the claim (e] that this experience does not reveal any neces-
sity in the objects, in particular, what in view of Hume's use
of the term 'necessity' is closely related to this, that the experi-
ence of constant conjunction does not provide materials for any
rational inference from cause to effect (or vice versa) in a new
instance. The third line of thought is the psychological theory
of belief, summed up in the thesis (/) that belief consists in the
liveliness given to an idea by an associative link with a present
impression.
That is, the main structure of the argument is as shown in
Diagram (i). (In this, d is shown twice because it is both used
along with e and f to support b and also used along with b to
support a.)

Diagram (i)

That the main structure of Hume's argument consists in the

bringing together of these three lines of thought to support
what I have called his main conclusion is confirmed by his own
summary of his argument in the Treatise, which I quote, merely
inserting the letters of reference:
Before we are reconciled to this doctrine, how often must we repeat
to ourselves, that (c} the simple view of any two objects or actions,
however related, can never give us any idea of power, or of a con-
nexion betwixt them: that (d) this idea arises from the repetition
of their union: that (e) the repetition neither discovers nor causes
anything in the objects, but (f) has an influence only on the mind
by that customary transition it produces: that (b) this customary
8 HUME'S ACCOUNT
transition is therefore the same with the power and necessity; which
are consequently qualities of perceptions, not of objects, and are
internally felt by the soul, and not perceived externally in bodies?11

This account is equally confirmed by Hume's lengthier sum-

mary of his argument in the Abstract; there the first line of
thought is found on pages 13 and 14, the second is foreshadowed
on page 12 and developed on pages 14 and 15, and the third
is given on pages 16 to 21, while the conclusion is stated several
times in slightly different forms in those pages.
In the Treatise (Book I, Part III) the first line of thought is
presented partly at the beginning of Section 6 and partly at the
beginning of Section 14 (though associated points are made in
Section 3); the second line of thought occupies the central part
of Section 6; the third begins at the end of that Section, and,
with a wealth of illustration, fills Sections 7 to 10, while Sections
i i to 13 further illustrate it by attempting to explain reasoning
about probabilities (in the narrow, ordinary, sense, not in the
broad one in which 'probability' covers all inferences concern-
ing matters of fact) in terms of this same psychological theory of
belief. Section 14 puts the whole argument together and draws
the conclusions.
In the first Enquiry, Section IV, Part I develops the first line
of thought, stressing point c; Section IV, Part II gives the
second line of thought; Section V gives the third; Section VII,
Part I contributes further points to the first line of thought, and
Section VII, Part II again puts the whole argument together.
What I have given so far is only the broad structure of Hume's
argument. We may now fill in some details.
Leading up to c, in the first line of thought, is the argument
that (g) cause and effect are distinct existences and the ideas of
a cause and its effect are also distinct, so that (h) a particular
cause is always conceivable without its effect, and a particular
effect without its cause. From g it also follows—in a branch
argument—that (i] there can be no demonstrative proof of the
principle that every event has a cause, and this is further sup-
ported by detailed criticisms (j) of specific attempts to prove
this.12 But c needs, and is given, further support in addition to
k. We might be able to observe something about C that would
11
Treatise, I, in, 14, Selby Bigge, p. 166. « Treatise, I, HI, 3.
 HUME'S ACCOUNT 9
tell us a priori that it would produce E (or something about E
that would tell us a priori that it had been produced by C),t even
if C were conceivable without E and vice versa: the C-E
sequence might be necessary a priori without being analytic.
This possibility is excluded partly by a challenge to anyone to
produce an example of such detection of power or necessity,
and partly by a search through different areas of experience
for an impression that might give rise to the idea of power
or necessity.13 The negative result is that (K) no impression
of power or necessity is supplied by any observed relation or
sequence. Also, to support d, c needs the help of the claim that
(/) causal knowledge is actually acquired in cases where we
experience constant conjunction. Inserting these details, we can
give in Diagram (ii) a fuller picture of the first line of thought.

Diagram (ii)

The second line of thought could be similarly expanded. It

turns upon the problem-of-induction dilemma. The conclusion
(e) that the experience of constant conjunction does not provide
materials for any rational inference from cause to effect (or vice
versa) in a new instance is supported by the claim that any such
rational inference would have to rest on the principle that un-
observed instances must resemble observed ones, or that nature
is uniform, while this principle can be established neither by
demonstration (since its falsity is conceivable) nor by 'probable
reasoning' (since that would be circular). This argument clearly
has a further detailed structure, but I need not analyse it
because it has been very thoroughly analysed by D. C. Stove1*
whose conclusions I shall simply take over and apply. I shall
refer to the whole problem-of-induction dilemma as m.
" Treatise, early part of Section 14; Enquiry', Sect. VII, Pt. I.
14
'Hume, Probability, and Induction1 in Philosophical Review, Ixxiv (1965), re-
printed in Hume, ed, V. G. Chappell. Stove develops his analysis more fully in
Probability and Hume's Inductive Scepticism (1973).
10 HUME'S ACCOUNT
The third line of thought is developed at great length in the
Treatise, and again there would be scope for a detailed analysis
of the subsidiary arguments. But it may be sufficient to note
one important premiss, that (n) belief is not a separate idea but
a peculiar feeling or manner of conceiving any idea; it is on
this, together with the whole body of examples (o) where belief
seems to be produced by association, that thesis/rests.
The structure of the whole argument, in as much detail as we
require, can then be shown by Diagram (iii) with the following
'dictionary':
(a) Causation in the objects, so far as we know, is only regular
succession.
(b) Necessity is in the mind, not in objects.
(c) The simple view of objects gives us no idea of power or
necessary connection, i.e. no knowledge of one 'object' on
its own allows causal inference to another 'object',
(d) Causal knowledge and inference, and the idea of necessary
connection, arise purely from the experience of constant
conjunction.
(e) This experience neither reveals nor produces any necessity
in the objects; that is, it does not provide materials for any
rational inference from cause to effect (or vice versa) in a
new instance,
(/) Belief consists in the liveliness given to an idea by associa-
tion with a present impression.
(g) Cause and effect are distinct existences, and the ideas of
a cause and of its effect are distinct.
(A) A cause is conceivable without its effect and vice versa.
(i] There can be no demonstrative proof that every event has
a cause.
(j) Specific attempts to prove that every event has a cause fail.
(k) No impression of power or necessity is supplied by any
observed relation or sequence.
(/) Causal knowledge is acquired in cases where we experience
constant conjunction.
(m) The problem-of-induction dilemma.
(n} Belief is not a separate idea but a feeling.
(o) Belief is produced by association.
We have already noted a curious duality in this argument,
 HUME'S ACCOUNT n
which comes out in the double formulation that we need to
give for such propositions as c, d, and e. The main topic is the
character and justification of inferences about matters of fact,
but bound up with this is the secondary topic of the nature of
causal connection. These are further linked by a very important
assumption which is implicit in a double treatment of necessity
or necessary connection. On the one hand this is taken to be

Diagram (iii)

the distinguishing feature of causal as opposed to non-causal

sequences. On the other hand, it is taken to be something that
would allow rational a priori inference, e.g. something such that
if we detected it holding between C and E> we should (if we
knew enough about C itself) be able to infer rationally that E
would follow C. This is clearest: when necessity is equated with
power: 'if the original power were felt, it must be known: were
it known, its effect must also be known; since all power is
relative to its effect.' And again *We must distinctly and par-
ticularly conceive the connexion betwixt the cause and effect,
and be able to pronounce, from a simple view of one, that it
must be followed or preceded by the other. This is the true
manner of conceiving a particular power . . .Vs
This second interpretation of necessity is used, with important
results, in the survey that establishes k. It is, for example, partly
because we do not know a priori which movements are under
the control of the will that we cannot be detecting necessity in
» Enquiry, Sect. VII, Ft I, Sclby-Bigge, p. 66; Treatise, I, m, 14, Selby-Biggc,
p. 161.
12 HUME'S ACCOUNT
the will-movement sequence. And again if we could detect
necessity in this sequence, the whole transition from will to
movement would have to be transparent to our understanding,
whereas in fact the intermediate links are utterly obscure.
Now it is a sheer assumption that these two interpretations
belong together, that the distinguishing feature of causal
sequences must be something that would justify a priori infer-
ence. Hume gives no argument to support it. Moreover, it has
no basis in the ordinary concept of causation. This ordinary
concept may include suggestions that there is some intimate tie
between an individual cause and its effect, that the one pro-
duces the other, that it somehow pushes or pulls it into exist-
ence, and the like. But none of these suggestions would carry
with it a justification of a priori inference. The notion of a neces-
sity that would justify a priori inference is indeed to be found
in rationalistic philosophy, for example in Locke as well as
Descartes.16 It has repeatedly cropped up in the thinking
of some scientists and philosophers of science, particularly in
the form of the view that no explanation is satisfactory unless
it is intelligible, unless it gets beyond brute fact and shows that
what does happen has to happen, and therefore, if only we
knew enough, could be firmly expected to happen. But it is a
rationalist scientists' and philosophers' notion, not part of the
ordinary concept of a cause.
We should, therefore, distinguish the two things which this
assumption has unwarrantably identified. We might call neces-
sityt whatever is the distinguishing feature of causal as opposed
to non-causal sequences, and necessity^ the supposed warrant for
an a priori inference, for example, the power which if we found
16
Locke (An Essay concerning Human Understanding, Book IV, Chapter III, § 13)
says that we can see 'some connexion' in mechanical processes, and he clearly
thinks (Book III, Chapter VI) that things have real essences which are the'founda-
tion* of all their properties, from which their qualities flow, and from which, if
only we knew those essences, we should be able to infer those qualities. The trouble
is merely that these real essences 'are so far from our discovery or comprehension',
that 'we know them not1. Similarly Descartes (Principles of Philosophy, II, §§ 37 ff.)
states such laws as 'Every reality, in so far as it is simple and undivided, always
remains in the same condition so far as it can . . ,' and 'A moving body, so far as
it can, goes on moving'. The behaviour of things will be intelligible in so far as it
can be shown to be a consequence of such laws as these, which are themselves
based on 'God's immutability'. Again 'if I move one end of a s t a f f . . . I easily
conceive the power by which that part of the staff is moved as necessarily moving
at one and the same instant all its other parts' (Regula IX).
 HUME'S ACCOUNT 13
it m C would tell us at once that C would bring about Et or
some corresponding feature which if we found it in E would
tell us at once that E had been produced by C, or some relation
between one kind of thing and another which told us that things
of those kinds could occur only in sequences, with a thing of the
second kind following a thing of the first. But once we have
drawn this distinction, we must concede that while Hume has
a strong case for asserting k about necessitya, he has no case at
all for asserting it about necessity!.
This point can be put in another way. I have interpreted the
passages we are discussing as presenting an argument in which it
is assumed that necessity is something that, if found, would
justify a priori inference, and in which it is concluded that this
necessity is not found but is felt and then improperly objectified.
But Hume also pretends to be engaged in a very different pur-
suit, a search rather than an argument. Necessary connection is
introduced as that of which the idea is the as yet unidentified
third element (along with contiguity and succession) in our
idea of causation, which will be identified only when we have
found the impression from which it is derived.17 But if this
second pursuit were to be conducted fairly, it would have to be
carried on in an innocent, open-minded way; Hume would
have to be prepared to accept as the sought-for impression
whatever seemed in practice to differentiate the sequences we
accept as causal from those we do not. As we have seen, this is
not how he proceeds: he starts with a firm conviction that
nothing can count as necessary connection unless it will some-
how support a priori inference,, His search for necessity! is
sacrificed to his argument about necessity2. His ostensible pro-
cedure is to look for the impression from which the idea of
power or necessity is derived in order to decide what this idea
is, since, on Hume's principles, the idea must be just a fainter
copy of the impression. But his real procedure is to examine
various sorts of observation which might be supposed to include
the required impression, and to reject these in turn on the
ground that they include nothing which could yield, as a copy,
such an idea of power as he has already postulated. And when
he finally reaches his thesis that: necessity is in the mind,
explained in terms of the psychological theory of belief,
" Treatise I, nij a, Seiby-Bigge, p. 77.
14 HUME'S ACCOUNT
associative links, and the 'determination of the thought', he finds
it convincing just because this, and perhaps this alone, could
supply a source, in the form of an impression, for what he has
from the start assumed the idea of power or necessity to be, a
licence to infer effect from cause or vice versa.
After this analysis and clarification of Hume's arguments we
can turn to evaluation. How good are they? What, if anything,
do they establish?
If we accepted Hume's conceptual framework, and assumed
that all kinds of thinking, or mental life generally, consisted
simply of having certain perceptions, certain impressions and
ideas in various combinations and sequences, then it would be
plausible to equate believing with having a lively idea. What
else, on these assumptions, could belief be? Hume is right in
saying (as in ri) that belief is not just another idea compresent
with the idea of whatever we believe (or believe in). But as
soon as we break free from these assumptions, this view becomes
utterly implausible. At any time there are large numbers of
things that I believe of which I have no idea at all, faint or
lively. In so far as I can identify ideas and compare their
degrees of vivacity, I seem to have very faint ideas of many
things in whose existence I believe and some relatively vivid
ones of what I class as fantasies. The theory that belief arises
from association with a present impression is equally shaky.
Some ideas of things I believe in are thus associated, but a great
many are not. And some things the ideas of which are strongly
associated with present impressions—for example, the events in
a work of fiction that I am now reading, the printed words of
which are before my eyes—are not believed. In so far as Hume's
conclusions rest on this full-blown associationist theory of
belief, then, they are ill founded. But he could fall back on the
much more defensible thesis that human beings, and some other
animals, have an innate or instinctive tendency to reason in-
ductively, to expect a conjunction or sequence of events that
they have noticed recurring a number of times—especially if it
has recurred without exceptions—to be repeated in the future
and to have occurred similarly in past unobserved cases. That
there is such an inductive propensity can be confirmed experi-
mentally for non-human animals (for example, the psycho-
logists' rats) and can be taken as an empirical fact in its own
 HUME'S ACCOUNT 15
right, detached from the wider theory of the nature and causes
of belief in general; this proposition, which we may call /',
would be sufficient for some of Hume's purposes.
What I have called Hume's second line of thought has been
thoroughly analysed and criticized by Stove, Also, though this
problem-of-induction dilemma has tremendous importance for
other parts of philosophy, it is only marginally relevant to our
topic of causation. For both these reasons I shall deal with it
only briefly. Hume's argument is that if it were reason that leads
us, after the experience of a constant conjunction, to infer a
cause from an effect or an effect from a cause in a new like
instance, reason would be relying on the principle of the uni-
formity of nature, that unobserved instances must resemble
observed ones, but that this principle cannot be established
either by 'knowledge*—because its falsity is conceivable—or by
'probability'—because that would be circular. It would be
circular because 'probability* rests always on causal relations,
and so on a presumption of the very principle of uniformity the
support for which we are considering. Now Hume's premiss
that 'reason' would have to rely on the principle of uniformity
holds only if it is assumed that reason's performances must all
be deductively valid: if it were suggested that an observed con-
stant conjunction of As with Bs probabilities that this new A will
be conjoined with a B, in terms of some logical or relational
probability as proposed by Keynes and Carnap, that is, that
some non-deductively-valid argument is none the less rational,
that its premisses really support though they do not entail its
conclusion, then this possibility would not be excluded by
Hume's argument, because such a probabilistic inference would
not need to invoke the uniformity principle which produces the
circularity that Hume has exposed.
Reasonable but probabilistic inferences, then, have not been
excluded by Hume's argument, for the simple reason that Hume
did not consider this possibility. Modern readers are tempted
to suppose that he did consider, ;and refute, it because he calls
what the second horn of his dilemma discusses 'probable
reasonings'. But this does not mean probabilistic inferences; as
he tells us, he is using 'probable* in a wide sense, covering all
reasoning about matters of fact, as opposed to demonstrative
'knowledge* based on the comparison of ideas, and the only
 16 HUME'S ACCOUNT
reasonings that this second horn condemns as circular are
deductively valid arguments in which the principle of uni-
formity is both a premiss and the conclusion. Hume can hardly
be blamed for not having considered a possibility which was not
clearly presented until 150 years after his death: but it is more
surprising that modern philosophers like Ayer and Popper
should suppose him to have refuted it.18
Proposition e, then, can be established only if we amend it,
reading 'deductively valid* in place of 'rational', and inter-
preting 'necessity' in such a way that the denial that constant
conjunction reveals necessity is equivalent to the denial that it
furnishes the materials for deductively valid inferences about
new instances. Let us call this amended version e'.
Returning to the first line of thought, we are forced to make
a more complicated evaluation. Propositions^, z, g> and h are
all clearly correct: neither the principle that every event has a
cause, nor any specific causal law, nor any singular causal
statement connecting particular events, is an analytic truth. Of
18
See, for example, K. R. Popper, Objective Knowledge, p. 89, 'Hume's answer .. .
as he points out, . . . remains negative even if our inference is merely to the prob-
ability of a connexion that has not been observed rather than to its necessity. This
extension is formulated in the Treatise, , , . The argument against probabilistic
induction is ... purely formal.1 Popper refers to Stove's article, but still has not
grasped its point; 'since Hume's argument is formal . . . Stove cannot be right.'
This ignores the fact that only a deductively valid 'induction' would require
the uniformity principle, as well as misunderstanding Hume's use of the word
'probability'.
Similarly, A. J. Ayer in Probability and Evidence, p, 5, in what purports to be an
exposition of Hume's argument, first correctly notes that the principle of uni-
formity is needed as an extra premiss to make the jump from 'All hitherto observed
As bear the relation R to Bs' to 'This A will have the relation R to some B' valid,
and correctly repeats Hume's claim that there can be no demonstrative argument
for this principle. But he goes on: 'Even if the principle cannot be demonstrated,
perhaps we can at least show it to be probable', and repeats Hume's circularity
objection, without noting that to make it sound we must assume that what is
being criticized is still an attempt to make the inductive jump deductively valid, and
that 'probable' does not cancel this concern with validity, but merely means
'resting on empirical information'. He then adds what he has admitted (privately)
to be his own extension and not a report of anything in Hume: 'The same objection
would apply to any attempt to by-pass the general principle of uniformity of nature
and argue that inferences from one matter of fact to another, though admittedly
not demonstrative, can nevertheless be shown to be probable. Again, this judge-
ment of probability must have some foundation. But this foundation can lie only
in our past experience....' To say this is simply to ignore the suggestion that there
is a logical, a priori, relation of support or probabilification. Against this suggestion
the same (circularity) objection does not arise.
 HUME'S ACCOUNT 17
course, we can find descriptions of causes and effects such that
the description of a cause is analytically connected with the
description of its effect, but this is a trivial point on which no
time need be wasted. What matters is that there are no logically
necessary connections between the events themselves, or between
any intrinsic descriptions of them, however detailed and com-
plete. On this Hume's argument is conclusive and beyond
question. But difficulties begin with k. We can concede that the
survey arguments of Section 14 in the Treatise and Section VII,
Part I in the Enquiry, coupled with the signal lack of success
of rationalist philosophers of science in taking up the challenge
to produce examples of synthetic a priori physical laws or
'intelligible* explanations, provide a very strong, though not
absolutely conclusive, case for the claim that we never observe
a necessityjj which would support a deductively valid a priori
causal inference. But it has not been nearly so conclusively
shown that no observations reveal a necessity2 that might sup-
port a probabilistic a priori causa! inference; for all that we have
so far seen, observable features of one event, taken on its own,
might probabilify a certain continuation or successor-event. And
of course it has not been shown at all that no observations
reveal what we have called necessity!- If we then distinguish
necessity2! as something that would support a deductively
valid a priori causal inference, necessity2 2 as something that
would support a probabilistic a priori causal inference, and apply
subscripts to k to correspond with the kinds of necessity men-
tioned in it, we must conclude that k± is quite unsupported,
^2.1 is verY strongly but not conclusively supported, while kz 2
is a fairly open issue. Similar qualifications, with similar sub-
scripting, must be applied to c and to d. In other words, dlt the
thesis that our knowledge of the distinguishing feature of causal
as opposed to non-causal sequence arises purely from the experi-
ence of constant conjunction, is quite without support; </2-1, the
thesis that if there were any deductively valid inference 'from
the impression to the idea' it would have to rest purely on the
experience of constant conjunction, is pretty firmly established;
but <4i{i, the thesis that a similar probabilistic inference would
have to rest purely on this experience, is a moot point.
When we combine ^8il, the only well-supported outcome of
this first line of thought, with our amended propositions e' and
i8 HUME'S ACCOUNT
/', we can conclude that there are no deductively valid infer-
ences 'from the impression to the idea', and that i/we have an
idea of a necessity that would support such inferences—that we
have such an idea has not been shown—it may well be explained
in Hume's manner as a projection of the inductive inferences
that we instinctively make.
But in view of the relatively weak support for </2 a, and the
amendment (that Stove has shown necessary) of e to e't we must
conclude that it is an open question whether, either before or
after the experience of constant conjunction, there may be
forceful probabilistic inferences from an observed cause to an
unobserved effect or vice versa. What in Hume's way of speak-
ing would come out as the corresponding kinds of necessity may,
for all that he has shown, be discoverable 'in the objects'; there
is as yet no need to explain them away in terms of instinctive
tendencies to infer, transferred improperly to external reality as
a result of the mind's propensity to spread itself upon objects.
And in view of the total lack of support for k± and the cor-
responding rfj, the question where and how we are to look for
the distinguishing feature of causal sequences is wide open: it
has not even been shown that anything in the area of constant
conjunction or regular succession is required for this purpose.
The search which Hume sacrificed to his argument must start
again from scratch.
The conclusion that Hume has pretty well established about
causal inferences—that the inference from an observed event to
another event as its cause or effect is never deductively valid,
even if we include in the premisses the fact that a great many
similar sequences have occurred—is not a surprising one; it may
not have been clearly realized before Hume pointed it out, but
it would now be accepted by most philosophers without dis-
pute. The more sweeping conclusions about such inferences
which he suggests but fails to establish—that these inferences
are not even reasonable or probable, that they are to be
ascribed to imagination, custom, and habit rather than to
reason, that it is out of the question to try to justify them on any
ground except that they are natural, instinctive, and unavoid-
able—are interesting and important. They raise those problems
about induction, probable inference, and the confirmation of
hypotheses which are still a major area of dispute and investiga-
 HUME'S ACCOUNT 19
tion for philosophers of science and inductive logicians. But to
pursue them directly would lead us away from the subject of
causation. I shall therefore now leave on one side what was,
as I have tried to show, Hume's main topic of inferences con-
cerning matters of fact, and concentrate on what was for him
only a secondary subject, the nature of causation.
Hume's conclusions on this subject can be summed up under
three headings: the idea of causation, causation as it really is
'in the objects', and causation so far as we know about it in the
objects.
According to Hume, our idea of causation combines three
elements: succession, contiguity, and necessary connection.
When he includes contiguity he does not mean that we require
every cause to be contiguous with its effect, but merely that
where this is not the case, cause and effect are thought to be
joined by a chain of intermediate items, where each such item
is the effect of its predecessor and causes its successor, and is
contiguous with both of these. But, as I have said, he hesitates
about including contiguity even in this sense. I think he is right
to hesitate. While we are happiest about contiguous cause-
effect relations, and find 'action at a distance' over either a
spatial or a temporal gap puzzling, we do not rule it out. Our
ordinary concept of causation does not absolutely require con-
tiguity; it is not part of our idea of causation in a way that
would make (<? caused E over a spatial, or temporal, or both
spatial and temporal, gap, without intermediate links' a con-
tradiction in terms. Hume did not hesitate about requiring
succession; I think he should have hesitated, but I shall discuss
this issue later.19 But Hume is quite confident that our idea of
causation includes an idea of necessary connection. Sceptical
though he is about our ascription of necessity to the real world,
he is not sceptical about our having this idea; his problems
concern only its precise character and its origin. We can, it is
true, find such remarks as 'We never therefore have any idea
of power. . . . We deceive ourselves in imagining we can form
any such general idea . , . we have really no distinct meaning,
and make use only of common words, without any clear and
determinate ideas'20 and 'the necessary conclusion seems to be
that we have no idea of connexion or power at all, and that
19
In Chapter 7. *° Treatise, I, ra, 14, Selby-Bigge, pp. 161-3.
30 HUME'S ACCOUNT

these words are absolutely without any meaning . . ,\ZI Such

remarks have led careless readers to think that Hume excluded
necessity even from our idea of causation; but this is a mistake.
These remarks express what would follow on a certain hypo-
thesis, or what, at a certain stage of the investigation, seems to
be the case; Hume goes on at once to explain that, and how,
these conclusions can be avoided. The whole tenor of his
thought is that we have this obscure idea, and that we must in
the end find an impression from which it is derived. Admittedly
even in stating his conclusion he says 'Either we have no idea
of necessity, or necessity is nothing but that determination of
the thought . . ,';" but it is clear which alternative he prefers,
Hume is very far from holding a regularity theory of the
meaning of causal statements. His answer to the question 'What
do we ordinarily mean by "C causes £"?' is that we mean that
C and E are such that E's following C is knowable a priori, m
view of the intrinsic character of C and E, so that the sequence
is not merely observable but intelligible. When at the conclu-
sion of the discussion in the Enquiry he says 'we mean only that
they have acquired a connexion in our thought', etc., he is
telling us what we can properly mean rather than what we
ordinarily mean. He thinks that the ordinary meaning is itself
mistaken, and calls for reform.
On this crucial issue, I think that Hume is right to assume
that we have what in his terminology, with my subscripting,
would be an idea of necessity!; we do recognize some distinc-
tion between causal and non-causal sequences, and what is
more we are inclined to think that, whatever this difference is,
it is an intrinsic feature of each individual causal sequence. But
I have argued that Hume is quite wrong to assume that our
idea of necessity! is also an idea of necessity2, that we ordinarily
identify the differentia of causal sequences with something that
would support a priori inference.
About causation as it really is in the objects Hume usually
says that it is regular succession and nothing more.
. . . we are led astray by a false philosophy . . „ when we transfer
the determination of the thought to external objects, and suppose
any real intelligible connexion betwixt them ... As to what may be
al
Enquiry, Sect. VII, Pt. II, Selby-Bigge, p. 74.
" Treatise* I, m, 14, Sclby-Bigge, p. 166,
 HUME'S ACCOUNT si
said, that the operations of nature are independent of our thought
and reasoning, I allow it; and accordingly have observed, that
objects bear to each other the relations of contiguity and succession;
that like objects may be observed, in several instances, to have like
relations, and that all this is independent of, and antecedent to, the
operations of the understanding. But if we go any further, and
ascribe a power or necessary connexion to these objects, this is what
we can never observe in them. . . .23
Sometimes, indeed, he speaks of secret powers and connections
of which we are ignorant. 'We can never penetrate so far into
the essence and construction of bodies as to perceive the prin-
ciple, on which their mutual influence depends.' But he may
well have his tongue in his cheek here, as he does when he
speaks about the deity. At most he is 'ready to allow, that there
may be several qualities, both in material and immaterial
objects, with which we are utterly unacquainted; and if we
please to call these power or efficacy, it will be of little con-
sequence . . ,'.24
It is about causation so far as we know about it in the objects
that Hume has the firmest and most fully argued views. In
effect, he makes three negative points, and one positive one.
The first negative point is that there are no logically necessary
connections between causes and effects as we know them, and
Hume is clearly right about this. The second negative point is
that what I have called necessitya is not known as holding
between causes and effects. This second exclusion is of synthetic
a priori connections, as the first was of analytic ones. It is this
second exclusion that is vital and controversial. It is pre-
supposed, for example, in Hume's very forceful argument
(Treatise) Section 6) that we cannot get over the inductive
problem by referring to powers which we infer to be present in
causes (for even if we could infer that a certain power was
associated with certain sensible qualities on one occasion, how
are we to know that the same power will be associated with the
same qualities on other occasions?). This assumes that the
power needs to be inferred, that it cannot be observed ('it
having been already prov'd that the power lies not in the visible
qualities'). But this 'proof was that 'such a penetration into
2J
Treatise, I, in, 14, Selby-Bigge, pp. 168-9.
*+ Treatise, II, HI, i and I, in, 14, Selby-Bigge, pp. 400 and 168.
aa HUME'S ACCOUNT
their essences as may discover the dependence of the one
[object] upon the other' would imply the 'impossibility of con-
ceiving any thing different. But as all distinct ideas are separ-
able, 'tis evident there can be no impossibility of that kind.'25
Hume is here attempting to derive his second exclusion directly
from his first, that is, to derive k in our diagram from h (or
from g}} and so to present c as having been established by g
alone. But of course this is invalid. Similarly, after explaining
'power' in the terms on which I have based my account of
necessity2, Hume goes on to say that 'nothing is more evident
than that the human mind cannot form such an idea of two
objects, as to conceive any connexion betwixt them, or com-
prehend distinctly that power or efficacy by which they are
united. Such a connexion would amount to a demonstration,
and would imply the absolute impossibility for the one object
. . . to be conceived not to follow upon the other.'*6 Here, too,
Hume invalidly argues that there cannot be a synthetic a priori
connection because it would be an analytic one. Yet he also
takes the trouble to survey the various places where it might be
held that such connections are found, and uses a challenge
argument, saying 'This defiance we are obliged frequently to
make use of, as being almost the only means of proving a
negative in philosophy'.27 This seems to show that he realizes
that he is now making a further point, which the separate con-
ceivability of causes and effects does not establish. If we discard
the arguments which confuse the second exclusion with the
first, we find that it is supported mainly by a series of assertions
that (in various fields) we do not in fact possess a priori know-
ledge of causal relationships, that we have to find, by experience
in each case—even in the cases of voluntary bodily movements
and the calling up of ideas at will—what effects each kind of
cause is actually able to bring about, together with a challenge
to anyone who disagrees to produce an example of a priori
causal knowledge. Again, there is the argument (already men-
tioned) that we could perceive necessity in the transition from
will to bodily movement only if we knew the whole process,
including 'the secret union of soul and body' and the succession
« Treatise, I, HI, 6, Selby-Bigge, pp. 90-1 and 86-7.
26
Treatise, I, in, 14, Selby-Bigge, pp. 161-2.
" Treatise, I, m, 14, Selby-Bigge, p. 159.
 HUME'S ACCOUNT 23
of events in muscles and nerves and 'animal spirits', which we
do not. These are strong but not absolutely conclusive argu-
ments against a known necessitya> if this is taken as necessity21}
something that would support deductively valid a priori causal
inference. They seem less strong if it is taken as necessity2,2>
something that would support probabilistic a priori causal in-
ference. It may, indeed, seem that Hume underplayed his hand
here, that instead of looking for examples of observed necessity
and challenging his opponents to produce some he might have
asked what could conceivably count as a perception of neces-
sity. How could we possibly perceive anything in one event
which would in itself point to anything else ? Or how could we
perceive anything in one individual cause-effect sequence which
would tell us that any other event which resembled this cause,
however closely, was either certain or even likely to be followed
by an event which resembled this effect? Surely anything that
we perceived would be, as it were, enclosed within itself, and
could not point beyond itself in any way. This would have been
a very forceful argument for Hume's second exclusion. It is in
fact a variant of the strong basic case against synthetic a priori
knowledge of an objective reality. But we should not decide
whether to accept it without qualification until after we have
weighed the further considerations that will be introduced in
Chapter 8.
Hume's third negative point is that we find nothing at all in
causal sequences except regular succession and (perhaps) con-
tiguity. This third sweeping exclusion is hardly supported at all;
but it is undeniably asserted. 'It appears that, in single instances
of the operation of bodies, we never can, by our utmost scrutiny,
discover any thing but one event following another, . . . All
events seem entirely loose and separate. One event follows
another; but we never can observe any tie between them.'28
28 Enquiry, Sect. VII, Pt. II, Selby-Bigge, pp, 73-4: cf. Treatise I, III, 2, Selby-
Bigge, p. 77. Hume's view is endorsed and defended by Ayer in Probability and
Evidence, pp. 10-18: 'I think that we can make sense of this issue, . . . if we under-
stand Hume to be denying that apart from relations of comparison there could be
anything more than spatio-temporal relations between events. This may sound
paradoxical at first hearing, but when we consider what it comes to we shall see
that it is at least very plausible. It amounts in fact to the claim, which is scientific-
ally most respectable, that everything that happens in the world can be represented
in terms of variations of scenery in a four-dimensional spatio-temporal continuum,.,
there is nothing in the situation that calls for anything more than the identification
8244058 B
34 HUME'S ACCOUNT
At the same time, Hume himself supplies examples which
falsify this claim: 'the reason, why we imagine the communica-
tion of motion to be more consistent and natural, not only than
those suppositions [that a body striking another might just stop,
or go back, or be annihilated, etc.], but also than any other
natural effect, is founded on the relation of resemblance betwixt
the cause and effect, which is here united to experience, and
binds the objects in the closest and most intimate manner to
each other . . /.a9 Of course, he is here arguing that such
resemblance does not constitute an observed necessity2, that it
does not justify an inference but only aids the association of
ideas. Still, it is something other than contiguity and succession
that we observe in some causal sequences. This falsifies the third
exclusion, which is made plausible only by its being confused
with the second, just as the second received illicit support from
confusion with the first. Again, Hume admits that when we set
bodies in motion we experience 'a nisus or endeavour', and
again that we have a special 'sentiment or feeling' when an
object in motion strikes us.30 These again are additional observed
features in some causal sequences. Hume brushes them aside as
'sensations, which are merely animal, and from which we can
a priori draw no inference'; that is, they cannot constitute neces-
sity^ but again this is only a confused reason for not admitting
them as exceptions to the third exclusion. It is clear that both
some sort of resemblance and some sort of nisus enter into the
process by which will or desire or intention leads on to bodily
movement; a pre-existing intention foreshadows the movement
which fulfils it, and there is often some feeling of effort. Modern
of various objects and the specification of their changing spatio-temporal rela-
tions.' But it is worth noting that Ayer (i) allows relations of comparison,
fii) introduces, though with a hint that this is unnecessary, enduring identifiable
objects, and (iii) like Hume, mixes up this third exclusion with the second: it is
for the latter that he really argues:'.. . either this supposed relation is a phenomenal
property, something detectable in the situation, or it is not. If it is, its existence
will enable us to draw no conclusion with regard to any event occurring at any
other time than the time at which it is manifested. If it is not . . . no terms could
then be observed to stand in this relation . . . our uncertainty about what will
happen cannot be removed by the introduction of this entirely spurious relation.'
Ayer makes out a strong case for what I have called Hume's second exclusion, but
not for the third. What this leaves room for, and why it is of interest, will be dis-
cussed in later chapters, particularly Chapter 8.
z
* Treatise, I, ni, 9, Selby-Bigge, pp. 11 i-ia.
a° Enquiry, Sect. VII, Pt. II, footnote, Selby-Bigge, pp. 77-8.
 HUME'S ACCOUNT 25
philosophers sometimes on this account argue that intentional
action cannot be a case of 'Humean' causation; but it is at
least as reasonable to argue the other way round, that this is a
case of causation and that Hume should therefore have with-
drawn his third exclusion. Modern philosophers are, however,
wrong if they think that intentions and actions are logically
connected in a way that would conflict with Hume's first
exclusion.31
In addition to these three negative points, Hume makes the
positive assertion that causation as we know it in the objects
involves regular succession. I have commented, right at the
beginning of this chapter, on the vagueness and the prima facie
implausibility of this claim. Whether all causal relationships
that we discover involve some sort of regularity of succession,
and if so what sort, are questions I must leave until a later
chapter.32 But in fairness to Hume it should be noted that he
does not say that we can discover a particular causal relationship
only by observing a number of sequences that instantiate it. On
the contrary he admits that 'not only in philosophy, but even
in common life, we may attain the knowledge of a particular
cause merely by one experiment, provided it be made with
judgement, and after a careful removal of all foreign and super-
fluous circumstances',33 and his 'Rules by which to judge of
causes and effects'34 include fairly crude anticipations of Mill's
methods of agreement, difference, and concomitant variation,
which again allow the discovery of a causal relationship by the
use of very restricted sets of observations. Of course, implicit in
these methods is an appeal to some general uniformity assump-
tion,35 and Hume himself explains that the 'one experiment' is
interpreted in the light of the principle—itself based on custom
—that 'like objects., placed in like circumstances, will always produce
like effects'. In other words, he is content to let 'custom' provide
a belief in certain general principles, but then to allow us to
« e.g. A. I. Melden, Free Action, p. 53; T. F. Daveney, 'Intentions and Causes',
Analysis, xxvii (1966), 23; the view is criticized but also defended in a revised form
by F. Stoutland, 'The Logical Connection Argument', American Philosophical Quar-
terly Monograph 4 (1970), 117-30 and by G. H. von Wright, Explanation and Under-
standing, pp. 93-4; it is discussed in Chapter 11 below.
JI 3
Chapter 3. * Treatise, I, HI, 8, Selby-Bigge, p. 104.
» Treatise, I,-m, 15, Selby-Bigge, pp. 173-4.
3!
Cf. my article, 'Mill's Methods of Induction', in Encyclopedia of Philosophy^
ed. P. Edwards, also Chapter 3 and the Appendix to this book.
26 HUME'S ACCOUNT
discover particular causal relationships by reasoning from those
principles together with a few relevant observations. Belief, it
seems, need not arise immediately from an associative link with
a present impression, but can be produced by rational argu-
ment so long as the premisses include some to which the
associative theory applies.
It is a curious fact that Hume's choice of regular succession
as the one acceptable feature of causation as it is known in the
objects may well be another result of his assumption that the
differentia of causal sequence must be something that will
justify causal inferences. For of course if we know that events
of kind C are always followed by events of kind E, and events
of kind E always preceded by events of kind C, then on observ-
ing a C we can validly infer that an E will follow, and on
observing an E we can validly infer that a C has preceded.
Regularity would do the job that Hume throughout assumes
that necessity should do. But of course only genuine, unqualified
regularity of succession will do this, whereas if Hume's doubts
about induction are justified all that we can know is a regularity
in all the cases so far observed. Only the latter could consistently
figure in Hume's account of causation as we know it in the objects,
The former, the unqualified regularity, might figure in an
account of causation as it is in the objects; but, if he were con-
sistent, this would fall under his condemnation of 'qualities,
both in material and immaterial objects, with which we are
utterly unacquainted'.36
We might, in other words, introduce a necessity3, defined as
something that would license causal inference—in both direc-
tions, from cause to effect and from effect to cause—but not,
like necessity^, as licensing such inference a priori, not, that is,
supposed to be discoverable in each single cause or effect on
its own. As I have shown, Hume undoubtedly sometimes
assumes that our ordinary idea of necessity!, our notion of what
marks off certain sequences as causal, is the idea of a necessity2.
But at times he uses only the weaker assumption that our idea
of necessity-L is that of a necessity3. Regularity (of some sort) is
admirably suited for the role of necessity3, and it is not as
epistemologically objectionable as necessity2. We can at least
say what regular succession would be like, which is more than
36
Cf. p. a i n. 24 above.
 HUME'S ACCOUNT 27
we can say about necessity^. Presumably this is why Hume is
prepared to use it in his reforming definitions, his account of
what we can properly mean by 'cause'. But if his doubts about
induction were well founded, not even necessity3 could be
known. We should be using a regularity in which we believed
to license a causal inference about a particular case, but the
belief in the regularity would itself be explained by Hume as a
product of the association of ideas, and hence the causal infer-
ence in turn is never really justified, but only explained.
Modern philosophers frequently accuse Hume of having con-
fused logic with psychology. His discussion certainly combines
logical and psychological elements, but it is disputable whether
they are confused. Admittedly his terminology can mislead us.
If he ascribes a certain operation to, say, 'imagination' rather
than 'reason', he may seem to be saying merely that it is per-
formed by one mental agency or faculty rather than another,
but in fact he will also be making a logical evaluation of the
performance: whatever 'reason' does will be rationally de-
fensible, what 'imagination* does will not. But we can, without
much difficulty, learn to interpret such remarks as these. What
is more important is that logical and psychological considera-
tions can properly contribute to the same investigation. If it
appears that we have a certain 'idea' or concept, but it is a
moot point whether it is true of, or genuinely applicable to,
the real things to which we are inclined to apply it, considera-
tions which bear on the question whether those real things
could conform to this concept will be relevant, so will con-
siderations that bear on the question whether we could have
observed such conformity, and so too will the possibility of
giving an alternative account of how we could have acquired
this concept if we did not in fact observe things conforming to
it. In particular, anyone who, like Hume, casts sceptical doubts
on ingrained everyday assumptions is committed to giving an
alternative account of how these beliefs could have arisen if
they were false. But such alternative accounts cannot fail to
appeal to psychological principles or theories.
On the other hand, Hume's resort to psychology was pre-
mature. As we have seen, he looked for a psychological explana-
tion of a very puzzling idea of necessary connection without
having properly shown that the kind of necessity which he was
28 HUME'S ACCOUNT
about to locate in the mind really was the key element in our
concept of causation. Before we look for a psychological explana-
tion of an inappropriate idea, we should investigate more care-
fully what idea we actually have and use, and then inquire
whether it is inappropriate.
Hume's discussion of causation, then, includes both strong
points and weak ones. Accepting his exclusion of logically
necessary connection and of necessity2 1} and postponing con-
sideration of necessity2.2> we have still to consider what sort or
sorts of regularity, if any, characterize causal sequences, what
within our ordinary concept of causation differentiates causal
sequences from non-causal ones, and what features—whether
included in this ordinary concept or not—over and above
'regularity' are to be found in causal relationships. We may
start with the job that Hume only pretended to be doing, of
identifying what we naturally take as the differentia of causal
sequences; that is, we may ask again what our idea of necessary
connection (necessityj) is in an open-minded way, without
assuming that it must be the idea of something that would
license a priori inferences.
 2

The Concept of Causation-

Conditional Analyses

ONE of the many problems with which Hume's discussion

leaves us is this: what is our concept of causal as opposed to
non-causal sequences; what do we (rightly or wrongly) take to
be their distinguishing feature? The obvious, naive, way of
tackling this question would be to compare two otherwise
similar sequences, only one of which we recognize as causal,
and see what distinctive comment we should make about the
causal one. Let us, then, compare the following sequences A
andB.
A: A chestnut is stationary on a flat stone. I swing a hammer
down so that it strikes the chestnut directly from above. The
chestnut becomes distinctly flatter than before.
B: A chestnut is stationary on a hot sheet of iron. I swing a
hammer down so that it strikes the chestnut directly from
above. At the very instant that the hammer touches it, the
chestnut explodes with a loud pop and its fragments are
scattered around.
I assume that we know or perceive enough about sequences A
and B to be able to say that the chestnut's becoming flatter in
A was caused by its being hit with the hammer, but that its
exploding in B was not caused by its being hit with the hammer.
(Among other things, the explosion was not due to the hammer's
pushing the chestnut into closer contact with the hot metal, or
anything like that.) I am not concerned, at the moment, with
how we know this, but merely with what we mean, what we are
saying when we say that the blow-flattening sequence was
causal but that the blow-exploding sequence was not.
I suggest that the obvious answer, the one that pretty well
30 CONDITIONAL ANALYSES
any intelligent but unprejudiced person would give to this
question, is that in A the chestnut would not have become flatter
if the hammer had not struck it, whereas in B it would have
exploded even if the hammer had not struck it. It is tempting
to say, then, that the contrast between these two contrary-to-
fact conditionals is at least an initial indication of what we
mean by calling the one sequence causal and the other not,
that is, that to ascribe what we called necessity! is to assert such
a contrary-to-fact conditional as the former. As Mill says, 'if a
person eats of a particular dish, and dies in consequence, that
is, would not have died if he had not eaten of it, people would
be apt to say that eating of that dish was the cause of his
death'.1 They would indeed.
Curiously enough, Hume says this too. In the Enquiry, after
giving the first, regularity, definition ('we may define a cause
to be an object, followed by another, and where all objects similar to
the first are followed by objects similar to the second'), he adds 'or in
other words where, if the first object had not been, the second never had
existed'.2 Now of course this variant is not the regularity defini-
tion in other words. (It does not even follow from the regularity
definition as given here, though on a certain interpretation of
the conditional it might be held to follow from, and even to be
equivalent to, a different regularity statement, namely 'where
all the objects similar to the second are preceded by objects
similar to the first'), But I shall not spend time on the questions
why Hume asserted this implausible equivalence, or why he
did not realize that this was a much better answer than the
one he had given in all his previous discussion to the question
'What is our idea of necessary connection?', but consider
directly the merits and the defects of this suggestion.
To simplify matters, I shall for the present be considering
only singular causal statements, especially statements about
individual event sequences, such as 'The hammer's striking the
chestnut caused the flattening of the chestnut*. I shall speak of
causes and effects (or results) as events, but 'event' is just a
general term that stands in for such items as 'the hammer's
1
System of Logic, Book III, Gh. 5, Sect. 3. Conditional analyses of causal state-
ments are defended and discussed by Ardon Lyon in 'Causality', British Journal
for the Philosophy of Science, xviii (1967), i-ao.
2
Enquiry, Sect. VII, Pt. II, Selby-Bigge, p. 76.
 CONDITIONAL ANALYSES 31
striking the chestnut*. (Anxieties about the exact ontology of
causation may be postponed for later consideration in Chapter
10.) Such phrases are nominalizations of singular event-
sentences or event-clauses; e.g. ''the flattening of the chestnut*
is a nominalization of'the chestnut became flatter'. The present
suggestion, then, is that a statement of the form 'X caused T'
means 'X occurred and T occurred and T would not have
occurred if X had not', it being understood that when we
instantiate *Z* and T* with particular event descriptions we can
express the suggested meaning most neatly by going back from
the nominalizations to the corresponding clauses or their nega-
tions: e.g. 'the striking of the match caused the appearance of
the flame' would, on this suggestion, mean 'the match was
struck and the flame appeared and the flame would not have
appeared if the match had not been struck*.
But is the suggestion correct?
First, are there cases where we would say that X caused T
but would not say what is proposed above as the meaning of
this claim? Well, clearly we cannot say that Xcaused Sunless
both X and T occurred, provided that we take 'occurred' in a
fairly broad sense, to include the presence or persistence of
standing conditions. But what about the counterfactual con-
ditional? Might we not say that the striking of the match caused
the appearance of the flame and yet admit that even if the
match had not been struck the flame would have appeared if,
say, this match had been touched by a red-hot poker? Certainly
we might, so the suggestion needs some modification at least:
let us insert 'in the circumstances', reading '. . . and in the
circumstances the flame would not have appeared if the match
had not been struck*. 'The circumstances' can be taken to
include the fact that this match was not in fact touched by a
red-hot poker at the critical time. And in general we can modify
our suggestion to read '. . . in the circumstances T would not
have occurred if X had not', the qualifying phrase being inter-
preted in the sort of way indicated by our example. An objector,
however, might say, *But what if, for example, the match really
was touched by a red-hot poker at the same instant that it was
struck? Then even in the circumstances the flame would have
appeared even if the match had not been struck.' True, but it
is not clear whether in this case: we would say that the striking
 32 CONDITIONAL ANALYSES
of the match caused the appearance of the flame. What we
would or should say in such cases of causal over-determination,
of fail-safe causes, will be considered later: for the moment we
need not take them as falsifying our modified suggestion.
Secondly, and more awkwardly, are there cases where we
would not say that X caused T but would say that X and 1C
both occurred and that in the circumstances T would not have
occurred if X had not? Would we not have to say the latter,
trivially, if X and T were identical? Provided that AT occurred,
we must admit that X occurred and X occurred and in the
circumstances X would not have occurred if X had not. But
events, unlike Spinoza's God, are not commonly said to cause
themselves. Equally, the penny would not have fallen heads-up
if it had not fallen tails-down; but we would not say that its
falling tails-down caused its falling heads-up. Again, the driver
would not in the circumstances have broken the law if he had
not exceeded the speed limit; but we would not say that his
exceeding the speed limit caused his breaking of the law. To
exclude all such counter-examples, we must say that, in addi-
tion to the meaning suggested above, 'X caused Y' presupposes
that X and T are distinct events, and, as the last two examples
show, this must be taken in a stronger sense than merely that
f
X' and 'Y' are logically or conceptually independent descrip-
tions. Indeed, it is not even necessary, any more than it is
sufficient for this purpose, that these should be logically inde-
pendent descriptions. It is legitimate, though trivial, to say that
X caused every effect of X> again I can say that my meeting
Tom on Monday caused my remembering, on Tuesday, my
meeting Tom on Monday. And so on. Logically independent
descriptions, then, are not called for; what is required is that
the cause and the effect should be, as Hume says, distinct
existences. It may be objected that this requirement is vague or
obscure, but it is not, I think, necessary for me to aim at any
great precision here. I am discussing only what causal state-
ments mean, and for this purpose it is sufficient to say that
someone will not be willing to say that X caused T unless he
regards X and T as distinct existences. But are there really no
limits to the possible descriptions of X and 1"? What about
irrelevant descriptions, ones that have, as we would say, nothing
to do with the causal relationship? Some seem acceptable: 'the
 CONDITIONAL ANALYSES 33
incident in the tennis court advanced the French Revolution.*
But consider 'Jane's eating fishpaste in the pantry caused her
feeling ill' and the corresponding suggested expansion 'Jane ate
fishpaste in the pantry and felt ill and in the circumstances she
wouldn't have felt ill if she hadn't eaten fishpaste in the pantry*.
We are inclined to comment that she felt ill either because she
ate too much fishpaste or because the fishpaste was bad, and in
either case it wouldn't have mattered where she ate it. This
raises the question (which will be discussed in Chapter 10)
about the extensionality of causal statements. For the present
it is enough to say that the issue is in doubt: we might want to
reject or deny both the proposed singular causal statement and
its expansion, or we might want to say that so long as it was
eating the fishpaste that made Jane ill, and it was in the pantry
that she ate it, they are both true but merely misleading. But it
seems that whatever we decide to say, we can say about both
the singular causal statement and its proposed expansion, so
that examples of this sort do not tell against the suggestion that
the latter shows what the former means. Gases where X and T
are collateral effects of a common cause create a more serious
difficulty. Labour's defeat at the election pleases James but
saddens John, who, as it happens, are quite unknown to each
other. Then James's being pleased does not cause John's being
sad, and yet we might well say that in the circumstances John
would not have been sad if James had not been pleased. Yet
there is a way of handling the words 'if and 'in the circum-
stances' that will defeat this supposed counter-example. Con-
struct a 'possible world' in the following way: take something
that is just like the actual world up to the point where, in the
actual world, James is pleased; keep out, from your possible
world, James's being pleased, but otherwise let your possible
world run on in the same manner as the actual world; only if
John does not become sad in your possible world can you say
'in the circumstances John would not have been sad if James
had not been pleased*. And if you can say this, then you can
also say that James's being pleased caused John's being sad (in
the actual world), But this special way of handling *if in the
circumstances' runs a grave risk of circularity if we want to
use this conditional as an analysis of causal statements, since
in explaining this handling we have had to use the notion of
 34 CONDITIONAL ANALYSES
letting the possible world run on. Yet without this special inter-
pretation of the counterfactual conditional, it seems that while
it may be required for the truth of the corresponding causal
statement, it is not sufficient for this.
Another difficulty is concerned with the distinction which we
are inclined to draw between conditions and causes. There may
be a set of factors which were, in the circumstances, jointly
sufficient and severally necessary for a certain result, and which
all occurred, as, consequently, did the result. Then we can say
of each of these factors that if in the circumstances it had not
occurred the result would not; but we may not be so willing
to say of each of them that it caused the result. Perhaps we
ought to say this, since they are all logically related to the result
in the same way: the situation as described is symmetrical with
respect to all the factors. But at present we are discussing not
what we ought to say but what our causal statements mean.
And there is no doubt that we tend to be a bit selective, to be
more ready to call some kinds of factors causes than others.
There are no firm rules governing this selection, but there are
some fairly systematic tendencies.
Thus we are more ready to say that an event caused a certain
effect than that a standing condition did: it was the spark
rather than the presence of inflammable material that caused
the fire. Even among events ones which are seen as intrusive
are picked as causes in preference to ones which occur within
some going concern: it was the severing of the artery rather
than the pumping of the heart that caused the loss of blood.
This preference shades into one governed by moral and legal
assumptions: what is normal, right, and proper is not so readily
called a cause as is something abnormal or wrong. This third
tendency may conflict with and override the first. Since it is
normal for people to be striking matches and lighting cigarettes
in a residential flat, but a gas leak is abnormal and should
not occur, we may well say that the explosion which wrecked
this block of flats was caused by the presence of a quantity
of gas rather than that it was caused by Jones lighting his
cigarette.
These matters may be clarified to some extent if we realize
that causal statements are commonly made in some context,
against a background which includes the assumption of some
 CONDITIONAL ANALYSES 35
3
causal fold. A causal statement will be the answer to a causal
question, and the question 'What caused this explosion?' can
be expanded into 'What made the difference between those
times, or those cases, within a certain range, in which no such
explosion occurred, and this case in which an explosion did
occur?' Both cause and effect are seen as differences within a
Held; anything that is part of the assumed (but commonly un-
stated) description of the field itself will, then, be automatically
ruled out as a candidate for the role of cause. Consequently if
we take the field as being this block of flats as normally used and
lived in, we must take Jones's striking a match to light his
cigarette as part of the field, and therefore not as the cause of,
or even a cause of, or as causing, the explosion. What caused
the explosion must be a difference in relation to the field, and
the gas leak, or the presence of the gas that had leaked out,
is the obvious candidate.
What is said to be caused, then, is not just an event, but an
event-in-a-certain-field, and some 'conditions' can be set aside
as not causing this-event-in-this-field simply because they are
part of the chosen field, though if a different field were chosen,
in other words if a different causal question were being asked,
one of those conditions might well be said to cause this-event-
in-that-other-field. Any part of the chosen field is decisively
ruled out as a cause; a more elusive point is that among factors
not so ruled out which are still severally necessary for the effect,
we still show some degree of preference on the grounds indicated
above. But I think that this can be taken as reflecting not the
meaning of causal statements but rather their conversational
point, the sorts of use to which they are likely to be put. We
often want to know what caused some event with a view to
saying how it could, and perhaps should, have been prevented:
pointing to the spark rather than to the presence of inflammable
material indicates what would have been the last chance of
preventing the fire. But it might be conceded that the state-
ment that the presence of this material caused the fire would be
3
This notion of a causal field was introduced by John Anderson in 'The
Problem of Causality', Australasian Journal of Psychology and Philosophy, xvi (1938),
reprinted in his Studies in Empirical Philosophy, and used to resolve difficulties in
Mill's account of causation, I also used it in 'Responsibility and Language',
Australasian Journal of Philosophy, xxxiii (1955), I43~59} to deal with problems of
moral and legal responsibility.
36 CONDITIONAL ANALYSES
as true as the statement that the spark caused it, and merely m
some ways less interesting. We can and do indeed distinguish
between triggering causes and predisposing causes, which shows
that standing conditions are not prevented from being causes
by the mere meaning of the noun, or of the verb, 'cause'.
Similarly we may agree that the collision was caused just as
much by Smith's driving straight ahead as by Brown's deviating
to his right without warning, but say that it is more important,
for moral and legal purposes, to draw attention to the second
of the two causal relationships.
The supposed distinction between conditions and causes can
be adequately accounted for in these two ways: an alleged
condition which is not called a cause, although if in the cir-
cumstances it had not occurred the result would not, either is
part of the field presupposed in the view taken by the speaker
of the result (and so is not a cause in relation to this field) or
is a cause, but mention of this fact happens to be irrelevant, or
less relevant than mention of some other cause of the same
result, to some current purpose.
There is admittedly some logical redundancy in the two treat-
ments offered here of conditions which we are reluctant to call
causes. We should get a neater account if, say, we assigned all
such conditions to the assumed causal field, and said that any-
one who withheld the title of cause from what he admitted to
be one of a set of severally necessary and jointly sufficient con-
ditions for some result was implicitly taking that condition to
be part of the field. Again, we should get a neater account if we
discarded the notion of a field, and interpreted the withholding
of the title of cause from any such necessary factor as reflecting
some conversational or other purpose of the speaker. But at
present my object is not so much to give a logically neat account
as to analyse our ordinary thinking about causal sequences, and
I believe that this does contain the two separate elements I have
tried to describe. On the one hand there is the implicit question,
'What caused such-and-such?', where 'such-and-such' is already
thought of as a change in a certain material or background
state of affairs; on the other hand, even among differentiae
within this background, we tend to downgrade some and at
least initially deny that they caused the result, though we might
reluctantly concede, if pressed, that they helped to cause it.
 CONDITIONAL ANALYSES 37
Another sort of apparent counter-example is easily disposed
of in the same way. It can be truly said of anyone now dead
that if he had not been born he would not have died, and
although it would here be pointless to insert 'in the circum-
stances', such an insertion would not make the statement false.
Yet we would not say that being born caused his death—that
is, not ordinarily: with sufficient ingenuity one can construct
a case for saying this. But the reason is merely that when we
look for a cause of someone's death, the event, this person's
death, is a change in a field which centrally includes this
person's being alive for a while, and hence (in the ordinary
course of nature) his having been born. As before, his being
born is part of the field, and therefore cannot be the cause of
his death.
The suggested meaning for 'X caused 1*' which we have been
considering is (X occurred and T occurred and in the circum-
stances, T would not have occurred if X had not'. A slightly
different formula would run '. . . and there were circumstances
such that in them T would not have occurred if X had not'. To
see the difference, suppose that the effect is that a certain part
of a certain house catches fire, that the presence of oxygen, the
absence of any extinguishing device, etc., are taken as parts of
the field, and that in relation to this field the following three
conditions, which were all present on the occasion in question,
were severally necessary and jointly sufficient for the effect: an
electric current in a certain wire (A), decayed insulation at a
point on that wire (B), and inflammable material near that
point (C). Then there were circumstances, namely B and C, such
that in them the fire would not have occurred if the current, A,
had not; equally there were circumstances, namely A and C,
such that in them the fire would not have occurred if the faulty
insulation, B, had not, and again circumstances, namely A and
B) such that in them the fire would not have occurred if the
inflammable material, C, had not. So by our modified formula,
A, B, and C would each equally count as causing the fire. But
someone who says that, for example, the faulty insulation, B,
caused the fire, may be thinking of A and C as 'the circum-
stances', may be presupposing them rather than saying that
there are circumstances in relation to which B was necessary.
Before he will agree that A caused the fire, he will have to make
38 CONDITIONAL ANALYSES
a switch of presuppositions, and take B and C, rather than A
and C, as 'the circumstances'. So whereas with our modified
formula there is no conflict at all between the three statements
that A caused the fire, that B did so, and that C did so—since
the three existential statements are entirely compatible—there
is, with our first formula, using the phrase 'in the circumstances',
a contrast of approach and presupposition between those who
assert these three statements, although no outright disagree-
ment. And this surely is the case. If three speakers put forward
these three alternative causal statements, they would be con-
scious of such a contrast of approach. I think, therefore, that
our first formula is the better for catching the force of a causal
statement in use. The modified formula might be said to yield
more clearly marked truth conditions for causal statements; but
I shall argue that what it yields are not, strictly speaking, truth
(or falsity) conditions. Still, it may be conceded that the modi-
fied formula represents a perhaps desirable tidying up of the
meaning of causal statements, whereas the original formula
comes closer to displaying the meaning they actually have.
There is, however, something surprising in our suggestion
that 1X caused Y> means, even mainly, that X was necessary in
the circumstances for T, Would it not be at least as plausible
to suggest that it means that X was sufficient in the circumstances
for Y? Or perhaps that it was both necessary and sufficient in
the circumstances? After all, it is tempting to paraphrase 1X
caused T3 with 'X necessitated T3 or (X ensured T', and this
would chime in with some of the thought behind the phrase
'necessary connection*. But if 'X necessitated T' is taken liter-
ally, it says that T was made necessary by X or became neces-
sary in view of X, and this would mean that X was sufficient
rather than necessary for T. Of course, an X which in ordinary
discourse is said to cause a Th practically never in itself suffi-
cient for that T, or even believed to be so; as before, we must
add the qualification 'in the circumstances'. And then the
suggested description seems to fit what we recognize as causes.
Taking our last example again, and letting the circumstances
include the electric current and the inflammable material, we
can say that the faulty insulation (which on these assumptions
caused the fire) was sufficient in the circumstances for it. And
similarly in sequence A in our earlier example the hammer-
 CONDITIONAL ANALYSES 39
blow was sufficient in. the circumstances for the chestnut's be-
coming flatter. But is sufficiency sufficient, or is necessity neces-
sary as well? Can we explain the fact that in sequence B the
hammer-blow did not cause the explosion on the ground that
it was not sufficient for it in the circumstances? In fact the
circumstances in themselves, without the hammer-blow, were
sufficient: does it not follow that the hammer-blow was trivially
and automatically sufficient in the circumstances, since any-
thing at all, or nothing, would have been so, provided that the
relevant circumstances were not tampered with? If we argued
that what is only thus trivially sufficient in the circumstances
cannot be properly so described, and that that is why it cannot
be said to cause the result, are we not covertly reintroducing
the requirement that a cause should be necessary in the cir-
cumstances rather than sufficient? What there is any point in
mentioning as sufficient in the circumstances will be necessary
in the circumstances as well.
This point seems to be confirmed if we try to expand the
phrase 'sufficient in the circumstances' into a conditional state-
ment. 1X will be sufficient in the circumstances for T may be
taken as saying 'Given the circumstances, if X occurs then T
will', and even as a non-material conditional this will count as
true provided that the circumstances do not change, X occurs,
and T occurs also.4 And then 'X was sufficient in the circum-
stances for T' will be equivalent to 'Given the circumstances, if
X occurred, then T did', and provided that the circumstances
referred to are the actual ones this will automatically be true
of any sequence in which X and Y actually occurred. Sufficiency
in the circumstances is, then, of no use for our present purpose
of finding the distinguishing feature of causal sequences; every
cause is sufficient in the circumstances for its effect, but so are
many non-causes for events which are not their effects.
But this is a weak sense of 'sufficient in the circumstances'.
Can we find also a strong, counterfactual sense which will not
apply to the antecedent in every actual sequence? This would
be that, given the circumstances, if T had not been going to
occur, X would not have occurred. This is a possible sense, and
what we recognize as causes are in general sufficient in the
* I have discussed non-material conditionals, both open and counterfactual,
and their truth conditions, in Chapter 3 of Truth, Probability, and Paradox,
40 CONDITIONAL ANALYSES
circumstances in this strong sense as well as in the weak one.
In the appropriate possible world in which the circumstances
are the same as in sequence A, but the chestnut does not become
flatter, the hammer-blow has not occurred. (In constructing
this possible world, we are, of course, taking over laws of work-
ing from the actual world.) And we cannot say the correspond-
ing thing about the non-causal sequence B. The statement 'If
in the circumstances of sequence B the explosion had not been
going to occur, the hammer-blow would not have occurred' is
not true or even acceptable. The supposition implicit in its
antecedent cannot be coherently considered in the light of the
actual world's laws of working, for, given these laws, in the
circumstances of sequence B the explosion was going to occur.
And if we take the antecedent as inviting us to consider some
change in those laws of working, there is still no reason why we
should not combine, with whatever coherent interpretation we
place on this supposition, the view that the hammer-blow still
occurred. It looks, then, as if the strong counterfactual sense of
'sufficient in the circumstances' will distinguish causal from
non-causal sequences, though the weak sense does not.
But granted that causes are in general sufficient in the cir-
cumstances, in this strong sense, as well as necessary in the
circumstances, for their effects, while neither relation holds in
non-causal sequences, we can still ask whether in calling some-
thing a cause we require both of these features or only one, and
if so which.
To clear up this problem, let us consider three different
shilling-in-the-slot machines, Kt L} and M. Each of them pro-
fesses to supply bars of chocolate; also each of them has a
glass front, so that its internal mechanism is visible. But in other
respects the three are different. K is deterministic, and con-
forms to our ordinary expectations about slot-machines. It does
not always produce a bar of chocolate when a shilling is put
in the slot, but if it does not there is some in principle discover-
able fault in or interference with the mechanism. Again, it can
be induced to emit a bar of chocolate without a shilling's being
inserted, for example by the use of some different object which
sufficiently resembles a shilling, or perhaps by poking at the
mechanism with pieces of wire. Inserting a shilling is neither
absolutely necessary nor absolutely sufficient for the appear-
 CONDITIONAL ANALYSES 41
ance of a bar of chocolate, but in normal circumstances it is
both necessary and sufficient for this. ('Necessary' and 'suffi-
cient' are here being used with reference to the machine's laws
of working, they describe general relations rather than rela-
tions between single occurrences. But it will be a consequence
of these general relations that, if on a particular occasion, in
normal circumstances, a shilling is inserted, a bar of chocolate
will come out, and further that the inserting of a shilling on
this particular occasion was both necessary in the circumstances
and sufficient in the circumstances in the strong sense for this
result.) L, on the other hand, is an indeterministic machine. It
will not, indeed, in normal circumstances produce a bar of
chocolate unless a shilling is inserted, but it may fail to produce
a bar even when this is done. And such failure is a matter of
pure chance. Z,'s failures, unlike jff's, are not open to individual
explanation even in principle, though they may be open to
statistical explanation. With L, in normal circumstances, putting
a shilling in the slot is necessary, but not sufficient, for the
appearance of a bar of chocolate. M is another indeterministic
machine, but its vagaries are opposite to Lys. M will, in ordinary
circumstances, produce a bar of chocolate whenever a shilling
is inserted; but occasionally, for no reason that is discoverable
even in principle, the mechanism begins to operate even though
nothing has been inserted, and a bar of chocolate comes out.
With M, in normal circumstances, putting a shilling in the slot
is sufficient, but not necessary, for the appearance of a bar of
chocolate.
Now on some occasion I put a shilling into K and a bar of
chocolate comes out. As I have said, putting in the shilling was,
in the circumstances, both sufficient in the strong sense and
necessary for this result, and we have no hesitation in saying
that it caused this result.
Again, I put a shilling into L and receive a bar of chocolate.
Putting in the shilling was, in the circumstances, necessary for
this result. It was also sufficient in the circumstances in the
weak sense, but not in the strong, counterfactual, sense. A
possible world, with the same laws of working as the actual
worjd, can contain the same circumstances, can lack the result,
and yet still contain the inserting of the shilling. The statement,
'Given the circumstances, if the chocolate had not been going
43 CONDITIONAL ANALYSES
to appear, the shilling would not have been inserted* is not
now acceptable. But would we say in this case that the inserting
of the shilling caused the appearance of the bar of chocolate?
I think we would. Our ordinary causal concept seems to require
that where the shilling is put in, the mechanism operates, and
a bar of chocolate appears, and would not have appeared if
the shilling had not been inserted, the insertion of the shilling
caused the appearance of the chocolate despite the fact that in
the circumstances even given that the shilling was inserted, the
chocolate might not have appeared.
Similarly, I put a shilling into M and receive a bar of choco-
late. Putting in the shilling was this time sufficient in the
circumstances, in the strong sense as well as in the weak sense,
for this result. In an appropriate possible world in which the
chocolate did not appear, the shilling would not have been put
in. But putting in the shilling was not, it seems, necessary in
the circumstances. Not only generally but also on this particular
occasion the chocolate might have appeared even if no shilling,
or anything else, had been put in. But there is room for dispute
here. Perhaps, it might be argued, the insertion of the shilling
may have been necessary in the circumstances on this particular
occasion, though not generally; it may be that on this occasion
the mechanism would not have operated if the shilling had not
been put in. Moreover, it may be possible to settle this issue.
We can, by hypothesis, see the works of the machine. If on this
occasion only the later stages of the mechanism operated, not
those earlier ones which are normally actuated directly by the
shilling, we can decide that the chocolate would have come out
anyway, that the insertion of the shilling was not necessary in
the circumstances. But if the whole mechanism operated, the
issue cannot be settled. Since M is, by hypothesis, indeter-
ministic, there is in principle no discoverable answer to the
question whether the chocolate would on this occasion have
come out if the shilling had not been put in, or, therefore, to
the question whether the insertion of the shilling was on this
occasion necessary in the circumstances for the result. And yet,
it seems to me, it is just this question that we need to have
answered before we can say whether the insertion of the shilling
caused the result. If the chocolate would not have come out if
the shilling had not been put in, then the insertion of the
 CONDITIONAL ANALYSES 43
shilling caused the result. But if it would have come out anyway,
the insertion of the shilling did not cause this, (This last ruling
prejudges a question about causal over-determination that has
still to be considered; but we shall reach an answer to this
question which agrees with the present ruling.) And, con-
sequently, if it is in principle undecidable whether the chocolate
would on this particular occasion have come out if the shilling
had not been put in, it is equally undecidable whether the
putting in of the shilling caused the appearance of the chocolate.
The contrast between the comments we have made about the
two indeterministic slot-machines L and M seem to show that
*X caused T* entails (X was necessary in the circumstances for
1"' and also, trivially, 'X was sufficient in the circumstances
for 1*J in the weak sense, but not the latter in the strong counter-
factual sense. This is not required, though in general it holds
for what we recognize as causes, at least in the physical sphere.
Indeterministic machines are pretty rare.
This conclusion is confirmed by what we are prepared to say
about human beings without prejudging the question whether
they are wholly deterministic or not. 'Tom's saying what he
did caused Bill to hit Tom' is acceptable even if we suppose
that Bill had free will, that in the circumstances even after
Tom's remark Bill might have controlled himself, provided that
we believe that in the circumstances Bill would not have hit
Tom if Tom had not said what he did. In this respect we treat
Bill in the way I have proposed to treat slot-machine L.
There are, however, several possible objections to the thesis that
'A'caused T' entails 'X was necessary in the circumstances for T'.
The first of these concerns what we may call quantitative
over-determination. In sequence A, the hammer-blow caused
the chestnut's becoming flatter. But the whole of the blow was
not necessary for this result, though it was more than sufficient:
a somewhat lighter blow would have sufficed. Even if part of
the hammer-head had been absent, this result would still have
come about. And so on. But this difficulty is easily overcome.
It is possible to go on to a functional dependence view of
causation, which would relate exact quantities on the cause
side with exact quantities on the effect side. This development
will be considered in Chapter 6. But for the present we are
dealing only with a fairly primitive causal concept, which
44 CONDITIONAL ANALYSES
treats events, states, and occurrences, as ordinarily recognized
and described, as causes and effects. From this point of view we
regard the hammer-blow as a unit, and simply do not consider
parts or subdivisions of it or quantitative alterations to it. The
alternatives considered are that I strike the chestnut in the way
described and that I do not. In constructing possible worlds, in
considering what might or would have happened, we either
plug in the harnmer-blow as a whole or leave it out as a whole.
Reducing it a little bit is simply not an option at present under
consideration. From this point of view the hammer-blow was
necessary in the circumstances: leave it out as a whole from
your possible world, and the chestnut remains round.
A much more serious objection concerns what we may call,
by contrast, alternative over-determination, or what have been
called fail-safe causes. Let us list some examples, some of which
are favourites with writers on moral and legal responsibility.
(i) A man is shot dead by a firing squad, at least two
bullets entering his heart at once, either of which would have
been immediately fatal.
(ii) Lightning strikes a barn in which straw is stored, and a
tramp throws a burning cigarette butt into the straw at the
same place and at the same time: the straw catches fire.
(iii) '. . . conditions (perhaps unusual excitement plus con-
stitutional inadequacies) [are] present at 4.0 p.m. that guarantee
a stroke at 4.55 p.m. and consequent death at 5.0 p.m.; but an
entirely unrelated heart attack at 4.50 p.m. is still correctly
called the cause of death, which, as it happens, does occur at
5,0 p.m.*
(iv) Smith and Jones commit a crime, but if they had not
done so the head of the criminal organization would have sent
other members to perform it in their stead, and so it would have
been committed anyway.
(v) A man sets out on a trip across the desert. He has two
enemies. One of them puts a deadly poison in his reserve can
of drinking water. The other (not knowing this) makes a hole
in the bottom of the can. The poisoned water all leaks out before
the traveller needs to resort to this reserve can; the traveller
dies of thirst.5
* I touched on this problem in 'Causes and Conditions', American Philosophical
Quarterly, ii (1965), 1245-64, csp. 250-2. Examples (ii) and (iv) are borrowed from
 CONDITIONAL ANALYSES 45
In each of these five cases, we cannot say of either of the
candidates for the role of cause that it was necessary in the
circumstances for the effect. If either of the two bullets had not
been fired, the man would still have died. If the lightning had
not struck, the straw would still have caught fire; and equally
if the tramp had not thrown the butt away. If Smith and Jones
had suddenly abandoned a life of crime, the same crime would
still have been committed by the other members of the gang;
but equally if those other members had not been ready to act,
the crime would still have been committed, as it was, by Smith
and Jones. If the heart attack had not occurred, the stroke
would have carried the man off at 5.0 p.m.; but if the condi-
tions for the stroke had been absent, the heart attack would still
have killed him as and when it did. If the can had not been
punctured, the traveller would have died of poison, perhaps
even sooner than he actually died of thirst; but if it had not
been poisoned, he would have died just as and when he did.
But though they have this common feature, these examples
fall into two groups. In (iii) and (iv) we have no hesitation in
making specific causal statements, that the heart attack caused
the man's death, and that the actions of Smith and Jones
brought about the criminal result. In (v) also it seems clear to
me that the puncturing of the can caused the traveller's death,
but as Hart and Honord say that 'it is impossible to give a
satisfactory answer to this question in terms of either B or C or
both causing A's death', and that: 'their mutual frustration of
each other's plan precludes us from saying that either caused
-4's death', I need to explain and defend my view of this case.
Where we have no hesitation in making causal statements we
can tell some more detailed causal story: we can say how the
heart attack caused the man's death, how Smith and Jones
committed the crime. But the rival story about the alternative
or reserve cause cannot be completed. The conditions for the
stroke did not actually lead to a stroke, and since there was no
stroke, no stroke led to death. The reserve members of the
K, Marc-Wogau, 'On Historical Explanation*, Theoria, xxviii (1962), a 13-33) *ke
latter coming originally from P. Gardiner, The Nature of Historical Explanation,
p. 101. Example (iit) is quoted from M, Scriven, review of E. Nagel, The Structure
of Science, in Review of Metaphysics (1964), Example (v) is based on a modification
by Hart and Honore (Causation in the Law, pp. 219-20) of a hypothetical case
devised byj. A. McLaughlin, Harvard Law Review, xxxix (1925-6), 149, 155 n. 25.
46 CONDITIONAL ANALYSES
gang remained in reserve. What we accept as causing each
result, though not necessary in the circumstances for that result
described in some broad way, was necessary in the circumstances
for the result as it came about.
This matter can be thoroughly clarified if we introduce here
a distinction which will be discussed more fully in Chapter 10,
between facts and events both as causes and as results or effects.
In (v) the puncturing of the can brought it about that the
traveller died of thirst, that is, it caused his dying of thirst
(though it prevented his dying of poison). But we cannot say
that the puncturing of the can brought it about that he died,
or caused his dying—since he would have died anyway, if it
had not been punctured. That he died, and that he died of thirst,
are distinguishable facts, and hence distinguishable results. So,
as long as we are dealing with fact-results, it is not surprising
that the puncturing of the can should have brought about the
second of these but not the first. But if we think of an effect as
a concrete event, then the event which was the traveller's death
was also his death from thirst, and we must say that the punc-
turing of the can caused it, while the poisoning did not. For a
concrete event effect, we require a cause, or causal chain, that
leads to it, and it is the chain puncturing-lack-of-water-thirst-
death that was realized, whereas the rival chain that starts with
poison-in-can was not completed. In this way I would defend
my judgement that the traveller's death—that is, the concrete
event—was caused by the puncturing of the can, but at the
same time explain the doubts felt by lawyers (and others) about
this case as due to the equally correct judgement that his dying
—that is, the fact that he died on this journey—was not caused
by the puncturing of the can.
These distinctions vindicate our general thesis that (X caused
T' entails 'X was necessary in the circumstances for T': appar-
ently conflicting answers arise from some uncertainty or equi-
vocation about what is to count as T, whether it is a concrete
event or a fact, and if so, which fact. If T is the fact that the
traveller died of thirst, then the puncturing of the can both
caused and was necessary in the circumstances for T. If T is
the fact that the traveller died on this journey, then the punc-
turing of the can neither caused T nor was necessary in the
circumstances for T. If T is the traveller's death qua concrete
 CONDITIONAL ANALYSES 47
event, which was, among other things, a death from thirst, the
puncturing of the can both caused and was necessary in the
circumstances for T.
In the dubious cases of alternative over-determination, such
as (i) and (ii), it is natural to reject such statements as 'This
bullet caused his death' and 'The lightning caused the fire'. In
these cases even a detailed causal story fails to discriminate
between the rival candidates for the role of cause, we cannot
say that one rather than the other was necessary in the cir-
cumstances even for the effect as it came about. Even if we take
the effect as a concrete event all that was clearly necessary for
it in the circumstances was in (i) the volley and in (ii) the
lightning/cigarette-butt cluster of events, and it is such clusters
that we can confidently take as causing these effects. 'But which
item in the cluster really caused (or "brought about") this
effect (or "result")?' is a sensible question in so far as it asks
for a discrimination between the alternative 'causes' by way of
the filling in of a more detailed account. But if no more detailed
correct account would provide the desired discrimination, this
question has no answer.
For instance, if, as m an example mentioned earlier, a match
is struck and touched with a red-hot poker at the same time,
and a flame appears, we can say that the striking and the
touching together caused the flame to appear, since if neither
of them had occurred in the circumstances no flame would have
appeared, but if the match was affected in both these ways
simultaneously, we cannot say that either by itself caused this.
Our ruling in these cases agrees with what we said about slot-
machine M, where a shilling was put in and a bar of chocolate
came out, but the bar might have come out even if no shilling
had been put in: if on this particular occasion a bar of chocolate
would (not merely might] have come out anyway, then the
insertion of the shilling did not cause its appearance. This is
a strange case of causal over-determination, because one of the
rival factors is a null one: the bar would, on the present hypo-
thesis, have come out spontaneously. But even a null alternative
in a case of alternative over-determination is enough to prevent
the t)ther alternative from causing the result.
So far, then, we can defend the conclusion reached earlier:
the statement that X caused T entails that X was necessary in
48 CONDITIONAL ANALYSES
the circumstances for T, but it does not entail that X was in the
strong sense sufficient in the circumstances for T. But the latter
often holds as well.
Something very like this is the main negative thesis of Pro-
fessor Anscombe's Inaugural Lecture: contrary to a long
dominant philosophical tradition, causes need not be 'sufficient'
for their effects, need not 'necessitate' them. Raymond Martin
also has defended a 'necessity thesis' while criticizing a 'suffi-
ciency thesis' which he finds in an earlier work of mine.6
But is there anything still to be said for the sufficiency thesis?
A corollary of its rejection is that causal consequence is not
formally analogous to logical consequence, legal consequence,
and so on. If Q, is a logical consequence of P, then P is logically
sufficient for Q^. If, given R, Q_ is a logical consequence of P,
then, given R, P is logically sufficient for Q. Similarly if Q is a
legal consequence of P, then P is, in the circumstances, legally
sufficient for 0.-*-*^ And this holds in O
general for other sorts of
consequences. But by our present account., Q may be a causal
consequence of P without P being causally sufficient for Q even
in the circumstances. This is strange, and yet it seems correct.
There being a radon atom here now is a causal consequence of
there having been a radium atom here a little earlier, but if
radioactive decay is a non-deterministic process there are no
circumstances such that in them the earlier presence of the
radium atom was sufficient (in the strong, counterfactual sense)
for the radon atom's being here now: the former might not
have decayed, and even if it had the latter might also have
decayed already. Similarly if Jim, who is colour-blind, and his
wife Alice, who carries no gene for colour-blindness, have a
daughter, Jane, who has a son, Tom, and Tom is colour-blind,
then Tom's being colour-blind is a causal consequence of Jim's
being colour-blind; but the latter gave only a fifty-fifty chance
that any son of Jane would be colour-blind.
On the other hand, if P caused (K we can surely say that Q,
occurred because P did; the latter is practically equivalent to
'Since P occurred, (^occurred5; a since-statement can be fairly
6
G. E. M. Anscombe, Causality and Determination; Raymond Martin, 'The
Sufficiency Thesis*, Philosophical Studies, xxiii (1975s), 305-11; Martin, who criticizes
my 'Causes and Conditions', American Philosophical Quarterly, ii (1965), 245-64,
sent me an earlier version of his paper in 1969.
 CONDITIONAL ANALYSES 49
aptly described, in Goodman's terms, as a factual conditional,
that is, as adding to the open non-material conditional 'If P
occurred, (^occurred' the presupposition that P did occur (and
hence that Q,did so too). Putting these steps together, we seem
compelled to take 'P caused Q,' as entailing the non-material
conditional 'If P occurred, Q, occurred', which would commit
its user to the counterfactual 'If Q,had not been going to occur,
P would not have occurred*, that is to the strong counterfactual
sufficiency of P for Q,.
Another argument on the same side is that a future causal
statement SP will cause Q or (P would cause (£ seems to claim
that P is sufficient in the circumstances for Q,, that if P occurs,
(£, will occur also, and that if Q, is not going to occur, neither
is P. Even if we were right in saying that a past tense singular
causal statement entails only that the cause was necessary, not
also sufficient, in the circumstances for the effect, this may be
characteristic of past tense uses rather than of causal statements
in general.
It seems, then, that the analysis of a tenseless singular causal
statement 'P causes (£ should include both the non-material
conditionals 'If P occurs, Q, occurs' and 'If P does not occur,
Q,does not occur', and therefore also their contrapositives, each
of which will change into a counterfactual if the speaker takes
its antecedent not to be fulfilled. But different parts of this con-
cept will be stressed in different settings. When P and Q, are
both known to have occurred, it is natural to lay stress on the
counterfactual form of the second conditional; but if their
occurrence is problematic, it is natural to emphasize the open
form of the first conditional, though the second is also in force.
The general notion of a cause is of something which is both
necessary and sufficient in the circumstances for its effect, but
where the cause and the effect have both actually occurred we
do not require that the cause should be sufficient in the strong
counterfactual sense.
Some writers who have recognized that causes need not be
sufficient for their effects have drawn the further conclusion that
our ordinary concept is primarily probabilistic. Thus Patrick
Suppes7 argues that 'the everyday concept of causality is not
sharply deterministic' from the evidence that we may say 'His
7
A Probabilistic Theory of Causality, p. 7.
50 CONDITIONAL ANALYSES
reckless driving is bound to lead to an accident', where ' "lead
to" conveys the causal relation between the reckless driving
and the predicted accident5 and yet ' "is bound to" means
[only] that the probability is high'. But this is a confusion.
Saying that A is likely to cause B does not put likelihood into
the causing itself: it could (though it does not) mean that A is
likely to necessitate Bt and I think it does mean that A is likely
to be, in some particular case, necessary in the circumstances
for B, What will fulfil this probabilistic prediction is an actual
crash for which the man's reckless driving was necessary in the
circumstances. Similarly Suppes is right in claiming that a
mother who says that her child is afraid of thunder does not
mean that a state of fright ensues whenever the child hears
thunder, but rather that there is a fairly high probability of
this; but when the child is actually frightened by thunder, the
thunder is necessary in the circumstances for its state of fright.
Thus it may often happen that the only fully explicit causal
generalization that a speaker is prepared to make is a prob-
abilistic one, and yet he may still be taking it that the causal
relation in any particular case that fulfils it is one of necessity,
and perhaps also sufficiency, in the circumstances. To go
straight from this evidence to a probabilistic theory is a mistake,
and one that results from holding on to the Humean doctrine
that causation is essentially general while admitting that causal
claims need not involve universal generalizations. Against this
I shall argue (in Chapter 3) for the primacy of singular as
opposed to general causal statements.
These points seriously undermine Suppes's project of analysing
causation as a whole in terms of probabilities. We can admit,
none the less, that there may be probabilistic or statistical laws,
and singular sequences of events which in some sense fulfil
them, and that both of these can be called causal in a broad
sense. These will be discussed in Chapter 9, with particular
reference to problems about the form and content of statistical
laws and the interpretation of the 'probabilities' associated with
them—problems which Suppes does not bring out or resolve.
For the present, however, we can affirm that a cause is ordinarily
taken to be necessary in the circumstances for its effect, and
perhaps sufficient in the circumstances as well.
We cannot, however, conclude that necessity-in-the-circum-
 CONDITIONAL ANALYSES 51
stances is the distinguishing feature of a cause. As we saw, one
collateral effect may be necessary in the circumstances for an-
other, but it is not said to cause the other. Also, if X caused Tt
we have seen that X may be in the strong sense sufficient in the
circumstances for T. If it is, then T will be necessary in the
circumstances for X\ but we shall not say that T caused X,
Such counter-examples show that another relation, which we
may call causal priority, is thought of as being required, along
with necessity-in-the-circumstances, for causing. Of a pair of
collateral effects, neither is causally prior to the other, and in
our other example, since X caused T, X was causally prior to T,
and T was therefore not causally prior to X. But it seems that if
any X is both necessary in the circumstances for and causally
prior to T, we shall say that X caused T't also, wherever we are
prepared to say that X caused T we are prepared to say that
X was necessary in the circumstances for and causally prior to
T. The distinguishing feature of causal sequence is the con-
junction of necessity-in-the-circumstances with causal priority.
But this conclusion is rather hollow until we have said more
about causal priority. A full discussion of this notion, and of the
whole problem of the direction of causation, will be under-
taken in Chapter 7; but a preliminary account will be in place
here.
The core of the notion of causal priority is that the world
has some way of running on from one change to another. But
we can speak with any accuracy only when we associate causal
priority with some kind of sufficiency or necessity. Thus X is
sufficient in the circumstances in the weak sense for and causally
prior to T provided that if X is put into the world in the cir-
cumstances referred to and the world runs on from there, T
will occur. Similarly X is necessary in the circumstances for and
causally prior to 1"provided that if X were kept out of the world
in the circumstances referred to and the world ran on from
there, T would not occur. And X is sufficient in the circum-
stances in the strong sense for and causally prior to T provided
that if X were put into the world in the circumstances referred
to and the world ran on from there, T would occur, Sufficiency-
in-the-weak-sense plus causal priority is shown by an experiment
on the actual world; but necessity and sufficiency-in-the-strong-
sense plus causal priority involve counterfactual claims, and
52 CONDITIONAL ANALYSES
therefore involve assertions about how the world would have
run on if something different had been done: they involve
thought about the independent running of a merely possible
world.
We have already illustrated this in suggesting a way out of
the difficulty about collateral effects. We thought of construct-
ing a possible world which was just like the actual world up to
the point where, in the actual world, the proposed antecedent
occurs, of excluding this antecedent from the possible world and
then letting that world run on: if and only if the result did not
occur in that possible world were we prepared to say that the
antecedent was in the special sense necessary in the circumstances
for the result. That Special sense' of 'necessary in the circum-
stances' is what we now recognize as the conjunction 'necessary
in the circumstances for and causally prior to'.
But this account needs some adjustment to make it generally
applicable. Our proposal to make the possible world just like
the actual one up to the point of time at which the proposed
antecedent X occurs in the actual world would have the un-
desired result of making temporal priority necessary, though
not sufficient, for causal priority. For if T had occurred before
X in the actual world, it would already have been included,
by the above proposal, in the possible world: we should have
left no chance that the exclusion of X from the possible world
might, as that world ran on, exclude T also. But temporal
priority is not conceptually necessary for causal priority. We
can coherently consider the possibility of backwards or time-
reversed causation. To leave room for this, we must modify the
suggested account. We must think of somehow excluding an X
from a possible world at a time J3 (where X occurred in the
actual world at £3) without prejudice to the question whether
a Y occurred in that possible world (as it did in the actual one)
at an earlier time £a. But to combine this with the notion of
letting the possible world run on, we must think of excluding X
indirectly, by excluding some other possible event W which
actually occurred at a still earlier time ^ and which is believed
to be necessary in the circumstances for X. Our possible world
is to have the same circumstances as the actual world, but to
diverge from it by the exclusion of W at ^; from there on it is
allowed to run on by whatever laws of working the actual
 CONDITIONAL ANALYSES 53
world obeys. This leaves open the possibility that there should
be among these laws a backward causal one such that this
indirect exclusion of X would carry with it the non-occurrence
of T. If T does occur in our possible world at £2» then X is not
necessary in the circumstance;; for and causally prior to it; but
if T does not occur in our possible world at tz, X may be so,
though there are other possibilities which it is not easy to
exclude. (These details are discussed in Chapter 7.)
The notion of necessity-in-the-circumstances-plus-causal-
priority, then, involves the thought of what would happen when
a possible world, constructed by some appropriate alteration
from the actual world, was allowed to run on. It presupposes
that the actual world has some laws of working—not neces-
sarily strictly deterministic ones—which can be carried over to
the possible world. But it does not require that in order to use
this notion we should know what those laws of working are.
I have explicitly added the notion of causal priority to that
of being necessary in the circumstances. But I must admit that
the terms 'necessary' and 'sufficient' are often so used as to
include a suggestion of causal priority; even where a cause is
sufficient in the circumstances in the strong sense for its effect
we find it strange to say that the effect is necessary in the
circumstances for the cause, just because the effect is not
causally prior to the cause. However, there are senses of 'neces-
sary* and 'sufficient' which do not include causal priority and
which are exhausted by the appropriate conditional analyses,
arid I think it makes things clearer if we use 'necessary' and
'sufficient' in these simpler senses, and introduce causal priority
as a further element.
In so far as we have used conditionals, especially counter-
factual ones, in our analysis of the concept of causation, that
analysis is incomplete until these conditionals have themselves
been explained. A view of conditionals has probably been
implicit in what I have already said', to make it explicit I want
simply to take over conclusions that I have reached elsewhere.8
A non-material conditional statement introduces a supposition
(the antecedent) and asserts something (the consequent) within
the scope of that supposition. The conditional 'If P, Q' can be
paraphrased by 'Suppose that P\ on that supposition, £?,'; or
8
In Chapter 3 of Truth, Probability, and Paradox.
54 CONDITIONAL ANALYSES
again by 'In the possible situation that P, dalso3. This account
holds for all non-material conditionals; a counterfactual adds
to this the suggestion that the antecedent does not hold in the
actual world. If so, there is liable to be a certain arbitrariness
in the choice of 'the possible situation that P', or in other words
of the particular possible world of which the consequent is being
asserted. It is a consequence of this analysis that only some
conditionals are capable of being true or false. A conditional
whose antecedent is fulfilled will be true if its consequent is true
and false if its consequent is false. But if the antecedent of a
counterfactual is, as its user believes, unfulfilled, it cannot be
true or false in this way. A counterfactual whose antecedent
entails its consequent can be allowed to count as true. But most
counterfactuals cannot achieve truth in this way either, and
will be neither true nor false, though they may be acceptable
or unacceptable, well or poorly supported, and so on.
We are led, then, towards two conclusions which are similar
to Hume's. Statements of singular causal sequence involve in
their analysis counterfactual conditionals which on the present
showing are not capable of being true; so the singular causal
statements cannot be true either.9 (They can still be false in
rather obvious ways: 'X caused T will be false, for example,
if either X or T did not occur.) Also, whether they can be true
or not, these counterfactual conditionals describe possible situa-
tions or possible worlds, they are concerned with suppositions
and their consequences or accompaniments, they do not
describe what actually occurred, let alone what was observed
to occur, in the actual individual sequence. They state what
would have happened, not what did happen. Necessity!, then,
the distinguishing feature of causal sequences, is not something
that can be observed in any of those sequences. The case for
saying this is even stronger than Hume's case for saying the
same about rtecessitya. Consequently a question analogous to
one of Hume's arises here: how do we acquire the idea of
necessity!, since we do not derive it from anything we observe
in the individual causal sequences to which we apply it? It
looks as if a psychological account is called for; as I said at the
» A similar conclusion is reached by A. J. Ayer, Probability and Evidence, pp. 132-9.
But for me this is only a tentative conclusion, which will be qualified at the end
of Chapter 8.
 CONDITIONAL ANALYSES 55
end of Chapter i, Hume's resort to psychology was not wrong
in principle but merely premature.
The item for which we must seek a psychological explanation
is not, as Hume thought, the idea of a support for a priori.
inferences from one distinct existence to another, but the sort
of thinking that is expressed by the counterfactual conditional,
c
lf in these circumstances X had not occurred, ¥ would not
have occurred either', coupled with the notion of X's causal
priority to 2", the thought that the actual world ran on from
X to T, and that the appropriate possible world would run on
from the exclusion of X to the non-occurrence of T. Let us take
the counterfactual first, leaving aside the priority. It can be
paraphrased as 'Suppose that in these circumstances X did not
occur; then T did not occur either' or 'Given these circum-
stances, in the possible situation that X did not occur, T did
not occur'. For the general capacity and tendency to make
suppositions, to consider possibilities that appear not to be
realized, I can offer no explanation beyond the obvious evolu-
tionary one; this capacity and tendency are of some value to
their possessors and could have been fostered by the ordinary
processes of natural selection. But it is worth stressing here that
although we express this kind of thinking in words, its develop-
ment need not have waited for the development of language.
It is not essentially tied to verbal expression, and there is some
reason to suppose that other, non-verbal, animals share this
capacity with us. Possibilities can be literally envisaged rather
than described.10 But, given that we have this general tendency,
can we explain the particular ways in which we employ it?
That is, can we explain how, having supposed X not to have
occurred in these circumstances, we decide what else to assert
within the scope of this supposition, or how to fill out our
picture of this merely possible situation? We can distinguish a
sophisticated and a primitive way of doing this. The sophistic-
ated way uses general propositions which we take to be con-
firmed by observations of the actual world, but which we feel
justified in extending beyond the instances in which they are
confirmed not only to other actual instances but to merely
possible ones which are related to the confirming instances in
10
Cf. Truth, Probability, and Paradox, p. 100.
56 CONDITIONAL ANALYSES
the same way that other actual instances would be.11 These are
combined with the supposition, and consequences drawn by
deductive reasoning from the combination are asserted within
the scope of the supposition. This sophisticated way of develop-
ing a supposition anticipates a well-known procedure of
'natural deduction'. The primitive way of doing the same job
relies not on this combination of inductive and deductive
reasoning, but on imagination and analogy. I have observed
another situation very like the present one, in which (unlike
the present one) no event of the X type occurred. I borrow
features from that other situation to fill out my imaginative
picture of the possible situation that in the present circum-
stances X did not occur. In particular, if no event of the T type
occured in that other situation, I see the possible situation that
X did not occur in the present circumstances as continuing
without the occurrence of T. The sort of observation that can
be immediately used in this imaginative analogizing is that pre-
scribed for Mill's Method of Difference: two otherwise very
similar instances, in one of which both the 'cause' and the
'effect' occur, and in the other of which neither of these occurs.
Such a pair of instances may be found in a number of different
ways, but the most obvious, and presumably the most fruitful
original source of the causal concept, is provided by the 'before
and after' observation. In an otherwise apparently static situa-
tion, one striking change (X) occurs, followed shortly afterwards
by another (T}. The situation before X occurred, when equally
T did not occur, provides the control case or negative instance,
while the later situation, in which both X and T occurred, pro-
vides the experimental case or positive instance. And in a quite
primitive and unsophisticated way we can transfer the non-
occurrence of Y from the before situation to a supposed later
situation in which, similarly, X did not occur, and form the
thought which is expressed by the statement 'If X had not
occurred, T would not have occurred', or, in other words,
(
X was necessary in the circumstances for Y'.
But this is only one part of our ordinary idea of necessary
connection. We have also to account for the other element, the
causal priority. It seems undeniable that this notion arises from
our experience of our own active interventions in the world. If
" Gf. Truth, Probability, and Paradox, pp. 117-18, and Chapter 8 below.
 CONDITIONAL ANALYSES 57
I introduce the change X into an apparently static situation and
then the other change Y occurs, I not only see X as in the weak
sense sufficient in the circumstances for Y, but also see X as
causally prior to T. I see the actual world as running on from
my introduction of X to the occurrence of T. If, further, I think
(for the reasons just outlined) of X as necessary in the circum-
stances for T, I see the possible world in which I do not introduce
X as running on without T's occurring. I seem then to have
introduced into the actual world not only X> but the whole
X-Y sequence; but I seem to have done so by operating
on the X end of it. The notion of the actual world's running on
from—not to—X, and of a possible world's similarly running on
from not-^, is derived from this complex experience of inter-
vening and then waiting for results. Rightly or wrongly, I see
myself as a free agent in introducing X, and therefore rule out
the possibility that anything else in the objective situation is
causally prior to X, that the world ran on to A'; I therefore
rule out also both the possibility that X and T should be col-
lateral effects of some common cause and the possibility that
T should have caused X, even if I see T, as I well may, as being
necessary in the circumstances for X.
We can, then, find natural and presumably instinctive ways
of thinking which would be expressed by the counterfactual
conditionals and assertions of causal priority which we have
given as the analysis of necessity^ of the distinguishing feature
of causal sequence. But just as our analysis of necessity! differs
from Hume's equation of it with necessity^, so our account of
the psychological mechanism involved is different. The key item
is a picture of what would have happened if things had been
otherwise, and this is borrowed from some experience where
things were otherwise. It is a contrast case rather than the
repetition of like instances that contributes most to our primi-
tive concept of causation.
My main concern in this chapter has been to analyse our
actual concept of causing, to formulate what we commonly take
to be the distinguishing mark of causal sequences, and then,
since this has turned out not to be an observed feature of those
sequences, to sketch an explanation of our thinking in this way.
In this analysis the major item is that a cause is thought to be
both necessary and sufficient in the circumstances for its effect,
58 CONDITIONAL ANALYSES
but that the sufficiency is less firmly required than the necessity,
particularly where the sequence is known to have occurred.
But whatever our actual concept may be, it is obvious that we
can construct causal concepts as we wish, which may be useful
for particular purposes. We can speak, if we like, of necessary
causes, and make it explicit that sufficiency is not required. Or
we can speak of sufficient causes, and make it explicit that neces-
sity is not required. Or we can make it plain that we are
requiring both necessity and sufficiency in the circumstances.
But the general pattern on which such concepts are constructed
is that we combine the appropriate non-material conditional (s)
with the notion of causal priority. But all such causal concepts
are somewhat superficial: there may well be further backing
for these conditional descriptions, further relations in the objects
that encourage us to speak and think in these ways.
 3
Causal Regularities1

O N E of Hume's legacies is the regularity theory of causation.

In his definitions, which aim at reform rather than analysis of
our ordinary concepts, he equates causation as it really exists
in the objects with regular succession. Such regularity is, as we
saw in Chapter i, well fitted to play the role of necessity3, that
is, of something that would license causal inference but not
a priori, not from a knowledge of the individual cause or effect
on its own.
It was argued in Chapter 2 that an initial analysis, at least,
of our ordinary causal concepts could be given in terms of
certain conditional statements,, especially certain counter-
factuals; in asserting singular causal sequences we are talking,
in part, not only about what has occurred but about what
would or would not have occurred had things been different,
1
The ideas underlying this chapter were originally presented in 'Causes and
Conditions', American Philosophical Quarterly, ii (1965), 245-64. I have tried to take
account of a number of discussions of this article, both published and unpublished,
notably the criticisms made by Jaegwon Kim in 'Causes and Events: Mackie on
Causation", Journal of Philosophy, Ixviii (1971), 426-41, In particular, I have tried
to clarify my account by maintaining, as I did not in the original article, a firm
distinction between types of events {'generic events', 'properties') and the indi-
vidual events that instantiate them, more or less as recommended in Part IV of
Kim's article.
The account of eliminative induction used in this chapter was first presented in
my article 'Mill's Methods of Induction', in Encyclopedia of Philosophy, ed. Paul
Edwards, vol. 5, pp. 324-32. A fuller statement of it is given in the Appendix.
The ultimate credit for much of what is said here must go to J. S. Mill, but in
recent years a number of philosophers have, independently of one another and
using different approaches, produced improved accounts which are essentially
alike even in respects in which they could not be derived from Mill. I have noticed
particularly those of Konrad Marc-Wogau, 'On Historical Explanation1, Theoria,
xxviii (1962), 213-33, and of Michael Scriven, in a review of Nagel's The Structure
of Science, Review of Metaphysics, xvii (1964), 403-24, and in 'The Logic of Cause*,
Theory and Decision, ii (1971), 49-66, but there are, I believe, more than a few
others. Such a convergence of independent approaches suggests that we may be
getting near to the truth.
60 CAUSAL REGULARITIES
that is, about some merely possible situations and events. We
have, I suggested, two ways of doing this, a primitive one and
a sophisticated one. The primitive one relies on imagination
and analogy, but the sophisticated one uses general proposi-
tions, which sustain the counterfactual conditionals. If we have
inductive reasons—reasons that carry us beyond the support-
ing observations—for believing that all situations of a certain
kind develop in a certain way, we find plausible the counter-
factual conditional statement that if this situation had been of
that kind it too would have developed in that way. Regularity
statements, if inductively supported, will sustain the condi-
tionals which an initial analysis of causal statements brings to
light. The meaning of causal statements is given by the condi-
tionals, but their grounds may well include the corresponding
regularities.
It is therefore appropriate to inquire how far a regularity
theory will go as an account of causation as it exists in the
objects. Regularity has at least the merit that it involves no
mysteries, no occult properties like necessitya. It is true that an
unqualified regularity, holding for unobserved as well as
observed instances, obviously cannot be observed: but we can
say quite explicitly what it would be. And there is no need to
introduce the mystery of a special sort of regularity, a 'nomic
universal', to account for the ability of causal laws to sustain
counterfactual conditionals.2 Yet there is some obscurity in the
notion of regular succession. Hume's account, as we noted, is
careless and imprecise: if the regularity theory is to be given a
fair trial we must begin by describing more accurately and in
more detail the forms of regularity which might count as the
whole or as part of causation in the objects.
Mill's account is a great improvement upon Hume's: he
explicitly recognizes a number of important complications. 'It is
seldom, if ever, between a consequent and a single antecedent
that this invariable sequence subsists. It is usually between a
consequent and the sum of several antecedents; the concurrence
of all of them being requisite to produce, that is, to be certain
of being followed by, the consequent.'3 We may put this more
* I have argued this point in Truth) Probability, and Paradox, Chapter 3, pp. 114-
19, but it is considered further in Chapter 8 below.
3 System of Logic, Book III, Ch. 5, Sect. 3.
 CAUSAL REGULARITIES 61
formally. There are certain factors—that is, types of event or
situation—which we can symbolize as A,B, C, etc.,, and the effect
(Mill's 'phenomenon') P occurs whenever some conjunction of
factors occurs—say the conjunction of A and B and C, which we
shall symbolize as ABC—but not when only some of these con-
juncts are present. All ABC are followed by P, but it is not the
case that all AB are followed by /*, and so on. (The references
to 'sequence', to 'following', of course mean that the 'consequent'
occurs, in each individual instance of the causal sequence, fairly
soon after all the 'antecedents' are assembled, and in the appro-
priate spatial region. There is some looseness in these notions,
which may reflect a real inadequacy in the regularity theory,
but let us postpone this objection and assume that we under-
stand well enough what counts as sequence.) Mill also points
out that there can be what he calls a plurality of causes.
It is not true that one effect must be connected with only one cause,
or assemblage of conditions; that each phenomenon can be pro-
duced only in one way. There are often several independent modes
in which the same phenomenon could have originated. One fact
may be the consequent in several invariable sequences; it may
follow, with equal uniformity, any one of several antecedents, or
collections of antecedents. Many causes may produce mechanical
motion: many causes may produce some kinds of sensation: many
causes may produce death. A given effect may really be produced
by a certain cause, and yet be perfectly capable of being produced
without it.4
In our symbolism, this means that we may have, say, not only
'All ABC are followed by /» but also 'All DGH are followed by
P'. Now the conjunction of these two propositions is equivalent
to 'All (ABC or DGH) are followed by P\ A plurality of causes
is tantamount to a disjunctive antecedent, as an assemblage of
conditions is to a conjunctive one: allowing for both, we have
a disjunction of conjunctions. Now suppose that there is a
finite set of assemblages of conditions that produce J°, say ABC,
DGH, and JKL. It may well be that P occurs only when at least
one of these conjunctions has occurred soon before in the right
region. If so, all P are preceded by (ABC or DGH or JKL}.
(There is, of course, no logical necessity that this should be so.
Events might occur in a disorderly way: P might sometimes
* Ibid., Book HI, Ch, 10, Sect, i.
62 CAUSAL REGULARITIES
occur without there having occurred, just before in the right
region, any assemblage of conditions which is always followed
by P. But at present I am considering cases where this is not
so. That is, we may have a pair of (roughly) converse universal
propositions, 'All (ABC or DGH or JKL) are followed by P*
and 'All P are preceded by (ABC or DGH or JKL}\
In discussing such forms of regularity, it will be convenient
to use the terms 'necessary condition' and 'sufficient condition'
in senses different from, though related to, the senses in which
these phrases were used in Chapter 2, There, a necessary condi-
tion, for example, was related to a counterfactual conditional;
i
X was a necessary condition for T' meant 'If X had not
occurred, Y would not', where 'X' and 'Y' stood for particular
events. But we are now using letters to stand for types of event
or situation, and 'X is a necessary condition for T' will mean
that whenever an event of type T occurs, an event of type X
also occurs, and 'X is a sufficient condition for T' will mean
that whenever an event of type X occurs, so does an event of
type T.
Then in the case described above the complex formula '(ABC
or DGH or JJfCL)' represents a condition which is both necessary
and sufficient for P: each conjunction, such as *ABC\ represents
a condition which is sufficient but not necessary for P. Besides,
ABC is a minimal sufficient condition: none of its conjuncts is
redundant: no part of it, such as AB, is itself sufficient for P.
But each single factor, such as A} is neither a necessary nor a
sufficient condition for P. Yet it is clearly related to P in an
important way: it is an insufficient but non-redundant part of an
unnecessary but sufficient condition: it will be convenient to call
this (using the first letters of the italicized words) an mus con-
dition.5
Mill includes in his assemblages of conditions states, that is,
standing conditions, as well as what are strictly speaking events.
He also stresses the importance of factors which we should
naturally regard as negative, for example the absence of a
sentry from his post. It may be the consumption of a certain
poison conjoined with the non-consumption of the appropriate
s
This terrrij 'inus condition', was introduced in 'Causes arid Conditions', having
been suggested by D. C, Stove; the term 'minimal sufficient condition' was used
by K. Marc-Wogau in 'On Historical Explanation', Theoria, xxviii (1963).
 CAUSAL REGULARITIES 63
antidote which is invariably followed by death. If a certain
type of event is symbolized as C, then not-C, or C, will be the
absence of any event of that type. It may be that although AB
alone is not sufficient for P—because ABC is regularly followed
by not P—ABC is sufficient for P. Mill is reluctant to call such
a negative condition as C a cause, but speaks instead of the
absence of counteracting causes; if C is needed as a conjunct
hi the minimal sufficient condition ABC, then C itself is a
counteracting cause.6
A further complication, for which Mill does not provide, is
the recognition of a causal field in the sense explained in
Chapter 2. The 'antecedents' and 'consequents' will not, in
general, be events that float about on their own, they will be
things that happen to or in some subject or setting. In discuss-
ing the causes of death, for example, we may well be concerned
with the dying of human beings who have been living in an
ordinary environment. If so, human beings and this environ-
ment together constitute the field. We shall not then regard
the facts that these are human beings, or that they are in this
ordinary environment, as causal factors: these will not figure as
conjuncts in such a condition as ABC. The causal field in this
sense is not itself even part of a cause, but is rather a back-
ground against which the causing goes on. If we sum up such
a field as F, and allow for the various points made by Mill,
we arrive at the following typical form for a causal regularity:
In F, all (ABC or DGH or JKL) are followed by P, and, in F,
all P are preceded by ABC or DGH or JKL).
For some purposes this may be simplified to
All F (ABC or DGH or JKL) are P and all FP are (ABC or
DGff or JKL).
That is, some disjunction of conjunctions of factors, some of
which may be negative, is both necessary and sufficient for the
effect in the field in question. But what then is the cause? Mill
says that 'The cause . . . philosophically speaking, is the sum
total of the conditions positive and negative',7 in other words
suc*h a conjunction as ABC; but if we go as far as this there is
no good reason why we should not go further and equate *the
« System of Logic, Book III, Gh. 5, Sect. 3. ? Ibid.
64 CAUSAL REGULARITIES
cause, philosophically speaking* rather with the complete dis-
junction of conjunctions, such as (ABC or DGff or JKL]. It is
this that is both necessary and sufficient for the effect (in the
field) as causes have often been assumed to be. And when
speaking from a regularity point of view, it will be convenient
to call this the full cause. But what is ordinarily called a cause,
or what is referred to by the subject of a causal verb, is practic-
ally never anything like this; rather, in general causal state-
ments, like 'The consumption of more than such-and-such a
dose of aspirin causes death', the cause is a factor such as is
represented, say, by M' in our formula, and in singular causal
statements like 'Taking this dose of aspirin caused his death'
the cause is an instance of some such type of event as is repre-
sented by 'A\ That is, what is typically called a cause is an
inus condition or an individual instance of an inus condition,
and it may be a state rather than an event,
A regularity theory that is to have any chance of being
defended as even a partial description of causation in the objects
must deal in regularities of this complex sort. And such a theory
has considerable merits. It seems quite clear that there are
many regularities of succession of this sort, and that progress
in causal knowledge consists partly in arriving gradually at
fuller formulations of such laws. Also, even these complex
regularities could play the role of necessity3, they could license
the sorts of causal inference that Hume thought so important.
If (ABC or DGH or JKL) is both necessary and sufficient for P
in F, and this is known or believed, then if an instance of ABC
in F is observed, an inference to the conclusion that an instance
of P will follow is in order, while if an instance of P m F is
observed, and there is reason to believe that neither DGff nor
JKL (as a whole) has occurred at the right time and place, an
inference to the conclusion that ABC has occurred and therefore
that an instance of A has occurred is in order. Complex regu-
larities still license inference from cause to effect and from
effect to cause.
Moreover, they will sustain the various kinds of conditionals
that come to light in the analysis of causal concepts. First, and
most important, is the one which states that the individual
cause-event was necessary in the circumstances for the effect,
for example Tf he had not eaten of that dish, he would not have
 CAUSAL REGULARITIES 65
8
died'. Such a conditional, I have argued, can be understood
as saying 'Suppose that he did not eat of that dish; then (within
the scope of that supposition) he did not die' or 'In the possible
situation where he did not eat of that dish, he did not die*.
Now if there is some regularity of something like the form sug-
gested above—though no doubt with many more disjuncts—
giving the 'full cause5 of death of human beings in an ordinary
environment, and one of the minimal sufficient conditions, say
ABC, is the conjunction of consuming at least such-and-such
an amount of a certain poison (which was in the dish in ques-
tion, but nowhere else) with not taking the appropriate antidote
and not having one's stomach evacuated within a certain time,
and this minimal sufficient condition was realized on this occa-
sion but none of the other minimal sufficient conditions was
realized, then from the supposition that on this occasion he did
not eat of that dish it follows that A did not occur on this
occasion; from this, together with the second half of the regu-
larity, of the form 'All FP are (ABC or etc.)', and the information
that none of the other disjuncts, summed up here as 'etc.',
occurred on this occasion, it follows that P did not occur in
F on this occasion; such an inference justifies the assertion,
within the scope of the supposition that he did not eat of that
dish, that P did not occur, that is, that he did not die, and
hence sustains the conditional 'If he had not eaten of that dish
he would not have died'.
Secondly, although I argued in Chapter 2 that we do not
always require that an individual cause should be sufficient as
well as necessary in the circumstances for its effect, I had to use
rather odd indeterministic examples in order to discriminate
between necessity and sufficiency in the strong sense. Most
individual causes, it appears, both are and are taken to be
sufficient in the circumstances in the strong sense: we can say
that if in the circumstances he had not been going to die, he
would not have eaten of that dish. This counterfactual condi-
tional is sustained, analogously with the previous one, by the
first half of the regularity 'All F (ABC or etc.) are P* together
with the information that both B and C were absent on this
occasion; for these together with the supposition that he did
not die entail first that none of the disjuncts occurred on this
8
Truth, Probability, and Paradox, pp. 92-108.
66 CAUSAL REGULARITIES
occasion, then that A did not occur (since EC did), and hence
that he did not eat of that dish. So 'He did not eat of that dish*
can be asserted within the scope of the supposition that he did
not die, and this is equivalent to the required counterfactual."
Thirdly, it is commonly held that a causal claim entails that
if a sufficiently similar antecedent had occurred on other occa-
sions (when in fact it did not) a similar consequent would have
occurred. This means, in this case, that if ABC had occurred in
F on some other occasion, P would have, and it is clear that
this counterfactual too is sustained, in much the same way, by
the first half of the suggested regularity.
On the other hand, although such complex regularities may
hold in the objects, we must admit that they are seldom, if
ever, known in full. Even in a matter of such intimate and
absorbing interest as the death of human beings, we cannot
confidently assert any complete regularity of this kind. We do
not know all the causes of death, that is, all the different closely
preceding assemblages of conditions that are minimally suffi-
cient for death. And even with any one cause, we do not know
all the possible counteracting causes, all the factors the nega-
tions of which would have to be conjoined with our positive
factors to make up just one minimal sufficient condition. Causal
knowledge progresses gradually towards the formulation of such
regularities, but it hardly ever gets there. Causal regularities as
known are typically incomplete; they have rather the form
All F (A . . . B . . . or D. . . R., . or . . .) are P, and all FP are
(A . . . n . . . or D . . . ff . . . or . . .)
where the dots indicate further as yet unknown conjuncts and
disjuncts that have still to be filled in. What we know are
certain elliptical or gappy universal propositions. We do not
know the full cause of death in human beings, but we do know,
about each of a considerable number of items, that it is an
inus condition of death, that, as we ordinarily say, it may cause
death.
The same knowledge that is expressed by such elliptical uni-
versjil propositions can be expressed, alternatively, by proposi-
tions in which second-order existential quantifications precede
the universal ones. Thus the knowledge that A is an inus
condition of P in F may be formulated thus:
 CAUSAL REGULARITIES 67
For some X and for some T all F(AX or T) are P) and all
FPzre (AX or T).
The suggestion that causal regularities as known are com-
monly of this sort may provoke two questions. Of what use is
such exiguous information? And how is knowledge of such a
curious form acquired? We can show, however—surprising
though this may be—that such knowledge can be of great value,
and that it may be discovered in just those ways in which we
do most commonly acquire causal knowledge.
First, this information still permits causal inferences (in both
directions), but makes them more tentative. Knowing that
something of the form (AX or T)—where A is known but X
and T are not—is both necessary and sufficient for P in F, we
may well have reason to believe that the X, whatever it may
be, is often present; if so, we can infer from an observed occur-
rence of A that P is fairly likely to follow. X will, of course,
include the negations of all the counteracting causes, whatever
they may be; and we may well have reason to believe that
though there may be many as yet undiscovered counteracting
causes they are absent, and their negations therefore are present,
on this occasion. There may be many undiscovered antidotes
to this poison, but this victim is unlikely to take any of them.
Again, without being able to specify T in any great detail, we
may have some reason to believe that it is likely to be absent on
this particular occasion. T, of course, may be partly known; it
may be known that some of the other disjuncts it covers are of
the forms #£, CW, etc., where B and C, say, are known, though
£ and W are not. If so, we can check that these disjuncts at
least are absent by discovering that B and C are; that is, we
can check that none of the other known causes (that is, inus
conditions) of P in F is present on this occasion. And then we
can infer, still tentatively but with some probability, from the
occurrence of P on this occasion, that A has also occurred.
Typically we infer from an effect to a cause (inus condition)
by eliminating other possible causes.
Since we can often infer, with probability, an effect from a
cause, we can similarly infer, in corresponding circumstances,
the absence of a cause from the absence of an effect. And since
we can often infer, with probability, a cause from an effect, we
can similarly infer the absence of an effect from the absence of
68 CAUSAL REGULARITIES
a cause. Consequently the gappy universal propositions, the
incompletely known complex regularities, which contribute to
such inferences will still sustain, with probability, the counter-
factual conditionals that correspond to these inferences, that is,
statements of the forms 'If A had not occurred, P would not'
and 'If P had not been going to occur, A would not'. And the
gappy universal in allowing inference from cause to effect will
equally sustain the subjunctive conditional that if this cause
occurred again in sufficiently similar circumstances, so would the
effect, while leaving it open and unknown just what such sufficient
similarity requires. Gappy universals, then, still sustain all the
types of conditionals commonly associated with causal statements.
Moreover, exiguous though it is, this information allows us to
engage, though of course with something less than complete
confidence, in the production and the prevention of effects by
way of causes. If X, whatever it may be, is often present, we
may hope, by introducing an instance of A, to bring about a P.
Equally, by eliminating a known cause, A, of P, we have done
something to prevent the occurrence of P, and of course if there
are several such known causes, say A, B, and C—that is, if the
necessary and sufficient condition has the form (AX or B% or
CW or . . .)—then we do the best we can to prevent P by pre-
venting the occurrence of A or B or C; just what X, £, and W
may be is then of no practical importance. And this, surely, is
how in many fields intelligent practice goes on. We operate
with, or on, factors about which we know only that they can,
and perhaps often do, help to produce certain results. We take
what precautions we can, but failure does not necessarily show
that a plan was ill-conceived; and there is equally such a thing
as undeserved success.
Secondly, the elliptical character of causal regularities as
known is closely connected with our characteristic methods of
discovering and establishing them: it is precisely for such gappy
statements that we can obtain fairly direct evidence from quite
modest ranges of observation. Of central importance here is
what Mill called the Method of Difference; but we can improve
on his formulation of it.
This is one of the set of methods of eliminative induction.
Like any other such method, it can be formulated in terms of
an assumption, an observation, and a conclusion which follows
 CAUSAL REGULARITIES 69
by a deductively valid argument from the assumption and the
observation together. To get any positive causal conclusion by
a process of elimination, we must assume that the result—the
'phenomenon' whose cause we are seeking—has some cause.
While we can make some progress with the weaker assumptions
that it has only a sufficient cause, or only a necessary cause, the
most significant results emerge when we assume that there is
some condition which is both necessary and sufficient for the
occurrence (shortly afterwards and in the neighbourhood) of
the result. Also, if we are to get anywhere by elimination, we
must assume a somehow restricted range of possibly relevant
causal factors, of kinds of event or situation which might in some
way help to constitute this necessary and sufficient condition.
Let us initially include in our assumption, then, a list of such
possible causes, possibly relevant factors, say A,, J5, C, D, £, etc.—
though as we shall see later, we do not in fact need a list. Even
if we had specified such a list of possibly relevant factors, it
would in most cases be quite implausible to assume that the
supposed necessary and sufficient condition we are seeking is
identical with just one of these factors on its own; we expect
causal regularities to involve both assemblages of conditions
and a plurality of causes. The plausible assumption to make,
therefore, is merely that the supposed necessary and sufficient
condition will be represented by a formula which is constructed
in some way out of some selection from the list of single terms
each of which represents a possibly relevant factor, by means of
negation, conjunction, and disjunction. However, any formula
so constructed is equivalent to some formula in disjunctive
normal form—that is, one in which negation, if it occurs, is
applied only to single terms, and conjunction, if it occurs, only
to single terms and/or negations of single terms. So we can
without loss of generality assume that the formula for the sup-
posed necessary and sufficient condition is in disjunctive normal
form, that it is at most a disjunction of conjunctions in which
each conjunct is a single term or the negation of one, that is,
a formula such as ''(ABC or Gfi or J}\ Summing this up, the
assumption that we require will have the form:

For some £, % is a necessary and sufficient condition for the

phenomenon P in the field /% that is, all FP are £ and all
70 CAUSAL REGULARITIES
F£ are P, and ^ is a condition represented by some formula
in disjunctive normal form all of whose constituents are
taken from the range of possibly relevant factors A, B, C,
Z>, E> etc.
Along with this assumption, we need an observation which
has the form of the classical difference observation described
by Mill. This we can formulate as follows:
There is an instance, 71} in which P occurs, and there is a
negative case, JV15 in which P does not occur, such that
one of the possibly relevant factors, say A, is present in /j
but absent from JVj, but each of the other possibly relevant
factors is either present in both /x and JV^ or absent both
from /t and from JV^.
An example of such an observation can be set out as follows,
with 'a' and 'p' standing for 'absent' and 'present':
P A B C D E
** P P P a a p-l etc
jVj a a p a a pj
Given the above-stated assumption, we can reason as follows
about any such observation:
Since P is absent from JVls every sufficient condition of P is
absent from JV1} and therefore every disjunct in £ is absent
from jVj. Every disjunct in £ which does not contain A
must either be present in both /j and JVj or absent from
both, since each of its constituents is either present in both
or absent from both; so every disjunct in £ which does not
contain A is absent from It as well as from JV^. But since P
is present in /j, and ^ is a necessary condition of P, /£ is
present in 1-^ Therefore at least one disjunct in £ is present
in /,_. Therefore at least one disjunct in £ contains A. And
it must contain A un-negated, if it is to be present in /j
where A is present.
What this shows is that ^, the supposed necessary and suffi-
cient condition for P in F, must have one of these forms: (A),
(AX], (A or 1"), (AX or T). That is, A is either an inus condi-
tion for P in Fy if the necessary and sufficient condition is of
the form (AX or T]—that is, if there are both other factors
 CAUSAL REGULARITIES 71
conjoined with A and other disjuncts as well as the one in which
A figures—or, as we may say, better than an inus condition, if
the necessary and sufficient condition has one of the other three
forms. Our assumption and the difference observation together
entail a regularity of the form
For some X and for some T (which may, however, be null),
all F (AX or T) are P, and all FP are (AX or T).
This analysis is so far merely formal. To justify my suggestion
that we can obtain fairly direct evidence for regularities of this
form from modest ranges of observation I must show that it is
sometimes reasonable to make the required assumptions and
that we can sometimes make the corresponding observations.
A number of points can be stated in support of this claim.
First, the actual listing of possibly relevant factors is not
needed in practice: this was only a device for formal exposition.
All that matters is that any features other than the one, A,
that is eventually established as an inus condition (or better)
should be matched as between I± and JV^, that there should be
no other likely-to-be-causally-relevant difference between these
two cases: what features they agree in having or in lacking is
irrelevant. In a causal inquiry in a field in which we already
have some knowledge we may, of course, already know what
sorts of item are likely to be relevant, and so can check that
these are matched between the two cases; but if an inquiry is
starting from cold in a field in which little is known, the only
available criterion of possible relevance is spatio-temporal
neighbourhood: we simply have to see that things are, as far
as possible, alike in the space-time regions where an instance
of A is followed by an instance of P (our j^) and where there is
neither an instance of A nor one of P (our JVX).
Secondly, there are at least two well-known ways in which
some approximation to the classical difference observation is
often achieved. One of these is the before-and-after observa-
tion. Some change, A, is introduced, either naturally or by
deliberate human action, into an otherwise apparently static
situation. The state of affairs just after this introduction is our
/!, the state of affairs before it is JVj. If this introduction is
followed, without any further intervention, by some other
change Pt then we can and almost instinctively do reason as the
72 CAUSAL R E G U L A R I T I E S
Method of Difference suggests, concluding both that this
instance of A helped to produce this instance of P, and that A
generally is at least, in Mill's terms, an indispensable part of the
cause of P. The singular causal judgement in such a case could,
as I said in Chapter 2, arise in a primitive way from imaginative
analogizing; but the corresponding general judgement requires
something like the pattern of reasoning that we have just been
trying to formalize; and of course once we have this general
causal judgement, it could in turn sustain the counterfactual
conditionals implicit in the singular judgement. The second
well-known approximation to the difference observation is the
standard controlled experiment, where what happens in the
'experimental case'—-our 7j—is compared with what happens,
or fails to happen, in a deliberately constructed 'control case'
which is made to match the experimental case in all ways
thought likely to be relevant other than the feature, A, whose
effects are under investigation.
Thirdly, it may seem to be in general difficult to satisfy the
requirement of the Method of Difference, that there should be
only one point of difference between 7t and JVX. But fortunately
very little really turns upon this. Suppose that two possibly
relevant factors, say A and B, had been present in 7X but absent
from JVj. Then reasoning parallel to that stated above would
show that at least one of the disjuncts in % either contains A
un-negated, or contains B un-negated or contains both. This
observation still serves to show that the cluster of factors (A, B]
contains something that is an inus condition (or better) of P in
F, whether in the end this turns out to be A alone, or B alone,
or both these, or the conjunction AB. Similar considerations
apply if there are more than two points of difference. However
many there are, an observation of the suggested form, coupled
with our assumption, shows that a cause—an inus condition or
better—lies somewhere within that cluster of features in which
/! differs from JVj. It does not, of course, show that the other
features, those shared by /x and JV15 are irrelevant; our reason-
ing does not, as some misleading formulations of the method
suggest, exclude factors as irrelevant, but positively locates
some at least of the relevant factors within the differentiating
cluster. This point rebuts a criticism sometimes levelled against
the eliminative methods generally, that they presuppose and
 CAUSAL REGULARITIES 73
require a finally satisfactory analysis of causal factors into their
simple components, which we never actually achieve. On the
contrary, any distinction of factors, however rough, enables us
to start using such a method. We can proceed, and there is no
doubt that discovery has often proceeded, by what we may call
the progressive localization of a cause. Using the Method of Differ-
ence in a very rough way, men discovered first that the drink-
ing of wine causes intoxication. The cluster of factors crudely
indicated by the phrase 'the drinking of wine' contains some-
where within it an inus condition of intoxication. Later, by
distinguishing various possibly relevant factors within this
cluster, and making further observations and experiments of
the same sort, they located a cause of intoxication more pre-
cisely—the consumption of the alcohol contained in the wine.
In a context in which the cluster of factors is put in or left out
as a whole it is correct to say, of any particular case, 'He would
not have become intoxicated if he had not drunk that wine*.
But in a context in which alcohol and certain other constituents
were put in or left out separately, it would be correct to say
rather 'He would not have become intoxicated if he had not
consumed that alcohol'. Did the wine make him drunk? At one
level of analysis, of course it did; but in relation to a finer
analysis of factors it was not strictly speaking the wine but the
alcohol it contained that made him drunk. In different con-
texts, different specifications of a cause or inus condition may
be equally correct.
Fourthly, it is instructive to contrast the Method of Difference
as a logical ideal with any concrete application of it. If the
assumption and the observation were known to be true, then
the conclusion, asserting a typical causal regularity, would be
established. No doubt the assumption and the observation are
never known with certainty, but it may still be reasonable to
accept them provisionally, and, if so, our formal analysis shows
that it is equally reasonable to accept the regularity conclusion.
In particular, once the assumption is accepted, we may well be
in a position to say that we cannot see any other difference that
might be relevant between I± and N^ and consequently that
we cannot see any escape from the causal conclusion.
Fifthly, we need not and in practice do not rely so heavily
on a single observation (with just one /t and one JV^) as our
74 CAUSAL R E G U L A R I T I E S
formal account might suggest. Of course there might, in such
a single pair of cases, be an unnoticed but relevant further
difference which undermined our reasoning. It might be that
our control case did not match our experimental case closely
enough, or, in a before-and-after observation, that some other
relevant change occurred at the critical time. But repeating the
experiment or the observation reduces this danger. If we can
add an /2 and an JV2, and an /3 and an jV3, and so on, and each
time the presence of A is the only noticed difference between
the two sets of antecedent conditions, it becomes progressively
less likely that any other relevant change occurred just at the
right moment on each occasion—unless, of course, this other
change is itself connected with A by what Mill calls some fact
of causation. But it is important to note that it is not the mere
repeated co-occurrence of A and P that supports the causal
conclusion; we are not moving over from the Method of Differ-
ence to the Method of Agreement; what we are relying on is
the repetition of a sequence which on each single occasion was
already, because of the Method of Difference pattern it appeared
to exhibit, prima facie a causal one. The repetition merely con-
firms this by greatly reducing the likelihood that some unnoticed
factor might be undermining this prima facie interpretation.
Sixthly, it is worth noting that the assumption required, while
it is of course a deterministic one, is much weaker than the
usual formulations of the uniformity of nature. We need not
assume that every event has a cause, but merely that for events
of the kind in question, P, in the field in question, F, there is
some, possibly very complex, necessary and sufficient condition.
It is true that we also need to assume that this condition is made
up from a range of possibly relevant factors that is restricted in
some way: if we have no previous causal knowledge in the
relevant sphere, we have to take spatio-temporal nearness as a
criterion of possible relevance. But this is not a final or absolute
assumption: if, using it as a working assumption, we reach some
conclusion, assert some causal regularity, but then this is dis-
confirmed by further observations, we can always relax this
working assumption and look a bit further afield for possibly
relevant differences. There is, no doubt, still a philosophical
problem about what justifies any such deterministic assumption,
however local and however weak. But at least this analysis
 CAUSAL REGULARITIES 75
makes it clearer what precise form, of assumption is needed to
back up our ordinary detecting and establishing of causal
regularities. In particular, I hope to have shown that while we
can agree with von Wright that 'in normal scientific practice
we have to reckon with plurality rather than singularity, and
with complexity rather than simplicity of conditions', this does
not mean, as he says, that 'the weaker form of the Deterministic
Postulate ... is practically useless as a supplementary premiss
or "presupposition" of induction'.9
Towards the end of Chapter 2 I said that although necessity^
the distinguishing feature of causal sequences, is not something
that can be observed, we can explain how a certain sort of
observation can set off a psychological process of imaginative
analogizing that can yield a singular causal judgement. We
can now add to this by explaining how a not implausibly strict
assumption, coupled with that same sort of observation, can
entail a causal regularity statement of the form we ordinarily
use, and that this statement in turn will sustain the correspond-
ing singular causal judgement.
It is a further merit of such an account of complex but in-
completely known regularities that it disposes altogether of a
type of objection that is sometimes brought against regularity
(or 'Humean') theories of causation in general. Geach, for
example, says that
. . . the laws that scientists aim at establishing are not de facto
uniformities, either necessary or contingent. For any alleged uni-
formity is defeasible by something's interfering and preventing the
effect. . . . Scientists do not try to describe natural events in terms
of what always happens. Rather, certain natural agents . . . are
brought into the description, and we are told what behaviour is
proper to this set of bodies in these circumstances. If such behaviour
is not realized, the scientist looks for a new, interfering agent. . . ,
And he goes on to argue that 'interference just cannot be logic-
ally brought into a uniformity doctrine of causality'; criticizing
some of Mill's statements, he says that Mill 'retreats into saying
that physical laws do not state what does happen, but what
would failing interference happen; but this is to abandon the
Humian position'.10
9
A Treatise on Induction and Probability, p„ 135.
10
Three Philosophers, by G. E. M. Anscombe and P. T. Geach, pp. 102-3
76 CAUSAL REGULARITIES
It will be clear from what has been said above that though
interference could not be brought into a doctrine of simple
uniformities, it is easily accommodated in a doctrine of complex
uniformities. Interference is the presence of a counteracting
cause, a factor whose negation is a conjunct in a minimal
sufficient condition (some of) whose other conjuncts are present.
The fact that scientists rightly hesitate to assert that something
always happens is explained by the point that the complex
uniformities they try to discover are nearly always incompletely
known. It would be quite consistent with an essentially Humean
position—though an advance on what Hume himself says—to
distinguish between a full complex physical law, which would
state what always does happen, and the law as so far known,
which tells us only what would, failing interference, happen;
such a subjunctive conditional will be sustained by an incom-
pletely known law. Moreover, the rival doctrine can be under-
stood only with the help of this one. What it would be for
certain behaviour to be 'proper to this set of bodies in these
circumstances', what Aquinas's tendencies or appetitus are,
remains utterly obscure in Geach's account; but using the
notion of complex regularity we can explain that A has a
tendency to produce P if there is some minimally sufficient con-
dition of P in which A is a non-redundant element. (This is,
indeed, not the only sense of the terms 'tend* and 'tendency'.
We could say that A tends to produce P not only where A con-
joined with some set of other factors is always followed by P,
but also where there is an indeterministic, statistical, law to the
effect that most, or some, instances of A, or some definite per-
centage of such instances, are followed by P, or perhaps where
an A has a certain objective chance of being followed by a P.11
These statistical tendencies are not reducible even to complex
regularities: if they occur, as contemporary science asserts, then
they constitute something different from, though related to,
strict deterministic causation. But they have little to do with
Geach's problem of interference.)
Does this improved and corrected account of causal regu-
larities, with all its merits, constitute a defence of a regularity
theory of causation? Can we identify causation with the holding
11
See Chapter 9 below. Objective chance is also discussed in Trutht Probability,
and Paradox, Chapter 5.
 CAUSAL REGULARITIES 77
of regularities of this sort? No progress can be made with this
problem unless we keep separate the three sorts of question,
what do causal statements mean, what do we know about
causation and how do we know it, and what constitutes causa-
tion as it is in the objects themselves.
It seems very clear that singular causal statements do not
mean that the sequences about which they are made are
instances of regularities of any sort. In the Humean tradition
there is a doctrine that they ought to mean this, but Hume him-
self, as we have seen, did not claim that they do so. As I have
argued in Chapter 2, the main part of our concept of the dis-
tinguishing feature of a causal sequence is expressible by a
counterfactual conditional, or, what comes to the same thing,
by the assertion that the cause was necessary in the circum-
stances for the effect, and the meaning of a singular causal
statement will be analysable into the conjunction of this with,
probably, some further claims or suggestions.
It is, however, sometimes said that causal statements are
implicitly general. It is easy to refute the claim that a singular
causal statement normally implies a simple regularity state-
ment, of the form that instances of a certain kind of event are
always, or even often, followed by instances of another kind of
event: the taking of a contraceptive pill may cause one woman's
death although millions of women have taken large numbers of
exactly similar pills and survived. Nor can a singular causal
statement imply a complex but: complete regularity statement,
since we commonly assert statements of the former kind when
we are quite unable to formulate any corresponding complete
generalizations. But perhaps singular causal statements imply
our elliptical generalizations, or the equivalent forms with
second-order existential quantifiers. It is not so easy con-
clusively to refute this suggestion, but since it is only in recent
years that a number of philosophers have approached a correct
formulation of the generalizations in question^ we must say at
least that this would be an implication of which most users of
singular causal statements can be only very vaguely aware. In
fact, I would go further and say, referring to what I called (at
the end of Chapter 2) a primitive and unsophisticated way of
arriving at counterfactuals and the associated causal judge-
ments, that a singular causal statement need not imply even
78 CAUSAL REGULARITIES
the vaguest generalization. This is true even of physical and
mechanical examples. One can judge that this (very hot) stone
was cracked by water being poured over it without being com-
mitted to any generalization, meaning only that the stone
would not in the circumstances have cracked had the water not
been poured on, and that this pouring was causally prior to the
cracking in the sense explained in Chapter 2. It is, of course,
even more obviously true of mental examples; I can judge that
Bill's warning shout made me stop short of the precipice
without generalizations of any sort being involved.
On the other hand, I have argued that there is a sophisticated
way of arriving at causal and counterfactual statements which
does involve elliptical generalizations: one can use something
like the Method of Difference to establish such a generalization,
which will then sustain the counterfactuals involved in the
singular causal judgement. But even here it is the generaliza-
tion that supports the causal statement, rather than the causal
statement that implies the generalization. Also, it is worth
stressing that the generalization here is of the elliptical universal
form, it does not say what always or normally or generally or
even often happens—for example, careful checks might make it
probable that a certain pill had caused someone's death even
if this was the very first time such a pill had had any ill effect
(though further cases of the same sort would confirm this con-
clusion), but the doctor who reached this conclusion would say
that the victim must have been unusually susceptible, that she
had some as yet unknown combination of characteristics which,
in conjunction with the consumption of the pill, would regularly
lead to death. Again, it is worth stressing that the generaliza-
tion need not be known in advance: it may be discovered and
(tentatively) established by the observation of the very sequence
of events about which the causal statement is made. Not even
the vaguest foreknowledge about what often happens is needed
to smooth the way even for a physical causal discovery. The
doctor (Sir Norman Gregg) who discovered that German
measles in pregnancy had caused eye defects in a number of
children had no previous reason for regarding this as anything
but the mildest of ailments, with no lasting effects. It is true
that previously known generalizations may contribute to a
causal conclusion, but they do so by supporting the belief that
 CAUSAL REGULARITIES 79
the other features present on this occasion would be unlikely to
have produced the observed result. If someone eats of a certain
dish, and then becomes ill, what helps to show that (something
in) this dish made him ill is the fact that everything else that
he ate and drank on the same day, and everything else that he
then did, were of kinds and in quantities that he regularly ate
and drank and did without becoming ill. Such previously
known generalizations work, indeed, in the way explained by
what Mill called the Method of Residues: in effect, they are
used to construct a negative instance, corresponding to our jVj,
instead of observing one, as in the classical form of the Method
of Difference," Even these previously known generalizations,
then, are only useful, not essential: an observed negative
instance that resembles sufficiently closely the positive one makes
them superfluous. No specific generalization, however vague,
then, needs to be known in advance in order to support the
interpretation of an observed sequence as causal: even for the
sophisticated way of handling this observation all that is required
is the assumption that what has happened is an instance of
some, probably complex, regularity, that some perhaps as yet
quite unknown and unsuspected uniformity is instantiated here.
The singular causal statement says that without A, on this
occasion (our /j), P would not have occurred; this is very often
supported by the observation that without A, on some other
similar occasion (our jVj), P did not occur. In the sophisticated
procedure, this 'did not' supports the 'would not' because it is
assumed that there is some underlying regularity of behaviour;
it is this assumption that justifies the transfer of the non-
occurrence of P from JVj to the suppositional reconstruction of
/j. In the unsophisticated procedure, the transfer is made
imaginatively, by analogy; but one could say that to be prone
to make such imaginative moves is somewhat like having an
unconscious belief that there is some underlying regularity
in the world. Even my judgement that I would not have stopped
if I had not heard Bill's shout involves a similar transfer from
what I know of my own thoughts and movements before he
shoujed to the hypothetical situation in which, a moment later,
I did not hear his shout. But it is only in these very tenuous
senses that singular causal statements, sophisticated or primitive,
12
See Appendix.
8o CASUAL REGULARITIES
physical or mental, are implicitly general, that they necessarily
assert or presuppose regularities of any sort.
But what about general causal statements? The statements
that heating a gas causes it to expand, that hammering brass
makes it brittle, and such related dispositional statements as
that strychnine is poisonous ajid that lead is malleable, can
indeed be interpreted as assertions that the cause mentioned or
indicated is an inus condition of the effect. But even here it
would be more appropriate to take the general statements as
quantified variants of the corresponding singular ones, for
example, as saying that heating a gas always or often or some-
times causes it to expand, where this 'causes' has the meaning
that 'caused' would have in a singular causal statement. How-
ever, the essential point is that singular causal statements are
prior to general ones, whereas a regularity theory of the mean-
ing of causal statements would reverse this priority.
The question whether regularities enter into what we know
about causation and into our ways of learning about it has
been answered incidentally in this discussion. The crucial and
outstanding question is to what extent such complex regularities
as we have described constitute causation as it is in the objects.
It is undeniable that we ordinarily suppose both that there
are some such regularities underlying many of the sequences
that we take to be causal and that in scientific inquiries, at
least in the physical and the biological sciences, we make pro-
gress towards fuller formulations of them. Whether these sup-
positions are correct can still be questioned. The methods so
far examined for establishing them rest upon assumptions of
uniformity which those methods cannot themselves establish:
to ask whether such success as they seem to have achieved can
be taken to have confirmed their assumptions, or whether those
assumptions can be justified or vindicated in some other way,
is to raise the fundamental philosophical problem of induction
which I cannot pursue further here; though I noted in Chapter
i (following Stove) that the reasons given by many of Hume's
successors for supposing it to be insoluble are poor ones.* But
assuming that this problem can be somehow solved or dis-
solved, that we are justified in placing some reliance upon our
ordinary methods of induction or of the confirmation of hypo-
theses, it seems very likely that there are in fact regularities of
 CAUSAL REGULARITIES 81
this complex sort. The only plausible alternative view is that
the physical world works merely by statistical laws (which we
shall consider in Chapter 9), and that these generate only
approximations to regularities of the form discussed here.
Even if we grant, provisionally, that such regularities are
involved in at least some cases of causation, we must still
question whether all cases of causation involve them, and
whether causation ever consists only in the holding of such
regularities.
The stock argument for a negative answer to this last question
is that we can point to regularities of succession that are not
causal: day regularly follows night, but night does not cause
day, and is not even a partial cause of day;13 the sounding of
factory hooters in Manchester may be regularly followed by,
but does riot cause, London workers leaving their work.14 Mill,
being aware of this problem, tried to solve it by saying that
causal sequences are not merely invariable but unconditional:
night is not unconditionally followed by day, because we can
describe changed conditions in which there would be night but
day would not follow. But this suggestion is in the end fatal
to the theory which it was designed to save. In the first place,
Mill himself has stressed that no ordinary causal sequences are
unconditional: what we commonly accept as causes are only
members of assemblages of conditions, positive and negative,
and only such a complete assemblage is unconditionally fol-
lowed by the effect. But this is a fairly superficial criticism.
There is a problem, to which we shall return, why some con-
ditional regularities should be not accepted as cases of the
earlier item causing the later, while other regularities, no less
conditional, are accepted as cases of causation. But there are
more fundamental objections to Mill's way out. We must dis-
tinguish de facto unconditional regularity from counterfactually
unconditional regularity. If night in, say, Glasgow is always in
fact followed by day in Glasgow (it is evening in Glasgow when
the earth eventually blows up), then this regularity is de facto
unconditional; Mill's protest that we know of conditions in
13
T. Reid, Essays on the Active Powers of Man; Mill, System of Logic, Book HI,
Ch. 5, Sect. 6.
14
G. D. Broad, The Mind and its Place in Nature, pp. 455-6. Broad's actual
example is of workers going to work, but the reverse is in some ways more con-
venient.
8a CAUSAL REGULARITIES
which night would not be followed by day (for example, those
in which the earth blows up just before daybreak in Glasgow)
means only that this regularity is not counterfactually uncondi-
tional. Mill is claiming, then, that to be causal a regularity
must be counterfactually unconditional. But while this may
throw some light upon our concept of causation, it cannot apply
directly to our present question whether regularity constitutes
causation in the objects. For the holding of a counterfactual con-
ditional is not a fully objective matter: we must go back from
the conditional to whatever grounds make it reasonable to
assert it or to suppose it to hold. Now if we find some 'causal
mechanism', some continuous process connecting the antecedent
in an observed conditional regularity with the consequent, we
may be able to sort out some counterfactually unconditional
regularities which underlie the conditional one. (This will be
discussed and illustrated in Chapter 8.) On the other hand, even
if we can find no mechanism, no continuous connecting process,
and even if we believe that there is none to be found, we may
still assert, and have reason for asserting, that there is some
counterfactually unconditional regularity. Russell at one time
postulated, though he later rejected, 'mnemic' causation, in
which an earlier event (an experience) is the proximate cause of
a later one (a memory).15 This would not mean that experiences
are unconditionally followed by rememberings, but it presum-
ably would mean that in an appropriate field (which would
have to include the survival, and the consciousness, of the
subject) some assemblage of conditions, including an experience,
was unconditionally followed by the corresponding memory,
but without any specific linking mechanism over and above the
mere survival of the intermittently conscious subject. This hypo-
thesis may well be false and is indeed, as Russell admitted,
extravagant, but it is not incoherent, and we could fairly easily
describe experiments and observations which might confirm it
by discontinuing alternative explanations of remembering.
Such a regularity, then, could be reasonably asserted to be
counterfactually unconditional; its instances would then fall
under our concept of genuine causal sequences, but all that was
present in the objects would be that pattern of regularities and
IS
The Andlysis of Mind, pp. 78-9; An Inquiry into Meaning and Truth, p. 297;
'Reply to Criticisms' in The Philosophy ofBertrand Russell, ed. P. A. Schilpp, p. 700.
 CAUSAL REGULARITIES 83

irregularities, 'agreements* and 'differences', which we reason-

ably took as confirming this hypothesis and as discontinuing its
rivals. Of course, this and all similar hypotheses may well be
false; it may be that causation in the objects always does in-
volve continuous processes, but we cannot say that it must do
so, that such continuity must be added, even to those regularities
which we reasonably take as counterfactually unconditional,
in order to make them causal.
However, even if Mill could so far defend the claim that
counterfactually unconditional regularity is sufficient for causa-
tion, we can refute this claim by using, in a slightly different
way, the stock counter-examples to the regularity theory.
Typically, these are cases of branched causal patterns.16 A
common cause, the rotation of the earth relative to the sun, is
responsible for night in Glasgow, that is, for a period of, say,
twelve hours of darkness there, and also for the ensuing day,
that is, for a period of, say, twelve hours of light. A similar,
though more complicated, account can be given of the Man-
chester hooters and the London workers. Generalizing, we have
the sort of causal pattern that; is roughly indicated by this
diagram:

Diagram (iv)

Here Cis the common cause, A one effect, and B the other; the
pattern is repeated over and over again, but each instance of A
occurs just before the associated instance of B. But of course
this linear diagram is only rough: in fact other conditions will
be conjoined with C to produce J, and others again to produce
B: C is presumably only an inus condition of A and of B. We
can concede to Mill that the A-B sequence is not uncondi-
tional. But it is not this that prevents A from being the (or a)
cause of B\ for we can find, underlying this, an unconditional
sequence which is still not causal. Suppose that the full cause
16
But these are not the only counter-examples. More thoroughly accidental
regularities will be discussed in Chapter 8.
84 CAUSAL R E G U L A R I T I E S
of A is (CXor T), and the full cause of B is (C£or W)t X and £
being present whenever this whole pattern is instantiated. Then
clearly AT is unconditionally preceded by C, while C£ is uncondi-
tionally followed (after a longer time interval) by B\ hence Af£
is unconditionally followed by B, though f% presumably is not.
There is an unconditional sequence in which the antecedent
is an assemblage of conditions of which A is a non-redundant
member, and the consequent is B. In more concrete terms, the
sounding of the Manchester factory hooters, plus the absence
of whatever conditions would make them sound when it wasn't
five o'clock, plus the presence of whatever conditions are, along
with its being five o'clock, jointly sufficient for the Londoners
to stop work a moment later—including, say, automatic devices
for setting off the London hooters at five o'clock, is a conjunc-
tion of features which is unconditionally followed by the
Londoners stopping work. In this conjunction the sounding of
the Manchester hooters is an essential element, for it alone, in
this conjunction, ensures that it should be five o'clock. Yet it
would be most implausible to say that this conjunction causes
the stopping of work in London. So the antecedent in even an
unconditional sequence may fail to cause the consequent. (And
this is not because the sequence is logically necessary, though
our description may have suggested this. Though I have spoken
of whatever conditions are sufficient for this or that, this is only
a way of indicating the concrete conditions T and £, whatever
they may be; T and £ are not themselves logically related to
A, B, and C, though our descriptions of them are; the sequence
in which AT£ is followed by B is logically contingent though
unconditional.)
Nor can this sort of counter-example be undermined by say-
ing that to be causal a sequence must be such as to be reasonably
taken to be counterfactually unconditional. For if the (CX or
T)-A and (C£ or W}~B sequences are counterfactually uncon-
ditional, so is the Af%-B one. This sort of counter-example
shows, too, that adding a causal mechanism, a continuity of
process, is not enough; for if there are such mechanisms or
continuities between C and A and between C and Bt there will
inevitably be a set of mechanisms, a resultant continuous pro-
cess, linking A with B.
But it is not too difficult to begin to see what the key addi-
 CAUSAL REGULARITIES 85
tional feature is that marks off genuine cause-effect sequences,
and that is lacking in this A-B counter-example. It is what we
spoke of in Chapter 2 as causal priority. In the branched
pattern, each instance of A, or of AY Z, is not causally prior
(though it is temporally prior) to the associated instance of B.
Each A is indeed related to its B by 'some fact of causation',
by what is roughly indicated by the arrows in the diagram;
but the C-A arrow is pointing the wrong way. The Af%~B
sequence is causally maintained, but Af% does not cause B.
Admittedly this is only a schematic answer, since we have not
yet discovered in what causal priority in the objects consists:
the account sketched at the end of Chapter 2, in terms of
conditionals and possible worlds, may help to identify our notion
of causal priority, but it falls far short of anything that could
be an objective description. But though it is elusive, causal
priority can hardly be non-existent. The regularity theorist
could rebut this last criticism only if he were prepared to say
that there is no difference in the objects between the causal
pattern represented, however crudely, by the above diagram
and that which would be represented thus:

Diagram (v)

—in other words, that there is no difference between a means

(or intermediate link in a causal chain) and a side-effect. But
this would be most implausible.
We can now return to our first criticism of Mill's suggestion,
to the point that some conditional regularities are accepted as
cases of the earlier item causing the later, while others are
not. While a number of considerations (of which some were
mentioned in Chapter 2, and others will be touched upon in
Chapter 5) restrict the application of the term 'cause' to some
inus conditions in preference to others, what is most important
in the present context is again the absence of causal priority.
The sounding of a hooter in Manchester is not causally prior
 86 CAUSAL REGULARITIES
to Londoners stopping work, while the sounding of a hooter in
London is, although both these hooter-stopping regularities
are equally conditional, both de facto and counterfactually.
In conclusion, then, regularity of the sort we have elucidated,
even if it is of a kind that can be called counterfactually uncon-
ditional, is not the whole of what constitutes causation in the
objects. Some causal mechanism or continuity of process may be
an additional and distinguishing feature of sequences in which
the earlier item causes the later, but whether it is or not it
seems certain that something which we can call causal priority
is such an additional distinguishing feature. The regularity
theory, even in its improved form, is not a complete account of
causation in the objects, and as we saw earlier it is not adequate
either as an account of what causal statements mean or of what
we know about causation. But to say this is by no means to
deny that causal regularities, complex and only partly known,
contribute something to the answers to all three of these
questions.
Those who find regularity theories inadequate commonly
insist on some intrinsic necessity in causal relations, or suggest
a 'genetic' view of causation, according to which causes pro-
duce or generate their effects and are not merely followed by
them, or both.17 But it is not enough to reiterate such words as
'necessity' and 'production' and 'generation'; we need some
clear account of what this necessity is, or of what producing or
generating can be. Moreover, this account must be able to
resist Hume's criticisms, to take up his challenges, to explain
how these key relations escaped his notice. Logical necessitation
between distinct occurrences or their features is ruled out, and
Hume's challenge to his opponents to point out anything like
what we have called necessity2 is not easy to meet. Nor, I think,
is it at all helpful to say that things have causal powers; the
concept of powers needs to be elucidated in terms of causation
rather than causation in terms of powers.18 Since what was
called, in Chapter i, Hume's third exclusion or third negative
point was poorly supported, there is no serious obstacle to the
17
For example, A. C. Ewing, Idealism^ pp, 151-87; W. Kneale, Probability and
Induction, pp. 70-103; R. Harre", The Principles of Scientific Thinking, Chapters 4, 10,
and n.
18
I have discussed powers and dispositionai properties in Chapter 4 of Truth,
Probability, and Paradox.
 CAUSAL REGULARITIES 87
description of some empirical relations that might be called
producing and generating; but the description needs to be
given and, if possible, related in some way to the counterfactual
conditionals that are at the heart of the ordinary notion of
causing. In Chapters 5 and 8 I shall try to resolve these
problems, by both borrowing from and criticizing the work, in
particular, of Ducasse and Kneale, while in Chapter 7 I shall
investigate that other essential supplement to regularity, the
direction of causation.
 4
Kant and Transcendentalism

HUME'S treatment of causation, as is well known, led on to

Kant's, indeed it provoked Kant's whole critical philosophy.1
Should we retrace this historical path, and consider whether,
despite Hume's arguments, there is some important a priori
element in causation? Some of our discoveries may have
diminished this temptation. In criticizing Hume's arguments,
in Chapter i, we saw that the notion of a necessity2, of some-
thing that would justify or support a priori inference from cause
to effect and vice versa, is not really part of our ordinary con-
cept of causation; consequently the failure to find any empirical
basis for such a notion need not drive us to search for an a priori
basis. We also saw in Chapter i that while Hume has shown
that experience and 'demonstration' together cannot establish
in a deductively valid way the principle that every event has a
cause, or any particular causal law, or even the conclusion of
any single causal inference, he has not excluded the possibility
that there might be forceful probabilistic inferences from an
observed cause to an unobserved effect or vice versa, or from
experience to particular causal laws, and even perhaps to the
general causal principle. Hume's empiricism has not after all
led to a scepticism so extreme that we should be strongly
tempted to abandon the empiricism. And having brought out
in Chapter 2 the patterns of conditional thinking that really
constitute the main part of necessity^ of what we ordinarily
take to differentiate causal from non-causal sequences, we found
it fairly easy to suggest a psychological source for them. On the
other hand, we also studied in Chapter 3 a more sophisticated
way of arriving at and handling the key conditional statements,
and this, we found, could be formulated in terms of methods
1
Kant, Prolegomena^ 357-60.
 KANT AND TRANSCENDENTALISM 89
which involved, among their premisses, almost totally unsup-
ported assumptions which, while not quite equivalent to the
old universal causal principle, yet had some affinity with it.
For these assumptions, at least, it is still tempting to seek an
a priori foundation or justification. Besides, there are two
further aspects of our ordinary idea of causation for which our
discussion so far has not adequately accounted: the notion of
some intimate tie between an individual cause and its effect,
and that of the direction of causation, the priority of cause to
effect.
As Kant saw it, Hume
challenged Reason . . . to . . . say with what right she thinks: that
anything can be of such a nature, that if it is posited, something else
must thereby also be posited necessarily; for this is what the concept
of cause says. He proved irrefutably: that it is wholly impossible for
reason to think such a conjunction a priori and out of concepts. For
this conjunction contains necessity; but it is quite impossible to see
how, because something is, something else must also necessarily be....
From this he inferred that Reason completely deceives herself with
this concept, in falsely taking it: for her own child, whereas it is
nothing but a bastard of the imagination fathered by experience.*
But Kant claimed that his own critical philosophy solved
Hume's problem not just for the special case of causation but
for a whole system of metaphysical concepts, showing that they
have, after all, their origin in 'pure understanding'.
But as an escape from the difficulties Hume encountered in
trying to find an 'impression' from which to derive 'the idea of
necessary connexion', Kant's treatment is disappointing. He
says practically nothing about what the concept of causation
contains. The Metaphysical Deduction hints that the 'pure
concepts of the understanding', the categories, correspond to
the logical forms of judgement, and in particular that the
category of cause and effect has something to do with the hypo-
thetical (conditional) form of judgement. But this, though
correct, is only a hint, and requires the sort of development
attempted in Chapter 2; Kant himself says that he purposely
omits the definitions of the categories.3 In the passage quoted
above he tells us that what the concept of cause says is that
* Prolegomena, 5257, trans. P. G. Lucas.
» Critique of Pure Reason, A 82 = B 108.
go KANT AND TRANSCENDENTALISM
something is 'of such a nature, that if it is posited, something
else must thereby also be posited necessarily'. But this is
ambiguous as between necessitya and necessity3. It says that
causation must be something which licenses an inference from
cause to effect, but not whether this inference must be based
wholly on something in the cause itself, on its own, or whether
it is enough that there should be some rule or law from which,
in conjunction with the cause, the effect could be inferred.
However, when Kant comes to a detailed discussion and defence
of something a priori about causation, this ambiguity is resolved
in favour of necessity3: in the Second Analogy, his concern is
only with the principle that 'All alterations take place in con-
formity with the law of the connection of cause and effect' ;4
in the first edition he was even more explicit: 'Everything that
happens . . . presupposes something upon which it follows
according to a rule.'5 Surprisingly, in view of the importance
which it would appear to have for his thesis that objective time-
order depends upon causation, Kant has little to say about
causal priority, and he has nothing to say about any intimate
tie between an individual cause and its effect. Causation in the
(phenomenal) objects is, for Kant no less than for Hume, just
regularity. All that is a priori is that there should be some
regularity covering every case. Nor did he have anything to
say (as Mill did) about what sort of regularity it was, about
assemblages of conditions and plurality of causes. His account
bears only upon the status of such regularity: it purports to
show that, and why, the rules in accordance with which changes
are seen as occurring are strictly universal, are not merely of
the form 'All Cs so far observed have been followed by £s* with
inductive extrapolations forever plagued by sceptical doubts.
But, as Kant says, the question is not merely 'whether the
concept of cause is correct, useful, and in respect of all know-
ledge of nature indispensable': even Hume did not doubt or
deny this. The question raised by Hume was whether this
concept 'is thought a priori by reason, and in this way has an
inner truth independent of all experience and hence . . . [is]
not limited merely to objects of experience',6 and what Kant
claims to have shown is that 'these concepts' (including those
of substance and community as well as that of cause) 'and the
s 6
* B 2311. A 189. Prolegomena, 259.
 KANT AND TRANSCENDENTALISM 91
principles drawn from them' (for example, that everything that
happens 'presupposes something upon which it follows accord-
ing to a rule') 'stand a priori before all experience and have
their undoubted objective Tightness, though admittedly only m
respect of experience'.7
I shall examine this claim; but also, since modern com-
mentators like Bennett8 and Strawson,9 after pointing to many
fallacies and obscurities in Kant's own arguments, have offered
other arguments in their place which are still Kantian in
spirit, use Kantian materials, and purport to establish con-
clusions which are at least close to part of what Kant tried to
establish, I shall also consider whether any such replacement
arguments yield important conclusions about causation,
There is a preliminary puzzle about the apriority of concepts
as opposed to principles. Even if a concept were a priori in the
sense of not being derived from experience, it is hard to see
how this would help. How would being 'thought a priori by
reason1 give it an 'inner truth'? Surely the crucial question is
whether the concept is applicable, either to experience or to
some kind of non-experienced entities, and it will be some
principle that states that it is so applicable. On the other hand
it is conceivable that such a principle should have some a priori
justification without being itself universal in form: there might
be good a priori reasons for expecting some alterations to take
place in accordance with some, possibly only approximate,
regularities, even if there were not good a priori reasons for
expecting all alterations to conform to strict laws. This would
not have suited Kant, both for temperamental reasons and
because it would not have supplied the desired foundation for
Newtonian science; but it is a live option for modern thought.
But how could such a principle be a priori? Not by being
demonstrable as an analytic or logical truth: that is a possibility
which Hume has, as Kant admits, conclusively ruled out. What
we need is a transcendental argument, to show that the prin-
ciple is necessarily presupposed in our experience. But this can
itself be taken in either of two ways, with or without Kant's
transcendental idealism. With the idealism, we could say that the
principle was necessarily true of the content of our experience
7 8
Ibid., 311. Jonathan Bennett, Kant's Analytic.
» P. F. Strawson, The Bounds of Sense,
92 KANT AND T R A N S C E N D E N T A L I S M
because that content itself had somehow been constructed in
accordance with it, and was proved by Kant's system to have
been so constructed. Without the idealism, on Strawson's
'austere interpretation*, the principle, or the corresponding
concept, would be merely 'an essential structural element in
any conception of experience which we could make intelligible
to ourselves'.10 Its being true of the world we know, the world
of science and common sense, would not be guaranteed by
anything we knew or could prove about that world itself. For
once we reject the transcendental idealism and say that the
world we are concerned with is genuinely objective, completely
real and not merely 'empirically real but transcendentaliy ideal',
nothing founded solely upon presuppositions of our experience
can be authoritative about it. All that would be shown would
be that it was idle for us to speculate about the possible falsity
of the principle in question, since a world for which it was false
would lie, for us, beyond the scope of investigation and beyond
the scope of coherent detailed speculation.
If such a principle were known a priori, how would this help?
Would this put any necessity into individual causal sequences,
or allow us to infer a particular effect from a particular cause?
For this would we not require a priori knowledge of particular
causal laws, such as Newton's law of gravitation? Yet Kant
explicitly denies, in the Critique, that these are known a priori.
An answer which Kant does not give explicitly, but to which
he seems to be committed, is that the a priori knowledge of the
general principle that every change is covered by some regu-
larity would somehow help us to discover and establish par-
ticular causal laws. Such an answer could be developed and
defended by constructing something like Mill's methods of
induction (or the improvements on them suggested in Chapter 3
and in the Appendix) and showing how particular causal laws
can be deduced from limited observations taken in conjunction
with an appropriate general assumption. In this way empirical
causal laws could be given a strict universality that enabled us
to apply them with certainty to unobserved instances, whereas
10
The Bounds of Sense, p. 68. The passage in Prolegomena, 259, partly quoted on
p. 90 above, shows that Kant himself would have been thoroughly dissatisfied
with such an austere interpretation: it would leave him saying no more than, in
his view, Hume had already admitted.
 KANT AND TRANSCENDENTALISM 93
observation and truly inductive reasoning alone could not give
them such strict universality. But Kant nowhere develops such
an answer. All he does is to argue for the apriority of the general
causal principle itself.
Arguments relevant to this problem are to be found in two
places in the Critique of Pure Reason, the Transcendental Deduc-
tion and the Second Analogy. Let us consider them in turn.
The Transcendental Deduction is, as Strawson says, not only
an argument but also 'an explanation, a description, a story'.11
In the latter respect it belongs to what he unkindly calls 'the
imaginary subject of transcendental psychology'.12 As Bennett
remarks,13 this chapter in the first edition was 'desperately ill-
written', arid the second edition version 'is only marginally
clearer*. Nevertheless, we can sort out at least the main drift
of the argument as distinct from the 'story', though its precise
interpretation is much more difficult. Kant himself provides a
summary: 14
The manifold given in a sensible intuition is necessarily subject to
the original synthetic unity of apperception, because in no other
way is the unity of intuition possible (§ 17). But that act of under-
standing by which the manifold of given representations . . . is
brought under one apperception, is the logical function of judge-
ment (cf. § 19). All the manifold, therefore, so far as it is given in a
single empirical intuition, is determined in respect of one of the logical
functions of judgement, and is thereby brought into one conscious-
ness. Now the categories are just those functions of judgement, in so
far as they are employed in determination of the manifold of a given
intuition (cf. § 13). Consequently, the manifold in a given intuition
is necessarily subject to the categories.
This is none too clear; but a paraphrase might run as follows:
(1) Any experience must include some plurality of sensory elements,
and it must be possible for all these elements to be seen as owned by
a single consciousness, by a self which is able to be self-conscious.
(2) Only judgements bring elements together in a kind of experi-
ence that is thus potentially self-conscious. (3) Any experience,
then, must be united by some form of judgement (some prepositional
form). (4) To fit into such a prepositional form is to be subject to one
of the categories. (5) Therefore, any experience is subject to the
categories.
11 12 I3
Op. cit., p. 86. Op. cit,, p. 97. Kant's Analytic, p. 100.
14
B 143.
94 KANT AND TRANSCENDENTALISM
Kant explains that this conclusion, that the categories are valid
a priori for all objects of experience, is meant to hold for any
experience based on sensible, not intellectual, intuition, whether
that sensible intuition is like ours or not; but (for us) 'only our
sensible and empirical intuition can give [the categories] body
and meaning'.15
It is clear that any categories for which this line of argument
is to yield a deduction must be closely linked with the forms of
judgement; that is, any determinate conclusion here would be
parasitic not only upon the Metaphysical Deduction as stated
but also upon that detailed development of it which Kant
purposely omitted. But, what is far worse, the conclusion could
at most be merely that every chunk of experience must be
subject to at least one category, must be unified in at least one
propositional form. There is nothing here to guarantee that any
chunk, or even any larger body, of experience should be subject
to all the categories, or even that every category should find
some employment within experience. For all that this line of
argument could show, a certain category, say that of cause and
effect, might have no application at all within our experience.
This argument, then, fails completely to establish or illuminate
the status of causation,
Kant does, indeed, argue in the Metaphysical Deduction that
since judgements of all twelve kinds can be made about any
object, it must be subject to all the twelve categories. But 'can'
is not enough. Unless all twelve forms of judgement are actually
and necessarily used in uniting each item of experience, there
is nothing to show that every experience must incorporate each
of the categories.
This criticism would stand even if we accepted steps (i) to (3).
In fact, they too are open to question. But this area of dis-
cussion becomes more important when we leave Kant's explicit
argument and consider some of the other lines of thought
which commentators have either found implicit in what Kant
says or themselves constructed in a Kantian spirit.
Thus Bennett concedes to Kant the premiss that 'all supposi-
tions about possible experience must concern experience which
is owned and which is accompanied by self-consciousness'.
He argues that although a non-reflective animal might have
» B 148-9,3 161.
 KANT AND TRANSCENDENTALISM 95
experience, and consciousness, without self-consciousness, we
could not envisage ourselves as having such experience.16 This
is too close for comfort to Berkeley's notorious argument
that we cannot without contradiction conceive trees existing
unconceived.17 On the contrary, we can quite coherently en-
visage, that is, contemplate the possibility, that we might have
consciousness without self-consciousness; though if we tried
actually to imagine ourselves being unselfconscious there would
be a contradiction between the imagining and the prescribed
unselfconsciousness. But even if we, unnecessarily, follow
Bennett and give Kant this premiss, how can he argue that 'if
I am self-conscious I must have a mental history which I can
bring under concepts'?18 Bennett first tries to illustrate Kant's
notion of intellectual synthesis by such examples as these: 'I am
aware of the presence of something cold, and of something
which hums; I consider the criteria for one thing's having two
properties . .. and conclude that the cold thing is the humming
thing, perhaps by identifying both with the tuning-fork which
I see in my hand.' Or again, I synthesize two book-appearances
by connecting them as appearances of a single book. In cases
like these, as Bennett says, there is or may be something that
could be described as an (empirical) act of synthesizing. But,
he insists, nothing like this is involved in one's awareness of
the unity of mental states in one's own mind. Kant seems to
say that where a unity is not produced by an empirical act of
synthesizing it must depend on a transcendental one, but
Bennett, like Strawson, finds this sort of transcendental psycho-
logy totally unacceptable. Rejecting this 'genetic' interpreta-
tion of transcendental synthesis, he suggests that there is a better,
'analytic' interpretation: 'when someone is aware of a unity
without having reasoned his way to this awareness, there is
still a "transcendental synthesis" in the sense that the unified
items must satisfy certain criteria for unity or identity, and the
person concerned must have a grasp of those criteria',19 and he
argues that this applies even to being aware of mental states as
one's own. Bennett suggests that on this analytic interpretation,
** '/Cant's Analytic, pp. 105-6.
" Principles, §23, and First Dialogue: I have examined the fallacy in 'Self-
Refutation; a Formal Analysis', Philosophical Quarterly, xiv (1964), 193-203.
18
Kant's Analytic, p. 124. " Ibid., p. 113.
96 KANT AND TRANSCENDENTALISM
* "transcendental synthesis" refers to a conceptual complexity
the grasp of which consists in a capacity for . . . empirical syn-
thesis'.20 He argues against Kant that the making of judgements
does not require such a capacity: a dog, for example, can make
judgements about the present without having 'the kind of
intellectual capacity Kant needs for the later stages of his
argument in the "Principles" chapter'.21 But he says that a
capacity for empirical synthesis is required for any awareness
of a unity which involves judgements about the past, and that
this is enough for Kant's argument, since 'Self-consciousness . ..
entails a capacity to judge about the past, and so—by my
argument in Rationality—entails the possession of a concept-
exercising language'.22
We can summarize this argument: (i) Coherently conceiv-
able experience requires self-consciousness. (2) Self-conscious-
ness requires the capacity to judge about the past. (3) This
capacity entails a command of concepts. (4) Therefore, any
coherently conceivable experience would involve the use of
concepts.
I have already queried the first step. The second, I think,
when used in conjunction with the first, involves an equivoca-
tion about self-consciousness. There is a difference between the
minimal self-consciousness, the reflective contemplation of an
experience that one is having as an experience, which is the
most that the (fallacious) Berkeleian argument for (i) would
even purport to show to be required, and the full-blooded con-
sciousness of a persisting self which would require judgements
about the past,
These criticisms may be reinforced by a general doubt about
the kind of argument proposed. The steps are intended to be
progressive, to constitute an advance. The various premisses of
the form 'X requires T! are therefore not meant to be analytic.
And yet if they are not analytic, by what are they supported?
Presumably by a kind of thought-experiment: one tries to
imagine coherently what it would be like to have X without Tt
but fails. I suggest that such failures (if not due to analytic
impossibilities, which will build up to a progressive argument
only with the help of equivocations) can result from, and
reveal, nothing more than the limitations of our imagination,
20
Kant's Analytic, p. 115. " Ibid., p. 116. " Ibid., pp. 116-17.
 KANT AND TRANSCENDENTALISM 97
the difficulty that we find in going far beyond the forms of our
actual experience. I want to suggest that there are various
possible forms of experience, to our coherent conceiving of
which there are only contingent, psychological, barriers. As
Bennett admits, there could be consciousness without self-
consciousness. Indeed even Kant admits this possibility, though
he disparages it: 'merely a blind play of representations, less
even than a dream'23—which, he says, would not then 'belong
to any experience'. As Bennett admits, though Kant does not,
there could be judgement without self-consciousness. Again,
there could be minimal self-consciousness without awareness of
oneself as a being with a past. And so on, in opposition to the
various synthetic but allegedly transcendental premisses of
whatever argument of this sort we are examining. Whereas the
aim of a transcendental argument (as Strawson, quoted above,
has said) is to reveal 'an essential structural element in any
conception of experience which we could make intelligible to
ourselves*, I suggest that it will succeed in revealing only a very
pervasive structural feature of our actual experience. Only the
limits of our imagination, along with equivocations which allow
the same verbal formulations to express now analytic truths,
now synthetic claims, and into which it is all too easy to fall
in a complex, highly abstract, and obscure discussion, make what
is pervasively actual seem essential.
It is, of course, obvious that even if we waived all these
objections and accepted Bennett's conclusion (4) above, that
any coherently conceivable experience would involve the use of
concepts, we would not have shown that any specific concept,
such as that of cause arid effect, is applicable to any, let alone
to all, of our experience. That task is reserved for the Analytic
of Principles, in particular for the Second Analogy.
Strawson explores and develops, more fully than Bennett,
Kant's anti-Cartesian notion that I can be aware of myself as
a single subject of experience only by being aware of a unified
world. Considering the Transcendental Deduction as an argu-
ment, he says that 'its fundamental premise is that experience
contains a diversity of elements (intuitions) which, in the case
of each subject of experience, must somehow be united in a
single consciousness capable of judgement, capable, that is, of
2J
Critique, A 112.
g8 KANT AND TRANSCENDENTALISM
conceptualizing the elements so united' and that 'its general
conclusion is that this unity requires another kind of unity or
connectedness on the part of the multifarious elements of
experience, namely just such a unity as is also required for
experience to have the character of experience of a unified
objective world, and hence to be capable of being articulated
in objective empirical judgements'.24 This formulation seems
to me to mingle intermediate steps with both the premiss and
the conclusion. The intended argument may be set out more
distinctly as follows, using others of Strawson's formulations:
(1) Any coherently conceivable experience must contain ele-
ments, experiences, which are 'subject to whatever condition
is required for it to be possible for [the subject] to ascribe them
to himself as his experiences'.25 (2) 'Unity of diverse experiences
in a single consciousness requires experience of objects' ;26 'for
a series of diverse experiences to belong to a single consciousness
it is necessary that they should be so connected as to constitute
a temporally extended experience of a unified objective world'.27
(3) If we ask 'How in general must we conceive of objects if we
are to be able to make judgements, determinable as true or
false, in which we predicate concepts of identified objects of
reference, conceived of as related in a single unified spatio-temporal
system?' or equivalently 'What in general must be true of a
world of objects, so conceived, of which we make such judge-
ments?', we shall in answering these questions 'simultaneously
find our a priori concepts or categories and the guarantee of
their application'. But these 'will be the schematized categories
we actually employ in the spatio-temporal world', not the un-
schematized ones which Kant 'envisaged as discoverable with
the help of the clue provided by formal logic',28 that is, as in
the Metaphysical Deduction, and as applicable to any sensible
intuition, however unlike our own. (4) Therefore, any coher-
ently conceivable experience must be such that the schematized
categories are applicable to it.
Or, sacrificing accuracy to brevity: (i) Coherently con-
ceivable experience requires unity in a single consciousness.
(2) What is united in a single consciousness must be experi-
ence of a unified objective world. (3) Experience of a unified
2 a6
+ The Bounds of Sense, p. 87. ** Ibid., p. 98. Ibid.
« Ibid., p. 97. 2* Ibid., p. 83.
 KANT AND TRANSCENDENTALISM 99
objective world must be of a world seen as subject to the
categories. (4) Therefore (as above).
(The remarks quoted in the fuller statement of (3) above
occur in what Strawson describes as a sketch of an over-hasty
argument. But his main comment is that it is undesirable that
the argument should immediately follow these lines, and that 'a
major part of the role of the Deduction will be to establish that
experience necessarily involves knowledge of objects, and hence
to displace that thesis from the status of prior definition, or
premised assumption, of the inquiry\29 Given that under-
pinning, these remarks can represent a step in a line of argu-
ment which Strawson thinks respectable, though it is not one
which Kant actually completed in the Transcendental Deduc-
tion.)
But are these premisses (i), (2), and (3) true? Strawson's (i)
is weaker than Bennett's, and might therefore be more de-
fensible : Strawson does not say that each item of experience
must be accompanied by self-consciousness, but only that it
must be possible for the subject to ascribe them to himself. Yet
I can see no convincing reason for accepting this. Admittedly
if there were (are?) creatures who had successive experiences
but were incapable of self-consciousness, then as Strawson says
'We seem to add nothing but a. form of words to [this hypo-
thesis] by stipulating that [these experiences] all belong to a
single consciousness' ;30 still, we have no difficulty in conceiving
that there should be such successions of experiences, even
though the attempt to conceive ourselves as having them is self-
refuting. But this issue is elusive; (2) gives rise to a much
plainer dispute. Surely there could be a kind of experience
which consisted of a sequence of impressions, with some memory,
some recognition of similarities, but which gave no ground for
any distinction between those impressions and their objects.
Strawson considers this objection carefully. 'Certainly concepts,
recognition, some span of memory would be necessary to a
consciousness with any experience at all; and all these would
involve one another. But why should the concepts not be
simply such sensory quality concepts as figure in the early and
limited sense-datum vocabulary? . , . The claim is that it is
quite conceivable that experience should have as its contents...
« Ibid., pp. 85, 87. » Ibid., p. 100.
ioo KANT AND TRANSCENDENTALISM
impressions which neither require, nor permit of, being "united
in the concept of an object".'31 But he replies that 'it must be
possible, even in the most fleeting and purely subjective of
impressions, to distinguish a component of recognition'; and
'Recognition implies the potential acknowledgement of the
experience into which recognition enters as being one's own'.3a
I am afraid that I simply cannot see this implication. Strawson
adds that the fact that this potentiality is implicit in recognition
'saves the recognitional component . . . from absorption into
the item recognized . . . even when that item cannot be con-
ceived of as having an existence independent of the particular
experience of it'.33 I would agree that the recognizing that an
experience is of a certain sort, or resembles some remembered
earlier experience, must be something other than the experience
itself: this recognizing, we might say, has the experience as its
object even where the experience itself has no object. It thus
introduces an awareness/object distinction where at first none
was allowed. But I cannot see that this has anything to do with
the potential acknowledgement of the experience as one's own. But
the crucial step is the next.
The minimum implied [by the potentiality of such an acknowledge-
ment] is that some at least of the concepts under which particular
experienced items are recognized as falling should be such that the
experiences themselves contain the basis for certain allied distinc-
tions: individually, the distinction of a subjective component within
a judgement of experience (as 'it seems to me as if this is a heavy
stone5 is distinguishable within 'this is a heavy stone'); collectively,
the distinction between the subjective order and arrangement of a
series of such experiences on the one hand and the objective order
and arrangement of the items of which they are experiences on the
other.34
Now this has some plausibility. The acknowledgement 'This is
mine'—that is, my experience, my idea, my impression—can
have point only where a contrast with what is not-mine—an
objective reality, or perhaps the experience of some other sub-
ject—is implied and can at least occasionally be implemented.
But as soon as we realize that this is what the 'acknowledge-
ment' involves, we must be more reluctant than ever to concede
3« The Bounds of Sense, p. 99. " Ibid., pp. 100-1.
34
« Ibid., p. 101. Ibid., p. 101.
 KANT AND TRANSCENDENTALISM 101
the previous claim that recognizing that an experience is of a
certain sort, or resembles some earlier one, carries with it the
possibility of acknowledgement of the experience as one's own.
Whereas Strawson says that 'the thesis of the necessary unity of
consciousness can itself be represented as resting on a yet more
fundamental premise—on nothing more than the necessity, for
any experience at all to be possible, of the original duality of
intuition and concept',35 I think that this last-mentioned neces-
sity cannot bear the weight of the superstructure Strawson
erects upon it.
Strawson sums up his criticism of 'the sense-datum theorist's
conception of a possible experience' as follows:
If such an experience were possible, then a series of corresponding
judgements of experience would be possible. But a set of such judge-
ments would yield no picture of a world of objects the relations of
which are distinct from the relations of experiences of them. Hence
it would provide no basis either for the conception of an experiential
route through such a world or for the isolation of the subjective,
experiential component in individual judgements. Hence it would
provide no basis for the necessary self-reflexiveness of experience,
which is in its turn the essential core of the possibility of self-
ascription of experiences. Hence the theorist has not succeeded in
producing a description of a possible experience.36
The first three sentences here can be accepted as merely
spelling out the sense-datum theorist's hypothesis. We can
agree, too, that the hypothetical experience would provide no
basis for self-ascription in the sense in which this involves a
contrast between 'mine' and 'not-mine'. But we must hesitate
over 'necessary self-reflexiveness'. Granting that any conceivable
experience will somehow include the duality of intuition and
concept—or, more simply, particular items being, and being
seen as being, of certain sorts—that it must be reflexive in the
sense of containing a 'component of recognition', it does not
follow and it is not true that it must be reflexive in the sense of
making self-ascription possible.
There is, I believe, a fundamental flaw in Kant's argument
which survives in Strawson's reconstructions. Even though the
various 'representations' must in fact belong to a single conscious-
ness in the sense that they are remembered, compared, and so
K 36
Ibid., p. 97. Ibid., p. 109.
102 KANT AND TRANSCENDENTALISM
on, it does not follow that they must be seen as belonging to a
single consciousness in the sense in which this carries with it
the contrast between 'mine' and 'not-mine', subjective and
objective, and methods of at least sometimes implementing this
contrast.
But what has all this to do with causation? Directly, very
little. As Strawson says, 'for anything detailed or specific by
way of conclusion, [Kant's argument] depends entirely on the
derivation of a list of categories from the forms of judgement.
We can place no reliance on this derivation.'37 Even if we could
rely on this derivation, Kant's argument, as we have seen,
would at best prove that every experience embodies some cate-
gory, and does not guarantee any application at all for any
specific category, such as causation. The conclusion which
Strawson says we are left with reason for entertaining
favourably is 'that any course of experience of which we can
form a coherent conception must be, potentially, the experience
of a self-conscious subject and, as such, must have such internal,
concept-carried connectedness as to constitute it (at least in
part) a course of experience of an objective wo'rld, conceived
of as determining the course of that experience itself'.38 This is
still a long way from a vindication of the category of causation.
Even if we accepted what Kant, or Bennett, or Strawson hopes
to extract from the Transcendental Deduction, the vindication
of any such specific category would be left for the Analytic of
Principles. Yet it is not quite beside the point to draw attention
to weaknesses in the preliminary argument. If Kant or Bennett
or Strawson were right, the Transcendental Deduction would
at least set the stage for the Second Analogy: it would lead us
to suppose that the transcendental method had already yielded
some general requirements, and that a further pursuit of the
same method might well yield some more specific ones. But if
my criticisms of these preliminary arguments are correct, the
prospect for detailed successes is much less promising.
Kant's argument in the Second Analogy is very repetitive,
the same point being made over and over again in various
forms, sometimes directly and sometimes by way of a reductio
ad absurdum, but unlike the Transcendental Deduction it is not
tortured or tortuous, and there is no great difficulty in under-
j8
« The Bounds of Sense, p. 117. Ibid.
 KANT AND TRANSCENDENTALISM 103
standing what he says. He sets out to prove that all alterations
(in the phenomenal world, the world of science and of common
sense) take place in accordance with causal laws: 'Everything
that . . . begins to be, presupposes something upon which it
follows according to a rule.'39 His proof may be summarized
as follows: If I see a ship move downstream, I have a succession
of subjective impressions. But if I survey the various parts of a
house I also have a succession of subjective impressions. What
makes the difference between those sequences of impressions
which I interpret as representing an objective process or
sequence of events and those which I interpret as representing
the coexisting parts of a persisting object? It is that the order
in which the impressions succeed one another is in the one case
determined and irreversible, it is a necessary order, whereas in
the other case the order in which the impressions succeed one
another might well be reversed. That is, a sequence of impres-
sions which I interpret as representing an objective succession
must conform to some rule which prescribes an objective order
from which the subjective order is derived. Consequently, it is
only conformity to such a rule which enables me to establish
an objective succession; hence all alterations in the pheno-
menal world, which we take to be objective, must take place
in accordance with rules.40
This argument as stated provokes some objections, by
examining which we can bring out its substance more clearly.
There is what Strawson calls the simple-minded objection that
the subjective order of impressions in Kant's ship example is
not in fact irreversible: ships move upstream as well as down-
stream, and even a ship which is pointing downstream may be
carried upstream by the tide or towed upstream by a tug which
I cannot see. Kant must reply that all he means is that when
on a particular occasion I do see a ship move downstream I am
not then free to have the impression of its lower position before
I have the impression of its higher one. But there are cases
where I do have such freedom. There is what Strawson calls
the more sophisticated objection that if I am, say, watching a
man, chopping through a log on the far side of a valley from
me, I shall see the halves of the log fall apart before I hear
the last stroke of the axe, whereas if I had been close by I
« A 189. <° Precis of A 189-94, B 833-9-
104 KANT AND TRANSCENDENTALISM
should have received these impressions of sight and sound in
the reverse order. To meet this Kant must, as Strawson sug-
gests, reply that the order of perceptions will be irreversible
provided that they are equally direct and in the same sensory
mode.41 But even if these cracks can be papered over, the argu-
ment seems to proceed, as Strawson says, 'by a non sequitur of
numbing grossness'.41 The fact that the order of impressions (on
a particular occasion) was not one which I was free to reverse
does not imply that either they, or an objective sequence of
which they are perceptions, took place in accordance with a rule.
Kant's argument turns upon a blatant fallacy of equivocation,
a play upon two different senses of the word 'necessary'. An
order which is necessary as not being reversible in the particular
case need not be necessary as in accordance with some universal
rule or law.
The fallacy is indeed so gross that we may wonder how Kant
can have been guilty of it. There are several things that help
to explain, if not to excuse, it.
First, the fallacy stands out only when, in response to the
'simple-minded objection', we have pinned down the kind of
irreversibility involved in such an example as that of the ship.
Another kind of irreversibility may have clouded the issue,
though it would not really help Kant's case. Kant may have
been thinking vaguely of the kind of sequence where one always
has a perception of one sort before, not after, one of another
sort: the perception of a spark, say, precedes and does not
follow that of an explosion. Of course, as we have noted in
Chapter 3, we do not really encounter such unqualified simple
regularities. But even if we did, the key step in the argument
would still be unsound. Kant would be saying, in effect: it is
only where my perceptions of ,4s have so far been regularly
followed, not preceded, by my perceptions of J5s that I
can interpret each (perception-of-an-^4)-(perception-of-a-jB)
sequence as a perception of an objective A-B sequence, there-
fore it is only because there is a rule which ensures that As
are followed by Bs with strict universality (not just so far) that
I can see each B as following its A in an objective time-order.
And even if the premiss were true, this would still be a non
sequitur. The problem of how to distinguish an objective from a
41
The Bounds of Sense, p. 135. <2 Ibid., p. 137.
 KANT AND TRANSCENDENTALISM 105
subjective succession contributes nothing to the bridging of the
inductive gap between regulairity-so-far and an unqualified
universal rule.
Secondly, if on some particular occasion a ship has moved
downstream, so that (given the stipulations that exclude our
simple-minded and sophisticated objections) I had to have the
impression of its higher position before having the impression
of its lower one, this necessity-for-me arises indeed from 'some
fact of causation', namely from the causal links between the
ship's being in each position and my impression of it in that
position, conjoined with the simple fact that its higher position
is earlier than its lower one. But although causal necessity thus
comes in somewhere, it does not come in where Kant wants it,
namely between the ship's occupancy of its higher and of its
lower position, or between some events that underlie and deter-
mine this sequence. Even if the sequence of these two situations
were the purest accident, the necessity-for-me, the irreversi-
bility of the order of my impressions (given our stipulations)
would be no less rigid.
But thirdly, and most significantly, there is at least a hint of
a second argument on which Kant may be relying. Suppose
that I am presented with a sequence of impressions; they come
in a subjective time-order, but I have not even a prima facie
reason for interpreting this subjective order as a copy or even
an indication of an objective order; how then can I put them
together in something that I can treat as an objective order of
events, either the same as or different from the subjective order
of the impressions? Surely I can do so only if there is some
intrinsic Tightness about one possible order as contrasted with
any other. It is as if I were handed the pieces of a linear jigsaw
puzzle in some order which might or might not be that in
which I was to put them together. Surely I could put them
together with confidence in one order rather than another only
if either their shapes, or the bits of pictures on them, indicated
which piece would join on to which, and which way round.
And if I were presented with several sequences of impressions,
each sequence in a subjective order, I could put together as
representing objective sequences of events only those sequences
of impressions among whose members I could detect some
intrinsic Tightness of some possible order.
io6 KANT AND TRANSCENDENTALISM
This clearly is a different argument from the one we have
been discussing, and it does not incorporate the non sequitur that
Strawson finds so shocking. But it has other fatal flaws. The
story told obviously misrepresents our empirical intellectual
history. We are, fortunately, not confronted in infancy with
sequences of impressions whose subjective order is out of all
relation to the objective time-order of events: while some are
indeed successive perceptions of coexisting parts of persisting
things, others are successive perceptions of the successive phases
of processes in the right order. And we are helped in deciding
which are which—or rather, in fairly automatically arriving at
a view of the world in which such a decision is implicit—by the
fact that the impression-sequences that belong to coexisting
parts are reversible in the simple sense that impressions of
recognizably the same parts often come in different orders (wall-
window-door, window-wall-door, door-wall-window, and so
on) while at least some impression-sequences that belong to pro-
cesses exhibit a fair degree of repetition of similar phases in the
same order, they are 'irreversible' in the sense considered three
paragraphs back. To save the present argument, Kant would
have to retreat from empirical into transcendental psychology,
claiming (as he would be very willing to do) that the human
mind—not in infancy but somehow out of time altogether—is
confronted with a task analogous to that of our jigsaw puzzle,
of constructing a world of objects from a manifold of repre-
sentations, and for all we know it may get no help in this
from any initial subjective time-order. But then this argument
requires the transcendental psychology, the 'genetic' interpreta-
tion of intellectual synthesis, about which modern commenta-
tors are so sceptical. But even if we waived this scepticism, the
argument would still fail. For clearly it requires that there
should be some intrinsic Tightness about particular sequences,
in other words, that particular causal laws should be knowable,
indeed known, a priori. They would constitute the design in
accordance with which we had (transcendentally) constructed
the phenomenal world. But this conflicts with the actual pro-
cedures of science, especially with the Newtonian science whose
foundations Kant thought that he was making secure, and with
what he explicitly says in the Critique. 'Certainly, empirical laws,
as such, can never derive their origin from pure understanding.'
 KANT AND TRANSCENDENTALISM 107
'To obtain any knowledge whatsoever of these special laws, we
must resort to experience.' It is true that he says also that 'all
empirical laws are only special determinations of the pure laws
of understanding', and that 'The laws of nature, indeed, one
and all, without exception, stand under higher principles of
understanding. They simply apply the latter to special cases
[in the field] of appearance.'43 These last remarks are unclear,
and they might mean that particular causal laws are only
applications of higher pure laws to concrete situations, that a
particular law—perhaps of the form 'There are As and they are
regularly followed by Bs'—is a logical consequence of the con-
junction of a pure law with the proposition that there are As,
so that the latter is the only information for which we need to
resort to experience. If this were so, then the pure laws would
in themselves be sufficient to constitute the design in accord-
ance with which we had (transcendentally) constructed the
time-order of events. But even if Kant meant this, it could not
be reconciled with the actual procedures of science. A more
defensible interpretation, to which, as I have said, Kant seems
to be committed, would be illustrated by our account of the
Method of Difference and of the other eliminative methods of
induction.44 A particular law of nature 'stands under' a general
principle of uniformity if it can be inferred from such a prin-
ciple in conjunction with the empirical fact that certain related
groupings of features occur, such as those set out as I% and jVj
in our classical difference observation. But now the law of
nature is not even knowable a priori, and Kant's argument
would again fail.
This second argument, then, will not do. Neither it, nor the
previous one, nor the conjunction of the two, will really serve
Kant's purpose, but the deficiencies of each may have been
partly concealed from him by the simultaneous operation of
the other within his thinking.
Bennett distinguishes two arguments in the Second Analogy,
'the object/process argument', which is our first, and 'the
ordering argument', which is roughly equivalent to our second.45
43
A 127, B 165, A 159, B 198; cf. Strawson, op. cit., p. 116. Bennett, op. cit.,
pp. 157—9. In the Opus Postumum Kant says that some particular laws are knowable
a priori, but this is not his view in the Critique.
4 4S
* Chapter 3 and Appendix. Kant's Analytic, pp. aaa-g.
io8 KANT AND TRANSCENDENTALISM
He says that the latter 'occurs in fragments, just below the
surface of Kant's text', that 'several parts of the text become
more intelligible if they are seen to be coloured by a subliminal
awareness of the ordering argument'. I would agree, but since
it is below the surface, the precise interpretation of this argu-
ment is open to dispute. Bennett frees it from dependence upon
transcendental psychology, upon a 'genetic' interpretation of
intellectual synthesis, and also from implausible empirical
claims about the intellectual achievements of infants, by relating
it to memory: how am I to recollect digging in the garden
before having a bath? The suggestion is that to do this 'I must
be able to appeal to objective considerations' such as 'the be-
haviour of clocks, or the fact that I always have a bath after
gardening, or the fact that the mud which I recall washing off
in the bath could only have come from my digging in the
garden'. As an interpretation of Kant, I find this implausible
(though I accept Bennett's point that the ordering argument,
unlike the object/process one, applies as much to the order of
experiences as to the order of objective events); but how good
is it as an argument in its own right? As Bennett says, the thesis
that remembered temporal order depends on such objective
considerations does not hold without exceptions: he mentions
three ways in which the order of a pair of events can be simply
recalled. First, we may 'recall a specious present containing both'.
Secondly, we may recall a sequence of happenings, A-B-C-D-E
say, where each contiguous pair is recalled as belonging to one
specious present. Thirdly, 'One may simply recall that X pre-
ceded T by recalling a time when one experienced T while
recalling X.' Still, as he says, these exceptions leave a lot of
work to be done by the objective, causal considerations. But,
Bennett rightly concludes, this thesis will not suffice to establish
Kant's principle that all alterations take place in accordance
with laws. It merely illustrates what Hume would have been
happy to admit, that causation in addition to being 'to us, the
cement of the universe', constitutes, to us, the reinforcing rods
as well.
Most of what Kant says in the later paragraphs of the Second
Analogy merely restates the same two basic arguments though
with some picturesque touches. 'Now since absolute time is not
an object of perception, this determination of position cannot be
 KANT AND TRANSCENDENTALISM 109
derived from the relation of appearances to it. On the contrary,
the appearances must determine for one another their position
in time, and make their time-order a necessary order.'46 That
is, the Newtonian time-axis is not visible, with a series of dates
marked off along it, nor do events come to our perception with
their dates stamped on them. But this only raises the question
how we arrive at an objective time-order: to get an answer Kant
must use one or other, or both, of the arguments already discussed.
Kant does, however, mention a difficulty which is of some
interest.47 How is he to reconcile his thesis that causation sup-
plies objective time-order with the fact that causes are some-
times simultaneous with their effects, indeed that 'the great
majority of efficient natural causes' are so. Kant's reply that it
is the order, not the lapse, of time that matters, and that the
order can persist even where the interval between cause and
effect is a vanishing quantity, is unconvincing. He correctly
illustrates the sort of experiment that reveals a direction of
causation: 'if I lay a ball on the cushion, a hollow follows upon
the previous flat smooth shape; but if (for any reason) there
previously exists a hollow in the cushion, a leaden ball does not
follow upon it'. This shows, indeed, how causal priority is
established between simultaneous cause and effect (the ball
resting on the cushion, and the hollow) where these are as it were
continuations of events between which there is a temporal
succession (the ball being placed on the cushion and the cushion
sinking in the middle). But such examples still falsify the thesis
that causal order is what constitutes objective temporal order.
Kant also discusses 'the law of the continuity of all alteration',
and attempts an a priori deduction of it. What he actually says
is of little value, but this point and the previous one show that
he was aware of certain aspects of causation that call for further
attention, such as I shall attempt to give in Chapters 7 and 8.
Strawson points out the crucial fallacy in the object/process
argument; he does not bring out any version of the ordering
argument; but he does make 'another attempt' at an a priori
vindication of causality. 'If the direct arguments of both second
and third Analogies fail, ... it does not follow that the problem
of the Analogies cannot be advanced at all, and advanced in
something like a Kantian direction, with the help of materials
46
A 200 = B 245. 47A 202-3 = B 247-9.
 MO KANT AND TRANSCENDENTALISM
which Kant puts at our disposal.'48 The argument which
Strawson suggests may be summarized as follows: (i) If there
were no currently unperceived, but perceivable, objects co-
existing with objects of actual perception, there would be no
effective distinction between objective and subjective time-
orders.49 (2) There can be such unperceived coexisting objects
only if we perceive some things as relatively persistent things
of certain kinds, in some spatial or quasi-spatial framework of
relations.50 (3) If we also have the idea of perceived objective
succession or change (that is, if we can sometimes conceive a
change in our perceptions as the perception of a change) our
concepts of persistent objects must be concepts of objects which
are capable of changing without changing out of all recogni-
tion: there may be some random changes, but only against a
background of regular persistences and alterations.51 (4) Some
changes in perceptions 'exhibit a regular correlation with
change of the observer's position and his sense-orientation . . .
Without some such correlation it is impossible to see how
notions of enduring and reidentifiable objects of changing per-
ception could secure application in the observer's experience.
But the possibility of this correlation in turn seems to depend
upon changes and persistences in the world of objects being
themselves subject to some kind and degree of order and
regularity.'" (5) Our concepts of objects, therefore, '[are] neces-
sarily compendia of causal law or law-likeness, carry implica-
tions of causal power or dependence. Powers, as Locke remarked
. . . make up a great part of our idea of substances. More
generally, they must make up a great part of our concepts of
any persisting and re-identifiable objective items. And without
some such concepts as these, no experience of an objective
world is possible.'53
My fundamental unease about this argument is the same as
that which I expressed about the Transcendental Deduction. Is
not Strawson misreading pervasively actual features of our
experience as essential? On (5), for example, we can concede
that powers, dispositional properties, do make up a great part
48 4 50
The Bounds of Sense, p. 140. « Ibid., p. 141. Ibid., p. 148.
51 52
Ibid., pp. 143-4. Ibid., p. 144.
SJ
Ibid., pp. 145-6; Locke's actual words (Essay, Book II, Chapter 23, Sections
8, 10) are:'... powers make a great part of our complex ideas of substances ...'.
 KANT AND TRANSCENDENTALISM in
of our concepts of objects: but must they do so? Could not our
experience of an objective world have been one of objects each
of which persisted for a while, with little change, then ceased
to exist, which did not interact much, and whose changes and
ceasings to be were unpredictable? A Lockean would reply
that if these are perceived objects, they must at least have regular
powers to produce 'ideas', 'impressions', or something of the
sort, in us. But many philosophers have tried, however per-
versely, to get along without a causal theory of perception: it
would be strange if it now turned out to be an a priori truth.
It may be a quibble, and it certainly would not affect the
general outcome of Strawson's argument, but is even (2) strictly
true? Could not a sufficient degree of regularity in successive
phases of continuous processes enable us to dispense with per-
sistent things? And if my criticisms of Strawson's reconstruction
of the Transcendental Deduction have any force, we cannot
now use as a premiss the penultimate conclusion of that argu-
ment, that any coherently conceivable experience must be of
a unified objective world.
But what is perhaps most worth noting is how little of Kant's
grand scheme would be left even if Strawson's arguments went
through. Strawson offers only 'the necessary but loosely woven
mesh of our concepts of the objective'. Kant's 'two absolute
thoughts—of strictly sufficient conditions for every objective
change'—are not necessary thoughts but only natural hopes, 'a
pressing to the limit of those truly indispensable but altogether
looser conditions' for which Strawson has argued.54 We have
not perfect conformity to a precisely derivable set of schematized
categories, but just some persistence and some regularity in a
world that we can treat as objective.
But perhaps this is not so bad. We do not feel, as Kant did,
the need to lay philosophical foundations for Newtonian physics.
Modern science, in some of its aspects, would be positively
embarrassed by an a priori proof of strict universal determinism,
and yet might welcome a presumption of some regularity, some
persistence.
More serious is the change in the status of the conclusion.
Having rejected Kant's transcendental idealism, Strawson can
claim that even these modest regularities are necessary only
«« Ibid., p, 146.
iia KANT AND TRANSCENDENTALISM
in the sense that we cannot coherently conceive a mode of
experience without them. But is this any sort of safeguard,
any defence against scepticism? For example, does it protect
us against Hume's scepticism about induction? Surely not. The
Humean sceptic can still say 'Induction seems to have worked all
right so far, some inductive conclusions have been further con-
firmed, many of our more carefully reached predictions have
been fulfilled; but that only shows that the world so far has
shown a degree of regularity that matches the Strawsonian
requirements; from now on, perhaps, everything will go wrong.'
And it is not much comfort to be able to reply. 'You can't say
in detail, or even imagine, what it would be like if everything
went wrong.' It is true that sceptics are commonly unjustifiably
selective in their scepticism: they are liable to suppose that the
regularities which are involved in the continued existence and
broad relative location of things, in our own existence, and in
our perceiving of what is there, will survive, but that those
which we more immediately classify as causal may go wrong.
If so, they can be convicted of arbitrariness; but the thorough-
going sceptic is not so easily defeated.
There is, nevertheless, some point in the exposure of selective
scepticism. Since we can show that presumptions of some degree
of regularity are involved in our ordinary concepts of things
and of their largely dispositional properties, we can criticize as
arbitrary any unwillingness to grant a similar presumption in
favour of what we need as the assumption in some particular
application of the Method of Difference or of one of the related
methods. For, as we have seen in Chapter 3, such an applica-
tion does not require the complete uniformity of nature, but
only some local regularity, and we can use this as a working
assumption so long as there is some initial reason for adopting
it, without needing to claim that it is certain and indeed while
being prepared to give it up if the empirical evidence is hostile.
But it seems to me that to make this point one needs only to
bring out the extent to which causal regularities are involved
in our actual picture of a world of objective persisting things,
not to add Strawson's disputable claim that the conditions for
which he has argued are indispensable in the sense that any
coherently conceivable experience must satisfy them.
I conclude that neither Kant's explicit arguments nor the
 KANT AND TRANSCENDENTALISM 113
general Kantian approach contribute much to our under-
standing of causation. The attempt to establish the general
causal principle as valid a priori for the objects of experience
(and so to vindicate the universality of particular causal laws
in a way that would be proof against Humean doubts about
induction) has failed, and the claim that weaker principles of
regularity can—without invoking transcendental idealism—be
shown to be still in some way necessary turns out to be question-
able and in any case not very useful. And, as we noted, Kant
does not even attempt a transcendental deduction of those
other still puzzling features of our ordinary concept of causa-
tion, the notions of causal priority and of an intimate tie
between individual cause and individual effect, nor do his
recent commentators.
There is, indeed, one clue—no more than that—which might
be worth following. Kant's chapter on Schematism offers an
obscure solution to an equally obscure problem, namely how
the pure categories, which are supposed to have been generated
by the forms of judgement, can apply to sensory appearances.
The solution is that time somehow mediates between the
intellectual and the sensible and produces the schematized
categories of substance, causation, and the like. These are the
categories which we actually use, and these, because they
essentially involve time, will fit on to the objects of sensory
experience. But related to Kant's obscure problem there is, for
us, a more pressing question. We have found that certain con-
ditional statements are prominent in the analysis of causal
beliefs—these are the counterpart to Kant's attempt to derive
the pure category from the hypothetical form of judgement. We
have been able to relate these to certain psychological processes
and also to certain (complex and incompletely known) regu-
larities. But we have not yet linked them with the notion of a
continuous closely knit process, with our spatio-temporal pic-
ture of a bit of causing going on. Perhaps if we did so link them
we should have a counterpart to Kant's schematism.
Another point of some interest might be developed from
Kant's suggestion in the Second Analogy that time is somehow
dependent upon causation. As we have seen, he is concerned
only with time-order; but a much stronger relation can be
shown with respect to duration, to the metrical aspect of time.
ii 4 KANT AND TRANSCENDENTALISM
Newton, in a famous passage, claimed that absolute time flows
equably;55 but as Newton himself admitted, we cannot see it
flowing and cannot therefore measure it directly. If we observe
three successive instantaneous events A, B, and C, we cannot
directly inspect the durations between A and B and between
B and C, and so decide whether these intervals are equal or
whether one is longer than the other. So of course we have to
use what Newton calls relative time; 56 in other words, consult
a clock. Keep a pendulum swinging continuously and count
the swings between A and B, and between B and C, and com-
pare these numbers. Another, perhaps better, clock is the rotat-
ing earth. See how often it rotates, or through what angle it
rotates, relative to the sun or to a fixed star, between A and B
and again between B and C. But of course the swinging of the
pendulum, or the rotating of the earth, is (in a broad sense) a
causal process. The presumption that if, and only if, the absolute
time intervals A-B and B-C are equal, our clock will go through
equal numbers (or quantities) of its characteristic performances
is one example of our general presumption of uniformity. But
causation also plays a further role. Our actual clocks are only
prima facie clocks; some of them, as Newton says, are certainly
inaccurate, and all of them may be.57 But then do we just
arbitrarily choose one prima facie clock rather than another as
our criterion of the equality of durations? Certainly not. Or
do we take votes, and rely on majority verdicts, or average the
measurements supplied by all available prima facie clocks?
Not quite, though we may give the consilience of a number
of clocks some provisional, interim authority. Rather, if two
clocks disagree about the relative length of two durations, we
look for a causal explanation of the changed behaviour of at
least one of them—for example, if clock X measures A-B and
B-C as equal, while clock T measures A-B as shorter than B-C,
we might explain this by finding that the pendulum of T has
55
Principia (ist edition), p. 5: Tempus absolutura verum et Mathematicum,
in se et natura sua absque relatione ad externum quodvis, aequabiliter fluit,
alioque nomine dicitur Duratio.
56
Ibid.: relativura apparens et vulgare est sensibilis et externa quaevis Dura-
tionis per motum mensura, (seu accurata seu inaequabilis) qua vulgus vice veri
temporis utitur.
" Ibid., p. 7: Possibile est ut nullus sit motus aequabilis quo Tempus accurate
mensuretur.
 KANT AND TRANSCENDENTALISM 115
contracted because it became colder round about time B, or
that X, which is driven by a mainspring, was fully wound at A
and has since been running down. We correct the measure-
ments supplied by a clock where we find a causal explanation
for a change in its behaviour; that is, we no longer take its
rotations or oscillations or whatever it may be as simple and
direct indications of the equable flow of absolute time.
This is what we in fact do. It is not, indeed, the only thing
that we could do. It has a more plausible claim than any
alternative procedure I can think of would have to give us
progressive approximations to accurate measurements of abso-
lute duration. There are admittedly still problems about the
philosophical interpretation of what we thus achieve. But this
procedure illustrates, far more strikingly than anything to
which Kant draws attention, the way in which causal notions
and causal discovery compensate for the lack of a visibly
graduated Newtonian time-axis. But this too, it seems to me,
shows not that causation is in any worthwhile sense a priori or
transcendentally guaranteed, but only that it is a very important
pervasive structural feature of our actual experience.
We might make one other small move in a Kantian direction.
Though I insisted that Kant's second or ordering argument,
taken empirically, misrepresents our intellectual history, and
that in that history we are helped in arriving at our view of the
world, in discriminating between processes and persisting com-
plex objects, by the actual 'subjective' order of impressions, we
may concede that we are helped also by certain innate tendencies
or propensities in favour of some kinds of simplicity and regu-
larity. An infant looking round a room has, it is plausible to
suppose, some inborn reluctance to interpret what it sees as
disorderly sequences—sometimes window followed by wall fol-
lowed by door, sometimes door followed by wall followed by
window, and so on. If the impressions come like this, they are
more readily interpreted as showing one door, one window,
and so on, each persisting much as it is; but if the infant has
impressions that fit into the pattern of the same sort of process,
the same type-sequence of phases, repeated on different occa-
sions, then it will interpret these impressions as of processes.
There may well be something, then, that could be described
as a priori forms in terms of which empirical impressions are
n6 KANT AND TRANSCENDENTALISM
interpreted and ordered. But this kind of apriority is no
guarantee of necessary truth. The causal beliefs that result
when our actual sequences of impressions are interpreted in
accordance with these a priori forms might still be 'bastards of
the imagination fathered by experience'. Before we can say
whether we can add to such innate propensities any genuine
presumption in favour of some kind and some degree of regu-
larity we must pay attention to those aspects of the causal
relation which Kant brushed aside: how, if at all, causal laws
differ from other universal truths; the link between individual
cause and effect; and the direction of causation. When we have
thus clarified the concept whose applicability is at issue we shall
be better able to decide whether it is only an arbitrary pre-
judice, a quirk of our imaginations, to suppose that such rela-
tions actually hold.
 5
Common Sense and the Law

ONE of the best studies that have been made of ordinary causal
concepts is that by Hart and Honord in Causation in the Law.
Their discussion deals not only with legal applications of causal
notions, but also with the common-sense concepts and their
ordinary use, concentrating particularly on singular causal
statements. It will therefore be a good test of the account which
I have tried to develop in Chapters 2 and 3 to see how well it
can either accommodate the points made by these authors or
defend itself against them.
Hart and Honore" stress 'the possibility that the common
notion of causation may have features which vary from context
to context, that there may be different types of causal inquiry,
and that there may not be a single concept of causation but
rather a cluster of related concepts';1 they argue, in fact, that
these possibilities are realized.
In this cluster, the fundamental division is between those
ordinary causal statements which assert 'that one physical
event was caused by another such event or by a human action',
and 'statements of interpersonal transactions [which] involve
the notion of a person's reason for acting'. For the first class
(which I shall refer to as physical) Mill's analysis is often at least
approximately correct, but for the second (which I shall call
interpersonal) something radically different is needed. However,
there are also 'cases where causal language is used to indicate
relationships different from either of these two main types but
with important affinities to them'.
Where Hart and Honore" find only a cluster of related con-
cepts, I would rather say that there is a single basic concept of
causing to which various frills are added: there is one common
1
Causation in the Law, p. 17.
 ii8 COMMON SENSE AND THE LAW
kind of causing but with different accompaniments. Since we
have no well-established rules for counting concepts, this may
seem to be a dubious distinction; but we shall find that it
covers some substantial differences.
Hart and Honore" recognize Mill's improvements on Hume—
in admitting standing conditions, as opposed to events, as
causes, and in allowing for assemblages of conditions and for
plurality of causes, with the resulting complexity of generaliza-
tions. But they insist, rightly, that 'Mill's standard of "invariable
and unconditional sequence" cannot be met', that his doctrine
'presents an idealized model' even for the physical causation to
which it is more or less appropriate.2 Our account has allowed
for this by noting, first, that causal regularities as known are
elliptical, and, secondly, that Mill's requirement of uncondi-
tionality was a misguided attempt to deal with the difficulty
that regularities of succession may fail to be causal, the correct
resolution of which involves the direction of causation.
Hart and Honor6 say that 'Mill's description of common
sense "selecting" the cause from a set of conditions is a suggestio
falsi so far as ... simple causal statements are concerned; for,
though we may gradually come to know more and more of the
conditions required for our interventions to be successful, we do
not "select" from them the one we treat as the cause. Our
intervention is regarded as the cause from the start before we
learn more than a few of the other necessary conditions. We
simply continue to call it the cause when we know more.'3 This
is at least part of the truth; it is allowed for both by the analysis
suggested in Chapter 2, that a cause is 'necessary (and perhaps
also sufficient) in the circumstances' whatever those circum-
stances may be, and by the thesis of Chapter 3 that causal
regularities as known are elliptical, but in varying degrees.
They also argue that 'The line between cause and mere condi-
tion' (whose philosophical importance Mill denied) 'is ...
drawn by common sense on principles which vary in a subtle
and complex way, both with the type of causal question at
issue and the circumstances in which causal questions arise . . .
two contrasts are of prime importance . .. [those] between what
is abnormal and what is normal in relation to any given thing
2
Causation in the Law, pp. ai, 41.
3
Ibid., p. ag; the point is also made more fully on pp. 42-3.
 COMMON SENSE AND THE LAW 119
or subject-matter, and between a free deliberate human action
and all other conditions'.4 Developing the first of these they
say that '. . . normal conditions (and hence in causal inquiries
mere conditions) are those . . . which are present as part of the
usual state or mode of operation of the thing under inquiry:
some . . . will also be familiar, pervasive features of the environ-
ment, . . . present alike in the case of disaster and of normal
functioning'. By contrast, 'What is abnormal . . . "makes the
difference" between the accident and things going on as usual.5*
Features which are normal in this sense are the ones which in
Chapter 2 above were relegated to the causal field, and there-
fore not allowed to count as causes. Hart and Honor6 stress the
relativity to context of this contrast between the normal and
the abnormal.6 'If a fire breaks out in a laboratory or in a
factory, where special precautions are taken to exclude oxygen
during part of an experiment or manufacturing process, since
the success of this depends on safety from fire, there would be
no absurdity at all in such a case in saying that the presence of
oxygen was the cause of the fire. The exclusion of oxygen in
such a case, and not its presence, is part of the normal function-
ing.' Again, 'The cause of a great famine in India may be
identified by the Indian peasant as a drought, but the World
Food authority may identify the Indian government's failure
to build up reserves as the cause and the drought as a mere
condition.' As I have put it, different fields may be chosen for
causal accounts of the same event. What is normal may depend
upon man-made norms: the gardener's failure to water the
flowers caused their dying, being a deviation from a routine
practice as well as a duty, but the failure of other people to
water the flowers was normal and hence not a cause.7 Hart and
Honore1 also show that 'A deliberate human act is ... most
often a barrier and a goal in tracing back causes ... it is some-
thing through which we do not trace the cause of a later event
and something to which we do trace the cause through inter-
vening causes of other kinds.'8 The presence of arsenic in a man's
body caused his death, but someone's putting the arsenic in his
food (which caused this cause) also caused his death, though
causing is not always taken as transitive. (If lightning caused a
5
+ Ibid., p. 31. Ibid., pp. 33-3. « Ibid., p. 33.
' Ibid., pp. 35-6. * Ibid., p. 41.
lao COMMON SENSE AND THE LAW
fire, the atmospheric conditions which caused the lightning are
not said to have caused the fire.) On the other hand, even if
something caused the poisoner to act as he did, we would not
say that this something caused his victim's death.
Now these are quite real distinctions, and they are important
for an analysis of the ordinary uses of causal terminology, which
in turn may be, as Hart and Honord argue,9 rightly relevant to
legal decisions, though they are not the only considerations that
affect the limits of responsibility. But I would stress that though
our account has not developed them, it has left room for them,
in noting that 'an alleged condition which is not called a
cause . . . either is part of the field presupposed in the view
taken by the speaker . . . or is a cause, but mention of this
fact happens to be irrelevant, or less relevant than mention of
some other cause of the same result, to some current purpose'.10
Deliberate human actions are particularly relevant as causes
just because they are the focus of interest with respect to
responsibility and various forms of control. But since even the
choice of a field is relative to a purpose or a point of view, and
since even apart from this what we recognize as a cause, rather
than a mere condition, commonly depends on what we know—
or what we knew first—or what is closely related to our interests,
there is much to be said for Mill's refusal to distinguish 'philo-
sophically speaking' between causes and conditions. As an
analysis of ordinary language, this would be wrong; but from
a theoretical point of view, as an account of causal processes
themselves, it would be right. What is not a cause in relation to
one field may be so in relation to another. A deliberate human
act is related to a physical result that it helps to bring about in
just the same way that other factors may be, namely as 'neces-
sary (and perhaps sufficient) in the circumstances', as an in-
stance of an inus condition, and as being causally prior to that
result.
Of much greater moment is the claim made by Hart and
Honord that statements of interpersonal transactions, involving
the notion of a person's reason for acting, incorporate a quite
different causal concept from that of which Mill's analysis is
approximately correct. But their main ground for this claim is
that generalizations play a different role in physical and in
I0
» Especially in their Chapters 3 and 4. Chapter a, above.
 COMMON SENSE AND THE LAW lai
interpersonal cases '. . . the assertion that one person, for
example, induced another to act is not "covertly" general even
in the modified sense discussed in Section IV. Generalizations
have a place here but a less central one.'11 I think, however,
that this is a false contrast, and that it is due to their having
conceded too much to Mill in the physical sphere. Singular
causal statements in this sphere are not, I shall argue, covertly
general even in the modified sense allowed by Hart and Honored
Criticizing Hume, Hart and Honore" say that 'the lawyer
approaches the general element inherent in causal statements
in a different way, which disguises it: when it is suggested that
A is the cause of B he is apt to ask as the first question, would B
have happened without A? And though in fact, in order to
answer this question, we have ... to use our general knowledge
of the course of nature, it looks as if this general knowledge
were just part of the evidence that a particular event was the
cause of another, and not constitutive of the very meaning of
the particular causal statement.'12 Yes; but it is not only the
lawyer who is apt to ask this first question, but also the ordinary
man, and (as we have seen) even Mill, and even, in an un-
guarded moment, Hume himself. It is the negative answer to
this question, and nothing either explicitly or implicitly general,
that I have put at the centre of the analysis of singular causal
statements. I have shown (in Chapter 3) that it is only in very
tenuous senses that any singular causal statements are implicitly
general. Also, granted that general knowledge may somehow
be part of the evidence that this particular event of type A
caused that particular event of type B, what general knowledge
plays this role, and just how does it work? I have argued in
Chapter 3 that an elliptical generalization, saying that A is an
inus condition (or better) of B, will sustain the counterfactual
conditionals involved in this singular causal statement. If an
appropriate elliptical generalization is known, it will indeed
make it easier for us to interpret this particular A-B sequence
as causal, and I grant that in physical cases such generaliza-
tions are often known in advance. As Hart and Honore" say,
'When we assert that A's blow made £'s nose bleed or A's
exposure of the wax to the flame caused it to melt, the general
knowledge used here is knowledge of the familiar "way" to
" Causation in the Law, p. 48. ™ Ibid., p. 14.
iza COMMON SENSE AND THE LAW
produce . . . certain types of change.'13 But, as I have shown,
even in physical cases no such generalization need be known in
advance: the elliptical generalization can be discovered and
(tentatively) established by the observation of the very same
individual sequence the conditionals about which it may be
said to sustain. We do not, then, need it as evidence. The
generalizations that are far more vital as providers of evidence
are those which inform us of the irrelevance of various other
changes in the spatio-temporal neighbourhood of the observed
sequence, for example, that B's nose is unlikely to have been
made to bleed by C"s whistling or by the cold draught as D
opened the door, or that the wax is unlikely to have been
melted by the mere movement that brought it near to the
flame. This distinction is vital, because general knowledge
related in the same way to the singular causal judgement is
relevant even in the interpersonal cases. How do we know that
it was A's threat that induced B to open the safe? It is not
enough that A's gesture with a loaded revolver gave B a reason
for opening the safe; we want to say that this was his reason for
doing so. This causal statement too will have an analysis that
includes the counterfactual 'B would not have opened the safe
on this occasion, in these circumstances, if A had not threatened
him'. To be able to say this, we need to know that B is not
prone to open the safe gratuitously in the presence of strangers
or in response to an unauthoritative request. This is general
knowledge that we probably have about B, and that B almost
certainly has about himself, and both B himself and we as
spectators are implicitly relying on such knowledge when we
say that it was the threat which induced him to open the safe.
It may be objected that B himself, at any rate, has no need to
appeal even to this general knowledge, that he knows that
until A produced the revolver he had no thought of opening
the safe, and so knows directly that on this occasion he would
not have opened it if the revolver had not been displayed,
whatever he may or may not have done, or might or might
not do, on other occasions. We can concede this; but this is no
more than an example of the imaginative analogizing, the auto-
matic transfer from an earlier situation to a reconstructed later

" Causation in the Law, p. 29.

 COMMON SENSE AND THE LAW 123
one, which we have already recognized as an unsophisticated
way of arriving even in physical cases at the counterfactual
conditional 'If in these circumstances X had not occurred, Y
would not'. Not only B himself but we as spectators can think
in this way in an interpersonal case; but so can we in a physical
case. And just as we may need, in a physical case, to back up
this unsophisticated imaginative performance by reliance on
the general knowledge that regularly when in similar circum-
stances no event of type X occurs, no event of type T occurs
either, so in the interpersonal case B himself, and a fortiori we
as spectators, might need to back up our immediate imagina-
tively based judgement with the general knowledge that B does
not open safes gratuitously. It is true that we need no generaliza-
tion about B's regularly opening safes when threatened; but equally
we do not need the corresponding generalization in the physical
case. The general knowledge that a blow can make a nose bleed
facilitates our interpretation that on this occasion A's blow made
B's nose bleed, but we can interpret a quite novel sequence as
causal if we have the general knowledge which with the help
of observation rules out rival explanations. Hart and Honore"
themselves concede that 'what we recognize as a possible reason
is not independent of how in a very broad sense most people
a c t . . . we recognize as a reason for action . . . something which
is relevant to some purpose known to be pursued by human
beings'.14 Indeed, I think they concede too much here. Given
appropriate evidence, we could recognize a novel human pur-
pose, and hence could recognize something quite bizarre as
giving someone a reason for acting. The correct point is merely
that our general knowledge of human purposes facilitates inter-
personal causal interpretations of what we observe, and this is
strictly analogous to what we can say about physical cases.
Once we see just how general knowledge of different sorts may
or may not come in, we find no systematic difference between
physical and interpersonal types of case which would justify us
in speaking of two different causal concepts. In both spheres,
the elliptical generalizations which tell us that Xs can cause Ts,
that the X event-type is an inus condition of the T event-type,
aid the interpretation that this X caused this T; in neither sphere
is advance knowledge of such a. generalization required. In both
«« Ibid., p. 53.
184 COMMON SENSE AND THE LAW
spheres the general knowledge of the other sort, that which
tells us that the other features of the situation were not sufficient
to produce an event of type T, may be needed to back up the
causal interpretation; yet in both spheres an initial causal
judgement is often made in an unsophisticated way, without
actual use of such general knowledge, the 'T would not have
occurred' being reached by transfer, by imaginative analogiz-
ing, from like, especially immediately preceding, states of
affairs. In other words, once we complete the job of refining
and correcting Mill's account of causal statements and causal
relations in the sphere in which that account is recognized by
Hart and Honored to be appropriate to some degree, we enable
it to cover also causal statements and causal relations in the
interpersonal sphere, where Hart and Honore1 think that some-
thing radically different is required.
I am, of course, here verging upon a well-known group of
controversial topics, whether desires and other 'motives' can be
said to cause the correlated actions, whether an explanation of
an action in terms of the agent's reasons is a causal explanation,
and so on. I think the affirmative answer to these questions is
correct, and that the popular arguments for the opposite view
are inadequate, and indeed often rest on an inadequate account
of causation. That is, I think that these arguments will be in-
directly weakened, undermined, by the account of causation
which I have been developing. However, I am now directly
concerned not with these questions, but with the rather different
question whether if A induces B to do something by offering
him a reward or a bribe, or by threatening him, A's doing
what he does causes B's doing what he does in the same sense in
which a blow causes a nose-bleed. For the rival view of Hart
and Honore" is not that such inducing is non-causal, but merely
that it exemplifies a different causal concept. In rejecting their
view, I am not of course saying that there are no systematic
differences between 'physical' and 'interpersonal' cases. In any
variety of inducing, the result comes about by way of the second
agent's desires and choices, and some desire or at least readi-
ness to desire and some capacity in this agent to understand
the offer or threat must be either part of the field or conjoined
factors in the assemblage of conditions, and nothing like these
will be operative in ordinary physical causation. What I am
 COMMON SENSE AND THE LAW 195
saying is that inducing is still causing in the same basic sense,
but with these additional features.
Hart and Honore" mention, as another kind of exception,
even in the 'physical' sphere, to the Millian account of causa-
tion, 'cases . . . where an omission or failure to act is identified
as the cause of some (usually untoward) event'. As they say,
'Our concern here is to show that the omitted precaution would
have averted the harm, not that when it is omitted harm always
results.'15 But this is an exception only to the way that we have
already rejected of incorporating generalizations in the analysis.
An omission that is recognized as a cause is 'necessary in the
circumstances' in just the same way that a positive act or event
or state is: to say that the precaution would have averted the
harm is to say that if the omission had not occurred, the harm
would not have occurred either.
Hart and Honore" discuss another kind of interpersonal trans-
action, the giving of opportunities. 'A man who carelessly
leaves unlocked the doors of a. house . . . has provided the
thief with an opportunity to enter and steal the spoons though
he has not caused him to steal or made him steal.'15 The leaving
of the doors unlocked may well have been necessary in the
circumstances for the stealing of the spoons, and causally prior
to it, it will then be a (partial) cause of the latter in terms of our
analysis; is this not, then, a difficulty for our view? I should
reply that this is another instance of the same basic causal
relationship; in a philosophically interesting sense it is causing,
the problem is merely why we are reluctant to use causal
language here. In fact we should be happy to say that the
stealing of the spoons was a (causal) consequence of the leaving
of the doors unlocked, though we should not say that it was
an effect of this, or, as Hart and Honore" point out, that it
caused the thief to steal or made him steal. Surely these last
expressions are avoided because they too strongly suggest
inducement or coercion or overwhelming impulse, whereas in
this case the thief had, presumably, already set out to steal
something when he came upon the unlocked door. And there
are, as Hart and Honord point out, fine distinctions between
'consequence', 'effect', 'result', and so on. 'The use of the term
"effect" is ... fairly definitely confined to cases where the
« Causation in the Law, p. 47. «' Ibid., p. 55.
ia6 COMMON SENSE AND THE LAW
antecedent is literally a change or activity . . . and where the
event spoken of as the effect is a change brought about in a
continuing thing or person. . . . "Consequences" has a much
wider application, . . . [but] it is not normally used of the
terminus of a very short series of physical. . . changes.'17 Along
these lines we may perhaps be able to account for the differential
use of particular causal words and phrases. It would be possible
to speak of distinct sub-concepts of effect, consequence, and the
like, but unless these differential usages showed a fair degree of
persistence through time and over a multiplicity of dialects, and
had close counterparts in a number of languages, to speak even
of sub-concepts would be misleading and not very useful. But
however we describe these fine distinctions, what matters is that
they rest upon features which are additional to the basic causal
relationships, and do not replace these: How long was the
series of physical changes ? How hard would it have been for
the agent to do otherwise? Did the interpersonal cause start
what we see as a single course of action or feed into it some way
along? The only trouble is that the very word, 'cause', which we
most naturally adopt for the common basic concept, tends to be
already used for what is at best a sub-concept. But this is merely
linguistically embarrassing, not philosophically significant.
However, in trying to formulate this basic concept I have
used such phrases as 'necessary in the circumstances'. Hart and
Honord argue that there are (at least?) 'three different ways,
hidden by the general expression "necessary", in which a cause
may be said to be a necessary condition of the consequence'.18
If so, perhaps my phrase gives a misleading impression of unity.
The three senses are, first, that a condition may be 'one of a
set of conditions jointly sufficient for the production of the con-
sequence : it is necessary because it is required to complete this
set'; secondly, it may be 'a necessary member of every set of
sufficient conditions of a given type of occurrence'; thirdly, it
may be necessary on the particular occasion, whereas what is
necessary in the first sense may, through over-determination
(fail-safe causes), not be necessary on the particular occasion.
The second sense is not relevant: a type-event which is a
(partial) cause of some other type-event may in fact be neces-
sary for it in this stronger sense, but it is never required to be
17 l8
Causation in the Law, pp. 25-6. Ibid., pp. 106-7.
 COMMON SENSE AND THE LAW 137
so, in any established concept of a cause. I have argued, in
effect, in Chapter 2 that the third is the key sense of 'necessary
in the circumstances', that the first is not strong enough, and
I have shown how we can accommodate those cases of over-
determination where we do regard one factor rather than
another as the cause. This plurality of senses of 'necessary',
therefore, does not undermine the unity of my account.
There is a major issue in legal theory with which Hart and
Honore" are concerned. Granted that not everything which
would count as a (partial) cause of some outcome in terms of
what I have called the basic common concept is actually
recognized as a cause, or still less as a 'proximate' as opposed
to a 'remote' cause, is such an item's being recognized as a
(proximate) cause prior or posterior to a legal decision? The
traditional view is that there is some genuine issue, prior to any
legal decision or policy, about whether something that was
admittedly a condition sine qua non of the outcome caused it or
not, or was or was not a proximate cause. The 'modern' view
is that whether a condition sine qua non counts as a (proximate)
cause is always a matter of legal policy. The difference is
brought out by two examples. 'In the State of New York . . .
if a fire negligently started spreads to buildings, damages may
be recovered only in respect of the first of the buildings affected.'
But the claim of the owner of even the first house against the
negligent fire-starter would be rejected if 'it transpired that
just as the fire was flickering out a third party deliberately
poured paraffin on the embers'.19 On the modern view, the
rejection of a house-owner's claim on either of these two grounds
(that it was the second house to which the fire had spread, or
that a third party had intervened with the paraffin) is simply
a matter of legal policy, the fire-starting being a condition sine
qua non of the actual damage in either case. But on the tradi-
tional view, while a New York court's rejection of a claim on
the first ground would be a matter of legal policy, a claim
would be rejected on the second ground because in the case
described the original fire was not the cause of the damage
(the paraffin-pouring was). While they recognize that modern
criticism has produced 'advances towards simplicity and
clarity',20 Hart and Honors' still find some merit in the
2I>
«» Ibid., p. 84. Ibid., p. 103.
 iaB COMMON SENSE AND THE LAW
traditional view. And clearly there are differences between the
two kinds of case. Though we can see the point of the New York
doctrine, it would be in no way surprising if other jurisdictions
followed different rules, and allowed damages to be claimed
in respect of buildings other than that first affected. But it
would be somewhat more surprising if courts anywhere awarded
damages against a man who negligently started a fire which
would have died out harmlessly unless someone else had deliber-
ately poured on paraffin. Yet is this not still to be explained in
terms of a policy, though one that is broader, more fundamental,
and more widely embraced? In any such case we are looking
for someone to blame (and to bill) for the damage, and a
deliberately harmful act, being both more open to deterrence
and seen as worse than a negligent one, attracts to itself blame
that might initially be shared between them. The negligent
starting of the fire was in the actual course of events a non-
redundant factor, and hence undeniably, by our account, a
(partial) cause of the damage. But we cannot ignore the fact
that it could easily have been replaced: if the paraffin-pourer
had not found some convenient embers, he would presumably
have struck a match. Though the negligent act and the deliber-
ate one were both partial causes of the damage, very natural
and widespread human purposes will lead us to pin the cost
upon an act which, if it had not occurred, would not have been
likely to be replaced in the causal network rather than upon
one for which a substitute would almost certainly have been
found.
We can in this way resolve the question whether one cause
or causal factor can have more weight than another.21 If two
factors are each necessary in the circumstances, they are equally
necessary; one inus condition cannot be more of an inus condi-
tion than another. But, as the last example shows, one can be
more easily replaceable than another. The elimination of one
11
In 'Causes and Conditions', American Philosophical Quarterly, ii (1965), 353,
I said: 'Each of the moments in the minimal sufficient condition, or in each
minimal sufficient condition, that was present can equally be regarded as the
cause. They may be distinguished as predisposing causes, triggering causes, and
so on, but it is quite arbitrary to pick out as "main" and "secondary" different
moments which are equally nonredundant items in a minimal sufficient condition,
or which are moments in two minimal sufficient conditions each of which makes
the other redundant.' This view has been criticized but it can be defended in the
sense explained here.
 COMMON SENSE AND THE LAW 129
might have diminished the likelihood of the result (as seen from
some ordinary point of view incorporating less than complete
knowledge of the situation) more than would the elimination
of the other. Also, time differences may well be important for
some purposes: the 'last clear chance rule' has been used in
law to assign responsibility for negligence to 'the party who
failed to take the last clear chance of avoiding the harm'.22
Hart and Honore" rightly argue that this rule, as applied by
the courts, does not embody any common-sense principle of
causation; it goes beyond the principle that a later voluntary
or reckless act, or gross or extraordinary negligence, 'negatives'
a causal connection between someone else's prior conduct and
subsequent harm. But all such judgements can be seen as the
expression of various purposes and policies: in themselves all
actual causal factors are equally causes of the result for which
each was necessary in the circumstances, but from some human
point of view one such factor may be selected as the more
important cause, or even as the sole cause.23
But other examples where the courts have, with some reason,
distinguished between a condition sine qua non and a cause
require a different treatment. Hart and Honore" quote a case
where the owners 'in breach of a Hongkong ordinance sent a
ship to sea without duly certificated officers. The master, though
uncertificated, was a perfectly competent man of long experi-
ence, but the ship was in fact involved in a collision when he
was the officer of the watch and guilty of negligent navigation.'
The judge held that the owners' breach of the ordinance was
a cause of the collision 'in the sense of it being a causa sine qua
non . . . because . . . Sinon, the uncertificated officer, was in fact
the officer of the watch at the time . . .', but that since he was,
22
Hart and Honore, op. cit., pp. 201-7.
2
' On the other hand quantitatively measurable effects and the functional
dependence concept, to be discussed in Chapter 6, do allow a more objective
weighting of causes. It could be that three-quarters of the increase in the cost of a
product was due to a rise in the price of raw materials and only one-quarter to
a rise in wages.
Less coherent selective judgements can also be found. Thus a psychiatrist who
discovered that 'a widely used anti-depressant drug could cause temporary loss of
speech in some patients' is reported (Guardian, 20 December 1979) as saying: 'All
but one of the patients recovered when treatment was stopped . . . reports of
patients being "struck dumb" are alarmist and sensational. . . . It is not the fault
of the drug but idiosyncrasies of certain patients. Vast numbers of people have the
drug without any effect at all.'
130 COMMON SENSE AND THE LAW
by all the evidence, competent, there could not be 'any causal
connection between the fact of his not having a certificate and
the fact of his negligent navigation'.24 What this amounts to is
that Sinon's being in charge caused the collision, and he was
an uncertificated officer; but it was not an uncertificated
officer's being in charge that caused the collision. His being
uncertificated would have been causally relevant if it had meant
that an incompetent man was in charge, and this had caused
the collision, but on the evidence this was not so. We are here
verging upon the question (to be examined in Chapter 10) of
the extensionality of causal statements and the related question
whether the phrases that identify events or states which enter
into causal relationships are to be read as perfect or as imperfect
nominalizations of the corresponding verbal forms. A concrete
event describable by the perfect nominalization 'an uncerti-
ficated officer's navigating' did cause the collision, but it was
not its being an event of this type that caused it, as would be
asserted if we said, with an imperfect nominalization, 'an un-
certificated officer's having navigated caused the collision': the
event's being of this type was causally irrelevant. But it is quite
misleading to draw a distinction here between a cause and a
condition sine qua non. The actual man's, Sinon's, navigating
was both a cause and a condition sine qua non; that this concrete
event was of the type an uncertificated officer's navigating was neither
a cause nor a condition sine qua non; the judge's reasonable
decision drives no wedge between a factor's being necessary in
the circumstances and its being recognized as a cause either by
common sense or by the law.
Another such wedge is suggested by Hart and Honore" in
their discussion of accidents caused by a motorist's excessive
speed.25 Where the speed leads through traceable stages to the
driver's losing control and hence to a collision, or where it
accounts for the great violence of the impact and so for the
amount of injury or damage, there is no difficulty: the speed
is said without qualification to have caused the accident or the
harm. But it is different where a driver's excessive speed was
a condition sine qua non of an accident only because, if he had
driven more slowly, he would not have arrived at the critical
time at the place where this particular accident occurred. Here
24
Causation in the Law, pp. 112-13. " Ibid., pp. 114-16.
 COMMON SENSE, AND THE LAW 131
we do not want to say, for legal purposes, that the speeding
caused the accident; yet in terms of our basic analysis, it did.
Hart and Honore* quote the view of the American courts that
in these problematic cases the earlier speeding was not causally
connected with the accident, sometimes backed up by the argu-
ment that there would have been no accident if the driver had
gone faster still! It is true that in a case of this sort it is not
exceeding the speed limit as such which has caused the accident,
but exceeding it to just that particular degree; but to deny a
causal connection here is absurd. However, I find equally un-
satisfactory the view of Hart and Honore" that the right distinc-
tion between the causally necessary condition and the mere
condition sine qua non is that in the problematic, mere sine qua
non, cases 'We could not show that the sequence of events,
even when broken down into stages, exemplified some trace-
able general connexion between speeding and [accidents]'.
Admittedly these cases do not exemplify any general connection
between speeding and accidents; but each stage of the process
does exemplify some general connection, for example between
being at one place at one time, moving at eighty miles an hour,
and being at a place four miles distant three minutes later, and
again between one car's meeting another at an intersection and
both cars, and their occupants, being damaged. But in any
case I have argued that the exemplifying of a broad generaliza-
tion is not an essential feature of even physical causal con-
nections. What Hart and Honore themselves describe as a
sensible way of dealing with the matter, but then unaccount-
ably leave on one side, is to say that 'the purpose of rules against
speeding is to protect persons from injuries arising in certain
familiar ways which would normally be regarded as the risks
attendant on speeding'. In other words, although a statute may
verbally prescribe special penalties (and perhaps special rights)
in cases where accidents and injuries are caused by excessive
speed, it is intended only to guard against (or to compensate
for) damage and injuries that arise from speeding in the direct
and familiar ways. In the bizarre cases the speeding is still a
cause, as well as a condition sine qua non, of the damage, but
not in the way with which the statute is concerned.
Hart and Honore" may well be right, as we saw earlier, in
drawing attention to ways in which ordinary usage restricts the
132 COMMON SENSE AND THE LAW
term 'cause' more narrowly than our basic analysis, and in
saying that the law should take some account of these ordinary
restrictions. But while these restrictions will not, of course, be
relative to specific local legal policies, they still depend upon
the broader human interests and purposes that commonly give
point to causal statements, and they seem not to drive any
philosophically important wedge between a cause and what is
necessary in the circumstances, over and above the refinements
of that notion that we considered in Chapters 2 and 3.
One further matter of interest is the contrast that Hart and
Honore draw between the use of the general verb 'cause' and
such simple transitive verbs as 'push, pull, bend, twist, break,
injure'. We use simple verbs where the process 'consists of an
initial immediate bodily manipulation of the thing affected
and often takes little time', but we speak of 'cause' and 'effect'
where the effect is a secondary change brought about by some
primary change in what we directly manipulate: 'we cause one
thing to move by striking it with another'.26 It may be true
that we tend not to use the words 'cause' and 'effect' of these
immediate manipulations, but the converse thesis would not
hold: there are plenty of simple transitive verbs which describe
the indirect bringing about of 'secondary' changes: 'kill',
'poison', 'cure', 'calm', 'bake', 'clean', 'water', and a host of
others. I can see nothing wrong in saying that whenever any
of these simple verbs is applicable, some causing is going on or
has gone on, and that this is causing in accordance with our
basic analysis. If A kills B, A does something which causes jB's
dying; if A waters the garden, A does something which causes
the garden's becoming wetter; if A bends a twig, he makes a
movement which causes the twig's ceasing to be straight. We
have simple verbs for some of these transactions but not for
others, and where we have a simple verb it will usually be
clumsy and pedantic to use the word 'cause'. But nothing vital
turns upon these linguistic differences.
Simple transitive verbs may, however, throw light on philo-
sophical questions about causation in another way. Problems
that have been with us since Hume include these: What is the
intimate tie between an individual cause and an individual
effect? Where and how, if at all, can we perceive causation?
16
Causation in the Law, pp. 26-7.
 COMMON SENSE AND THE LAW 133
It may be argued that these problems evaporate, or admit of
startlingly simple solutions, if we attend to the use not of the
general verb 'cause' but of the simple transitive verbs, especially
those of which Hart and Honore" were speaking. What need is
there to look for a further 'tie' or 'link' between my hand and
the cup it pushes? The coupling between the railway engine
and the truck it pulls is clearly visible. Hume has taught us to
look in vain for the power and necessity in causing, but is it
so hard to see a stone break a window? Thus R. F. Holland
says: 'Our understanding, then, of "the link between cause and
effect" is basically an understanding of physical contacts in all
their various kinds—-jolting, pushing, gashing, rubbing, gnaw-
ing, scooping, severing, grappling, getting entangled with,
absorbing, falling on top of, crushing, mixing, irradiating, and
so on (it is to be noticed that our words for the various sorts
of contact mostly indicate a pattern of motion and a degree of
force).'27 There is no doubt that it is in and from observations
and experiences of this sort that we get at least part of our
understanding of causation, but we can still ask just what
understanding we get, and how. If I see someone else peel a
potato, I see the relative movement of the potato and the knife,
I see the gradually lengthening strip of peel come up from the
surface of the potato, each new portion coming up just as the
leading edge of the knife reaches it; but as Hume would still
rightly protest, I don't see the knife making the peel come up.
And what I most obviously fail to see, though I do judge, is
that each bit of the peel would not have come up if the knife
had not moved in there. But what if I peel the potato myself?
Then in addition to seeing all that has been mentioned, I feel
my fingers pressing on the knife and on the potato, and I may
guess that the knife is also pressing on the potato and the potato
on the knife. As Holland says, there is force here as well as
motion. But to feel forces, and even to transfer them, in thought,
from one place to another (from the finger-knife and finger-
potato surfaces to the knife-potato surface) is still not to per-
ceive a counterfactual conditional. To arrive at this we still
need either the sophisticated or the unsophisticated procedure
discussed in Chapters 2 and 3, and the shift of attention from
" From an unpublished paper. Similar points are made by G. E. M. Anscombe,
Causality and Determination, pp. 8-9.
 134 COMMON SENSE AND THE LAW
causing in general to the various processes described by simple
transitive verbs has not diminished this need. It sounds absurd
to say, in the spirit of Hume, that I do not perceive myself
peeling a potato, but only imagine this, even in the most
favourable circumstances. But it is not absurd, but true, to say
that an element either of imagining or of discursive thinking
enters into what we would ordinarily call my knowledge that
I am peeling a potato. We should not fall back from the correct
parts of Hume's critical analysis to a mere acceptance of
common sense.
Admittedly we can, in the most ordinary sense of 'see', see
a stone break a window. But this seeing, like almost all seeing,
includes an element of interpretation. And, of course, we might
be wrong. It might be that a vacuum somehow suddenly pro-
duced on the inside of the window made it collapse a tenth of
a second before the stone arrived. The point of mentioning such
fantastic possibilities is not to promote scepticism, not to recom-
mend the conclusion 'Perhaps the stone did not really break the
window' on some perfectly ordinary occasion, but merely to
bring out the fact that something which we can call interpreta-
tion is involved, and to leave room for a further inquiry into
its nature, which the simple adherence to the common-sense
judgement or the ordinary use of language might preclude.
Whereas Holland deprecates the search for a general analysis
of causal statements and insists that understanding is to be
found in the variety of ordinary physical contacts, a general
theory of causation that incorporates some of the same prin-
ciples has been advocated by Ducasse.28 His main themes are
'that the correct definition of the causal relation is to be framed
in terms of one single case of sequence, and that constancy of
conjunction is therefore no part of it, but merely, under certain
conditions, a corollary of the presence of the causal relation',
and that this relation 'when correctly denned, is as directly
observable as many other facts, and that the alleged mysterious-
ness of the causal tie is therefore a myth due only to a mistaken
notion of what a tie is'.29
28
C. J. Ducasse, Truth, Knowledge and Causation, Chapters i to 4. The first of
these chapters reprints a paper 'On the Nature and the Observability of the
Causal Relation', which was originally published (in Journal of Philosophy, xxiii)
in 1936.
19
Truth, Knowledge and Causation, p. i.
 COMMON SENSE AND THE LAW 135
Ducasse assumes that 'if the occurrence of a particular change
sufficed to the occurrence of [another] it is then said to have
caused that other', and suggests that this 'sufficing' is to be
defined as follows:
Considering two changes, C and K (which may be either of the same
or of different objects), the change C is said to have been sufficient
to, i.e. to have caused, the change K, if:
1. The change C occurred during a time and through a space
terminating at the instant / at the surface S [the limit of a change of
a solid being a surface, not a point].
2. The change K occurred during a time and through a space
beginning at the instant / at the surface S.
3. No change other than C occurred during the time and through
the space of C, and no change other than K during the time and
through the space of K.
More roughly, but in briefer and more easily intuited terms, we
may say that the cause of the particular change K was such particular
change C as alone occurred in the immediate environment of K immediately
before.™
Several things are to be noted about this definition. Ducasse's
first thesis, that causation is to be defined first for the singular
sequence, and that any constancy or regularity is a secondary
matter, anticipates what I have argued in Chapter 3. He is also
right in arguing that causation can be detected in a single
sequence: 'one single experiment is sufficient in principle to make
evident that [one particular change] did cause [another]. For
repetition of the experiment is called for, if at all, only to make
sure that the experiment was conducted strictly and its outcome
observed accurately.'31 Ducasse's definition is, of course, an
3°
31
Ibid., pp. 3-4.
Ibid., p. 27. This quotation cornea from a chapter which reprints an article
published in 1965 (in Philosophy and Phenomenological Research, xxvi). In my own
'Causes and Conditions', published (later) in the same year, but written without
knowledge of Ducasse's account, I said, about my analysis of the Method of
Difference: 'The importance of this is that it shows how an observation can reveal
not merely a sequence but a causal sequence', and, about the repetition of experi-
ments, 'it is not the repetition as such that supports the conclusion that the dipping
causes the turning red, but the repetition of a sequence which, on each single
occasion, is already prima facie a causal one. The repetition tends to disconfirm
the set of hypotheses each of which explains a single sequence of a dipping followed
by a turning red as a mere coincidence, and by contrast it confirms the hypothesis
that in each such single sequence the dipping is causally connected with the
change of colour.' These points are echoed in Chapter 3 above.
136 C O M M O N SENSE AND THE LAW
account of what it is for one change immediately to cause another;
but there would be no difficulty in covering indirect causation
by denning cause as the 'ancestral' of the relation analysed by
Ducasse. As Ducasse says, he has in effect turned Mill's Method
of Difference into a definition of cause, whereas for Mill it is a
roundabout method of ascertaining something beyond what is
observed. But he is so using what from Mill's point of view is
only one of the possible ways of finding suitable 'instances' (our
/! and JV^); for Ducasse, the before-and-after observation is
radically different from the comparison of an experimental case
with a control case, and it is only the former that is taken up
into his definition.32 Also, he has tightened up and made
explicit what I called in Chapter 3 the use of spatio-temporal
neighbourhood as a criterion of possible causal relevance.
What may be more surprising is that Ducasse has dispensed
with what I call the assumption: it is only (in a refined form)
the difference observation that he has turned into a definition.
No doubt this is because he does not want a conclusion in the
form of a generalization, for which an assumption would clearly
be required. Another way of looking at his definition would be
to say that he has equated causation with contiguity and suc-
cession, dispensing with Hume's third factor, necessity; but
strictly speaking his replacement for Hume's necessity is the
requirement that C be the only change contiguously preceding
K and that K be the only change contiguously succeeding C. It
is a consequence of Ducasse's definition that he can reinstate
the sharp distinction between a cause and a condition which
Mill denied: only an event, not a state or standing condition,
can be a cause. (He does not, indeed, restrict causes to changes,
for he allows what, following Dr. Charles Mercier, he calls
unchanges, that is a state's enduring for some time unchanged,
also to be causes.)33 But the clear and simple distinction thus
drawn will not coincide with that which, as Hart and Honord
say, 'is ... drawn by common sense on principles which vary
in a subtle and complex way . . .'.34
What might seem to be a difficulty for Ducasse is raised by
the question, 'What if several changes take place in the same
spatio-temporal region, say that of C in the definition?' For
" Truth, Knowledge and Causation, pp. 7-8. " Ibid., pp. 5, 29.
34
See above, pp. 118-19.
 COMMON SENSE AND THE LAW 137
example (his own example) 'at the instant a brick strikes a
window pane, the pane is struck . . . by the air waves due to
the song of a canary near by',35 Consistency forces him to say
that the cause of this particular breaking of this pane is the
total change contiguously preceding it, including the air waves
as well as the brick's impact. He admits our reluctance to say
this, but explains it as a confusion between the questions ' What
did cause, i.e., what did then and there suffice to, the occurrence of that
concrete individual event?' and 'which part of what did suffice would
be left if we subtracted from what did suffice such portions of it as were
unnecessary to such an effect?' This second question, Ducasse
admits, 'is usually that which we mean to ask when we inquire
concerning the cause of an event'; but he argues that 'it is not,
like the first, really an inquiry after the cause of one individual
concrete event strictly as such. It is, on the contrary, an inquiry
concerning what is common to it and to the causes of certain other
events of the same kind. This much of a generalization, indeed, is
indissolubly involved in the mere assigning of a name to the
cause and to the effect perceived; although it is not involved in
the merely perceiving them.'36 But there is a tangle of problems
here. We can indeed perceive that the total contiguously pre-
ceding change C 'sufficed'—was sufficient, given the standing
conditions—for K in what I called in Chapter 2 the weak sense
of 'sufficient in the circumstances', just because K did occur.
But whether it was also sufficient in the circumstances in the
strong, counterfactual, sense, that if K had not been going to
occur, C would not, is not so easily decidable. And yet an answer
to this counterfactual question need not involve any generaliza-
tion. Similarly, an answer to Ducasse's second question, which is
equivalent to my 'What was necessary in the circumstances for
K', that is, 'What occurred such that if it had not occurred K
would not?' need not involve any generalization. I rely on the
discussion in Chapter 2 for a proof that an answer to the latter
(and perhaps also to the former) counterfactual question is
involved in the meaning of an ordinary singular causal state-
ment, and on that in Chapter 3 for a proof that this counter-
factual element is not, as Ducasse holds, an infection from a
general inquiry about what is common to a number of causal
sequences. Ducasse's contrary argument rests upon a refusal to
35 Truth, Knowledge and Causation, p. II. 36 Ibid.
138 COMMON SENSE AND THE LAW
consider counterfactual conditionals: 'to say that the song of the
canary was unnecessary . . . is to say only that in another case,
otherwise similar, where the song did not occur, an effect of
the same sort, viz., breaking, nevertheless did occur'.37 Another
actual case, that is: he eschews possible or imaginary cases, and
hence what is necessary or unnecessary can only be a matter of
general truths, of universally or existentially quantified state-
ments.
Another serious difficulty concerns the uniformity of causality,
which Ducasse discusses with Professor Richard M. Gale.38
Gale questions Ducasse's thesis 'that every event of necessity must
have a cause from which it follows in a uniform manner'; quot-
ing from another work by Ducasse a definition which is formu-
lated a little differently from that quoted above, but essentially
equivalent, he says that 'the crucial question is whether this
definition either includes or entails that causality is uniform'.39
Ducasse argues that his definition does entail that causality is
both uniform and universal. Distinguishing 'causally sufficient'
and 'causally necessary' from 'logically sufficient' and 'logically
necessary', he explains that if in a completely determinate state
of affairs S a change C occurs at moment M, and immediately
afterwards there occurs in S another change E, these being the
only changes that occur in S during the period' that contains
them both, then C was both causally sufficient and causally necessary
for E, and (what is equivalent to this) E was both causally
necessitated by and causally contingent on C.40 If the italicized terms
are thus defined, it does indeed follow that every event has a
cause which is both sufficient and necessary for it, by which it
is necessitated and on which it is contingent: something, some
total change or unchange, must have been going on in the
spatio-temporal region contiguously preceding any event (with
the possible exception of the beginning of the universe in time).
Causality (so defined) is thus universal; the key question is
whether it must also be uniform. Gale argues that it need not.
37
Truth, Knowledge and Causation, pp. 12-13.
3* Gale's paper is in Philosophy and Phenomenological Research, xxii (1961), 92-6;
Ducasse's reply appeared in the same volume and is reprinted as Chapter 4 of
Truth, Knowledge and Causation.
" Op. cit., p. 92, reported by Ducasse, op. cit., p. 29: the definition used is
from Nature, Mind, and Death, p. 106.
+0
Truth, Knowledge and Carnation, p. 32.
 COMMON SENSE AND THE LAW 139
Suppose that there is, at some other time and place, a state of
afiairs •$" just like S, and at some moment M' a change C' occurs
in 5" which is quantitatively and qualitatively identical with C;
Gale says that this supposition, together with what we have
already said about S, C, and E, does not logically necessitate
that there will occur, immediately after M', an only change E'
quantitatively and qualitatively identical with E.41
Gale is surely right in this. Ducasse's reply is so curious that
it must be quoted in full:
I submit, however, that this contention is self-contradictory and
therefore invalid, for it supposes that the requirements for causal
necessitation of E in S, as these were defined, are strictly met, and
yet that this does not logically necessitate that occurrence of E in S is
then causally necessitated.
More explicitly, the fact which logically necessitates that causation
of £ in 5 by CinS, as denned, is uniform, is that the definition of such
causation makes no reference at all either explicitly or implicitly to
any particular date or point in space. Hence the supposition that C
in S did, in the sense defined, cause E in S yesterday, and yet that
identical occurrence of S and of C in S today might not, in the sense
defined, cause identical occurrence of E in S is incongruous exactly
as would be the supposition, for instance, that a prime number is
defined as one divisible without remainder only by itself and unity
and that 7 is such a number, and yet that 7 might be prime today
but perhaps not tomorrow. In both cases alike, particular date or
place in space, merely as such, is totally irrelevant.42
Something is very wrong here. If causally sufficient and the
other three relations are defined strictly as holding between the
concrete individual events C and E in S in the way suggested,
then the quantitative and qualitative identities between these
three items and C', E', and S' will not logically necessitate that
C' is causally sufficient for an event E' in S', and so on. Ducasse
seems, inconsistently, to be using '£", 'E', and 'S' as names for
types of events and a type of state of affairs—he does not here use
such distinctive symbols as 'C", '£", and 'S'\ But then the
statement that C is causally sufficient for E in S will already be
a generalization, asserting that any event of the (fully deter-
minate) type C in any state of affairs of the (fully determinate)
** Gale, op. cit., pp. 93-4, reported in Ducasse, op. cit., p. 33. I have slightly
changed the symbolization.
« Op. cit., pp. 33-4.
140 COMMON SENSE AND THE LAW
type S will be contiguously followed by an event of the (fully
determinate) type E; and this is clearly not something that can
be observed on a single occasion. Or perhaps he is influenced
by the strong, counterfactual, sense of 'sufficient', though this
would be inconsistent with his definitions. Alternatively, as the
end of the passage quoted suggests, he is simply assuming that
space and time are causally indifferent; but this is equivalent
to the traditional principle of the uniformity of nature, and
(however natural) it is an additional assumption, which again
goes beyond what can be observed on any occasion or in any
finite collection of events. If, in whatever way, we incorporate
the uniformity into the definition of cause, the proof given
above of its universality at once becomes invalid; though any
event must be contiguously preceded by something, there is
nothing to show that whatever contiguously precedes it here
will be contiguously followed by like events elsewhere. Ducasse's
argument in this passage seems to be another member of the
family of invalid proofs of the causal principle so decisively
criticized by Hume.
It would, then, be better if Ducasse went back to the inter-
pretation of his definition as one strictly for the single sequence,
and abandoned the view that 'constancy of conjunction' is even
'a corollary of the presence of the causal relation'.
Even so, we could not accept his account as an analysis of
the ordinary concept of cause and effect for the reasons brought
out by the example of the window pane and the canary: the
element of counterfactual conditionality, of the cause's being
necessary in the circumstances for the effect, is part of our
ordinary notion of a singular causal sequence, and it cannot be
brushed aside as an intrusion of the general into the singular,
as involving a concealed comparison of this individual sequence
with others.
But though this account fails as an analysis, might it be
defended as a proposal for reform? Is it a better concept than
our ordinary one? Its merits are undeniable. By eliminating the
counterfactual element, it removes the essential (though only
partial) subjectivity of the ordinary concept, the feature which,
near the end of Chapter 2, led us towards the conclusion that
singular causal statements cannot be true, do not describe what
actually occurred, let alone what was observed to occur. Causa-
 COMMON SENSE AND THE LAW 141
tion as Ducasse defines it is a relation which does literally occur
and can be observed. (It is, of course, no part of his case to
claim that such observation is infallible. We may think that
some change C was the only one that occupied the right spatio-
temporal region when some other, unnoticed, one occurred
there too. All he claims, and all he requires, is that causation
should be as observable as many other facts, that it should not
be essentially mysterious or inaccessible.) His account gives us
at once a link between individual cause and effect which other
approaches left obscure. It also preserves in an overwhelmingly
simple way the centrality of the Method of Difference in causal
discovery: 'although sequence is not eo ipso consequence, yet,
when sequence occurs under the special conditions distinctive
of what is properly called an experiment—whether one performed
by man or by Nature—then, eo ipso, the sequence is causal
sequence'.*3 It also builds upon the fact that the clearest cases of
causation are most naturally described, not by the word 'cause',
but by simple transitive verbs.
But it also has serious demerits. What it offers may be all
right as far as it goes, but something more seems to be needed.
Now that we have shown that it does not have constancy of
conjunction even as a corollary, the causal relation suggested
by Ducasse cannot in itself serve as the foundation for causal
inference: it could not be, 'to us, the cement of the universe'.
Equally, it would not be a basis for recipes and precautions.
A connected but perhaps even more serious point is that it is
insufficiently selective. We should like to have a concept that
allowed us to say that, even in this individual case taken on its
own, it was the brick and not the air waves from the canary's
song that broke the window, that if a copper wire conducts
heat and electricity at once, it may be the electricity, not the
heat, that rings a bell, but the heat, not the electricity, that
makes some piece of metal expand. It is easy to see how we
get this selectivity once we bring in counterfactual conditionals
and generalizations and comparisons of one sequence with
another, for example, the 'experimental case' and 'control case'
form of the difference observation as opposed to the before-
and-after one. It is not so easy to see how we can get it while
remaining strictly within the bounds of the single sequence as
43
Op. cit., p. 27.
142 COMMON SENSE AND THE LAW
it is (not as it might have been). Yet there is something to be
found even here. The breaking of the pane—that is, the actual
movements of the glass and of the fragments of glass—is a
continuation of the movement of the brick in a way in which it
is not a continuation of the air waves. But where a delicate
glass bowl is shattered by a loud musical note of just the fre-
quency at which that bowl naturally vibrates, the movements of
the glass are a continuation of the air waves. The processes
involved in the ringing of the electric bell are a continuation
of the flow of electrons in the wire, but not of the molecular
acceleration which we recognize as a flow of heat. At least in
such favourable cases, then, we can find a tie which binds an
individual cause and an individual effect together more closely
and more selectively than does the relation defined by Ducasse,
and which is still in principle observable. We have still to con-
sider how widely applicable, among what we ordinarily regard
as cases of causation, this notion is, and also whether there is
any connection between it and the counterfactual element in
our ordinary causal concepts.
This final suggestion, that a modification of Ducasse's account
may yield an observable link between an individual cause and
its effect which is also selective, which allows for a distinction
between the causally relevant antecedent features and the rest,
is a hint for further inquiry. In the main this chapter has been
devoted to testing the accounts developed in Chapters 2 and 3
by comparing them with those given or suggested by Hart and
Honore", Holland, and Ducasse. I have argued that a unitary
basic analysis of causal judgements can be defended, that
failures of coincidence between this analysis and the use of
explicitly causal terminology, and some of the complications of
that terminology, can be explained without sacrifice of this
basic unity, but that this analysis has to retain the puzzling
counterfactual element which is an obstacle to our saying that
causal relations can be simply and directly observed in single
sequences.
 6
Functional Laws and
Concomitant Variation

'THE law of causality . . . is a relic of a bygone age, surviving,

like the monarchy, only because it is erroneously supposed to
do no harm.'1 Already in 1912 Bertrand Russell was arguing
that the concepts of causation and of causal law, to which we
have been paying such close attention, were from the point of
view of science out of date, perhaps even incoherent, and had
been or should be replaced by the notions of functional relation
and of differential equations.
Russell's lively, brief, rather dogmatic discussion makes a
large number of points against what he regards as current mis-
conceptions, by no means all of equal importance or of equal
merit. Some of them contribute to 'soft determinism', to the
'reconciling project' first advanced by Hume,2* the thesis that
the conflict between free will and determinism is largely illu-
sory. Some are negative claims about the direction of causa-
tion : past and future do not differ in principle with respect to
determination; it is merely by definition that an effect is sub-
sequent to its cause; there would be one-way determination only
if there were a many—one correlation, various alternative causes
being correlated with exactly the same effect—for example, if
various alternative mind-states went with exactly the same
brain-state, there would be a one-sided dependence of brain on
mind. These views will be considered in Chapter 7. But first I
shall examine Russell's criticisms of the view that science either
is or should be concerned with regular sequences of distinct
1
'On the Notion of Cause', Proceedings of the Aristotelian Society, xiii (1912—13),
1-26; reprinted in Mysticism and Logic, to which all page references are given: the
passage quoted is on p. 180.
2
Treatise, II, m, I and a, and Enquiry, Sect. 8.
144 FUNCTIONAL LAWS
events, and the theory of functional relations with which he
would replace it.
One main criticism has the form of an antinomy, though it
is not explicitly so presented. The thesis is that, if there were
causal sequences of events, there would have to be a finite time-
interval between cause and effect. For, if they were contiguous,
either the cause or the effect or both must . . . endure for a finite
time [presumably, because two instantaneous items, if contiguous,
would be simultaneous, so that the one would not succeed the other];
indeed . . . it is plain that both are assumed to endure for a finite
time. But then we are faced with a dilemma: if the cause is a process
involving change . . . only the later parts can be relevant to the
effect, since the earlier parts are not contiguous to the effect, and
therefore (by the definition) cannot influence the effect. Thus we
shall be led to diminish the duration of the cause without limit,
and however much we may diminish it, there will still remain an
earlier part which might be altered without altering the effect, so
that the true cause, as defined, will not have been reached ... If, on
the other hand, the cause is purely static . . . then, in the first place,
no such cause is to be found in nature, and, in the second place, it
seems strange—too strange to be accepted, in spite of bare logical
possibility—that the cause, after existing placidly for some time,
should suddenly explode into the effect. This dilemma, therefore, is
fatal to the view that cause and effect can be contiguous in time;
if there are causes and effects, they must be separated by a finite
time-interval . . .3
To prove the antithesis, that there can be no time-interval,
Russell argues that however short we make it,
Something may happen during this interval which prevents the
expected result. I put my penny in the slot, but before I can draw
out my ticket there is an earthquake which upsets the machine and
my calculations. In order to be sure of the expected effect, we must
know that there is nothing in the environment to interfere with it.
But this means that the supposed cause is not, by itself, adequate to
insure the effect. And as soon as we include the environment, the
probability of repetition is diminished, until at last, when the whole
environment is included, the probability of repetition becomes
almost nil.4
But both these arguments are unsound. In the thesis argu-
ment, contiguity without simultaneity would require only that
4
Op. cit., pp. 184-5. Op. cit., p. 187.
 FUNCTIONAL LAWS 145
either cause or effect should endure for a finite time, and the
most that Russell's dilemma could show is that the cause cannot
so endure. But, more significantly, the first horn of the dilemma
fails. There is nothing in the definition he has quoted, and
certainly nothing in the general notion of causation as regular
contiguous succession, to show that the 'earlier parts' of the
cause-process are irrelevant or 'cannot influence the effect'. On
the hypothesis considered, it is not true that an earlier part
might have been altered without altering the effect, for there
will be, as he mentions, 'causal relations between its [the
cause-process's] earlier and later parts'. It is true that smaller
and smaller final bits of this process could be taken as the
cause, but there is no need to say that there is a 'true cause'
which has not been reached until the trimming off of earlier
parts of the cause-process, and its diminution to its final phase,
have been carried to the limit,
The antithesis argument also fails. Russell's 'environment',
which is our 'causal field', needs only to be described generally
and largely negatively in order to ensure that the effect will
regularly follow the proposed cause. For the penny in the slot
regularly to produce a ticket, we need the absence of earth-
quakes, bombs, and intrusions from outer space, but it does not
matter in the least what in detail the non-intrusive surroundings
are like, and there is no improbability about repetitions of the
suggested cause within this broadly defined field.
For all that Russell's arguments show, then, there could be
regular causal sequences of events either with or without a
time-interval. Without a time-interval, at least one, and pre-
sumably both, of the cause and the effect would be processes
having duration, contiguous in the way that Ducasse's defini-
tion5 explains. With a finite time-interval, the freedom from
intervention is secured by the fact that it is only within some
broadly defined field that the regularity of sequence holds. And
in fact there are regularities of succession, as we noted in
Chapter 3, though our knowledge of them is almost always
incomplete.
But Russell has a second, and more telling, criticism, namely
that science is not much concerned with such regularities. They
belong to 'the infancy of a science'; 'every advance in a science
i Above, p- 135.
146 FUNCTIONAL LAWS
takes us away from the crude uniformities which are first
observed into greater differentiation of antecedent and con-
sequent, and into a continually wider circle of antecedents
recognised as relevant'.6
The last phrase here refers only to the point which we have
already taken over from Mill about conjunctions of factors,
'assemblages of conditions'. But Russell's main point is that the
laws of an advanced science take a form different even from
that of the complex regularities we discussed in Chapter 3, a
form which is illustrated by Newton's law of gravitation. 'In
the motions of mutually gravitating bodies, there is nothing
that can be called a cause, and nothing that can be called an
effect; there is merely a formula. Certain differential equations
can be found, which hold at every instant for every particle of
the system, and which, given the configuration and velocities
at one instant, . . . render the configuration at any other earlier
or later instant theoretically calculable.' And he unkindly sug-
gests that philosophers have remained faithful to the old 'law
of causality' only because most of them are unfamiliar with the
mathematical idea of a function. We should replace 'same
cause, same effect' with 'sameness of differential equations'.7
The nearest approach we could find, in non-mathematical
language, to a 'law of causality' that is actually discoverable
in the practice of science would be: 'There is a constant relation
between the state of the universe at any instant and the rate of
change in the rate at which any part of the universe is changing
at that instant, and this relation is many-one, i.e. such that the
rate of change in the rate of change is determinate when the
state of the universe is given.'8
Nothing (.urns, of course, upon the particular Newtonian
gravitational form that Russell has chosen as an illustration,
but it will be simplest to continue to use this to illustrate our
discussion. In it, certain second derivatives with respect to
time, namely the acceleration of each particle, have, at any
moment, values which are functions of the distances and masses
at that moment of all other bodies in the system. Given the
masses and distances, the accelerations are fixed and calculable,
but not vice versa.
6
Op. cit., p. 188. ' Op. cit., pp. 194-5.
8
Op. cit., p. 195; Russell has a strange idea of non-mathematical language.
 FUNCTIONAL LAWS 147
Now there is no doubt that science has formulated quite a
number of laws of this general kind. It is, perhaps, only for
Newtonian gravitation that the reference to the state of the
universe is in order; as Russell himself points out, what scientists
actually deal with is rather some 'relatively isolated system1.
But with this qualification it is plausible to suggest that as
sciences leave their infancy and become 'advanced', they come
to be concerned more with laws of this functional sort, or come
nearer to being 'reduced' to sciences which already deal in
such functional laws. Let us call what such laws describe
functional causation, and, in contrast, let us call what we have
been dealing with so far neolithic causation, so as to maintain
Russell's analogy with the institution of monarchy. I have
already argued, against Russell, that neolithic causation can be
coherently presented. But I would argue also that neolithic
and functional causation are much more closely related than
Russell's account might suggest, and that this makes it natural
to extend the old term 'causation' to cover the subject of
functional laws.
First, I shall show that Russell's causation can be seen as a
limiting case of causal sequence of distinct events as described,
for example, by Ducasse. In the gravitational illustration, the
analogue of the effect is the acceleration of each particle.
Mathematically, this is represented as acceleration at an
instant. But 'acceleration at an instant' makes sense only as a
limiting description of changes of velocity over periods of time.
Talking about the acceleration-effect at the instant / amounts
to a special way of talking about changes in velocity over
longer or shorter periods of time starting at t. A Russell type
effect, then, is just the limit approached by a Ducasse type
effect—a process starting at the instant where cause and effect
are contiguous—as the time-length of this process is decreased
towards zero. It is obvious that this holds not only for the
gravitational example but wherever the functional analogue of
a neolithic effect is a derivative, of whatever order, with respect
to time.
A second, closely related, point is that a functional law will
entail the holding of neolithic laws in appropriate fields. A
differential equation containing time-derivatives, of whatever
order, can in principle be integrated with respect to time, and
148 FUNCTIONAL LAWS
will then tell us what later states of the system will regularly
follow such-and-such earlier ones. Galileo's law of terrestrial
gravitation is a functional law, but it entails such neolithic laws
as that stones thrown up into the air in an ordinary 'empty'
environment will subsequently fall to the ground. If functional
effects could not be thus integrated with respect to time to
yield actual changes, functional laws would be of little interest
or use.
Thirdly, a neolithic causal regularity of the complex form
discussed in Chapter 3 can be seen as a special case of a func-
tional law where the variables take only one of the two crude
values 'present' and 'absent', which we could write as 'i' and
'o'. If the presence at t of some disjunction of conjunctions of
actually relevant factors or their negations, such as (ASC or
DGH or JKX], is both necessary and sufficient in the field F
for the presence at a slightly later time t-{-8t of P, then we could
say that the change of value of P between t and t-}-8t was a
function of the values at t of A, B, C, D, G, H, J, K, and L.
This last point may look like a mere formality, but it leads
on to a fourth and much more significant link between neo-
lithic and functional causation. The Method of Difference is,
as we have seen, the principal method for discovering and
establishing neolithic causal regularities, while variants of it,
such as Mill's Method of Residues, and related devices like the
Method of Agreement and some of the things covered by the
name of the Joint Method, can also play some part.9 What
Mill calls the Method of Concomitant Variation is pretty
clearly a device of the same general sort as these; but I shall
show that it is not one method, but several, including counter-
parts of the Methods of Difference, Agreement, and so on, that
it works with the functional concept of causation very much as
they work with the neolithic concept, and hence that reasoning
closely analogous to that which identifies neolithic causes does
some, but only some, of the work of identifying functional
causes.
The 'Fifth Canon' by which Mill introduces this method is
far too loose to be helpful: 'Whatever phenomenon varies in any
manner whenever another phenomenon varies in some particular manner,
is either a cause or an effect of that phenomenon, or is connected with it
9
See Appendix.
 FUNCTIONAL LAWS 149
10
through some fact of causation.'' In fact by an equivocation in the
use of 'whenever' between 'whenever without restriction' and
'whenever within the observation' it mixes up a new, functional
definition of 'cause' with the principle of a method, or methods,
for discovering such causes. The definition hinted at is, of
course, a development of a regularity definition, and like others
of this class leaves on one side the question of the direction of
causation. We must now think of the various 'phenomena' as
items that have some kind of magnitude. Carrying over from
our previous discussion the notion of a field (F) and the expecta-
tion that causes will in general be complex, we obtain, as the
typical form of a functional causal regularity, the statement
that the magnitude of P in the field F is always such-and-such
a function of the magnitudes of the factors, say, A, B, C, and D,
which we can write
Pr=f(A, B, C,D).
Where such a regularity holds, we may say that the complete
set of actually relevant factors, A, B, C, Z), together with the
function/which shows in what way they are relevant, constitutes
the full cause in this functional sense of P in F, while we may
call each of the various actually relevant factors, A, B, C, and D,
a partial cause. The full cause will be that function of the mag-
nitudes of the various relevant factors on which the magnitude
of P in F wholly depends, while a partial cause will be a factor
on whose magnitude that of P in F partly depends. (Since we
are leaving aside the problem of direction, 'depends' has here
a purely mathematical sense.)
But this says only what concept of a cause is now being used:
where does the method of discovery or proof come in? Clearly,
what is required is an ampliative method, one in which we
argue from a covariation in some cases or over some limited period
to a regularity of the kind just indicated, which is believed to
hold for unobserved cases as well. We argue from 'whenever
within the finite range of observations' to 'whenever absolutely'.
As we saw, Mill obscured this argument with an equivocation.
It is also clear that there are two distinguishable tasks, that of
identifying the various partial causes and that of determining
the functional relation/, of deciding just how the partial causes
'» System of Logic, Book III, Gh. 8, Sect. 6.
15° FUNCTIONAL LAWS
affect the magnitude of P. The completion of the first task
would yield a mere list of terms, that of the second a mathe-
matical formula. It is in the performance of the first task that
we can find the closest analogy with the other eliminative
methods.
If an ampliative method is to be formally valid it will, of
course, require an assumption as well as an observation. The
general pattern of reasoning will be this: we assume that there
is some regularity of the kind indicated, relating the magnitude
of P in the chosen field Fto the magnitudes of some members of
a range of possibly relevant factors (we can work formally with
a list of such possibly relevant factors, but as before this is only
an expository device, in practice no list is needed); we observe
over some limited period certain patterns of variation and non-
variation of P and the possibly relevant factors; and we infer
that certain of the possibly relevant factors are actually relevant,
that they are in fact partial causes in the sense defined.
A counterpart of the Method of Difference would identify
a certain factor, say A, as actually relevant by observing cases
in or between which P varies while A varies but all other
possibly relevant factors remain constant. The assumption
would be that the magnitude of Pin the field /'wholly depends
in some way on the magnitudes of one or more factors X, X',
X", etc., where each of these is identical with one of the possibly
relevant factors A, B, C, D, E, etc. Then if it is observed that
over some period or range of cases (in the field F) P varies
while A varies but B, C, D, E, etc. all remain constant, it
follows that one of the factors X, X', X", etc. is identical with A,
It is not shown, of course, that B, C, D, E, etc. are irrelevant;
but it is shown that the full cause cannot be made up of them
alone, for, if it were, the magnitude of P would have remained
constant over that period when the magnitudes of all of B, C,
D, E, etc. remained constant. That is, this assumption and an
observation of this form together entail that A is an actually
relevant factor, a partial cause. A series of experiments, in each
of which one possibly relevant factor is varied at a time, the
others being held constant, could in accordance with this
method pick out partial causes one by one.
A counterpart of the Method of Agreement would identify a
certain factor, say A, as actually relevant by observing cases in
 FUNCTIONAL LAWS 151
or between which both P and A remain constant while other
possibly relevant factors vary. Like the original Method of
Agreement,11 this requires a stronger assumption than does its
Difference counterpart if any conclusion is to be entailed. If we
assume that (in the field chosen) the magnitude of P depends
on that of just one of the possibly relevant factors A, B, C, D, E,
etc., and that it is responsive to all changes in that actually
relevant factor, then the observation indicated will show that
this one actually relevant factor must be A: it cannot be B, or
C, or D, or E, and so on, because each of these has varied while
P has not. But if we make the weaker and in general far more
plausible assumption that the magnitude of P depends upon
those of one or more of the possibly relevant factors, nothing
follows from the observation indicated. For although B and C,
say, have varied while P has remained constant, it is still
possible that the magnitude of P is a function of those of B
and C, since the function might be such that the variations in
B and C compensated for one another, adding up to a zero
result. And sets of more than two partial causes are similarly
not ruled out.
Suppose, on the other hand, that the magnitudes of P and
of A both remained constant over some considerable period,
during which the magnitudes of all the other possibly relevant
factors varied independently. Then if we could say that these
other factors had varied in all conceivable ways, it would
follow that no function of any set of them would have remained
constant, and therefore that they must all be irrelevant, that
the magnitude of P must be correlated with that of A alone-
all the time, of course, in the field in question. This would be
another counterpart of the Method of Agreement, it would
require only our weaker and more plausible form of assumption,
and it would have the great merit of providing a way of closing
the set of actually relevant factors, whereas our counterpart of
the Method of Difference only picked out relevant factors one
by one, and did not positively exclude any as irrelevant. Un-
fortunately the proposed observation is, speaking strictly, an
unattainable ideal; there must in principle, for any finite range
of observed variations in B, (7, D, E, etc., be functions of their
magnitudes which remain constant: no actual pattern of
" See Appendix.
158 FUNCTIONAL LAWS
variation can include variation in all conceivable ways. However,
we can have a probabilistic method which approximates to this
ideal. If P and A are observed to remain constant while there
is a great deal of apparently random and independent variation
in all the surrounding conditions—other, of course, than what
has been taken as included in the field F—then in effect all
functional hypotheses that we can actually think of to relate P
to combinations of these surrounding factors are ruled out, and
by contrast it is at least probable that the magnitude of P
depends, in F, on that of A alone. And it is obvious that a similar
argument could be used to show the irrelevance of, say, D, E,
and so on if they vary randomly while, say, A, B, and C, as
well as P, remain constant: the probabilistic evidence for the
closure of the set of actually relevant factors can operate after
any number of partial causes have, by repeated appeals to
our counterpart of the Method of Difference, been included
in that set.
It is, I think, undeniable that many actual investigations con-
tain parts which could be regarded as applications of these
methods to the task of picking out (all) the partial causes, in
this functional sense, of some phenomenon in some field. In
practice this task is likely to be mixed up with the different,
second, task of determining just what the functional relation-
ship is, which will be a matter of testing hypotheses and fitting
curves to data. When we note that P and A vary while all other
possibly relevant factors remain constant, we shall naturally try
to say also how P varies as a function of A, and if we find that B,
and C, say, are also actually relevant factors, we shall look for a
formula that covers all the observed covariations. Nevertheless,
these two tasks can be distinguished, and when we study the
logic of such investigations it is appropriate to distinguish them,
and to attend separately to the different devices that aid their
fulfilment. And when we do so, we find close formal and logical
analogies between the assumptions, observations, and conclu-
sions of the original Methods of Difference and Agreement
which establish neolithic regularities and those of their Con-
comitant Variation counterparts which identify functional
causes.
The Method of Concomitant Variation, then, is not, as Mill
seems to have thought, just a second-best way of discovering
 FUNCTIONAL LAWS 153
the same old sorts of causal relationship in situations where, for
some reason, it was difficult to secure the total absence of this
or that factor: it implicitly introduces a new, though systematic-
ally related, form of causal connection; it is a bridge between
neolithic and functional causation.
This fact, together with the three other links we found earlier
between neolithic and functional causes, makes it plausible to
regard the functional notion as a refinement and extension of
the neolithic causal concept rather than as an entirely new
concept: the magnitude of each partial functional cause is
necessary in the circumstances for the magnitude of the effect.
(The sovereignty of parliament is perhaps an analogous refine-
ment of the neolithic institution of monarchy, and one for
which the Newtonian age was also responsible.) It is worth
noting that this refinement resolves a difficulty which we en-
countered in Chapter 2, that, for example, a hammer-blow is
said to have caused a chestnut's becoming natter although quite
so hard a blow was not necessary in the circumstances. Working,
as we were in Chapter 2, with the fairly crude neolithic concept,
we could explain this by saying that we were there regarding
the hammer-blow as a unit, something which was either present
as a whole or absent as a whole. Now, with our refined func-
tional concept, we can take account of the quantity of momen-
tum, of how hard the blow was and of how hard it needed to
be, and we can say either that the actual momentum was
necessary (and sufficient) in the circumstances for the actual
degree of flattening, or that what caused such-and-such a
flattening was the occurrence of a blow of at least so much
force. The supposed difficulty arose from a confusion between
these different correct comments, that is, from an attempt to
speak in neolithic and functional terms at once.
Russell's first criticism, then, does not show the neolithic
causal concept to be incoherent, and while his second criticism
may show that that concept on its own is likely to be left behind
by the advance of science, we can see that what that advance
calls for is a development and extension of that concept, not
its replacement. The old concept—allowing, of course, for
complex and incompletely known regularities—will continue
to be applicable in many spheres, even if neolithic causation
is to be seen as ultimately resting upon laws of the functional
154 FUNCTIONAL LAWS
sort; and even where we are dealing with functional relations,
very similar patterns of thought and inquiry are in order. It is,
indeed, very plain that whereas Russell thought that causality
was out of date in 1912, causal concepts are, sixty years later,
constantly being used in our attempts to understand perception,
knowledge, and memory, and to clarify our thought about
action, responsibility, legal claims, purpose, and teleology.
As Bunge points out,u Russell's view that functional depen-
dence should replace causation was widely shared. It was put
forward by Mach, and adopted by Schlick and the Vienna
Circle. Bunge remarks that 'functional dependence is poorer
than causation, since it does not include the concept of genetic
connection through a process'.13 This could be misleading. It is
true that functional relationships need not be genetic, that func-
tions can express inert uniformities of coexistence: but, as we
have seen, those that include time-derivatives yield, by their
integration, laws of change and succession. Functional formulae
can neatly express the interdependence of quantities and mag-
nitudes, the results of two-way causation between persisting
things or features of things, but of course neolithic causation
allows for this too. The asymmetrical relation of causal priority,
the unique direction from cause to effect, does not preclude the
state of A at t helping to bring about a change in the state of B
between t and t-\-8t, while the state of B at t helps to bring about
a change in the state of A between t and t -f-8£. In other respects,
the functional dependence view is only a variant of a pure
regularity theory of causation, and appealed to positivists for
that reason. And what it crucially leaves out is what regularity
theories in general leave out: the direction of causation and
any adequate treatment of the problems of necessity and the
'causal tie'. But if the functional concept is interpreted non-
exclusively, as neither abolishing the neolithic concept nor
precluding all non-regularity aspects of causation, it can be
welcomed as a refinement and extension of older notions.
But having recognized the need for this extension of the con-
cept of cause and effect, we may be led to consider two others.
»12 M. Bunge, Causality, pp. 91-8.
15
Ibid., p. 98. Bunge is, no doubt, referring to the fact that functional depen-
dence does not conceptually require the sort of continuity of process that is discussed
in Chapter 8 below.
 FUNCTIONAL LAWS 155
One concerns the relation between an earlier and a later phase
in a continuous self-maintaining process. A top, or the earth,
say, rotates continuously, whether at a changing or at a con-
stant rate, from ^ through tz to ts. Having been spinning between
*! and tz, it goes on spinning after tz. Why? Because nothing,
or nothing much, interfered or tended to stop it. Given that it
was spinning, the cause of its continuing to spin was the absence
of interference. But in relation to some wider field, can we not
say that it went on spinning after tz because it was already
spinning before t2? Can we say, then, that the rotation from /x
to tz caused the rotation from tz to ta? We would, it is true, not
ordinarily say this, whereas we would say that the friction
between the top and the air, and between its point and the
ground, caused its slowing down, and that tidal friction is
similarly changing, very slowly, the earth's speed of rotation.
We ordinarily look for and recognize a cause of a change in a
process, rather than for the mere continuance of the process.
And yet, as we saw, we can speak of a negative cause, an
absence of interference, as causing its continuance. Also, where
a process needs some sort of input to keep it going we are ready
to speak of a cause: the pull of the engine causes the continued
movement of the trucks. If we include the various frictions in
'the circumstances', the pull of the engine is necessary and
sufficient in the circumstances for the train's not slowing down;
conversely, if we include the engine's pull in 'the circumstances',
the frictions are necessary and sufficient in the circumstances
for the train's not speeding up; these two partial causes together
are responsible for the constancy of the train's velocity. Simi-
larly, where there is (practically) no friction, the absence of
friction and of various other interferences is necessary and suffi-
cient in the circumstances for the earth's continuing to rotate
at (practically) the same speed. But all these causes, even where
their effect is some constancy of behaviour, and even where the
cause is itself negative, are in an obvious sense intrusive or
transeunt: they come in, or refrain from coming in, from out-
side, in a way that an earlier phase of the same rotation (or
other movement) does not. However, while it seems strange to
call this earlier phase a cause, and while our reluctance to do
so reveals something about our actual concept of causing, there
are analogies which would justify our extending the existing
156 FUNCTIONAL LAWS
concept to cover this. If we do not include the rotation between
<! and tz in the field, it is quite clear that, in 'the circumstances'
which include neither any appreciable friction nor any new
impulse to rotate, the rotation between ^ and t% was necessary
and sufficient for the rotation between <2 and ta. Besides, the
relation between earlier and later phases of a continuous self-
maintaining process is very like that between cause and effect
as defined by Ducasse. A very minor difference is that succes-
sive phases make contact with one another over a three-
dimensional volume, whereas his cause and effect make contact
only over a two-dimensional surface. The earlier phase of a
self-maintaining process surely brings about, or helps to bring
about, the later phase. If the concept of cause and effect does
not yet cover them, it should: we can recognize immanent as
well as transeunt causation.
But if this extension is accepted, can we reasonably stop
there? Should we not extend the concept still further to in-
clude the relation between the earlier and later phases of the
existence of any material object? Is not any physical object a self-
maintaining process or a cluster of such processes, helped to
some extent by a balance of inputs and outputs, of interactions
with neighbouring things? Is its continued existence not very
like the continued movement of the train, which is largely self-
maintaining but is also kept up by the balance between friction
and the pull of the engine? It is very easy to find circumstances
such that the existence of a pen here from ^ to tz is necessary
and sufficient in the circumstances for the existence of a pen
here from tz to t3.
No doubt if we thus stress the analogies which justify these
extensions of the existing concept we should also try to explain
why the existing concept is so restricted, in much the same way
that in Chapter 4 we noted and tried to explain restrictions on
the use of the term 'cause' even among events and states that
are undeniably inus conditions of some result.
I suggest that the explanation lies in the associations between
'cause', 'because', and 'why'. It is especially where we are
inclined to ask why-questions that we will be ready to accept
because-statements as answers to them, and we are particularly
ready to call causes those items that can be described by clauses
introduced by 'because'. Now the persistence of many sorts of
 FUNCTIONAL LAWS 157
object is something that we take very much for granted. With-
out going all the way with Strawson, we can admit that this
is a very pervasive and fundamental structural feature of the
world as we know it. We are little inclined, then, to ask why
this pen (in ordinary circumstances) continues to exist. We
might well ask why there is a pen here now if there had not
been one anywhere in the neighbourhood immediately before,
or, even if there had been one here immediately before, if it
had survived, say, an atomic explosion, but not if such a pen
had been here immediately before and no catastrophes had
intervened. Such persistences are assumed and used in answer-
ing other why-questions rather than made the subject of why-
questions themselves. The earlier phase of a thing's existence is
not, then, looked upon as causing the later phase because the
later phase is not something for which an explanation seems to
be required when we know, or assume, that the thing has just
continued to exist.
The continuance of a process, for example a rotation, is,
indeed, neither so familiar nor so thoroughly built into our
ordinary world picture as is the persistence of objects. Our
experience suggests that many processes quickly run down and
stop when left to themselves, and if anything keeps on working
we look for a sustaining cause. Even if we know that the top
has been spinning from tl through <2 to ta, we can therefore still
ask why it has been spinning from tz to ta. But if we do ask
this, the answer 'Because it was spinning from t1 to t^', is still
likely to appear disappointing. It will merely provoke us to ask
further, 'But why was it spinning then?' We are not satisfied
until we come to an impulse that started the process off or to
an input of energy that keeps it going. This is another case
rather like that to which Hart and Honore drew our attention,
that we trace causes back through intermediate causes to a
deliberate human act, but not through that act to its causes.
Here we are in so much of a hurry to trace the cause of the
later phase of a process back through the earlier phase to some
energy input that we do not recognize the intermediate cause,
the earlier phase of the self-maintaining process, as a cause at
all.
But all such attempts at an explanation of the restrictions of
our existing concept of cause and effect point to fairly superficial
I58 FUNCTIONAL LAWS
differences, to respects in which the same basic pattern of
relations may present itself differently to our knowledge, beliefs,
expectations, and practical interests. If we attend rather to
what is there 'in the objects', it is the common pattern that
impresses us, and it is this that justifies the extension of causal
terms and causal concepts to cover not only all event and state
inus conditions (or better) but also functional causes and the
earlier phases of persisting objects and self-maintaining pro-
cesses.
On the other hand, if we accept both the kinds of develop-
ment and extension of causal concepts that have been con-
sidered in this chapter, we should also take note of the differences
between them. According to our second sort of extension, what
something is next or does next depends upon what it has been
or upon what it has been doing; according to the first, a change
in what one thing does depends upon how other things are.
This contrast carries others with it. In the second case the
effect typically resembles the cause, being indeed some kind of
continuation of it; in the first case it may well not. In Russell's
gravitation example, there is no resemblance between the
acceleration of one body and the distances and masses of the
surrounding bodies. This example can, indeed, be used to
bring out both sides of the contrast. The almost equal velocity
of a body just before and just after / illustrates our second kind
of extended 'cause' and 'effect', while the body's acceleration at
t, as a function of the distances and masses of other bodies,
illustrates our first. The two problems which I have mentioned
from time to time take different forms in these different con-
texts. The tie between individual cause and individual effect is
obvious in the cases covered by our second extension: it is
simply the continuity of activity or existence between arbitrarily
marked off phases. But in the cases covered by our first exten-
sion it is not clear what sort of link we could even hope to find.
The problem of the direction of causation, to which I shall turn
in the next chapter, arises in both cases, but in the one case
there is, while in the other case there is not, an obvious temporal
sequence of events.
And yet this contrast may have been pushed too far. There
is nothing in functional relationship as such to preclude re-
semblance or continuity. The Newtonian gravitational example,
 FUNCTIONAL LAWS 159
in which the function holds instantaneously between utterly
disparate factors, may be an unusually extreme one. There is
certainly a difference between interaction and the mere con-
tinuance of some self-maintaining process, but interaction can
itself involve continuous but intersecting processes. Transeunt
causation can share a feature which immanent causation most
clearly displays: this will be of interest when we return, in
Chapter 8, to the problem of the necessity of causes.
 7
The Direction of Causation1

THE relation between cause and effect is commonly taken to

be asymmetrical: a cause is related to its effect in some way in
which the effect is not related to its cause. We have been using
the term causal priority as a name for whatever it is that marks
this distinctive direction; but so far this phrase has been little
more than a name for an unknown somewhat. To make it much
more than a mere name we must try to answer the two ques-
tions, 'What is our concept of causal priority, of the direction
of causation?' and 'What, if anything, constitutes causal
priority in the objects? In what real respects, if any, is the
causal relation asymmetrical?'
On a simple regularity theory, a cause is both necessary and
sufficient for its effect in such a sense that it follows automatically
that the effect is equally sufficient and necessary for its cause:
whenever an event of type B occurs an event of type A has
preceded it and whenever an event of type A occurs an event
of type B will follow. This symmetry is unaffected by the
improvements and complications in the account of causal
regularities given in Chapter 3. A full cause, as there defined, is
still both necessary and sufficient for its effect in some field,
and in that same field the effect will be both necessary and
sufficient for the full cause—all these statements of necessity
and sufficiency being translatable into the appropriate universal
propositions. But what about the relation between a partial
cause, or inus condition, and its effect? This relation is not
symmetrical: it is logically possible that A should be an inus
condition of B while B is not an inus condition of A, But neither
1
I discussed this problem in Section 8 in 'Causes and Conditions' American
Philosophical Quarterly, ii (1965), and more fully in 'The Direction of Causation"
Philosophical Review, Ixxv (1966), 441-66, but the present chapter embodies con-
siderable changes of view.
 THE DIRECTION OF CAUSATION 161
is it asymmetrical, and in fact where A is an inus condition of
B, B is usually also an inus condition of A. If (in the field F)
there is some X and some T such that all (AX or T) are B and
all B are (AX or T), it follows at once that all BT are A, and
there will usually be some £ such that all (B'f or £) are .4
while all A are (.6T or £). So this relation will in general fail
to introduce any asymmetry, any distinctive direction.
It is not surprising, then, that regularity theorists tend to
rely on time to supply the asymmetry: if the type of event A
is necessary and sufficient for the type of event B, and vice
versa, or again if each type is an inus condition of the other,
then events of whichever type the events of which regularly
precede the associated events of the other type are the causes.
Causal priority is just temporal priority under another name.
But this view is at variance at least with our ordinary con-
cepts.
For one thing, we seem ready to accept causes that are
simultaneous with their effects—for example, Kant's leaden ball
resting on a cushion and being causally responsible for a hollow,
or, in a Newtonian gravitational set-up, the acceleration of one
body at / depending causally upon the masses and distances of
other bodies at that same t. We even seem able to conceive
that there might be backward causation, with the effect pre-
ceding its cause in time. Certainly many people have speculated
about the possibility of some sort of direct precognition, of a
literal foreseeing of future events, and some have believed that
they, or others, have had this power. But such a literal seeing
of the future would mean that the future event was already
affecting the precognizer. We need not, of course, defend the
view that precognition, or any other form of backward causa-
tion, or even simultaneous causation, actually occurs. It is
enough for our present purpose that they have been conceived,
for this would show that our concept of causal order is not
simply that of temporal order, that we have some notion of a
relation of causal priority which can be variously directed with
respect to time.
A possible reply to this argument is that while we can talk
and speculate vaguely about these possibilities, we cannot con-
ceive them coherently. Backward causation, it might be said,
is like time travel. The concept, which we have, of just a single
162 THE D I R E C T I O N OF CAUSATION
time dimension makes literal time travel conceptually impos-
sible; but this has not prevented novelists from telling stories
about it. However, I shall argue later that we can dispose of
this reply by describing in some detail conceivable experi-
mental results which, if they were obtained, would be coherently
interpreted as evidence for backward causation.2
Another reason for insisting on a conceptual distinction
between causal and temporal priority is that these relations
seem to have different logical structures. The direction of time
characterizes the dimension as a whole, whereas the direction
of causation characterizes each process or sequence on its own.
Suppose, for example, that events A, B, C, D, E form a series
in this order in time, but perhaps not in this temporal direction
—I mean that it may be that A comes first and E last, but
again it may be that E comes first and A last; what is settled is
that B comes between A and C, C between B and D, and so on.
Suppose further that events D' and E' are simultaneous with
D and E respectively. Then as soon as we fix the time direction
between any two of these events, we fix it for every pair. If B
comes before C, then A comes before B, D before E, and D'
before E'; but if C comes before B, then all the other relations
are correspondingly reversed. Once a time-direction has been
given to any pair of events, it has been given to the system as
a whole. Not so with causal direction. It belongs, if it belongs
1
There has been a vigorous controversy about the possibility of backward
causation, started by the symposium between M. Dummett and A. Flew 'Can an
Effect Precede its Cause?' in Proceedings of the Aristotelian Society, Supplementary
Volume XXVIII (1954), 27-62. See also Max Black, 'Why Cannot an Effect
Precede its Cause?', Analysis, xvi (1955-6), 49-58, reprinted in Models and Metaphors;
A. Flew, 'Effects before their Causes?—Addenda and Corrigenda', Analysis, xvi
(1955-6), 104-10; M. Scriven, 'Randomness and the Causal Order", ibid., xvii
(1956-7), 5-9; D. F. Pears, 'The Priority of Causes', ibid., xvii (1956-7), 54-
63; A. Flew, 'Causal Disorder Again' ibid., xvii (1956-7), 81-6; R. M.
Chisholm and R. Taylor, 'Making Things to have Happened1, ibid., xx (1959-
60), 73~8; W. Dray, 'Taylor and Chisholm on Making Things to have Happened",
ibid., xx (1959-60), 79-82; A. J. Ayer, The Problem of Knowledge, pp. 170-5; M.
Dummett, 'Bringing about the Past', Philosophical Review, Ixxiii (1964), 338-59;
S. Gorovitz, 'Leaving the Past Alone', ibid., 360—71; R. M. Gale, 'Why a Cause
Cannot be Later than its Effect', Review of Metaphysics, xix (1965), 209-34. I have
drawn freely on points made by all those who have contributed to this controversy,
though my main purpose is not to consider whether causes could be later than
their effects, or could be found to be so, but simply to discover what concept we
have of causal priority and what if anything, constitutes causal priority 'in the
objects'.
 THE DIRECTION OF CAUSATION 163
at all, to each ordered pair of events on its own. B may be
causally prior to C, and yet D and E may be causally irrelevant
to one another, or again E might be causally prior to D. It
seems conceivable that although D and D' are simultaneous,
and E and E', the direction of causation might be from D to E
and yet from E' to D'. Of course this suggestion that causal
direction characterizes processes separately is a direct denial of
the view that it is simply temporal direction in disguise. An
adherent of the latter view would say that when it is settled
that D is temporally prior to E, although they may be causally
unrelated, it follows at once that if they are causally related, D
is the cause and E the effect. But what I am maintaining is
that this seems not to be how we ordinarily see this matter.
This apparent difference of logical structure would also be a
difficulty for a view which is the reverse of that which we have
been considering, and which makes temporal priority causal
priority in disguise. (This second view is implicit in discussions
which are ostensibly about the direction of time, time's arrow,
or temporal asymmetry, but whose detailed concern is with
asymmetries in specific kinds of causal process.)3 It is hard to
see how the pervasive directedness of the whole time dimension,
which is undoubtedly part of our concept of time, could be
constituted by the local asymmetries of particular processes.
This view would have to be seen as involving a revision of our
concept of time, as saying that time does not really have the
pervasive directedness we ascribe to it. All that is pervasive is
its unidimensionality; given this, we can borrow an asymmetry
from particular processes and impose it upon the whole dimension.
How are these two views related to one another? We are
tempted to say that they cannot both be right: we cannot,
without vicious circularity, analyse the concept of causal
priority wholly in terms of temporal priority and also analyse
the concept of temporal priority wholly in terms of causal
priority. Yet it would be possible, however implausible, to say
that direction is imposed on the time dimension by a borrowing
from some special sorts of causal process, and that the supposed
direction of causation in all other sorts of causal process is
3
For example, H. Reiphenbach, The Direction of Time; K. R. Popper, 'The
Arrow of Time', Mature, clxxvii (1956), 538; A. Grunbaum, 'Temporally-
asymmetric Principles, etc.', Philosophy of Science, xxix (1962), 146-70.
164 THE DIRECTION OF CAUSATION
merely that of time. This would mean, in effect, that there is
an essentially causal priority of some causes to their effects,
but that the causal priority we ascribe to other causes consists
merely in their lying, as we might say, on the same side of their
effects in the time dimension as do those special causes whose
relation to their effects is intrinsically asymmetrical.
A third reason for insisting on a conceptual distinction be-
tween causal and temporal priority is that we regard causes as
explaining their effects in a way in which effects do not explain
their causes, and the mere fact of temporal priority would not
account for this. A corollary is that if two events occur close
together this may be explained by their being collateral effects of
a common cause, but not by their being joint causes of a single
effect. The same principle applies if two features, or two kinds of
event, tend frequently to occur together. A tendency for whi te hair
and pink eyes to go together may be explained by these features
being both produced or encouraged by a single gene; but we could
not explain a tendency for high humidity and falling barometric
pressure to go together by their being joint causes of the falling
of rain. It may be objected that such joint causes are not really
time-reversed images of collateral effects. Since, given some of
the circumstances, the common cause is necessary and probably
also sufficient for one effect, and given others of the circum-
stances, it is necessary and probably also sufficient for the
other effect, the time-reversed image of this situation would
be one where, given some circumstances, one cause was sufficient
and probably also necessary for the common effect, while, given
others of the circumstances, the other cause was sufficient and
probably also necessary for that same effect. The nearest we
could come to this would be a case of causal over-determination.
And we could still use this example to make our point: we
do not regard the co-occurrence of a pair of over-determining
causes—say the flash of lightning and the throwing of the
cigarette butt in example (ii) in Chapter 2—as being explained
by the fact that they do between them over-determine their
common effect—the fire in the straw. But, the objector may
say, the cases are still not parallel. Given all the circumstances,
the common cause, C, may be sufficient—in the strong, counter-
factual, sense—for each of its collateral effects, E1 and Ez, and
it is only when it is so that it explains their co-occurrence. But
 THE DIRECTION OF CAUSATION 165
the common effect, E, cannot in this sense be sufficient for each
of its over-determining causes, C1 and Ca. For if it were, each
of them would be necessary in the circumstances for it; but
neither of them is necessary in the circumstances just because
the other also occurs. (The letters C, E1} etc. here refer to
individual events.)
But the extreme difficulty which we encounter in setting up
a time-reversed image of the commonplace pattern of events in
which one cause has two collateral effects itself illustrates the
point that we work all the time with a concept of causal direc-
tion that is not reducible to temporal direction. We are happy
to say that, given the circumstances, C was sufficient for Et and
for Ez: if El had not been going to occur, C would not have
occurred, and so, since C was necessary in the circumstances for
EZ, Ez would not have occurred either. And the same holds
with El and E2 exchanging places. We are willing to accept
the claim that there is no causally possible world which includes
all the circumstances and Elt but not £2, or vice versa. But in
the proposed time-image of this case, we are not prepared to
say that Cj and (72 were each necessary (as well as sufficient)
in the circumstances for E. If C\ was sufficient for E, then Ca—
even if it was in fact simultaneous with Ct—could not also be
necessary for E, in the circumstances. We are convinced that
there must be possible worlds which include the relevant cir-
cumstances and Clt but not C2, and vice versa. Cx and C2
cannot be tied to co-occurrence by their relationships to a later
event E, whereas EI and E2 can be tied to co-occurrence by
logically analogous relationships to an earlier event C. But if
we cannot keep the rest of the conceptual pattern while reversing
the time-direction, we must be using a notion of some other
asymmetry which we believe to be somehow linked with the
asymmetry of time. And this must be something other than the
corresponding regularities, for these are time-reversible. We
cannot plausibly lay it down, a priori, that Q and C2 cannot be
both necessary, as well as sufficient, in the circumstances for E
in the regularity sense of 'necessary' and 'sufficient'. It must be
possible that Q, C2, and E and the relevant circumstances
should be such that whenever in those circumstances an event
of the Cx type occurs an event of the E type follows, and that
whenever an event of the E type occurs, an event of the C2
166 THE DIRECTION OF CAUSATION
type has preceded, so that an event of the C1 type never occurs
without one of the Cz type occurring also, and vice versa.
These arguments show that we have some concept of a causal
priority that is something more than temporal priority, and
that is not constituted by regularities; but they do not yet tell
us what this concept is. Does Russell's suggestion that one-way
determination can be nothing but many-one correlation help
us here? The suggestion is that if we have a law which says that
whenever an event occurs of any one of several types, say A, B,
C, D, an event of type E occurs also, and that whenever an
event of type E occurs, an event occurs either of type A or of
type B or of type C or of type D, then if an event of type A,
say A-L, and one of type E, say Elt occur appropriately together,
satisfying both halves of the law, we shall say that it is
AI that is causally prior to £a; and it is suggested that only a
relation of this form will fulfil our concept of causal direction.
Now it is true that in such a case we can say not only that At
was sufficient for E1} in a strong, counterfactual, sense, but also
that A1 was necessary in the circumstances for Elt since no
event of type B or C or D occurred; equally we can say that E{
was necessary for A^\ but we cannot say in the strong sense
that El was sufficient for Alt since E1 might well have been
associated not with an A but with a B or a C or a D. So there
is a sense in which we can agree with Russell that the occur-
rence of A1 determines that of E1} but not vice versa. But this
has nothing to do with our ordinary concept of causal priority.
Any attempt to base causal priority on such a relationship
would, for instance, be at once overruled by temporal order.
Suppose that the law ensured that any E would be closely
followed by either an A or a B or a C or a D, and that any one
of these would be closely preceded by an E; none the less
when E! occurred and was closely followed by Av we should
say that E^ had brought about At, despite the known fact that
£s could bring about 5s or Cs or Ds as well as As. This case
would in fact be analogous to that of our partially indeter-
ministic slot-machine L in Chapter 2. The existence of a law of
this form does not in the least tempt us to talk of backward
causation, of the later At bringing about the earlier Et. So
Russell's one-way determination does not produce a corre-
sponding assignment of causal priority; and on the other hand
 THE DIRECTION OF CAUSATION 167
we are often ready to assign causal priority when we have no
thought of anything like Russell's relationship. No doubt we
are at least dimly aware of Mill's plurality of causes, of the
fact that the effect, described in whatever, perhaps fairly broad,
way we are thinking of it, could have been brought about by
alternative causes instead of the actual one; but equally we are
aware not only that this same cause would have had alternative
outcomes in other circumstances, but that even in the same
circumstances a cause which would still fall under the descrip-
tion under which we are thinking of this one might well have
other effects. And where, in scientific contexts, we describe the
relevant field, the cause, and the effect so accurately that there
is no room left either for alternative effects of this cause or for
alternative causes of this effect, we still confidently assign causal
priority to the earlier event. It may be true that if we were
thinking of causation purely in terms of regularities, but had
advanced from neolithic to functional regularities, and had
adopted the special feature of Russell's Newtonian example
that the function related simultaneous and instantaneous values
of the relevant quantities, then we could find nothing but
Russell's one-way determination to mark off causes from effects;
but this procedure would have even less to do with our ordinary
concept of causal priority than Hume's idea arid impression of
necessity2 have to do with our ordinary concept of what marks
off causal from non-causal sequences; Russell, like Hume,
would be advocating a conceptual reform.
In order to express the strong, counterfactual, sense in which
a cause may be—though it need not be—sufficient for its effect
I have used statements of the form 'If E had not been going to
occur, C would not have occurred'—for example, 'If the chest-
nut had not been going to become flatter, the hammer-blow
would not have been struck'. But this, even where it is accept-
able, where we are prepared to expand it into the claim that
in the appropriate possible world in which the chestnut does
not become flatter the hammer-blow is not struck, is obviously
much less natural than the converse counterfactual, 'If the
hammer-blow had not been struck, the chestnut would not
have become flatter'. This contrast in naturalness is clearly
connected with the fact that the hammer-blow is causally,
not merely temporally, prior to the flattening. If we were
i68 THE DIRECTION OF CAUSATION
entertaining a precognition hypothesis, we should find it quite
natural to say that if the later, target, event had not been going
to occur the foreseer would not have had the experience he did.
But we can hardly use this contrast in naturalness as an analysis
of the notion of causal priority: we must go further and try to
understand why we find one antecedent-consequent order in
a conditional linking two (possible) events more natural than
the other.
A more promising approach is by way of the notions of
control and 'effectiveness': the direction of causation is con-
nected with the way round in which we might use a causal law
as a recipe. A cause is effective as a means of bringing about its
effect. The striking of a match on a matchbox is an inus condi-
tion of the appearance of a flame, and vice versa. But I can use
the striking as a means to the flame (if the other conditions are
right), whereas although I can produce a closely similar flame
independently (say, by applying another, already lighted,
match to this one) I cannot use the flame as a means to the
previous striking of the match on the box.
But this brings out one of the queerest aspects of the hypo-
thesis of backward causation. One might suppose that if there
were a time-reversed exact analogue of causation, with causal
priority opposite to temporal priority, then one could use a
later event as a means to the bringing about of an earlier one:
in one of the stock examples, putting up a particular picture on
Tuesday would be a way of bringing it about that a precognizer
had had a corresponding experience on Monday, and even had
made, on Monday, a drawing that resembles this picture. And
this seems absurd.4 If there were such a thing as backward
causation, it would somehow have to stop short of offering us
means of bringing about the past.
The causally prior item, then, seems to be the one which we
can directly control, and by way of which we can indirectly
control the other. Causes are effective, effects are not.
4
This point is used, by Flew, in Analysis, xvii (1956-7), 81-4, to argue that what
Scriven offers as an example of backward causation would not be causation because
it leaves out 'effectiveness'. Gale similarly (op. cit., p. 217) takes 'effectiveness' to
be an essential feature of a cause, saying that 'a cause is that which can be used
for the purpose of making, or bringing it about that, something else happens',
and concluding from this that 'the ordinary concept of causality is anthropo-
morphic'.
 THE DIRECTION OF CAUSATION 169
Can we, then, simply equate causal priority with effective-
ness? This suggestion would fit in well with one of the main
uses to which we put the notion of causal priority. If there is
some evidence of a causal connection between two items—for
example, if there is a significantly high positive statistical cor-
relation between self-confidence and skill in the performance of
a certain task—then until we know the direction of causation
we cannot base any practical policy on the causal connection.
We want to know whether it is self-confidence that increases
skill or skill that generates self-confidence. Only if the former
relation holds can we hope to increase people's skill by en-
couraging their self-confidence.
Can we link this suggestion with the above-mentioned con-
trast in naturalness between our converse counterfactuals? The
bringing down, or not bringing down, of the hammer is thought
of as having been directly in someone's power. The possible
world envisaged in the conditional 'If the blow had not been
struck, the chestnut would not have become flatter' could have
been the actual world, and the way of making it actual would
have followed the order of antecedent and consequent in the
conditional: the supposition which the antecedent introduces
corresponds to the first change the agent would have intro-
duced. The possible world envisaged in the converse condi-
tional 'If the chestnut had not been going to be flatter, the
blow would not have been struck' may well be the same one
(though it need not be), but there was no way of making it
actual except by operating first on what appears here as the
consequent. This counterfactual seems unnatural, because the
supposition-consequent order in it reverses the order in which
an agent would have to operate. This does not make the counter-
factual improper, but it does make it awkward.
It might seem to be a difficulty for this suggestion that such
effectiveness, and hence causal priority thus analysed, applies
in the first place to regularities, to general causal statements or
laws, since it is these that we can use as recipes. However, it is
easy to see how a notion thus based can be extended so as to
apply to various kinds of singular causal sequence. If a sequence
is still in the future then (if C is necessary in the expected cir-
cumstances for E) we can say that someone by preventing C,
or by failing to bring C about, will thereby prevent E or fail
i?o THE DIRECTION OF CAUSATION
to achieve E; and if C is also sufficient in these circumstances
for E, we can say that someone could in principle bring about E
by means of doing or bringing about C. And about a past
sequence we can say similarly that if someone had prevented or
failed to bring about C, he would thereby have prevented or
failed to achieve E, and (if C was also sufficient for E) it may
be that someone did bring about E by means of C.
It is true that this analysis makes singular assertions of causal
priority doubly anthropocentric, first because they come out as
conditionals (which, I have argued, essentially express sup-
positions, and are not in general capable of being simply true) 5
and secondly because the topics of those conditionals are the
actions of human, or something like human, agents. Can this
be reconciled with the fact that we apply our causal concepts,
including that of causal priority, to cases that are far beyond
human control, and to events that took place long before the
human race existed, indeed that had to take place if the human
race was later to exist? The hypothesis that the solar system
was produced by the approach of another star to our sun in-
volves the supposition that this approach was causally prior to
the formation of the planets; but we can surely entertain this
hypothesis while believing that there were no agents in existence
at the time, and certainly none that could have used the rele-
vant laws as a recipe, none for whom it would have been
possible to make a set of planets by bringing one star close to
another. Still, it may well be that we are here extending the
application of the notion of causal priority by an implicit use
of conditionals, we are in effect saying that the star-approach/
solar-system sequence was one such that if there had been an
agent with the requisite power, etc. Nevertheless, this would be
a rather surprising conclusion. We do not commonly think that
we are referring even to hypothetical agents when we see causes
as related asymmetrically to their effects in some more than
merely temporal way. And if this is what in the end emerges
from an analysis of our concept of causal priority, it will raise
all the more sharply the separate ontological question whether,
since this concept of priority cannot be literally satisfied by 'the
objects', there is some other thoroughly objective directional
feature in causation. It certainly seems that our voluntary
5
Cf. Truth, Probability, and Paradox, Chapter 3.
 THE DIRECTION OF CAUSATION 171
actions give us our primary, direct, awareness of causal
priority. It is not unlikely that something that belongs essentially
to them is at least the core of our concept of causal priority.
Nevertheless, there may be something of which we are then at
least dimly aware, something objective and not essentially tied
to human agency and intervention.
Von Wright has adopted just such an analysis of causal
priority.6 'I now propose the following way of distinguishing
between cause and effect by means of the notion of action: p is
a cause relative to q, and q an effect relative to p, if and only if
by doing p we could bring about q or by suppressing p we could
remove q or prevent it from happening.' To the objection that
causation holds outside the scope of human action he gives two
replies. Though the eruption of Vesuvius was the cause of the
destruction of Pompeii, and men cannot make volcanoes erupt,
there are distinguishable events within the complex event, the
destruction of Pompeii—stones hitting men on the head, roofs
collapsing under loads, and so on—which are of kinds with
which we are familiar from experience 'and which are such
that the cause-factor typically satisfies the requirement of
manipulability'. Action-conferred priority in analogues of the
component events somehow constitutes the priority in the
complex, non-manipulable event of which they are components.
And although causation 'operates throughout the universe—•
also in spatial and temporal regions forever inaccessible to man',
von Wright insists that 'to think of a relation between events
as causal is to think of it under the aspect of (possible) action'.
This analysis raises all the problems I have mentioned, and,
as von Wright's subsequent discussion shows, other difficulties
as well. To explain the asymmetry of the causal relation in
particular cases, he says that 'In the cases when I bring about q
by doing/*, p is the cause and not q\ even if on other occasions
an event of the same type q causes an event of the type/*, namely
when I bring about p by doing q; but this distinction (if p and q
are simultaneous) 'requires that there is some basic action, i.e.
an action which we can do "directly" and not only by doing
something else'. If/» were a basic action, such as raising my arm,
it would be the cause and q the effect even if p and q were
simultaneous. But this assigning of causal priority to basic
6
Explanation and Understanding, pp. 70-81.
i?2 THE DIRECTION OF CAUSATION
actions has paradoxical consequences. There may well be
neural events or processes which are both necessary and suffi-
cient for any muscular activity, and which slightly precede in
time the rising of my arm. Then, if raising my arm is the basic
action, it must cause the associated necessary and sufficient
neural processes: by raising my arm I can bring it about that
these neural events have occurred. Von Wright accepts this as
a genuine case of 'causation operating from the present towards
the past',7 and of course this makes backward causation ex-
tremely common. But he also admits that a neurophysiologist
might produce or suppress neural events in such ways as to
make my arm rise or to prevent me from raising it. He reconciles
these facts by saying that there are different 'closed systems', in
one of which the rising of my arm, when I raise it, is causally
prior to the neural events, and in the other of which the neural
events, manipulated by the neurophysiologist, are causally
prior to the rising of my arm. But it is very tempting to say that
on an occasion when I actually raise my arm, so that by von
Wright's account the neural events are the effect, they are
causally prior in that a neurophysiologist could have interfered
with them and could thereby have prevented my arm from
rising. The talk of closed systems allows von Wright to say that
only one direction of causation is actually operated in any single
case, but he has obviously abandoned the assumption that a
specific direction of causation holds between the objective
events. Such an objectivist assumption would require that if the
neurophysiologist could have brought about not-q by suppress-
ing p, then p was in fact causally prior to q, and it would then
be contradictory to say that since I actually brought about p
by doing q, q was also in fact causally prior to p.
These further difficulties, the acceptance of frequent back-
ward causation and the abandonment of objective causal order,
should make us even more dissatisfied with the proposed
analysis of causal priority. Surely von Wright has mistaken the
experience which gives us our primary awareness of causal
priority for the relation of causal priority itself. It is no doubt
by performing 'basic actions' and seeing what they bring about
that we originally detect the asymmetrical relation between
7
The argument is due to R. M. Chisholm, 'Freedom and Action', in Freedom
and Determination, ed. K. Lehrer.
 THE DIRECTION OF CAUSATION 173
individual effects and causes, but there is no need to say that
this relation just is being-brought-about-by-means-of, or that
to think of a relation between events as (unidirectionally)
causal 'is to think of it under the aspect of (possible) action'.
This analysis, promising though it appeared, is only a first
attempt; we must try to go beyond it. If we succeed, we may
be able to conclude that the neural events are objectively
causally prior to my arm's rising, even though my raising of my
arm is 'basic' within my experience, since it is what I see myself
as just doing, and not as bringing about by way of doing some-
thing else. It is as misleading to say that I bring about these
neural events by raising my arm as it would be to say, when
I am driving my car, that by going round a corner to the left I
bring it about that the track-rods move to the right.
Let us try another approach, examining a type of experiment
that could conceivably be used to provide evidence for back-
ward causation.8 By so doing, we may not only defend the
claim made earlier that this possibility can be coherently
conceived but also, by discovering what would count as evidence
for it, find out more about our concept of causal priority.
On Monday a person who claims to have powers of pre-
cognition is asked to draw a copy of a pattern that is going to
be displayed in a certain place the next day; on Tuesday the
pattern is produced by some randomizing device and duly
displayed; and, let us suppose, the precognizer's drawing turns
out to be exactly like the random pattern which it was intended
to copy. Prima facie, this result is evidence that this person
really has precognized the pattern. Of course, several pre-
cautions will have to be taken if this evidence is to have any
weight. We must check that the drawing has not altered or been
altered to bring it into agreement with the pattern, that it still
is on Tuesday just as it was drawn on Monday. We must
exclude various possible ways in which the supposed pre-
cognizer, or his drawing, might have controlled, or fed informa-
tion into, the alleged randomizing device. Again, a single
favourable result, however surprising, might be written off as
an accident; but if the experiment were repeated a number of
8
This is based on Scriven, op. cit. An important feature of Scriven's experiment,
the use of several precognizers, is temporarily omitted but will be brought in later,
to make a distinct point.
174 THE DIRECTION OF CAUSATION
times with all the appropriate precautions and continued to
yield favourable results, then this would begin to constitute
weighty evidence that precognition was occurring. And if we
conclude that it is precognition, we are not merely saying that
if the pattern had been different the drawing would have been
correspondingly different in a sense in which the converse
counterfactual—that if the drawing had been different the
pattern would have been correspondingly different—would
have been equally true. We are saying also that the pattern is
responsible for the details of the drawing being as they are,
that it has determined the drawing and not vice versa; in
other words, that the pattern is causally prior to the drawing.
In this experiment, nothing depends essentially upon the use
of a human precognizer. We could describe a conceivable
experiment a favourable result of which might give evidence
of backward causation without precognition by replacing the
supposed precognizer by a machine which included, say, a
photographic film. If the film, when developed, regularly
showed a picture that closely resembled the pattern sub-
sequently produced by the randomizing device, then, given all
the appropriate precautions, we should have evidence that this
photographic machine was being affected by the subsequently
produced pattern. In particular, we need not make any use of
the fact that a human precognizer would be trying to draw on
Monday a pattern that was to be shown, in some place known
to him, on Tuesday, that he was directing his attention in a
specific way. Nor need we even make use of the fact that a
photographic plate, though itself devoid of purpose, is com-
monly intended, by those who make and use it, to record things.
We could replace the plate by, say, iron filings on a flat sheet
of glass, which are found to exhibit, on Monday, a pattern very
like that which is displayed nearby on Tuesday. However, I
shall continue to discuss this form of experiment with reference
to a precognizer's drawing, though it is to be understood that
nothing turns upon this special feature.
But how would a successful experiment of this sort provide
evidence for backward causation? It is not the randomness of
the pattern, in the sense in which this is an intrinsic feature of
the pattern, which tends to show its causal priority; for the
drawing, being very like the pattern, will share this sort of
 THE DIRECTION OF CAUSATION 175
randomness with it. This sort of randomness, in both items, is
important only because it makes it unlikely that two similar
random figures should have come into existence close together
but independently. Reduced to essentials, all that we have here
is a surprising degree of agreement in detail, so good that it is
hard to set even a single case aside as a coincidence. In each
single case there is prima facie evidence of some sort of causal
connection, and this is made much stronger by repetition. But
this evidence is in itself symmetrical. How then do we go from
'some sort of causal connection' to backward causation ? Only
by ruling out explanations in terms of forward causation.
There are two different methods by which the defender of
the experiment may try to achieve this. First, he could take
care to exclude various specific ways in which the drawing
might have influenced the pattern—locking the drawing up as
soon as it is made, making sure that no photograph or descrip-
tion of it gets around, putting the precognizer into a twenty-
four hour coma as soon as he has made his drawing, and so on.
But these specific exclusions may not be enough. They leave
us with a choice between some mysterious, hitherto unknown,
variety of forward causation and the equally hitherto unknown
and even more mysterious possibility of backward causation.
The former would seem to be the lesser evil. The defender gives
his hypothesis more initial plausibility by speaking of pre-
cognition, that is, by assimilating his hypothesis to the familiar
process of visual perception which, as we know, can reproduce
a complicated pattern in much of its detail. But since pre-
cognition would have to be utterly different from ordinary see-
ing, which uses normal causal mechanisms, this analogy does
not really support the precognition hypothesis against its rival;
foreseeing would be too unlike seeing to be prima facie any
better an explanation than 'some mysterious forward causal
connection'. And in any case even the initial plausibility dis-
appears if we replace precognition, as suggested above, by
something like our iron filings on a glass sheet.
Secondly, the defender of the experiment may try to exclude
all forms of forward causation, even unknown ones, either by
making the pattern that appears on Tuesday the product of
one or (preferably) more than one free human choice, or by
making it the product of what is in a. very strong sense a random
i?6 THE DIRECTION OF CAUSATION
device, something which operates indeterministically, by objec-
tive chance. Its nucleus, say, is something like a Geiger counter
with some radioactive material nearby. In either case he is
claiming to have terminated the regress of causal antecedents
of the pattern so that it cannot reach back to the drawing.
Of course the fact that there is something indeterministic in
the pattern's causal antecedents would not prevent the draw-
ing from being causally prior to it. A grandfather's genes help
to determine his grandchild's character even if it is only by
way of various free, indeterministic, choices that the grand-
child comes into existence at all. Similarly the drawing might
influence the pattern even if something indeterministic allowed
the pattern to be produced. What is wanted is an indeter-
ministic source for the pattern's having just those features,
rather than others, which constitute its too-close-to-be-fortuitous
resemblance to the drawing. The defender's argument then has
this general form:
A and B are too alike for their resemblance to be coincidental;
so there is some causal connection between them. Therefore
either A is causally prior to B or B is causally prior to A. But
B's origin is indeterministic in a way that precludes the tracing
back to A of the causation of the features relevant to this
resemblance. So A is not (in the relevant respect) causally prior
to B. Therefore B is causally prior to A.
Thus set out, the argument has an obvious flaw. The re-
semblance might be explained by A and B having, in the
respects relevant to their surprising resemblances, a common
cause rather than by either being causally prior to the other.
But this possibility is also blocked by the tracing of B's possession
of the relevant features to an indeterministic source. This pre-
vents not only the tracing back of their causation to A, but also
the tracing back of their origin to any C which could also be
causally responsible for A.
Spelling out the argument in this way reveals two further
things. First, the argument assumes that where there is an
explanatory causal connection between any two items there
must be some causal priority relation(s), either one item being
causally prior to the other or some third item being causally
prior to both. Secondly, it assumes that we already know the
direction of causation between one or more free choices and
 THE DIRECTION OF CAUSATION 177
their outcome or between an indeterministic device and its
product. This experiment, then, cannot be claimed to yield a
primary awareness of causal priority; it cannot purport to con-
tain what Hume would call an impression of causal priority; so
it does not, after all, throw much light on this idea. Even if it
can be defended, this experiment does not in itself reveal what
our concept of causal priority is or what is our fundamental way
of recognizing it. But it has made one small contribution: we
have seen that any indeterministic device would do instead of
a free agent; there is some hope after all of detaching causal
priority from human agency and 'effectiveness'.
Can we say, then, that the basic notion of a causally prior
item is of something that just happens: something pops up from
nowhere and events run on from there? (Of course not every-
thing to which we ascribe causal priority is itself seen as such a
sheer beginning. If X is causally connected with T, and T with
£, then if X is a sheer beginning, it is causally prior to T, but for
this very reason T is also prior to /£. Relative causal priority is
transmitted linearly from items that are seen as having absolute
causal priority, that is, causal primacy.) This would seem to
agree with our preliminary sketch of the notion of causal
priority in Chapter 2. But an objection is that we seem to be
trying to define causal priority in terms of causal primacy, and
causal primacy in terms of the negation of causal priority: the
item which just happens, which pops up from nowhere, is
simply something to which nothing is causally prior. And it is
not, of course, something to which nothing is causally linked:
it is typically the beginning of a chain of causation. Can we
then ascribe causal primacy to something merely because it is
at one end, and not in the middle3 of a chain of causation? No,
because it is quite conceivable that a causal sequence should
die out, that there should be a last event in the chain rather
than, or as well as, a first, but the last event would not be seen
as having causal primacy or as transmitting causal priority to
its neighbours. This line of thought, then, seems to yield nothing
new. We are driven back to agency and 'effectiveness'. We
seem to have to say just that X is causally prior to T where, if
an intervention had been applied to X, T would have been
different, and if we are asked what an intervention is we can
only say that it is the sort of thing we are sometimes aware of
 178 THE DIRECTION OF CAUSATION
making: we can give instances which will indicate the relevant
sort of experience.
If this were all that we could say about the concept of causal
priority it would be profoundly disappointing. Let us, then,
see if anything more can be extracted from discussion of the
precognition experiment. Something may be revealed by an
important objection to the claim that any result of such an
experiment, however favourable, could support a hypothesis of
backward causation.
Suppose that the first part of an experiment has been per-
formed, and a certain drawing has been made on Monday.
According to the precognition hypothesis, this is literally a
copy of a pattern that is to be produced on Tuesday. But then
suppose that someone decides to stop or destroy the device that
would have produced the pattern, so that no pattern is pro-
duced after all. The drawing cannot be a copy of a pattern
which never comes into existence, and thus on this occasion
the precognition hypothesis is false. But on every occasion, after
the drawing is made, it is possible that someone or something
should intervene so that the corresponding pattern fails to be
produced. Consequently, it cannot on any occasion be the
pattern that is responsible for the details of the drawing: the
precognition hypothesis must be false even for those occasions
when the device is not stopped, when the pattern actually is
produced and turns out to be just like the drawing.'
This objection rests on the assumption that past and present
events and states of affairs are fixed and settled and unalter-
able, whereas at least some future ones are still to be fixed, as
yet undecided. The argument is that if at any time A is fixed
while B is still unfixed, B cannot be causally prior to A, because
at this time things might still be so decided that B does not
occur. Can this objection be countered in any way? We may
dismiss the fantastic suggestion that past events may be not
fixed but still subject to decision; but the reply is still available
that some future events may already be fixed, and that these
alone can be precognized or have other effects in the present.
The objection tells only against (and indeed is fatal to) a sup-
posed case of backward causation where the occurrence of the
* Cf. Flew's three articles and Black's article mentioned p. 162, n. 3 above.
This is what Flew calls the 'bilking experiment'.
 THE DIRECTION OF CAUSATION 179
supposed cause is still undecided at a time when the supposed
effect has already occurred. But this undermines the argument
used above in attempting to extract a positive conclusion from
the experiment. Its defender was there using the assertion that
B was produced (on Tuesday) by free human choices or by an
indeterministic device to preclude any tracing of B's causal
origins back to A or to any earlier cause of A, say C. But we now
see that what was there used to show that neither A nor C was
causally prior to B can equally be used to show that B is not
causally prior to A. If B really is undecided on Monday then
its resemblance to A must be fortuitous, however implausible
this may seem: we have ruled out the possibility of causation
in either direction between them or to both from a common cause.
The defender of the experiment might try to meet this
objection by saying that B could (already by Monday) be
fixed in something stronger than the trivial logical sense that
if, as it turns out, it does occur on Tuesday it was, on Monday,
already going to occur. He could suggest that B may be fixed
in a sense strong enough to block the bilking experiment with-
out having deterministic causal antecedents. I do not think
that this can be reconciled with B's being, in the relevant
respects, the product of a genuinely free choice as we conceive
such choices; but it might be reconciled with B's being the
product of an indeterministic device. The full freedom of choice
seems to require a metaphysical non-fixity in its object that
goes beyond the mere absence of causal determination. Alterna-
tively, the defender might fall back from the second to the first
of the two methods mentioned above, abandoning the require-
ment that B should be produced, on Tuesday, by something
indeterministic, allowing B's occurrence on Tuesday, with the
relevant features, to have been already fixed by Monday by
some sufficient cause C, and merely trying to check that C does
not affect A by any kind of forward causation and that A does
not affect the chain of causes by which C brings about B. He
is then, as we saw, left with a choice of hypotheses, and that of
some mysterious forward causation may be the lesser evil; but
as we shall see later the defender of the experiment has one
further resource.
However, the possibility of precognition, or of backward
causation generally, matters less to us than what the arguments
i8o THE DIRECTION OF CAUSATION
for and against this possibility reveal about the concept of
causal priority. The defender's difficulties arise from his trying
to meet two requirements at once: that the causally prior item
should lie, in some line of causation, between the causally pos-
terior item and some intervention, and that the causally
posterior item should not be fixed at a time when the causally
prior one is not fixed. The second is required for the priority
itself, the first for any detection of the priority; the difficulty
lies in securing that both requirements should be known, or
even reasonably believed, to be met when the causally prior
item comes later in time than the causally posterior one.
But let us attend for the moment to the second requirement.
It is, so far, just that an effect cannot be fixed at a time when its
cause is not fixed. But this has the corollary that if A and B are
causally connected in a direct line, then if B is fixed at a time
when A is not fixed, it must be A which is the effect and B the
cause. It is a point in favour of this account that while it clearly
relates causal order to temporal order—the easiest way of
getting B fixed while A is unfixed is to have B occur before A,
with room for some indeterminacy between them—it does not
simply reduce one to the other, but mentions this further
feature of fixity. Moreover, it leaves open at least the logical
possibility of backward causation, however difficult it might be
to acquire firm evidence for this: it is possible that, although A
occurs at <2 and B at the later time t3, B should have already
been fixed at the earlier time t^ by the occurrence then of a
sufficient cause for B, while A was still unfixed at tv We leave
room for backward causation but not for bringing about the
past: we cannot at ta bring about A at t%, since B must already
have been fixed before ta if it is to be a cause of A. But it is con-
ceivable that we might at £t bring about A at <2 by way of B
at t3 and the backward link from B to A: we might bring about
something in the future by means of a line of causation part of
which is time-reversed.
This account of causal priority in terms of fixity and unfixity
is closely related to that in terms of effectiveness and inter-
ventions. We said that X is causally prior to Y where, if an
intervention had been applied to X, Y would have been different,
but not vice versa. Now if T had been fixed at a time when X
was not, an intervention could in principle have been applied
 THE DIRECTION OF CAUSATION 181
to X at that time; but if it had been, T, being fixed already,
would not have been different. That is, the fixity/unfixity
pattern which rules out a certain direction of causation rules
out automatically the corresponding effectiveness. Conversely,
if at a time when X is not fixed Y is not fixed either, and X is
necessary in the circumstances for T, then intervening to pre-
vent X, or failing to intervene to bring about X, will prevent the
occurrence of T.
But does the account in terms of fixity add anything to that
in terms of effectiveness ? In particular, does it free us from the
deplorable anthropocentricity of talk about interventions? Or is
our notion of unfixity simply that of something's being in
principle open to being brought about or prevented by a
deliberate act? Not quite. If events have preceding sufficient
causes in the strong, counterfactual, sense then they are fixed as
soon as these causes have occurred; and this will hold if there
are regularities of the form 'Whenever in the field F the con-
junction ABC occurs, P follows'. If, on the other hand, there
are events which occur but not in accordance with such regu-
larities, not as necessitated by any preceding sufficient causes—
though they may have had preceding necessary causes, and
again may have had causes in the sense defined by Ducasse,
'such particular change as alone occurred in the immediate
environment . . . immediately before'—then it is natural to
regard these events as unfixed until they occur. While it would
be very difficult to obtain conclusive evidence that a certain
event lacked any preceding sufficient cause, it would not be
difficult to find some evidence for this, and the hypothesis that
events of certain kinds commonly occurred without sufficient
(or 'necessitating') causes would be as much open to confirma-
tion as scientific generalizations usually are. In other words,
there could be knowledge about a class of (presumptively) un-
fixed events which was not derived purely and directly from
our experience of free choices and interventions. Radioactive
disintegrations are a case in point.
It is true that an event which was thus unfixed until it
occurred could not be the cause in a bit of backward causation:
for if it were necessary in the circumstances for some preceding
event, that preceding event would be sufficient in the circum-
stances—in the strong, counterfactual, sense—for it, so that it
 182 THE DIRECTION OF CAUSATION
would have been fixed as soon as that earlier event occurred.
(This is another variant of the argument about the bilking
experiment: it was for this reason that we had to choose, as a
conceivable backward cause, an event which was unfixed not
right up until its occurrence (at t3) but only much earlier, at
tv) In any instances of causal priority that we identified directly
with the help of these unfixed events the causes would also be
temporally prior to their effects.
It may be objected to this account of causal priority in terms
of fixity that it would make it impossible for a sufficient cause
to be causally prior to its effect: if X is a sufficient cause of T,
then T is fixed as soon as X is (unless the relevant circumstances
are still unfixed). Whether this is so depends on how we clear
up a vagueness that has so far been left in this account. We have
said that if X causes T, there cannot be a time when X is un-
fixed and T is fixed. But do we want to add that if X causes T
there must be a time when X is fixed and T is unfixed? It is
the latter that would deprive sufficient causes of causal priority.
Since our notion of causal priority seems not to exclude suffi-
cient causes, we should not add this second condition to our
analysis. But then how do we detect the priority of a sufficient
cause? The unfixity of the relevant circumstances may reveal it.
Otherwise we must fall back either on our experience of inter-
ventions, or on the continuity of causal processes. Even if X is
sufficient for T, it may be plain enough that it was to X, not
to T, that the intervention was applied; although a spark was
sufficient in the circumstances for an explosion, it was the spark
that I fairly directly made, say by striking flint and steel together.
Or again, even if X was sufficient for T, it may be plain that
some continuous process of movement or change, starting from
some other event £ that was unfixed until it occurred, led on
to X and only through X to T. But 'basic actions' as we see
them are not an infallible guide to the direction of causation.
In the example discussed by von Wright, though my raising of
my arm is a 'basic action', there can be no doubt that the
associated neural processes were fixed at a time when the
rising of the arm was still unfixed: there was a (very short)
stretch of time into which interferences of various sorts might
have intruded, so it was the neural processes that were causally
prior to the rising of the arm.
 THE DIRECTION OF CAUSATION 183
We seem to have found, then, in this notion of fixity a basis
for the concept of causal priority which does not reduce it
merely to temporal priority, which leaves a small gap into
which backward causation (but not the bringing about of the
past) might conceivably be fitted, but which will not (as an
account solely in terms of effectiveness would) proliferate very
brief displays of backward causation, and which frees causal
priority from complete anthropocentricity, while still allowing
us to explain both why it is our active interventions that mainly
give rise to this concept and why temporal and causal priority
tend so strongly to go together. But let us now leave this account
aside, and consider some other suggestions that have been made
about asymmetries in causal processes.
There is a class of processes, pointed out by Popper as being
temporally asymmetrical, of which an example is the system of
circular waves spreading out from a centre, as when a stone is
dropped into a pool. The reverse process, with circular waves
contracting to a centre, would be equally compatible with the
laws of physics, but it never, or hardly ever, occurs, whereas
expanding wave systems are extremely common.10 As Popper
says, the reverse process with contracting waves * would demand
a vast number of distant coherent generators of waves, the
coherence of which, to be explicable, would have to be shown
. . . as originating from one centre'. That is, we have here a
situation in which A is a single event (the disturbance at the
centre of the pool), B is an occurrence which involves a sort of
coherence between a number of separated items (a wave motion
all round the circumference of the pool), and there are causal
laws in accordance with which A could lead to B, but B could
equally lead to A. 5, however, is here in need of explanation in
a way in which A is not; consequently A could satisfactorily
explain 1?, whereas before B could explain A the coherence in
B itself would need to be explained by reference to some other
single event C. (We could explain the contracting wave pattern
easily enough if it arose from the reflection, at the edge of a
circular pool, of a wave that had previously spread out from a
stone dropped at the centre, or if a flat circular ring, just inside
the Circumference of the pool, were connected with a handle
10
K. R. Popper, 'The Arrow of Time', Nature, clxxvii (1956), 538, also Nature,
clxxviii (1957), 382, and dxxix (1958), 1297.
184 THE D I R E C T I O N OF CAUSATION
and moved up and down as a whole.) Examples of this kind
establish a direction of explanation, from some central occur-
rence to the coherence of a number of separated items: explana-
tion runs to situations involving dispersed order from ones
which do not. Let us say that the central disturbance as com-
pared with the expanded wave is prior with respect to the dispersal
of order. If we add the principle that a cause explains its effect
in a way in which the effect does not explain the cause, it
follows that there can be a satisfactory causal explanation only
if the causally prior item is also prior with respect to the dis-
persal of order. If the opposite were the case, if the item exhibit-
ing dispersed order were causally prior to the only related
central one, the two requirements for explanation would be in
conflict with one another. These points are, however, not directly
relevant to the direction of time. It is clear that it is logically
possible that a particular process which is the reverse of a normal
one—that is, one in which what is prior with respect to the
dispersal of order comes later in time, in which the order-
dispersal direction is opposite to the order-dispersal direction
of most processes in the same tirne-continuum—could occur
without there being any normal process available to explain
its starting point. This must be so, since to get such a reversed
process we need only take something like a contracting wave
that in fact results from the reflection, at a circular edge, of a
previously expanding one, and wipe out the earlier, expanding,
phase. Nor would such a reversed process be intrinsically in-
explicable, To explain it we need only to establish, or assume,
that the item which was prior with respect to the dispersal of
order was also causally prior, despite the fact that it came later
in time; that is, such a process would be explained in terms of
backward causation, If in fact such a reversed process never
occurs without a previous normal process to set up its initial
conditions, this merely exemplifies the general truth that pro-
cesses which would call for explanation in terms of backward
causation do not actually occur, though they are not logically
ruled out.
But why is it so plausible to say that a cause explains its
effect, but not vice versa? Surely this is due to the link between
causal priority and fixity. The explanatory account retraces the
order in which the effect was either brought, or allowed, into
 THE DIRECTION OF CAUSATION 185
existence, the way that it came to be fixed. Although we may
be able to infer a cause from a previously known effect, this
would show only thai the cause occurred: it would smooth the
way to our knowledge of the cause, but not to its existences
whereas if we explain an effect by reference to a cause we see
how the way was smoothed to the existence of the effect. And
above all, if the effect was unfixed when the cause was fixed,
we cannot explain the cause's actually being there by reference
to something which still might not have happened; what is
explained must depend upon what explains it, so the latter
cannot have been ontologically less solid than the former.
An example of dispersed order, then, such as the outer
circular wave, is a set either of collateral effects or of joint
causes of the central event. Because of its complex coherence,
it is intrinsically improbable and in need of explanation. Since
causes explain effects, and not vice versa, the central event will
explain the dispersed order only if it is its cause, not if it is its
effect. But the central event could still in principle explain
the dispersed order as the co-occurrence of its effects even if
the central event occurred later in time, provided that we could
say that the central event was fixed at least as soon as the
scattered effects were, that they came to be fixed by way of it.
And this points to what could be another kind of evidence for
backward causation.
In many cases what is, or could be, a set of collateral effects
is an example of dispersed order, and therefore is in need of
the explanation that is provided by the tracing of them to a
common cause. But we might find them all related, by con-
tinuous processes which conformed to time-indifferent physical
laws, to some later central event, and we might fail to find any
similar relation to any earlier central event, even after looking
very hard for one. Then, unless there were some strong reason
for saying that that later central event could not have been
fixed by the time the dispersed order items were, it would be
tempting to speculate that the central event was causally prior
to the dispersed order and so explained it, by a time-reversed
causal process. Indeed we can greatly strengthen our pre-
cognition experiment by adding this feature to it." Suppose
that instead of just one alleged precognizer we have a number,
11
Scriven, op. cit., makes use of this feature.
186 THE DIRECTION OF CAUSATION
all making drawings on Monday of the pattern that is to be
displayed on Tuesday, and suppose that all the drawings closely
resemble the pattern that is later produced. Then the mutual
resemblances between the drawings of this set are themselves in
need of explanation; the set of drawings is itself an example of
dispersed order. If there was no communication between the
precognizers, and no other common cause occurring earlier in
time can be found, then there is a case for taking the subsequently
produced pattern as the common cause, and it is plausible to
say that a successful experiment of this modified form would
give strong support to a hypothesis of precognition. Not, of
course, overwhelming support: there is always the possibility
of some other, unknown, earlier central event influencing all
the precognizers. But the introduction of dispersed order, of
the set of many separate precognizers, makes forward causation
from the drawings to the pattern useless as a rival hypothesis: such
joint or over-determining causes might explain the resemblance
of the pattern to the drawings, but it would be powerless to
explain the resemblances of the drawings to one another.
But if we found—as we do not—many examples of the sort
illustrated by a contracting wave pattern without any discover-
able earlier central event to explain the coherence of the many
dispersed 'generators', these would be even better evidence of
backward causation than the precognition experiment, just
because they would display, as it does not, continuous processes
conforming to time-indifferent laws and connecting the dis-
persed order items with the later central event.
We find, then, not that priority with respect to the dispersal
of order is causal priority, but that it is evidence for causal priority,
and we can see why it should be so. If the two coincide, if
the central event is also the causally prior event, the situation
is explicable, whereas if they fail to coincide it is bewildering:
we have a situation which is in need of explanation, but no
explanation can be given.
This resolves what would otherwise have been a difficulty.
If, as I have argued, causal priority is not to be simply identified
with temporal priority, it becomes a synthetic proposition that
they always (so far as we know) coincide, that all, or at least
most, causation is forward causation. But where do we find the
data for this inductive conclusion? If our detection of causal
 THE DIRECTION OF CAUSATION 187
priority had been limited to experience of our own voluntary
actions, then even if causal priority always coincided with
temporal priority here, it would, be rash to conclude that they
coincide in processes of all other kinds as well. But since
priority with respect to the dispersal of order is evidence of
causal priority, we can and do have far more extensive evidence
that the directions of time and of causation normally coincide.
It is often suggested that the increase in entropy in any iso-
lated system constitutes a general temporal asymmetry in
causal processes. But this is not so.12 Let us consider the stock
example of a rigid chamber containing gas, initially with part
of the gas, at a higher temperature than the other part, con-
centrated at one end of the chamber; the two parts gradually
mix and the gas approaches a uniform temperature throughout
the chamber. If we consider this chamber as persisting for a
very long time, with no loss of energy from the chamber as a
whole, we must say that there is an enormous number of possible
arrangements of the gas molecules, and that of these the over-
whelming majority will be disordered—that is, will represent
fairly uniform distributions of temperature. Ordered arrange-
ments with a distinctly higher temperature at one end of the
chamber than the other will be extremely improbable; never-
theless, given a sufficiently long time they will occasionally
occur. It follows that among the (rare) occurrences of order,
the change from order to disorder will be practically universal,
whereas among the (very common) occurrences of disorder,
changes from disorder to order will be extremely rare. Yet in
this very long run changes from disorder to order will occur,
and will occur with the same absolute frequency as changes
from order to disorder. Examples of this sort, then, do not yet
exhibit any asymmetry of causal pattern in relation to the
direction of time. Given a record of the statistical behaviour of
any closed system over a very long period, one could not infer
which was the forward time-direction.
What is true, however, is that if such a system is put into a
low entropy state by some intervention from outside and is then
left to itself, if it is in effect a closed system from then on, it is
overwhelmingly likely that its entropy will increase with time.
11
Cf. A. Grunbaum, Philosophical Problem* of Space and Timet Chapter 8, esp.
pp. 241-2.
188 THE DIRECTION OF CAUSATION
But this seems to have nothing to do with our concept of causal
priority. It is a mathematical fact that, of all the possible dis-
tributions of molecular positions and velocities in what would
count as a relatively low entropy state, most will lead to states
with higher entropy and very few to states with still lower
entropy. Then if, by and large, such low entropy states as are
actually set up (by interventions) are fairly representative of
the range of possible distributions, it is a consequence that most
of them will in fact display subsequent increases in entropy.
This consequence is not itself a mathematical truth, but neither
is it utterly contingent; it is in an obvious sense very much to
be expected. Yet I can find no indication that this expectable
truth about statistical closed systems is incorporated in any way
into our concept of causal priority. Even when we extend the
notion of cause, as I argued in Chapter 6 that we should, to
include the successive phases of self-maintaining processes, we
have no tendency to see the cause-phase as distinguished from
the effect-phase by greater intrinsic orderliness or lower entropy.
The Second Law of Thermodynamics seems not to have been
built into our causal concepts, either traditional or innate.
A complication, however, is that the dispersal of order also
enters into the sort of example used to illustrate the increase of
entropy. The state of the gas a short time after the two parts have
begun to mix will be, with respect to the general distribution
of temperature, and hence with respect to the availability of
energy for external work, a fairly disordered one; nevertheless
it will be different from most equally disordered distributions of
molecular positions and velocities in that if each velocity were
simultaneously reversed, the gas would return in a short time
to the very unusual, highly ordered arrangement with the part
at one end of the chamber at a higher temperature. In other
words, this is a rather special sort of disorder, and while the
occurrence of disorder in general is intrinsically highly probable
and calls for no special explanation, the occurrence of this
special sort of disorder does call for explanation, and is indeed
to be explained as the outcome of a recently-previous ordered
state. In fact, this special sort of disorder is another example of
what we have been calling dispersed order. The coherence is
now concealed within what superficially looks like any other
disordered arrangement of molecules; but there is still an
 THE DIRECTION OF CAUSATION 189
intrinsically improbable coherence between the positions and
velocities of the various molecules, which is brought out by
considering what would happen if all the velocities were re-
versed; this requires explanation and is explained by the lawful
descent of this from the recent obviously ordered state. Once
again, then, we have a definite direction of explanation, to a
dispersed order from a non-dispersed order to which it is related
by laws. The mixing process, then, is explicable on its own, in
terms of ordinary forward causation, because the temporally
prior state is also prior with respect to the dispersal of order, so
that a coherent explanation can be given if we assume that this
earlier state is also causally prior. But a change in the opposite
direction, a process of separation into two parts at different
temperatures, because of the initial improbability of its initial
state, its dispersed order constituting a sort of position and
velocity distribution that is very rare among states of equal
entropy, could not be explained on its own without assuming
backward causation; yet since this initial state is only im-
probable, it could be explained, going outside this process itself
to the whole history of our closed system, as something that
occurs very infrequently in a long period of time; and this
explanation would not require backward causation. In other
words, we can find an asymmetry of Popper's type in the sort
of example that is used to illustrate the increase of entropy; but
this asymmetry is quite different from the entropy increase
itself, and, unlike the latter, has the conceptual links already
traced with explanation, fixity, and the direction of causation.
The general principle is that otherwise improbable concomit-
ances and co-occurrences can be explained as collateral effects
of a common cause, as dispersed order lawfully related to a
causally prior non-dispersed order; but it must be a common
cause, not a shared effect, and the collateral effects must not
be fixed at a time when the proposed common cause is still
unfixed.
I have made a number of attempts to come to grips with and
to clarify the concept which we undoubtedly have of causal
priority. What seems to emerge from this long and often
frustrating survey is this.
Our concept of causal priority is not merely that of temporal
priority; nor does it reduce to any asymmetry in the form of
igo THE DIRECTION OF CAUSATION
causal regularities—it is not, for example, Russell's concept of
many-one correlation; nor is it the concept of increasing entropy
or of anything like that. A first approximation to an analysis of
I
X was causally prior to Y\ where X and T are individual
events, is 'It would in principle have been possible for an agent
to prevent Y by (directly or indirectly) preventing, or failing
to bring about or to do, X'; if we speak instead of types of
event, Jf-type events are causally prior to l"-type events if in
circumstances where an X-type event is necessary for a 3^-type
one, it is in principle possible for an agent to prevent the occur-
rence of a l"-type event by preventing the occurrence of any
X-type one, or if in circumstances where an Jf-type event is
sufficient for a 2*-type one, it is in principle possible for an
agent to bring about a l"-type event by means of an X-type one.
However, we can carry the analysis a stage further and reduce
its dependence on the notion of agency, as follows. Suppose
that X and T are individual events, and X is seen as necessary
(and sufficient) in the circumstances for T, so that the basic
requirement for the judgement that X caused T is met. Then,
despite this, X was not causally prior to T if there was a time
at which ^was fixed while X was unfixed. If, on the other hand,
X was fixed at a time when T was unfixed, then X was causally
prior to T. Again, if X was not fixed until it occurred, then even
if T also was fixed as soon as X occurred (given, of course, that
X was necessary in the circumstances for 1"), X was causally
prior to T. And further, if there is some line or chain of causa-
tion, some continuous causal process, linking X and T and some
other event £ so that X was between Y and £, and if % was not
fixed until it occurred, then X was causally prior to Y*
Again, if there is some central event A, and somehow causally
connected with it some dispersed order item B—that is, some
collection of separate events Blt J52,53, etc., whose co-occurrence
is intrinsically improbable and calls for explanation—so that
A is prior to B with respect to the dispersal of order, and there
is no other central event that can be related to B1, B2, B3,
etc., so as to explain their co-occurrence, then this is evidence
that A is causally prior to B; the reason being that A is apparently
the only available explanation for the co-occurrence, but it can
explain this only if it is the causally prior item.
This, then, or something like this, is our concept of causal
 THE DIRECTION OF CAUSATION 191
priority. But does it have any application? Does any such
directedness actually belong to objective causal processes? As
I have said, the counterfactual conditionals into which we
initially analyse singular causal statements cannot be literally
true of the objective sequences, but these can be replaced by
regularities which would sustain them. Similar difficulties under-
mine the objective applicability of our first approximation to
an analysis of the priority itself. But the further analysis, in
terms of fixity and unfixity, could have objective application
provided only that there is a real contrast between the fixity of
the past and the present and the unfixity of some future events,
free choices or indeterministic physical occurrences, which be-
come fixed only when they occur. Now we certainly do not
know that there are no such events; we do not know that strict
determinism holds; but neither do we know that it does not
hold, though the balance of contemporary scientific opinion
is against it. So we had better be content with hypothetical
judgements here: if determinism does not hold, the concept of
causal priority which I have tried to analyse will apply to the
objects, but if determinism holds, it will not. If you have too
much causation, it destroys one of its own most characteristic
features. Every event is equally fixed from eternity with every
other, and there is no room left for any preferred direction of
causing.
This conclusion will not, indeed, follow from a popular
slightly mitigated, only almost total determinism. If
. . . thefirstMorning of Creation wrote
What the Last Dawn of Reckoning shall read13
then that first creation or coming into existence was presumably
an event which was not fixed until it occurred, and causal
priority will be linearly transmitted from it: X will be causally
prior to Ty in terms of our analysis, if X lies between creation
and T in a continuous line of causation. But although on this
hypothesis causal priority would indeed belong to the objects,
all the experiences in and through which we have acquired the
concept would in this respect be spurious.
But if total determinism holds, and there was not even a first
creative event, our present concept of causal priority will not
« E. FitzGerald, Rubdiydt of Omar Khayydm.
iga THE DIRECTION OF CAUSATION
be true of the real world. Of course, we can go on applying it,
basing such applications on our false beliefs about the previous
unfixity of human choices and perhaps of some physical events;
but it will be a widespread, systematic, error. Other directional
concepts, however, will still have application. There will be
the direction of entropy increase from most low-entropy states
of what are, for a time, closed systems. There will be, in some
(especially functional) regularities, Russell's one-way deter-
mination based on many-one correlations. Neither of these, I
have argued, has anything to do with our present concept of
causal priority. Again, obviously, there will still be the direction
of time, which we see (but on the present hypothesis wrongly)
as almost perfectly correlated with the direction of causation.
And there will still be the direction of the dispersal of order,
which we now take as evidence for something else, but which,
if that something else is a fiction, will remain as an interesting
directional feature of causal processes in its own right.
 8
The Necessity of Causes

WE may now come back to what Hume called *one of the most
sublime questions in philosophy', namely that of the power or
efficacy of causes, of the necessity that has been thought to
reside in the causal relation. We may begin by reminding our-
selves of the different jobs that necessity has been expected to do.
It has been thought of, both by Hume and by some of his
critics, as something that would either justify or replace induc-
tion, something which, if only we could detect it, would give
us knowledge of how nature works even in unobserved and
future cases. In this sense, Russell's principle of the permanence
of laws1 does the work of necessity. Closely related to this is the
notion, in Hume, of what we have labelled necessitya, some-
thing which if it were detected in one (cause) event would
license a priori the inference that a certain other (effect) event
would follow, with perhaps a counterpart which if detected in
an effect event would license a priori the corresponding inference
to its cause. This is closely related to the former notion, since
knowledge of such a property would entitle us to assert the
universal hypothetical that if a property of either sort occurred
anywhere at any time, the appropriate effect would follow (or
the appropriate cause would already have occurred). Thirdly,
there is the weaker notion of what we have labelled necessity3,
of something which would license causal inference in new cases
but perhaps not a priori', it is in this sense that Hume was pre-
pared to redefine necessity as 'the constant union and con-
junction of like objects3.* Necessity in this sense neither justifies
nor replaces induction; rather, knowledge of it, or belief in
it, is the typical product of a piece of inductive reasoning.
Fourthly, there is the notion of what we have called necessitylt
2
* Mysticism and Logic, p. 196. Treatise, II, HI, a.
 194 THE NECESSITY OF CAUSES
of something that marks off causal from non-causal sequences.
Fifthly, especially under the title of 'natural necessity', there is
the thought of something that distinguishes causal or nomic
regularities (or universal propositions or generalizations) from
merely accidental ones. Sixthly, though this is closely linked
with the fourth notion, is the idea of an intimate tie that joins
each individual effect to its individual cause. Seventhly, there is
the belief that it is in some sense necessary that every event
should have a cause, which Hume contrasts with the belief that
such particular causes must have such particular effects.3 The
necessity of causes in this sense would be something that re-
quired there to be a causal law behind any event. This, of
course, joins hands with the first notion in so far as a general
causal principle, once established, might help to validate or
justify the drawing of particular inductive conclusions from
appropriate observations or bodies of evidence. And, finally,
there is always in the background the concept of logical neces-
sity. This will not do any of the seven jobs, though it plays a
part in association with necessity3; an unrestricted statement of
regularity or constant conjunction along with an adequate
description of initial conditions will logically necessitate the
assertion that such-and-such an effect will follow.
I shall examine first regularity-type theories of necessity and
then theories of a radically different kind in an attempt to
reach a defensible view of this subject.
The conclusion reached at the end of Chapter 3 was that the
meaning of causal statements (singular or general) could not
be analysed in terms of regularities, but that the complex
regularities there examined constituted part of causation in the
objects, and (in their elliptical form) constituted part of causa-
tion as we know it in the objects; what we there found to be
needed, over and above complex regularity, to constitute causa-
tion in the objects was causal priority and perhaps some under-
lying causal mechanism. We have now dealt with causal
priority; what remains to be considered is whether objective
causation includes anything more, anything over and above both
(complex) regularity and what we have found to constitute the
direction of causation, perhaps something indicated by talk
about an underlying mechanism, perhaps not.
3
Treatise, I, m, a.
 THE NECESSITY OF CAUSES 195
How well can a regularity theory cope with the seven jobs
for necessity listed above? It does not help with the first, that
is, it does not in itself justify or replace induction. The second
job was that of necessity2, and of course any regularity approach
simply denies that there is any such thing, and it thus excludes
something that might have been able to justify induction. Yet
it would be wrong to say that a regularity theory makes the
problem of induction insoluble.* In Hume's argument, what
makes this problem seem insoluble is quite a different step, his
misleading use of the problem-of-induction dilemma, and as
our analysis of his argument in Chapter i shows, that dilemma
does not in any way rest on the regularity doctrine, but instead
that doctrine is ultimately supported, in part and at several
removes, by the dilemma. Regularity is, as we have noted, well
fitted for the third job, that of necessity3, and we have seen in
Chapter 3 that even complex and incompletely known regu-
larities can still do this job. It will also do the fourth job, that
of necessity^ of marking off causal from non-causal sequences:
according to a regularity account, a single sequence is causal
if and only if it instantiates a regularity. But does it do this job
properly? Does it draw the line in the right place? If we use
both halves of such complex regularities as we have formulated
it will admit as causal only those sequences in which the earlier
item is both necessary and sufficient in the circumstances for
the effect. To allow for the point made in Chapter 2, that we
do not always require strong, counterfactual, sufficiency, we
should have to read the regularity account as admitting, as
causal, sequences which instantiate the second half of such
complex regularities as we have formulated—that which has
such a form as 'In the field F, all P are preceded by (ABC or
DGfi or JKLy—even if no corresponding first half, of such a
form as 'In the field F, all (ABC or DGff or JKL) are followed
by P', is in force. On the other hand, it is this first half of a
regularity that would perform the more frequently mentioned
task of necessity3, that would, in conjunction with some descrip-
tion of the cause, logically necessitate the effect; the second
half of a regularity would perform only a less popular task;
it would, in conjunction with some description of the effect,
4
Or even that it is a regularity theory that gives rise to this problem. Cf. H. R.
Harre, The Philosophies of Science, p. 117,
196 THE NECESSITY OF CAUSES
logically necessitate the previous occurrence of the cause. There
is tension here between the ordinary requirements for causal
sequence and the philosophical tradition: once it is assumed
that causation has something to do with logical necessitation it
is tempting to make the causally prior item at least part of the
logically prior item, to put the cause on the logically necessitat-
ing side. But then it will be different halves of a regularity that
do the more frequently mentioned task of necessity3 and supply
the most vital part of necessity^ Even apart from this, in order
to get the right sequences marked off as causal we must add
the proviso that there should be the right relation of causal
priority and also allow, as in Chapter 2, for the fact that some
members of over-determining sets of causal factors are causes
while others are not. These are minor and not particularly
troublesome adjustments. But the major issue is whether, even
after they have been made, the regularity account picks out
the right sequences as causal. This depends on its success in
doing the fifth job. For if there are or can be regularities which
are altogether accidental (and do not merely have the causal
priority arrows wrongly placed), it seems that their instances
will count as causal by the regularity definition, while they
would be intuitively set aside as non-causal. This is a problem
which we must discuss at length. A purely regularity theory
simply refuses to do the sixth job; it denies that there is any
tie between the individual cause and effect over and above their
instantiating of the right sort of regularity; again, we shall con-
sider whether it is right to do so. The regularity approach is
also unhelpful in connection with the seventh job: there seems
to be no reason why the instantiating of regularities should
pervade the universe, and we have examined and rejected, in
Chapter 4, both Kant's arguments for the strong thesis 'Every-
thing that happens . . . presupposes something upon which it
follows according to a rule' and arguments of a Kantian type
put forward by modern thinkers in support of similar but weaker
conclusions. We shall consider whether some other (or further)
account of what causation in the objects is lends more weight
to the expectation that events should have causes.
The problem raised by the fifth job for necessity, that of
distinguishing causal from Accidental regularities, is the great
difficulty for any regularity theory of causation. We can
 THE NECESSITY OF CAUSES 197
distinguish a general and a specific form of this difficulty. In
his classic discussion of the character of natural laws3 Krieale
considers four possible views, that such laws are 'principles of
necessitation', that they are restricted universals, applying only
to limited regions of space and time, that they are unrestricted
but still only contingent universals, describing what always
and everywhere actually occurs, and that they are prescrip-
tions, maxims, or rules. The third of these is the Humean
regularity view; against it Kneale brings what he-regards as
the conclusive objection that laws of nature entail counter-
factual conditionals, while contingent, actual universals do not.
For example, if there had been a law of nature that all dodos
have a white feather in their tails, this would have entailed the
counterfactual conditional that if there had been any dodos
other than those that actually existed, they too would have had
a white feather in their tails, whereas the contingent universal
proposition that all actual dodos had a white feather in their
tails does not entail this counterfactual conditional.6 But even
if this specific difficulty were resolved, there would still be a
more general difficulty: we want to draw some distinction be-
tween laws of nature on the one hand and merely accidentally
true universal propositions on the other (even if it is not the
entailing of counterfactuals that distinguishes them), and it is
hard to see how a regularity theory can explain this. In a later
article7 Kneale gives examples of conceivably true but only
accidental universals which we should not count as even derived
laws of nature: 'there has never been a chain reaction of
s
W. C. Kneale, Probability and Induction) pp. 70-103.
6
Ibid., p. 75. Kneale speaks of contrary-to-fact conditionals: others have sug-
gested that nomic universals are distinguished by their entailing subjunctive con-
ditionals. This is less apt, for the term 'subjunctive conditional* is ambiguous
between counterfactuals, that is, those whose form indicates that the speaker
takes their antecedents to be unfulfilled {'If there had been any . . .', 'If there
were any . . .') and those which merely indicate that the speaker takes their
antecedents to be unlikely ('If there were to be any . . .'). The latter are only a
species of open conditionals, and even an accidental universal will entail all the
corresponding open conditionals. If it is a cosmic accident that all past, present,
and future As are 5s, it still follows that if there is an A (in such-and-such a place)
it is a B and that if there were to be an A—however unlikely—then this too would
be a B. It is only the conditionals whose antecedents are unfulfilled that even look
like the exclusive property of nomic universals.
7
'Natural Laws and Contrary to Fact Conditionals', Analysis, x (1950), re-
printed in Philosophy and Analysis, ed. Margaret Macdonald, pp. 226-31.
ig8 THE NECESSITY OF CAUSES
plutonium within a strong steel shell containing heavy hydro-
gen and . . . there never will be' and 'no human being has ever
heard or will ever hear this tune' (said by a musician who has
just composed an intricate tune on his deathbed).
A Humean might respond in either of two ways to this
general difficulty: he might try to draw, in his own terms, the
required distinction between nomic and accidental universal
truths, or he might, heroically, maintain that it was only an
unwarranted prejudice that there should be any such distinc-
tion. He might argue that in a deterministic universe everything
that happens happens in accordance with causal laws, so that
any allegedly accidental universal truths like those suggested
by Kneale hold, if they hold at all, in consequence of laws, and
should therefore be regarded as laws themselves, but derived
laws. This is not a very good argument, but we need not bother
about its other weaknesses since it is completely destroyed by
the reflection that we have no right to assume that the universe
is deterministic. Suppose that we manufacture a number of
atomic bombs, all alike in all relevant respects, each of which
will explode if and only if a nuclear disintegration of a certain
kind A occurs spontaneously before a nuclear disintegration of
some other kind B occurs within the same core—a reaction of
type B will in effect defuse the bomb. Some of the bombs
explode and some do not, but it is a matter of pure chance
which do, that is, whether in each bomb an ^(-reaction occurs
before a 5-reaction. Suppose further that it just happens that
a red spot has been painted on all and only the bombs that in
fact explode. Then we have a true universal proposition of just
the right form to count as a causal regularity making the having
of a red spot painted on the outside an inus condition of such
a bomb's exploding, and therefore necessary and sufficient in
the circumstances for this: whenever there are the other con-
ditions, and the red spot, the bomb explodes, but whenever
there are only the other conditions, without the red spot, it
does not. Also the fixity relations are right to allow the spot-
painting to be causally prior to the explosion in each case.
Our heroic Hurnean will, then, say that this is a causal
regularity, a law of nature. But, ex hypothesi, its truth is not a
consequence of the truth of any other laws: it was a matter of
pure chance, at least so far as all the laws of atomic disintegra-
 THE NECESSITY OF CAUSES 199
tion are concerned, that just those particular bombs blew up.
If the red-spot/explosion universal is a law, it is a new funda-
mental law of nature in its own right. Well, the Humean could
still say this, defending it on the ground that nothing can be
found in those other universal truths which we happily accept
as fundamental laws of nature that relevantly distinguishes
them from this one. But most Humeans would prefer, if they
could, to find some grounds, compatible with their system of
ideas, which would draw a line between regularities like this
one, which could be called accidental, and those which would
be counted as causal laws or laws of nature.
But let us go back to the specific difficulty, to Kneale's argu-
ment that laws of nature entail counterfactuals while contingent
universal truths do not. If a counterfactual conditional were
merely the conjunction of a material conditional with the
denial (or a hint of the denial) of the antecedent, then indeed
this problem would not arise: contingent universal truths of
the most accidental sorts would entail counterfactuals too. But
then all counterfactuals, if their antecedents were in fact un-
fulfilled, would be trivially true; this analysis of counterfactuals
removes the objection only by missing the point of the dis-
criminating use of some counterfactuals but not others: we want
to say *If this bit of potassium had been exposed to air it would
have burst into flame' but not 'If this bit of potassium had been
exposed to air it would have turned into gold*. I have argued
instead for a 'suppositional' analysis of conditionals, including
counterfactuals.8 The first of the two just given is seen as saying:
'Suppose that this bit of potassium has been exposed to air;
then (within the scope of this supposition) it has burst into
flame.' An almost equivalent way of treating it would equate
it with 'In the possible (but not actual) situation that this bit
of potassium was exposed to air, it burst into flame'; but I have
insisted that such talk about possibilities (or possible worlds)
must not be taken too literally: to talk about them is still only
to talk about our supposings and how we develop them;
possible worlds other than, the actual one have no fully objective
existence. It is a consequence of this analysis that counter-
factuals whose antecedents are, as the speaker suggests, unful-
filled, and whose antecedents do not entail their consequents,
8
Truth, Probability, and Paradox, Chapter 3.
aoo THE NECESSITY OF CAUSES
cannot be true in a strict sense; I would therefore not want to
say that even the most respectable counterfactual was entailed
by a law of nature. Nevertheless I would say that it was reason-
able to assert some counterfactuals, notably those that could be
said to be sustained by laws of nature, and not at all reasonable
to assert others. Since it is a law of nature that potassium when
in contact with oxygen ignites, it is reasonable to assert the
first of our two counterfactuals above; the law sustains this in
that when we combine the supposition that this bit of potassium
was exposed to air with the law and with the fact that air con-
tains a fair proportion of oxygen we can validly infer from their
conjunction the conclusion 'This bit of potassium burst into
flame', corresponding to the consequent of our conditional. But
why does not the truth of the singular material conditional
'{This bit of potassium was exposed to air) D (it turned into
gold)'—which is true just because its antecedent is false—
equally sustain our second counterfactual? The conjunction of
it with the supposition that this bit of potassium was exposed
to air equally entails the conclusion 'This bit of potassium
turned into gold'. The obvious answer is that it is not reason-
able to combine, with that supposition, a proposition which we
believe to be true only on the ground that that supposition is
false, and attempt to draw any conclusion from their con-
junction. To reason in that way would amount, in effect, to
using a statement and its denial as joint premisses in a serious
argument, and though according to the rules of the classical
propositional calculus such a pair of contradictory premisses
validly entails any conclusion at all, no one supposes for a
moment that this is a sensible way of arguing. Again, an example
of an accidental universal truth that does not sustain counter-
factuals would be 'All the coins in my pocket are shiny', which
seems neither to entail nor to sustain the counterfactual 'If that
other coin had been in my pocket it would have been shiny'.
The reason why it fails to do so is plain: it is not reasonable to
retain, along with the contrary-to-fact supposition that that
other coin is in my pocket, a universal proposition which is
believed solely on the strength of a complete enumerative check
of all the coins actually in my pocket. The required argument,
'Suppose that that other coin is in my pocket; all the coins
in my pocket are shiny; so that other coin is shiny', though
 THE NECESSITY OF CAUSES aoi
Formally valid, is not one that can be sensibly used in the circum-
stances envisaged: the second premiss is accepted about the
actual world on grounds that make it absurd to retain it within
the scope of the supposition of the first premiss, and so apply
it to a certain possible situation, for that first premiss totally
undermines our reason for accepting the second. The accidental
universal, accepted solely on the strength of a complete enumera-
tion, therefore fails to sustain a counterfactual for a reason
rather like (but still a little different from) the reason why the
material conditional mentioned fails to sustain the counter-
factual that if this bit of potassium had been exposed to air it
would have turned into gold.
This explanation9 shows why accidental universals do not
sustain counterfactuals; but we must still explain why laws of
nature do so. Clearly, it must be because it is reasonable to
retain a law of nature along with a contrary-to-fact supposition.
But how can it be, if the law of nature itself does no more than
describe a contingent and actual state of affairs? If the sup-
position takes us to a merely possible world, how can it be
reasonable to carry with us, in our baggage for this journey, a
statement that merely describes the actual world? The answer
lies not in any peculiar content of the law of nature, any
stronger-than-contingent link which it reports between its
antecedent and its consequent, but in the sort of reason we
have for believing it. Either this law itself, or some higher law
or set of laws from which it is derived, or perhaps, indeed, both
this law itself and one or more higher ones, will be supported
by some sort of inductive evidence: or, what comes to the same
thing, it or they or both will be confirmed or corroborated by
some observations or experiments,, We need consider only the
simplest case, where this law itself is either directly confirmed
or inductively supported. If it is, say, the law that potassium
when in contact with oxygen ignites, the evidence will be that
other samples of potassium, in contact with oxygen under vary-
ing conditions, have ignited, that experiments have been made
in which potassium has at first been kept away from oxygen
and has not ignited, and then, some oxygen having been brought
* Which I gave originally in 'Counterfactuals and Causal Laws', in Analytical
Philosophy, ed. R. J. Butler, pp. 66-80, and again in Truth, Probability, and Paradox,
pp. 114-19.
202 THE NECESSITY OF CAUSES
into contact with it, but nothing else, so far as we can tell, that
could possibly be relevant having changed, the potassium has
burst into flame, and so on. Now consider how this body of
evidence is related to the following other cases: first, some
potassium will be brought into contact with oxygen tomorrow;
secondly, some potassium was brought into contact with oxygen
yesterday, but we have not heard what happened on that occa-
sion; thirdly, someone may be bringing some potassium into
contact with oxygen just now in the next room; fourthly, sup-
pose that this bit of potassium, which has not in fact been ex-
posed to oxygen, had been so exposed. The evidence does, by
hypothesis, support the proposed law. It does, therefore, give us
some reason to believe that the potassium in the first case will
ignite; equally, that the potassium in the second case did
ignite; and equally that if in the third case the potassium in
the next room is now being exposed to oxygen, it is igniting.
But that body of evidence bears the same logical relation to the
fourth, the counterfactually supposed case, as it bears to the
other three. It therefore makes it reasonable for us to assert,
within the scope of that counterfactual supposition, that the
potassium ignited. And it therefore makes it reasonable for us
to use the counterfactual conditional, the use of which is merely
an abbreviation of the performance of framing the supposition
and asserting the consequent within its scope: if this bit of
potassium had been exposed to oxygen, it would have ignited.
Despite the fact, then, that a counterfactual conditional of
the kind that emerges in the analysis of singular causal state-
ments cannot be true and cannot be entailed by the merely
actual regularity which can be true, it can be as well supported
by the evidence for that unrestricted regularity as are the condi-
tionals about actual but unobserved cases.
This explanation rests on just three assumptions: that what
we regard as laws of nature are supported by evidence which
makes it in some degree reasonable to believe that unobserved
cases will obey the law; that such making reasonable depends
upon some formal relations between the body of evidence and
the new propositions which it supports; and that conditionals,
especially counterfactual ones, are to be understood in the way
I have suggested. The first two of these are very difficult to
deny; for the third I have argued at length elsewhere.
 THE NECESSITY OF CAUSES 303
If this explanation is correct, then it is somewhat misleading
to say even that laws of nature sustain, let alone entail, counter-
factuals. Strictly speaking, it is not the laws themselves that do
this but the evidence by which they are ultimately supported—
a curious latter-day vindication of Mill's thesis that it is reason-
ing from particulars to particulars that is important.10 The
evidence has not done all its work when it has warranted our
acceptance of the law as an unrestricted but contingent uni-
versal truth about the actual world: it continues to operate
separately in making it reasonable to assert the counterfactual
conditionals which look like an extension of the law into merely
possible worlds. But they only look like this; since what is
really going on is what I have set out above we should resist
the temptation to take uncritically the suggestion that the
counterfactual describes one or more possible worlds.
But once we have understood what is going on, could we
after all concede Kneale's point and recognize natural law
statements, or nomic universals, as a queer special type of
statement, taking the 'statements' here to incorporate the
(reasonable) readiness to assert counterfactual applications of
these generalizations? 'A asserts a nomic universal* will then be
shorthand for 'A asserts a contingent actual universal statement
on (ultimately) inductive evidence and is also ready to assert
the counterfactuals which are related to that evidence in the
same way as are (actual) instances of the universal statement'.
If we could adhere to this understanding of the phrase, it
would be harmless. But for that very reason, it would no longer
justify Kneale's further claim that laws of nature are principles
of necessitation, that in asserting them we are somehow claim-
ing that there is, in what goes on in the world, something like
a logical requirement (although we do not know enough to be
able explicitly to see it as such or to know it a priori). Moreover,
the natural law statements, thus understood as incorporating
counterfactual applications, could no longer be held to be, in
a strict sense, true. Their function would be not only to describe
the world—certainly not only the actual world, and there are
not really any merely possible ones—but also in part to express
an admittedly reasonable way of handling suppositions: quite
a creditable performance, but still something that people do,
10
System of Logic, Book II, Ch. 3, Sect. 3.
204 THE NECESSITY OF CAUSES
This specific difficulty, then, can be completely resolved so
that it no longer tells against the thesis that natural laws, in so
far as they have any claim to truth, are contingent unrestricted
universals. But, as we have seen, this still leaves us with the
corresponding general difficulty: there still seems to be some
line to be drawn between such laws and those other contingent
universals which we say are, if they are true, only accidentally
true.
Some light may be thrown on this problem by the distinction
Mill draws between uniformities of succession and \miformities
of coexistence.11 If not only all the coins in my pocket but all
the coins in every pocket were shiny, that is if, among coins,
the properties of being in a pocket and being shiny uniformly
coexisted, this would still not be a causal law; for causal laws,
Mill thought, are uniformities of succession, they describe
regular ways in which events, or states, follow one another in
time. But, first, the sort of functional relationship stressed by
Russell between, for example, accelerations and masses and
distances would appear to be a uniformity of coexistence, and
yet it would be a law and, in the sense defended in Chapter 6,
a causal law. Secondly, Mill does not and could not claim that
the distinction between sequence and coexistence coincides
with any distinction between laws and non-laws. In rather
subtle and careful discussions he argues that the uniform co-
existence of the effects of a common cause can be regarded as a
derivative law, but that the uniform coexistence of the effects
of different causes cannot, since they will depend on the non-
lawful coexistence of their causes.12 Keeping to our example,
we may say that if the going together of the properties of being
in a pocket and being shiny could be causally explained, say
by the fact that the owners of pockets always chose shiny coins
to put in them, together with the fact that a coin if shiny when
put into a pocket, would remain so, then this uniformity of
coexistence would be a derivative law; but if the causal histories
of these two properties were independent, so that this uni-
formity was derived, via two separate causal chains, from a
mere going together of some ultimate cause of shininess with
some ultimate cause of pocketing, itself not explainable as the
" System of Logic, Book III, Ch. 5, Sect. 9, also Ch. 22.
» Ibid., Book HI, Ch. 5, Sect. 9 and Ch. 22, Sect. 3.
 THE NECESSITY OF CAUSES 205
coexistence of joint effects of any further cause, then our uni-
formity of coexistence would not be a law. And Mill argues that
though many of the coexistences of properties of natural kinds,
for example of chemical elements, may in this sense be deriva-
tive laws, there must be some ultimate, unexplainable, co-
existences of properties. But he admits that among these
ultimate coexistences of properties there may be some invari-
able uniformities—the ultimate particles of which the universe
is composed may exhibit invariable, but just because they are
ultimate, unexplainable, combinations of properties. These Mill
will allow to be (a peculiar sort of laws of nature',13 but he
insists that they are very different from causal laws: there is
'one great deficiency, which precludes the application to the
ultimate uniformities of coexistence of a system of rigorous
scientific induction, such as the uniformities in the succession
of phenomena have been found to admit of. The basis of such a
system is wanting; there is no general axiom, standing in the
same relation to the uniformities of coexistence as the law of
causation does to those of succession.'14
Forceful and illuminating as these comments are, they have
in one way only deepened the problem. For we now seem to
have three classes of contingent universal truths: purely acci-
dental ones, such as 'All the coins in my pocket are shiny' is
and as 'All the coins in every pocket are shiny' might be;
fundamental laws of the coexistence of properties in the ultimate
material constituents of things; and causal laws. And though
we have distinguished these, we have not seen clearly just how
they differ from one another. Even if Mill is right in saying
that causal laws are partly supported by a universal law of
causation, and that this makes them easier to discover by such
experimental procedures as those formalized by the Method of
Difference, whereas the fundamental laws of coexistence have
no such support, and can be established only by simple enumera-
tive inductions, we have not yet found any intrinsic difference
between the laws of these two classes that would explain this
epistemological inequality; nor have we yet seen what marks
off both classes of what Mill is willing to call laws from purely
accidental generalizations.
" Ibid., Book III, Ch. 32, Sect. 3.
«+ Ibid., Book III, Ch. 22, Sect. 4.
ao6 THE NECESSITY OF CAUSES
Regularity theorists are naturally tempted to say that what
marks off a nomic generalization from an accidental one is its
being explainable by, that is derivable from, higher or more
fundamental laws. To say this is, of course, to use again the
same sort of move that they have used to distinguish causal
from non-causal individual sequences. A single sequence is
causal if and only if it is an instance of some law; similarly a
generalization is nomic if and only if it is derivable from some
higher laws. But, as our example of coins in pockets showed,
what is still only an accidental uniformity may be derivable
from some laws in conjunction with another accidental uni-
formity of coexistence (the going together of the ultimate causes
of shininess and of pocketing). So we would have to say that a
generalization is nomic if and only if it is derivable from some
higher laws alone. This has the surprising but perhaps toler-
able consequence that Kepler's Laws are not laws: that the
planets move (approximately) in ellipses is not a consequence
of Newton's gravitational laws alone, but only of those in con-
junction with what Mill calls collocations, here the fact that
the parts of the solar system started off with just such initial
positions and velocities, or that they have just such positions and
velocities at any particular moment from which we start our
calculations. Different initial conditions combined with New-
ton's laws would make the planets move in, say, hyperbolas
rather than ellipses, or, if they were too large or too close
together, in paths that did not even approximate to any conic
sections.
But the fundamental objection to this move is that it explains
the lawfulness only of derived or derivable laws. Unless we
postulate, quite gratuitously, an. infinite regress of possible
derivations, we shall have to say that there are some basic laws
which, just because they are basic, are not derivable from any-
thing else, and yet which must be laws if derivation from them is
to make other things laws. The regularity theorist may at this
point be tempted to turn his criterion upside down and say
that these basic laws are laws not because they are derivable
from laws, but because other laws are derivable from them.
But will this leave us with any discrimination? Any true uni-
versal proposition at all will entail others, and pairs of universal
propositions so related as to be joint premisses of syllogisms will
 THE NECESSITY OF CAUSES 307
together entail more: will they not then count as laws by this
definition and confer the same status on the conclusions de-
rived from them? A possible reply is that the basic laws of a
system must be of very wide application; if we take this way
out we shall say that: what makes basic laws laws is their being
contingently true universal propositions of very wide applica-
tion, and what makes derived laws laws is their being derivable
from such widely applicable universal truths alone. It is pre-
sumably this line of thought that makes Mill willing to recognize
the ultimate uniformities of coexistence of properties as laws,
though of an inferior kind.
This connects with the suggestion that an accidental universal
truth is one which holds only for some limited region of space
and time, or in the formulation of which proper names occur
non-eliminably. It can be no more than an accidental generali-
zation that all the coins in John Smith's pocket on 19 November
1972 are shiny. Yet this criterion is easily undermined; we can
presumably find uniquely identifying descriptions of John
Smith, and of this date, using only general terms; and yet if the
original statement about John Smith is only accidentally true,
so surely is its artificially generalized translation. Can we then
make the distinction turn not upon the kinds of words used but
upon the sort of fact that makes the universal statement true?
It has been pointed out that there is a general strategy, trace-
able in Comte, Mill, Reichenbach, and Popper among others,
of 'distinguishing universals which are true generally from those
which are true by virtue of what is the case locally*.IS But it
seems that no definition based on this strategy will capture our
intuitive distinction between laws and accidental generaliza-
tions. It will not do to say that something is not a law if it is
made true by what is locally the case, by what goes on within
some closed space-time region. It might well be that life occurred
only on earth, and only over a period of time closed at both
ends. If so, any laws about living organisms that are true would
be made true by what goes on in this closed region; yet we
would want to count them as laws. Might we say instead that
something is not a law if it is made true only by what is locally
the case, defending the law status of organic laws on the ground
15
W. Suchting, 'Popper's Revised Definition of Natural Necessity', British
Journal for the Philosophy of Sciencet xx (1969), 349-52.
ao8 THE NECESSITY OF CAUSES
that though they could in principle be established by a survey of
this closed region, they do not need to be, they are in principle
derivable from physico-chemical laws to whose instantiation
there are no spatio-temporal limits. But then if there were any
'emergent' laws about organic life, any universal truths about
how living things proceed and develop which were not reducible
without remainder to physico-chemical laws, these would still
be denied the status of laws; but it is far from clear that there
could not be such emergent general truths, and if there were
we should, I think, want to call them laws. And on the other
hand, could there not be cosmic accidents, propositions whose
universal truth could not be ascertained by a survey of any
closed region, but which none the less only happened to be
true? Consider any occurrence which is in fact unique, which
has no exact replica anywhere in the universe. (Each of us likes
to suppose that he himself, or some of his activities, is or are
such, but even if this is inexcusable vanity there must be plenty
of other examples.) Since it has occurred, it is not excluded by
any natural law or conjunction of laws. But equally this occur-
rence might not have occurred, in all its unique detail, and it
seems plain that it would not have occurred if some initial con-
ditions had been a bit different, or if some uncaused events, if
there are such, had gone differently, without any change in the
laws of nature. And the elimination of this unique occurrence
need not have ensured the emergence, somewhere else, of an
exact replica of it. Then it might have been true that no occurrence
of just this sort ever occurred^ and yet, by hypothesis, that there was
no conjunction of laws alone that ensured this. That is, there
might have been a cosmic accident of the form 'Nothing of
exactly such and such a sort ever occurs anywhere'; but it
would be only how the universe as a whole was that made this
true, not the state of any closed space-time region. Not being
made true only by what is locally the case is therefore neither
necessary nor sufficient for a contingent universaPs having
what we intuitively recognize as the status of a law.16
But alongside and intertwined with this vain attempt to
equate accidentality with spatio-temporal limitation there is
another and more fruitful idea. Conceding Kneale's point that
16
This is the conclusion reached also by Suchting in the discussion quoted
above, on slightly different grounds.
 THE NECESSITY OF CAUSES *og
'there may be true, strictly universal statements which have an
accidental character rather than the character of ... laws of
nature', and that 'the characterization of laws of nature as
strictly universal statements is logically insufficient and in-
tuitively inadequate', Popper suggests that the natural or
physical necessity in laws 'imposes structural principles upon the
world' while leaving 'initial conditions' free.17 Whereas on
Kneale's view, or Descartes's, a God who was limited by logical
necessities was free to choose the initial conditions of the world
but not its structure, en Popper's view such a God would have
two different choices to make, the first of 'structure', that is, of
how the world was to work, and the second of initial conditions.
He offers this definition (but with considerable reservations
both about its accuracy and about the importance of definitions
anyway, as contrasted with ideas): 'A statement may be said to be
naturally or physically necessary if, and only if, it is deducible from a
statement function which is satisfied in all worlds that differ from our
world, if at all, only with respect to initial conditions' This idea is
illustrated by an example: we may assume that the biological
structure of the moa organism is such that under favourable
conditions it might live for sixty years or more, but the actual
conditions under which all moas in fact lived were far from
ideal—perhaps some virus was present—so that no moa ever
reached the age of fifty. So, Popper says, 'All moas die before
reaching the age of fifty' will be a true universal statement, but
not a law of nature: the laws of nature would allow moas to
live to sixty, and only the accidental or contingent conditions,
such as the co-presence of the virus, prevented this. This co-
presence counts, presumably, as an initial condition or as an
outcome of some initial conditions. Still, the distinction between
structural principles and initial conditions is not yet clear: just
why does the biological structure of moas contribute to laws,
but the presence of a virus only to an accidental universal?
Perhaps the answer is that the term 'moas', defined just by that
structure, is the subject of the laws and non-laws under con-
sideration; if their biological structure condemned moas to a
life span of, say, less than seventy years, 'All moas die before
reaching the age of seventy* will be a (derived) law, but 'All
moas die before reaching the age of fifty' will not. But if we
17
The Logic of Scientific Discovery, Appendix xs pp. 426-41.
aio THE NECESSITY OF CAUSES
change the subject we can turn the latter too into a derived
law: 'All moas in the presence of virus x die before reaching
the age of fifty.'
We may be able to get a firmer grip on the important distinc-
tion if we contrast laws of working with what Mill called 'col-
locations^ and consider another, simpler, example. We have,
fixed on the surface of the earth, a plane inclined at an angle 0
to the horizontal, several feet long from top to bottom. Near
the top (left) edge are placed six similar balls, three red (/21? Rz,
/?3) and three blue (Blt JS2i B3] spaced a few inches apart in
the order Rt JSj J?a Bz R% JS3 from left to right. They are released
simultaneously, and all run down the plane in accordance with
Galileo's law, and pass in turn a point P near the bottom of
the plane. Let us consider the following statements about this
system:
(1) Wherever there is a red ball there is a blue ball immediately
to its right.
(2) Whenever a blue ball passes P a red ball passes P shortly
afterwards.
(3) Whenever a ball is released on this plane it runs down it in
accordance with the law a — g sin B.
(4) Whenever an object of mass m is near the surface of the
earth it is subject to a force mg in the direction of the
centre of the earth.
(5) Whenever an object of mass m is subject only to a force/
it has an acceleration f\m in the same direction as that of
the force.
(6) Between any two objects there is a force proportional to the
mass of each and inversely proportional to the square of
the distance between them.
How are we to characterize these various statements? (i) is
a pure collocation statement. It is a true contingent universal
statement about this small system. Its application is spatio-
temporally limited, but this is not the point. Even if something
similar were true of the whole universe, it would be no more
than a cosmic accident. (2) is a derived law of this system; it is
derived from the conjunction of (3) with the collocation state-
ment (i)— and a little further description of the set-up. But (3)
18
System of Logic, Book III, Ch. 12.
 THE NECESSITY OF CAUSES an
also is a derived law; it follows from (4.) and (5), along with
some principles about the resolution of forces, and so on, in
conjunction with some collocation statements, especially that
the plane is inclined at the angle B. And even (4) is in turn a
derived law, following from (6) together with a collocation
statement about the mass of the earth and the solution of the
mathematical problem of integrating the gravitational effects
of all the portions of a sphere. Of the statements listed, (5) and
(6) alone represent a system of basic laws of working, in this
case the Newtonian theory of motion and gravitation. But also,
of those listed, only (i) is a pure collocation statement (though
other relevant collocation statements have been hinted at).
(2), (3), and (4) are all mixed products, the result of combining
laws of working with collocations. But each of these can be
turned into a pure, though derived, law of working by generaliz-
ing it, by replacing references to the individual set-up with
descriptions of its relevant features. Thus if we substitute, in (4),
'a sphere of such-and-such a radius, mass, and so on' for 'the
earth', it becomes a mathematical consequence of the basic laws
alone. The same can be achieved with (3) if we insert both this
relevant general description of the earth and a description of
the relevant features of the plane, and with (2) if we add to all
this a description of the initial placing of the balls. That is, we
purify the mixed laws of their collocation component by put-
ting it explicitly into the antecedents of the general condi-
tionals. Thus we can sort out pure laws of working—some basic,
some derived—from both mixed laws and collocation state-
ments. We can do this in practice only where we take it that
we fully understand how the system in question works, as we
do with a Newtonian mechanical system.
Let us try to apply these distinctions to Popper's moa
example, where we do not actually know the relevant laws. That
all actual moas lived in the presence of virus x we can take as
a fact of collocation. 'All moas die before reaching the age of
fifty' will be a mixed law; so is 'All moas die before reaching the
age of seventy', since it is presumably the product of some more
basic biological laws along with facts about how moas are
actually built. But the latter could be turned into a pure though
derived law of working by replacing the subject 'moas' with
'creatures of such-and-such a structure' including in this description
212 THE NECESSITY OF CAUSES
all the features that are relevant to their life span.19 And then the
former too can be turned into a pure though derived law by
adding to this detailed description of the subject a further clause,
*in the presence of a virus with such-and-such characteristics'.
But here the notion of laws of working, basic or derived, is
that of a theoretical ideal: we do not actually know these
laws, and for that reason we do not know the relevant features
whose explicit mention could purify the mixed laws that we
actually deal with. But it is reasonable to suppose that there are
some, as yet unknown, pure laws of working, and that there is
a real objective distinction between these and collocations or
initial conditions, although we are not in a position to imple-
ment it. We can read Popper's definition, then, as calling
naturally or physically necessary all pure laws of working, basic
or derived, and all singular conditional statements which these
entail.20
Equipped with these distinctions, we can now say that in
our atomic bomb example the presence of a red spot on every
bomb in which an ^-reaction occurred before a ^-reaction was
only a (locally) universal collocation, as contrasted with the
probabilistic laws of working m accordance with which these
reactions occurred.
But given that we can draw this distinction, that we can leave
a post-Newtonian God with his two choices rather than the
Cartesian single choice, what is the point of the distinction?
Why should the laws of working be said to have natural
necessity? The analogy that Popper draws with Tarski's defini-
tion of logical necessity, that while what is logically necessary
holds in all possible worlds, what is physically necessary holds in
all worlds that differ from ours, if at all, only with respect to
initial conditions, seems not very illuminating. Since 'initial
conditions' are simply what we contrast with laws of working,
this means that what is physically necessary holds in all worlds
" This is, I think, how Popper intends this statement to be taken, since he
speaks of using 'moa' as 'a universal name of a certain biological structure'; but
he does not emphasize the need to mention all the relevant features.
*° I think that this treatment preserves the idea behind Popper's definition while
avoiding the objections raised by G. C. Nerlich and W. A. Suchting to its literal
formulation in 'Popper on Law and Natural Necessity', British Journal for the
Philosophy of Science, xviii (1967), 233-5. Popper replied to this, in the same volume,
pp. 316-21, but later in Logic of Scientific Discovery, and English edn., Addendum
(1968), on p. 441, expressed dissatisfaction with his reply.
 THE NECESSITY OF CAUSES 313
that have the same laws of working as our own, and any actual
feature could be turned into its own variety of necessity by a
similar device. For example, the existence of just such minds as
there are is spiritually necessary in that it holds in all worlds
that differ from ours, if at all, only in non-spiritual respects.
In the end Popper himself decides to rest no weight on this
analogy: he regards 'necessary' as 'a mere word'; 'there is not
much connection here with logical necessity'; the only con-
nection between causal connection and logical necessity is that
the material conditional a -> b follows with logical necessity
from a law of nature.21
One point that the distinction may have is that laws of work-
ing will typically be discovered by inductive and hypothesis-
confirming procedures, whereas initial conditions will be
discovered by complete surveys and will not be inductively
supported. (Of course initial conditions may be inferred from
subsequent conditions, themselves ascertained by survey, with
the help of inductively supported laws, but that is another
matter.) In particular, experiments interpreted with the help of
the assumption that there are some laws of working seem to be
needed to sort out pure laws of working from mixed laws that
21
Op. cit., p. 438. Similarly in 'Two Faces of Common Sense' printed as
Chapter 2 in Objective Knowledge, Popper says:
'At any rate, in the light of a conjecture we can not only explain cause and
effect much better than Hume ever did, but we can even say what the "necessary
causal link" consists of.
'Given some conjectural regularity and some initial conditions which permit us
to deduce predictions from our conjecture, we can call the conditions the (con-
jectural) cause and the predicted event the (conjectural) effect. And the con-
jecture which links them by logical necessity is the long-searched-for (conjectural)
necessary link between cause and effect.' (p. 91.)
That is, the only necessity involved is what we have called necessity,,: the con-
junction of the regularity-statement with the description of the cause entails the
prediction of the effect.
As a solution to the problem of'the long-searched-for . . . necessary link* this is
laughable: it leaves out everything that is difficult and interesting—the detailed
form of causal regularities, causal priority, and all the jobs mentioned at the
beginning of this chapter except the third and fourth. It explains cause and effect
not better than Hume did, but almost exactly as Hume did (e.g. 'Necessity . . .
consists either in the constant conjunction of like objects or in the inference of the
understanding from one object to another', Enquiry^ Sect. VIII, Pt. II, Selby-
Bigge, p. 97). As I suggested in Chapter i, Hume was probably attracted to
constant conjunction by its suitability for the role of necessity^ Philosophers after
Hume have gone on searching for the causal link just because they could not
believe that this was all there was to it.
214 THE NECESSITY OF CAUSES
depend partly on collocations: we check that something is
Independent of collocations by changing the collocations and
seeing if it still holds. If so, these laws of working will have the
right sort of backing to allow them to sustain counterfactuals,
whereas statements of collocations or initial conditions will
not. {Reichenbach, who realized that 'inductive verifiability'
was the key to counterfactual force, tried to define nomological
statements by requirements that would indirectly ensure induc-
tive verifiability, but his requirements seem too formal to
achieve what is wanted, in particular to exclude cosmic acci-
dents and to include laws whose application is in fact limited
to a closed space-time region.)22
However, mixed laws no less than pure ones are discovered
and supported by inductive and hypothesis-confirming pro-
cedures, and they also can therefore be said to sustain counter-
factuals. These will not be distinctive features of the pure laws
of working which we should like to mark off as naturally
necessary.
But is there anything more about laws of working that would
justify their being singled out as necessary? Here we may turn
to a theory of necessity which is radically different from the
regularity-type views we have been examining, namely that
which Kneale presents and defends against Humean criticisms,
while using both what I have called the specific and the general
difficulties to attack the regularity theory.
The view which Kneale defends he tentatively ascribes to
Locke: it is that natural laws are 'principles of necessitation',
necessary in the same way in which necessary connections that
we are able to comprehend—for example, that redness and
greenness are incompatible—are necessary, although we are
not able to comprehend the natural laws, not simply because
our intellects are too feeble, but rather because 'our experience
does not furnish us with the ideas which would be required for
an understanding of the connexions we assert'.23 The essential
idea is that the things that enter into causal transactions, and
hence also the events that are their doings, have insides which
22
H. Reichenbach, Nomological Statements and Admissible Operations, especially
pp. ia~i3, 40,48; see also p. 94 for an account of'relative nomological statements'
which correspond to our 'mixed laws'.*
13
Probability and Induction, p. 71.
 THE NECESSITY OF CAUSES 215
we do not and perhaps cannot perceive; the necessary connec-
tions hold between these internal features. We cannot in fact
comprehend their necessity, but we could if we could perceive
the features between which the connections hold, and what
we would then comprehend would be something like the
incompatibility of redness and greenness. When we assert a
natural law we do not claim to know it as necessary, but we
conjecture that there is a necessary connection holding unper-
ceived between kinds of events which we identify by their
external features.
Kneale thinks that the Humean objection to this view is based
on a mistaken doctrine of perception, on the theory of impres-
sions and ideas, or of sense-data, and the various phenomenalist
systems erected on this basis. Only on this assumption, he
argues, would the fact that we can conceive a cause without its
effect, or vice versa, refute the suggestion that they are neces-
sarily connected in the sense proposed. We can conceive these
external features without those, and if the 'objects' in question
were wholly constituted by their external features that would
be the end of the matter. Consequently Kneale's main reply
to the Humean objections consists of a criticism of pheno-
menalism and a defence of realism about 'truths concerning
perceptual objects which are not open to inspection'. This
vindication of realism is, perhaps, less necessary now than it
was twenty-five years ago; I, at any rate, would accept this
conclusion without question. But I do not think that this takes
all the force out of Hume's doubits about objective necessities
of the kind proposed. As I have said in Chapter i, the main
weight of Hume's argument bears upon causation so far as we
know about it in the objects, and he explicitly (though perhaps
ironically) leaves room for 'several qualities, both in material
and immaterial objects, with which we are utterly unacquainted*
but which may be called power or efficacy.24 But he could have
argued that whatever qualities there may be in the cause-
event, however unperceived and unconceived, they cannot
logically-necessit&te the occurrence of an effect at all, let alone
an effect with such-and-such correlated qualities, just because
the cause-event and the effect-event are distinct existences.
As I said in Chapter i, there can be no logically necessary
2
« Treatise, I, in, 14, Selby-Bigge, p. 168.
ai6 THE NECESSITY OF CAUSES
connections between the events themselves or between any in-
trinsic descriptions of them, however detailed or complete, and,
we may now add, however internal and closed to inspection. But
this argument bars only logical necessity, entailment: Kneale
might be content to claim a necessity that falls short of logical
necessity but is illustrated by the incompatibility of redness and
greenness. Of course, there is no agreement about how this
incompatibility is to be analysed; but we need not go into this
question because this analogy, too, is inappropriate. Anything
like this can be only an impossibility of two features occurring
at the same place and time, in the same subject, or, if it has a
positive counterpart, a necessity of one feature occurring at the
same place and time as another, in the same subject. Nothing
in this example helps us to see even what could be conjectured
about a necessity linking two distinct occurrences.
To find anything worthy of the name of necessity in laws of
working, then, we must turn to what Kneale says about the
explanation of natural laws, not merely by their being derived
from 'higher' laws—the characteristic regularity-theory move
which, as we have seen, is unhelpful—but by their association
with transcendent hypotheses, such as wave or particle theories in
physics. These hypotheses introduce new terminology, they are
'concerned with things which are not observable even in prin-
ciple', and they concern only 'structures' which can be expressed
in the language of mathematics. Since the translations between
this new terminology and the ordinary language of material
objects are contingent, there is no prospect of our being able to
derive laws of nature about material objects from self-evident
truths alone, 'Although the connexions within the world of
transcendent entities posited by a theory may all be self-
evident, the relations between this world and the world of per-
ceptual objects remain opaque to the intellect, and it is only by
assuming these relations that we can explain our laws about
observables.'35 It is the concessive clause here that is vital; and
a few lines earlier Kneale has said that 'attempts are . . . made
to specify the hypothetical entities of the system in such a way
that any connexions between them required for the purposes
of the theory are intrinsically necessary'.
The suggestion, then, is that we advance beyond a view of
zs Probability and Induction, p. 97.
 THE NECESSITY OF CAUSES 317
causation as mere regular succession when we conjecture that
there really is some causal mechanism underlying the succes-
sion and explaining it. But must: this be a matter of conjec-
ture, of transcendent hypotheses? Would it not be even better
actually to uncover and observe the mechanism? And yet, if
we did so, would not Hume's criticisms again have force:
what, in the operation of a mechanism, however delicate and
ingenious, could we see except the succession of phases? It may
be to avoid this reply that Kneale stresses the transcendent
character of the hypotheses; but whether the mechanism is
observable or not, the question must still be answered: what is
there in it except the succession of phases?
For Kneale, it is the 'structures' and the possibility of mathe-
matical description of them that are vital. The cause-event is
identical with something unobservable that has a certain mathe-
matically describable structure,, The latter develops, necessarily,
into something else which is also unobservable and also has
some such structure. The latter something is identical with the
effect-event. The necessity of the development of the one struc-
ture into the other would be self-evident to us if we knew those
structures (and, presumably, if we were mathematically com-
petent enough). Science can help to conjecture what these
structures are, though not the correlative contents, not the
substances of which these are the structures. And the statements
identifying observable things and occurrences with unknowns
with such-and-such structures can never be more than con-
tingent conjectures, which are not themselves intelligible, but
are confirmed by their success in relating observed regularities
to intelligible structure-developments.
But how is a structure-development intelligible or necessary?
Physically it will be, say, a certain wave-pattern changing into
a certain other wave-pattern, or one assembly of particles, each
with a certain mass and position and velocity at one time,
changing into another arrangement of the same particles, each
retaining its mass but changing its position and perhaps its
velocity. Mathematically there will be some/orm of description
applicable to both the earlier and the later phases of the develop-
ing structure; the two particular descriptions of this form will
have identical values for some of the quantities (for example,
the total energy of the wave system may be conserved, or the
ai8 THE NECESSITY OF CAUSES
number of particles and also perhaps their mass and energy, or
at least the sum of both) while the values for other quantities in
the later phase will be derivable from the earlier values with
the help of the integration over time of some differential
equation, that is, some law of the functional dependence
variety. We may speculate, then, that what could count as
necessity or intelligibility here is some/orm of persistence.
Let us consider the simplest of all cases, a single particle, free
from interference, moving in a straight line in accordance with
Newton's first law (see diagram).

Diagram (vi)

There are several things we can say about this.

First, if the particle moves continuously from A to B, from
B to C, and from C to Dt these being all equal distances, in
equal times, it is in a very obvious sense keeping on doing
the same thing. It is not of course logically or mathe-
matically necessary that if it moves a distance d in a time t
between A and B it should move an equal distance d in the
succeeding interval t and in the same direction, from B to (7.
But it may well seem expectable that it should go on doing
as nearly as possible the same thing (of course, it has to do it
at a different place) if nothing intervenes to make it do anything
else.
Again we can argue, on the assumption that there has been
and will be no interference, that the motion from A to B has
produced the motion from B to C; but the motion from B to C
is just like the motion from A to B; so, if like causes produce
like effects, the motion from B to C will produce something
just like itself, that is, the motion from C to Z); if we knew that
it had gone smoothly from A to C in a period a£, we could
expect it to go from C to D in the next period t} all the time on
the assumption of no interference, of a constant uniform back-
ground.
Such an argument was given by d'Alembert, and has been
 THE NECESSITY OF CAUSES 219
26
criticized by Nagel, whose criticisms have been widely
accepted. They are, however, unsound in part, and for the rest
are easily met by a modification of d'Alembert's argument.
Thus he says, 'D'Alembert simply assumes that a force is
required to account for changes in the uniform velocity of a body
(where the state of rest is a special case of uniform velocity), but
that none is required to account for changes in the mere position
of a body. But this is to beg the entire question at issue.' But
d'Alembert doesn't assume this. He argues that 'a body cannot
be brought into motion by itself, since there is no reason why
it should move in one direction rather than another': that is,
he uses the same sort of symmetry considerations here as in the
argument below against a deviation from the straight line (say to
our Z)j). Nagel asks, 'Why should uniform velocity be selected as
the state of a body which needs no explanation in terms of the
operation offerees, rather than uniform rest or uniform accelera-
tion (such as motion along a circular orbit with constant speed)
or, for that matter, some still different state of motion of the
body (for example, the constancy of the time-rate of change of
the acceleration) ?' The answer is that a priori there would be
no reason for such a preference, but that d'Alembert needs no
reason, since he is not committed to an a priori preference. He
could ask, in our terms, what is the nearest that something can
be to keeping on doing the same thing? If it is at rest, staying
at rest is clearly the answer. But if it is already moving, that is
ruled out, and uniform motion is the next best possibility. Or
he could confine himself, as I have done in the text above, to
a case where the body has already moved uniformly from A
through B to C. If there has been and is no interference,
d'Alembert can then argue that what produced the B~C move-
ment was the A-B one, and that the B-C one, being just like
the A-B one (except for position), will produce something
exactly alike. This argument involves no prejudice in favour of
uniform velocity. We could argue equally well that if a spherical
body has rotated uniformly over a certain period it will go on
rotating, though this will involve a persistence of various
accelerations towards the centre. On the other hand, a body's
moving uniformly in a circular orbit is not in ^//"intelligible
26
The Structure of Science, pp, 175-8, quoting d'Alemb'ert's argument from
Traili de dynamique, Paris, 1921, vol. i, pp. 3-45.
220 THE NECESSITY OF CAUSES
in the same way as rectilinear motion is. Any small segment of
the orbit is obviously asymmetrical,, in a way that would cry out
for some further explanation—for example by the presence of
another body at the centre, or on the opposite side of the centre,
of the circular orbit. Nagel suggests that if a body has been
moving in a circular orbit subject to a central force, and that
force is then eliminated, d'Alembert's considerations of sym-
metry could lead to the expectation that the circular movement
would continue, rather than to the Newtonian expectation of
tangential movement. But this is clearly wrong. *As long as
whatever force has so far maintained the circular motion is
present, d'Alembert's reasoning would lead us to expect that
motion to be continued; but if the conditions are changed by
the elimination of the central force, they no longer point to any
conclusion at all. Nagel concludes that d'Alembert's argument
'is worthless simply because one can always exhibit in a given
state of motion of a body a variety of distinct symmetries and
asymmetries' with no logical grounds for preferring one to
another; but in the cases considered this judgement is simply
false, and only excessive devotion to empiricism has made it
palatable. NagePs further suggestion that Newton's axiom can
be construed as 'a concealed definition, a convention which
specifies the conditions under which one will say that there are
no impressed forces acting on a body' is therefore gratuitous.
Nagel does, indeed, mention what is a serious problem for
d'Alembert: he has assumed a Newtonian framework that
allows one to speak of absolute rest, motion, and so on of one
body at a time. D'Alembert would need either to vindicate this
assumption, or to recast his argument so that everything was
done in terms of relative motions. But the latter at least could
be done, at the cost only of some loss of simplicity.
Another comment of the same kind is that if after going from
A to B and from B to C the particle had then gone not to D
but, say, to Dlt this would have been prima facie surprising:
since Z)t is placed asymmetrically with respect to the line ABC,
we might say that if it were to go to Dl it might as well go rather
to jD2, similarly placed on the other side of the line; since it
cannot go to both, it ought not to go to either, but should
confine itself to the continuation of its straight path through Z).
Of course none of these arguments shows that the particle
 THE NECESSITY OF CAUSES 221
logically must continue to obey Newton's law. Innumerable
other sequences are conceivable, none is ruled out even by the
most complete, most 'internal', intrinsic description of the
motion from A to C. Yet they cast some doubt on Hume's claim
that 'If we reason a priori) anything may appear able to produce
anything' or even that 'Any thing may produce any thing*.2?
Perhaps it may; but at least in some cases some outcomes are
less intrinsically surprising than others; and yet this is a matter
of a priori considerations, it does not depend on any appeal,
either open or covert, to our actual experience of or familiarity
with processes of the kind in question.
What holds for this simplest case holds to some extent for
some less simple ones too. If we have several particles near one
another, each moving initially in some straight line at some
speed, it may be prima facie expectable that each will go on as
it is going, if they can do so without colliding. Of course, this
case is not as clear as the last: the different particles might
influence one another (as in fact, of course, they would): still,
it would be expectable that whatever they did to one another
would as it were be built on to a foundation of the continuance
of the separate motion of each. The persistence of a wave motion
is no doubt more complicated and less obviously intelligible
than that of the motion of a particle; but it has one great
advantage, that different waves can pass through one another,
adding up and cancelling out so far as their local effects are
concerned, but without damaging one another permanently.
We may suggest, then, that basic laws of working are, in
part, forms of persistence. In Chapter 6 I argued for extensions
of the concept of causing to include both the persistence of
objects and the persistence of self-maintaining processes, saying
that in both of these we could regard an earlier phase as a cause
and a later phase as an effect. I am now suggesting that this
sort of causing plays a larger part, underlying processes that at
the perceptual level are cases of unrelieved change, of a cause
being followed by an utterly different effect. A match is struck
on a matchbox and a flame appears: on the face of it this
effect has nothing in common with its cause. But if we were to
replace the macroscopic picture with a detailed description of
11
The more guarded claim is from Enquiry, Sect. XII, Pt. Ill, the less guarded
from Treatise, I, HI, 15.
aza THE NECESSITY OF CAUSES
the molecular and atomic movements with which the perceived
processes are identified by an adequate physico-chemical
theory, we should find far more continuity and persistence.
And this is part of the point of micro-reduction, of the tran-
scendent hypotheses of which Kneale speaks. Charles's Law
tells us that if the temperature of a quantity of gas in a con-
tainer of constant volume rises, the pressure on the walls will
rise proportionately; but there is no immediately obvious homo-
geneity between temperature and pressure. But if temperature
is identified with the average kinetic energy of molecules, and
pressure with their change of momentum in elastic collisions
with the walls of the container, both acquire a relation to the
mean speed of the molecules, and we can begin to see a con-
tinuity between them. A more obvious example is Fajan's Law
for radioactive disintegration, that emission of an alpha particle
moves an atom to a position two places lower in the periodic
table of elements. The heterogeneity of cause and effect dis-
appears when an alpha particle is identified as one with two
units of positive electrical charge and occupying a certain place
in the periodic table with having a certain charge on the
nucleus: the emission of this particle simply carries this amount
of charge away. What is called a causal mechanism is a process
which underlies a regular sequence and each phase in which
exhibits qualitative as well as spatio-temporal continuity.
On the other hand, there can be no question of reducing a
process of change wholly to some persistence: if the earlier and
later phases on the macroscopic level are different, there must
be equally real differences between the micro-states with which
they are identical; the most we can expect is that there should
be more persistence mixed with the differences than there
appears to be. Also, wherever there is interaction (say gravita-
tional or electro-magnetic attractions or repulsions) it seems in-
escapable that there should be a law of working which is not
just the persistence of anything: it is worth recalling that in
Chapter 6, along with the extension of the causal concept to
cover the persistences of objects and processes, where 'causes' and
'effects' resembled one another, we found appropriate another
sort of extension to cover functional dependences of which
the paradigm case was the gravitational law, in which the 'effect'
was an acceleration while distances and masses were the 'cause'.
 THE NECESSITY OF CAUSES 823
I have strayed rather far from the sort of theory Kneale
proposes. In relating natural to logical necessity and to the
incompatibilities of colours he is adhering to a rationalist tradi-
tion; instead of anything of this sort I find that his transcendent
hypotheses reveal only continuities and persistences; these are
an empirical counterpart of the rationalists' necessity. But we
can agree with Kneale that we can only conjecture that actual
laws of working will exhibit whatever this natural necessity
turns out to be. If laws of working are identified in the way
indicated by our discussion of Popper's theory, it is clearly a
synthetic claim, and not one that can be called a priori in any
strong sense, that they will also be forms of persistence.
We may weave in here some notions that have cropped up
earlier in our discussion. In Chapter 3 we considered the
possibility that some causal mechanism or continuity of process
might help to constitute causation in the objects; we have now
found a place for it, particularly as constituting a sort of neces-
sity that may belong to basic laws of working. In Chapter i we
noted Hume's hesitation over contiguity, and suggested that he
was right not to include this as an essential part of our ordinary
concept of causing. Yet contiguity, that is, a merely spatio-
temporal continuity of processes, may have a place in causa-
tion as it is in the objects, not so much for its own sake but
rather as making possible qualitative or structural continuity
as well.
Although I am offering this notion of persistence or of qualita-
tive or structural continuity as a hint of a kind of necessity that
may belong objectively to basic laws of working, it is worth
noting that something of the sort is to be found also in our
concept of causing. Michotte, investigating this concept from
the point of view of experimental psychology, found that we
have in certain cases an immediate perception or impression of
causation. His two basic cases are these. In one, an object A
approaches another object B; on reaching B, A stops and B
begins to move off in the same direction; here the observer gets
the impression that A has 'launched' B, has set B in motion.
In the other, A continues to move on reaching B, and B moves
at the same speed and in the same direction; here the observer
gets the impression that A is carrying B with it. In both cases
observers typically report that A has caused the movement of B.
224 THE NECESSITY OF CAUSES
Michotte argues that it is an essential feature of observations
that give rise to this causal impression that there should be two
distinguishable movements, that of the 'agent' A and that of the
'patient' B, but also that it is essential that the movement of
the patient should in some degree copy or duplicate that of
the agent.28 Where the effect is in this sort of way a partial
continuation or extension of the cause—Michotte speaks of
^ampliation du mouvement—it is, as Hume himself pointed out,29
particularly easy for us to see them as belonging together, to
form, imaginatively, the arrangement of ideas that is expressed
by the judgement that if the cause had not occurred, the effect
would not. This source of such thinking and judgement involves
only the intrinsic features of the single observed sequence; it
does not, like the Method of Difference reasoning, require any
general assumption. It is also different from, but can reinforce,
the sort of imaginative analogizing that we took in Chapters 2
and 3 to be a primitive counterpart of the use of the Method
of Difference followed by a sophisticated sustaining of a counter-
factual. Our ordinary concept of causation does not require this
qualitative or structural continuity, but such perceived con-
tinuity does contribute to the formation of that concept$ it is
one among several sources of the counterfactual conditional
thinking which is the core of the ordinary concept.
It is worth noting, also, that those cases of alternative over-
determination which nevertheless allowed us to pick out one
rather than another of the over-determining factors as the cause
were ones in which we could tell some more detailed causal
story, in which we could find links in a continuous chain con-
necting the preferred factor, but not its rivals, with the result.
Qualitative or structural continuity of process, then, may
well be something in the objects, over and above complex
regularity, which provides some backing for the conditional and
especially counterfactual statements that emerge in the analysis
of our ordinary causal concept. We can, as I suggested in
Chapter 4 that we might, thus relate the logical forms associated
with causal beliefs to our spatio-temporal picture of a bit of
causing going on—a connection which bears some resemblance
28
A, Micholte, La perception de la cataalM (translated by T. R. and E. Miles as
The Perception of Causality}, Introduction, pp. 17-19; Chapter 14, esp. p. 214.
'» Treatise, I, HI, 9, Selby-Bigge, pp. 11 i-ia (quoted in Chapter I above, p. 24).
 THE NECESSITY OF CAUSES 225
to Kant's schematism of the categories, an embodiment of forms
of judgement in a spatio-temporal pattern of sensory appear-
ances.
Remembering what we found, in Chapter 7, about the direc-
tion of causation, how it was associated with the contrast be-
tween fixity and non-fixity and with the way in which events
become fixed, we can see another important difference between
basic laws of working and collocations. Since laws of working
cover the ways in which (some) things become fixed, there is,
while not an absolute requirement, yet something of a pre-
sumption that there should be such laws. The universe needs
to know where to go next. Of course we must concede Hume's
point that it is in a way conceivable that there should be com-
plete chaos, that every occurrence should be random—though
we must also concede Strawson's point that we cannot say or
even imagine in detail what chaos would be like. But this would
be a very strange state of affairs, at variance with expectations
which are presumably instinctive and which are certainly in-
volved in and satisfied by our whole method of interpreting our
experience, whereas chaotic collocations would be harmless and
in no way out of the ordinary. Mill overstated the case in
claiming that the 'law of causation' is a general axiom which
somehow guarantees that there must be uniformities of succes-
sion, whereas there is no such axiom requiring uniformities of
coexistence,30 but at least there is a presumption in favour of
the former though not in favour of the latter. To this extent,
then, what I called the seventh job for necessity, the notion
that it is somehow necessary that every event should have a
cause, contributes in a weakened form to the distinctive neces-
sity of basic laws of working—and, of course, facilitates their
inductive establishment, which in turn explains their so-called
counterfactual force.
The dictum that 'the universe needs to know where to go
next' may require some explanation and defence. I am suggest-
ing that there is some truth in the notion that what happens
next flows from what is there already. The immediate future is,
so to speak, extruded by the present and the immediate past. Of
course all that is observed and all that ever could be observed
30
Quoted above, p. 205.
aa6 THE NECESSITY OF CAUSES
about this process amounts to a vast and intricate collection of
persistences, continuities, and complex regularities. But there
is perhaps also something more elusive, which my talk about
fixity and fixing is an attempt to capture. Ayer, endorsing what
I called in Chapter i Hume's third exclusion, suggests that
'everything that happens in the world can be represented in
terms of variations of scenery in a four-dimensional spatio-
temporal continuum'.31 Represented, yes; but is the repre-
sentation adequate? Does everything that happens in the world
simply consist of varied four-dimensional scenery? I want to
deny this, but it is very difficult to say just what the denial
means. The notion of flowing from, of extrusion, would mean
that there had to be some relations of the general sorts mentioned
above—persistence, continuity, regularity—that what happens
fits into such categories as the continued existence of objects,
motion, change (which involves also some persistence), growth,
development, decay, composition, disintegration, but not—or
at least not in most cases—annihilation or coming to be from
nothing. I am not, of course, suggesting that the laws or
regularities in accordance with which things come about must
themselves be intelligible, but only that if we are right to see
the future as flowing from the present we are justified in expect-
ing that there will be some ways in which it flows.
The implausibility of the 'four-dimensional scenery' account
can be brought out by considering how strange a world—or bit
of the world—we should have if the time dimension changed
places with one spatial dimension. Each persisting object or
continuant would turn into a very long but very short-lived
worm. The sudden changes that we encounter as we move
along some spatial straight line would become successive events
in the same place. The direction of explanation (for example,
from a common cause to joint effects) would run from, say,
south to north (but not from north to south) between simul-
taneous existents. It is obvious how serious a loss of intelligibility
such an interchange of axes would carry with it. It is worth
noting that this is a kind of transformation which the Special
31
Probability and Evidence, pp. 10-11 (quoted more fully above, pp. 23~4).At this
point he presents this just as an equivalent of Hume's thesis, but he says that it is
'scientifically most respectable' and 'at least very plausible', and he goes on to
defend it.
 THE NECESSITY OF CAUSES 237
Theory of Relativity neither imposes nor even allows. No
change of co-ordinate system within that theory will turn what
had been a line of simultaneity into a line of temporal succes-
sion in one place; no line representing the possible history of
any persisting tiling or causal process, however rapidly moving,
can become in some other co-ordinate system the locus of a set
of simultaneous events; possible light-paths constitute a rigid
barrier between possible causal succession and what can, from
any point of view, be seen as coexistence.
To say this is not merely to say, what is obviously true,
that time does not present itself in our experience as just
another dimension like the spatial ones. It is to say that if
what is now experienced as temporal were experienced as
spatial, and part of what is now experienced as spatial were
experienced as temporal, the result would be very strange. For
example, an ultimate uniformity of coexistence, say that a thing
of one sort should always shade off, in a northeny direction,
into a thing of a certain other sort, would strike us as a very odd
coincidence if there were no causal sequences, no laws of work-
ing through time, of which it was a resultant. To bring out the
strangeness, to us, of an exchange of roles between a spatial
and a temporal dimension is to display, by a contrast, some-
thing that is at least firmly built into our ordinary conceptual
scheme. But the facts also fit in with this scheme. Even if we
started, as we may well have, with an innate prejudice in
favour of something like it, we should not have been able to
work with it unless the data of observation had been amenable.
If we take the world just as Ayer's four-dimensional stage-
setting, it is remarkable that nearly all the long worms it con-
tains are temporal ones, occupying possibly causal lines, and
hardly any are spatial ones, occupying what could be lines of
simultaneity—even telephone wires, railway lines, oil pipelines,
and the like have a considerable temporal persistence as well
as spatial extension, and so come out not as worms but as thin
but two-dimensionally extended spatio-temporal sheets. Now
if there were nothing in the notion of what happens next
flowing from what is there already, it would be a surprising
coincidence that worms should be distributed through the four-
dimensional scene in the selective way that they apparently are,
whereas if there were something in this notion, the actual
aa8 THE NECESSITY OF CAUSES
distribution is just what we should expect. To this extent we
can defend a presumption that does something like what I
called the seventh job for necessity.
We can sum up this discussion by referring to the other six
jobs for necessity. In Chapter i, after distinguishing necessity2
from necessity^ I suggested that it might have a variety which
I called necessity2.2,something that would support a prob-
abilistic a priori inference, as contrasted with a necessitya <x
which would support a deductively valid a priori inference.
We have found nothing that could play the part of necessitya lt
but qualitative or structural continuity of process, along with
the persistence of objects, might count as necessity2i3. Such
continuities and persistences up to a point—for example, over
the stretch A-B-C in our diagram—seem to constitute some
presumption in favour of their own continuance. But this is
only a hint, which would need to be developed as part of a
wider discussion of induction. There might be something that
contributed a little to the second job, therefore, and hence also
to the first. The third job, that of necessity3, is done by the
regularity aspect of causal laws. But for the fourth job, that of
necessityx, we need not only regularity but also the natural
necessity that does the fifth job, the marking off of causal from
non-causal regularities. This has been our main topic in this
chapter, and we have located it first in the distinction of basic
laws of working from collocations and mixed laws which in-
volve collocations, and secondly in the qualitative and structural
continuity which processes that obey those basic laws seem to
exhibit, in the fact that they are forms of partial persistence,
once we move from the events at the perceptual level to the
'transcendent hypotheses' with which those events are, on a
scientific realist view, identified, and by which they are ex-
plained. This account has been built upon foundations provided
by Kneale's theory, but we have rejected his suggestions that
this natural necessity can be assimilated either to logical neces-
sity or to the incompatibility of colours; our continuity, as I
said, is an empirical counterpart of the rationalists' logical or
near-logical necessity. And of course these continuities and
partial persistences, while joining with the notion of basic laws
of working to do the fifth job, do the sixth job on their own:
it is they that would really constitute the long-searched-for link
 THE NECESSITY OF CAUSES 229
between individual cause and effect which a pure regularity
theory fails, or refuses, to find.
We can now, at last, reply to the sceptical suggestion made
near the end of Chapter 2, that since singular causal statements
involve in their analysis counterfactual conditionals which can-
not be true, they cannot be true either, and that necessity^ the
distinguishing feature of causal sequences, cannot be observed.
These comments hold, indeed, for statements of singular causal
sequence as we ordinarily mean them, for what is contained
in our concept of necessity!- But we can now offer a revised
account of what causation is in the objects. If our speculations
are correct, a singular causal sequence instantiates some pure
law of working which is itself a form of partial persistence; the
individual sequence therefore is, perhaps unobviously, identical
with some process that has some qualitative or structural con-
tinuity, and that also incorporates the fixity relations which
constitute the direction of causation. Here, then, we have what
was foreshadowed at the end of Chapter 5: a general charac-
teristic of causal processes, sometimes observable, sometimes
not, which constitutes a link between cause-events and effect-
events similar to but more selective than the relation defined
by Ducasse, since it relates specifically relevant causal features
to those features which constitute the result. This is an answer
to the question, 'What is the producing that Hume did not
see?' And it is easy to apply to this an explanation already sug-
gested in Chapter i: Hume failed to see it not, as Kneale
suggests, because he confined himself to 'impressions', but
because he was looking for something else, for an a priori
justification of inference, and gratuitously assumed that 'pro-
ductive quality' could be only another name for such a necessity.
These aspects of causation as it is in the objects—the holding
of pure laws of working which are forms of partial persistence,
and hence the qualitative or structural continuity in each
individual process that instantiates them—are related in various
ways to the important counterfactual conditionals. The induc-
tive evidence that supports or corroborates the laws justifies the
use of the counterfactuals which they seem to sustain, and the
continuities, when observed or inferred, give at least further
psychological encouragement and perhaps further justification
for their use. Thus the counterfactual conditionals which
230 THE NECESSITY OF CAUSES
emerge from the analysis of causal statements, though not
themselves true, are surrogates for clusters of statements which
can be true.
Of these objective aspects of causation some are sometimes
observed, for example in the experiences which, according to
Michotte, provide our basic 'perception of causality', and in
many of the experiences in which we learn to use common
transitive, causal, verbs, as Anscombe and Holland insist.
Where they are not observed, such features may be postulated
in transcendent hypotheses which are confirmed in the usual
scientific way. Other aspects of causation in the objects, in
particular the pure laws of working, are never the subject of
observation, but are reached by the framing and testing of
hypotheses, by the gradual sorting out of these laws from what
depends upon collocations or initial conditions. But that does
not make them any less objective.
 9
Statistical Laws

IT has become clear in several of the preceding chapters that

a study of causation need not presuppose causal determinism,
or even use it as a regulative idea. Chapter 2 showed that we
do not always require a cause to be a necessitating cause,
sufficient in the strong counterfactual sense, Chapter 3 that
even in those methods of discovering and establishing causal
relations that rely upon assumptions of uniformity, the form of
assumption required is not that of a 'law of universal causation'.
Chapter 4 revealed fatal flaws in Kant's attempts to establish
the general validity of the categories within our experience
and to show that, in the phenomenal realm of common sense
and science, every alteration must take place in accordance
with some law. In Chapter 5 we criticized Ducasse's attempt
to prove that ''every event of necessity must have a cause from
which it follows in a uniform manner'; in Chapter 7 we found
that universal determinism would undermine the contrast
between fixity and unfixity used in our preferred analysis
of the direction of causation; and nothing that Chapter 8
could uncover about natural necessity amounted to any
strong presumption that all events have sufficient causes.
Of course, none of these points amounts to a disproof of
determinism; but they suggest that we can get along with-
out it.
But if determinism goes, what takes its place ? The orthodox
answers are, freedom of the will in the realm of human choice
and action, and statistical laws in the realm of physics. I shall
not, in this book, be discussing freedom of the will, but I want
now to take up the question, Just what is a statistical law? Or,
so as not to prejudge the question whether there are any, Just
what would a statistical law be ?
832 STATISTICAL LAWS
Here, as elsewhere, I mean by the term 'law' an objective
law, not a law statement; 'laws' without further qualification
means what Bunge calls Iaws1} and defines as 'the immanent
patterns of being and becoming', as opposed to Iaws2, 'the conceptual
reconstructions thereof'.1
At first sight, the answer to my question may seem obvious:
take any universal law—perhaps of the complex regularity
form—substitute 'x per cent of for "all', and you have a statistical
law. A universal law states what always happens, a statistical law
what happens in some definite proportion of cases: universal
laws, if there are any, can be regarded as a limiting case of
statistical laws with x = 100.
Before criticizing this simple analogy, let us examine a corol-
lary of it. According to the Humean tradition, the only sense in
which an individual effect is necessary is that it is necessitated,
logically, by the conjunction of a contingent law and a cause.
(More accurately, the conjunction of the law statement with
some description of the cause-event logically necessitates the
statement that an effect of this sort occurs.) Analogously, this
tradition would suggest that an individual outcome, such as the
disintegration of a particular atom within some period of time t,
is probable only in the sense that it is probabilified, logically, by
the conjunction of a statistical law with an antecedent situation.
If the law is to the effect that of atoms of a certain sort in a
certain condition x per cent disintegrate within time t, then this
statement, in conjunction with the statement that this is an
atom of that sort in those conditions, probabilities to degree x
per cent the statement that this atom will disintegrate within
time t. We may wish to impose the restriction that no fuller
true description of the antecedent conditions, conjoined with
any other genuine law, should probabilify this outcome to any
different degree or entail either that it will or that it will not
occur. But the important point is that this tradition would
reduce the probability of individual events to logical probabili-
fication by statistical laws plus antecedent conditions just as
it reduces the necessity of caused individual events to logical
necessitation by laws of sufficient causation plus antecedent
conditions.
It is true that we have found reason to supplement the
1
M. Bunge, Causality, p. 249.
 STATISTICAL LAWS 233
Humean account of sufficient causation—as well as to insist
that our ordinary concept does not require a cause to be a
sufficient cause. What we have added is the direction of causa-
tion, the sorting out of pure laws of working, and the suggestion
that these are forms of at least partial persistence, that they
typically involve some qualitative or structural continuity. We
may well add these features also to our Humean picture of a
statistical law. Indeed it is particularly easy to add the direction
of causation to a statistical law, where the antecedent conditions
will be fixed before the outcome occurs, while that outcome will
presumably not be fixed until it actually occurs. And at least
in the central case of quantum mechanics, the mathematical
description in terms of ^-functions yields a kind of structural
continuity, though it is controversial whether these functions
can be interpreted realistically. But even when these features
are added, the probability of individual outcome events is still
being reduced to the logical probabilification of the correspond-
ing statements by statistical law statements plus descriptions of
antecedent conditions.
There are, of course, other concepts of probability, other
senses of 'probable' and its associates. For example, 'It is
probable that P (to a certain degree)' may mean 'There are
inconclusive reasons (of such and such strength) for believing
that P'. There is the 'range' or 'measure' concept of which the
classical definition in terms of a ratio of possibilities is a primitive
version. There are frequency concepts. And there is a notion
of objective chance or propensity.2 But the first three of these
would make no substantial difference here. If there are reasons
of some strength, but inconclusive, for belief about an indivi-
dual outcome that falls under a statistical law, this will be
just because the law and the antecedent conditions together
probabilify it; the measure concept comes in only as a way of
explaining and systematizing such probabilifications; the rele-
vant frequencies will be those within the class of events referred
to by the antecedent of the statistical law. Only objective
chance or propensity would give the individual outcome event
an intrinsic probability as distinct from the frequency in a class
of events and the consequential logical probabilifications. But,
2
I have examined these concepts and their interrelations in Chapter 5 of Truth,
Probability, and Paradox.
934 STATISTICAL LAWS
3
as I have argued, Hume's reasons for denying logical con-
nections between distinct occurrences would tell equally against
a corresponding intrinsic probability, an objective chance or
propensity. A Hurnean, then, would say that if antecedent
conditions ever probabilify, rather than necessitate, a certain
outcome, this relation will consist simply in this individual
situation's being an instance which falls under a statistical law,
just as any necessitating would have consisted simply in its
being an instance that fell under a universal law, a law of
sufficient causation.
But let us then go back to the question, Just what is a
statistical law? There is no difficulty or obscurity in the notion
of a statistical fact, for example that 30 per cent of dwellings
occupied in the United Kingdom at a certain date have no
bathrooms. But what constitutes a statistical law'? Is it a
statistical fact about the universe, differing from such a
statistical fact as that just mentioned only in being free from
spatio-temporal restriction ('in the United Kingdom at a cer-
tain date') ? Would it be a statistical law, if it were true, that
nine-tenths of all the human beings (defined, like Popper's
moas, by some biological structure) that ever have lived, are
now alive, or ever will live anywhere in the universe had, have,
or will have brown eyes ? This would be only a semi-uniformity
of coexistence. If, with Mill, we have recognized laws of co-
existence we might recognize statistical laws of the same sort,
which might arise in several ways.
First, it could be that events of type A, when conjoined with
ones of type B, regularly led to ones of type C, and that those of
type A, when conjoined with ones of type D, regularly led to
ones of type E, but it was a mere fact of collocation that in
x per cent of cases when an A was conjoined with a B, it was
conjoined also with a D. This, if Cs came about in no other
way, would have the consequence that x per cent of Cs had Es
associated with them. Thus a semi-uniformity of coexistence
could be derived from a mere statistical fact of collocation:
but Mill refused the title of law to those uniformities of co-
existence that had an analogous source.4
3' Truth, Probability, and Paradox,especially pp. 179-87 and 232-3; also in my review

of D. H. Mellor's The Matter of Chance, in Philosophical Quarterly, xxiii (1973), 85-7.

* System of Logic, Book III, Ch. 5, Sect. 9; cf. Chapter 8 above, pp. 304-5, n. 13.
 STATISTICAL LAWS 835
Secondly, it could be that some assemblage of conditions A
led uniformly, by a law of sufficient causation, to Cs, but only
by a statistical law to Es, producing Es only in x per cent of
cases; then again if Cs were produced in no other way, x per
cent of Cs would have Es associated with them. This semi-
uniformity of coexistence would be analogous to those uni-
formities of coexistence which Mill was ready to call laws: but
it would be only a derived law, and what is of greater interest is
the supposed law by which As produced Es in x per cent of cases.
Thirdly, just as Mill was willing to recognize ultimate uni-
formities of coexistence,5 so we must recognize the possibility of
ultimate semi-uniformities of coexistence of a statistical sort, not
derived in either of the two ways just mentioned.
However, the statistical laws that we need most to elucidate
are not any such semi-uniformities of coexistence, but laws of
working such as we used in the second case above, a law of the
form that As produce Es in x per cent of cases. It seems that it
would be a statistical law, if it were true, that 50 per cent of all
radium atoms, if left alone, not subjected to nuclear bombard-
ment and so on, disintegrate within a period of 1590 years
from the time at which they become radium atoms. But there
is still a difficulty. It might be that, in the whole history of the
universe, only 2" million radium atoms ever have come or will
come into existence, that the first half of these, the first a""1
million, all disintegrate within 1590 years of their becoming
radium atoms, while none of the other half ever have dis-
integrated or ever will. But if there were a statistical fact of this
sort about the universe, although this is a fact about changes
rather than coexistences, and although it is free from spatio-
temporal restriction, it seems that it would not constitute what
we mean by a statistical law. To be a law, it must in some way
apply to radium atoms severally, distributively, not merely
collectively. The proportionality of the outcome must be ex-
hibited repeatedly, it must in some way crop up wherever the
antecedent conditions occur. How can we reconcile this dis-
tributiveness, this severalty, with a merely statistical character?
It looks as if a statistical lav/ should have the form of a
universal quantification of some statistical fact, it should assert
that in all such and such, a certain statistical relation (about
5
Ibid., Book III, Ch. 22, Sect. 4, quoted above in Chapter 8, p. 205
236 STATISTICAL LAWS
changes, comings about) holds. But, if so, the simple analogy
between universal and statistical laws has broken down; the
latter will have some more complex form than 'x per cent of As
are (become, produce, lead to, and so on) Bs'.
'Where there are three doctors, there are two atheists.' In
whatever sense this might be true, it could not be that for
every set of three doctors, exactly two members of that set are
atheists. If there are four or more doctors in the world, then
however atheism is distributed among them it is mathematically
inevitable that there should be at least one set of three which
contains either more or fewer than two atheists. Similarly, it
cannot be true that for every collection of radium atoms exist-
ing at time t1} exactly half the atoms in the collection will have
decayed by ^+1590 years. It will be possible to find collections
all of whose members decay within this period, other collections
none of whose members decay, and others again with decay
ratios all the way up and down the scale between o per cent
and 100 per cent. No doubt one would be most unlikely to be
able to specify in advance one of these atypical collections, but
that is not to the point. We are trying to find a form for an
objective law of working, and any law-statement of the form
suggested above is necessarily false.
We seem able to do better with a Mendelian example. With a
plant species S, which displays contrasting genetically deter-
mined characteristics A and a, of which A is dominant, we
might try this formulation: 'In S, whenever a plant from a pure
strain displaying A is crossed with a plant from a pure strain
displaying a and one plant from the first filial generation (Fj)
thus produced is self-pollinated, exactly three-quarters of the
plants in the second filial generation (F2) thus produced will
display A.' Unlike the last, this suggested law-statement is not
necessarily false. But it is contingently false: most collections of
FI plants thus produced will not show this exact proportion of
As.
Nor will it do to substitute 'approximately' for 'exactly'. This
is too indeterminate for a statement of how nature actually
works in this field, and yet if any definite limits are set, for
example, if it is said that the proportion of As in F2 plants thus
produced will always lie between 70 per cent and 80 per cent,
the proposed law-statement will still be contingently false.
 STATISTICAL LAWS 337
It is worth noting why these proposed Mendelian laws are
only contingently false and not necessarily so. We are now
applying the universal quantification not to sets or collections
in general of the individuals of the kind in question, but to
groups of these individuals which are marked out independently
of our choices. The proposed law is about groups each of which
consists of all the individual Fz plants produced by the self-
pollination of one F± plant. So while a statistical law could not
be of the form 'For every collection of A, x per cent of the As
in the collection will be B\ it could be of the form 'For every
collection of As produced in such and such a way . . .'; it
could be a universal quantification over basic groups of a certain
kind—where each basic group might be, say, all the radium
atoms in a continuous lump of radium at time t^.
All the 'laws' of this form so far suggested are contingently
false. On the other hand, there could be genuine statistical
laws of this form. Let us construct an example, using only
slight distortions of some biological facts. Let us suppose that
the female golden eagle always lays exactly two eggs at a time
and that it always happens either that both chicks hatch out
or that both fail to do so. The two chicks in each brood then
compete for the food brought by the parents, and whichever
of the two has some slight initial advantage competes more
successfully and in consequence grows bigger and stronger than
its sibling. Eventually the chicks grow too big for the nest to
hold both of them, and the weaker of the two is pushed out
and dies. Assuming that eagle chicks are safe from every other
hazard, it will follow that of every pair of chicks just one
reaches maturity. Consequently a 50 per cent survival rate
holds for basic groups of golden eagle chicks, where the basic
group is the single brood.
An essentially similar example would be a law that has been
suggested for political triumvirates, namely that every trium-
virate is unstable, because it will always be to the advantage of
two of its members to combine against the third and exclude
him from power. If so, there will be a 33$ per cent casualty
rate among triumvirs, where the basic group is each particular
triumvirate.
In both these examples, a statistical truth results from a
certain pattern of competition within basic groups. But what is
238 STATISTICAL LAWS
essential is interaction, not necessarily competition. There
might be South Sea Island economies in which all the neces-
sities of life, and even all the known luxuries, were provided
easily by the activity of half the community, and in which
there was, in consequence, an amicable division of communities
into workers and idlers in equal numbers. But statistical laws
of this sort hold, if and when they do hold, because some par-
ticular pattern of interaction is displayed within each basic
group of the kind in question. These laws are always derived
laws; they result from interaction which goes on in accordance
with other, more basic, laws, which may well be universal,
deterministic ones. If there are any statistical laws of this sort,
let us call them interaction resultant laws, or IRLs. We could
put under this heading the fact that a heterozygous parent
produces, by meiosis, equal numbers of gametes carrying the
two genes that form a pair of allelomorphs: the 'interaction'
here is just the separation of the two halves of the cell. This
equal proportion of the sub-classes within this basic group
(gametes produced by a single individual) is a statistical out-
come of a process (cell division) which in this respect at least
is deterministic.*
But the most important examples of statistical laws of work-
ing are not of this sort: their statistical outcomes are not the
product of proportions thus causally maintained within any
basic groups. Even in the genetic process, only some parts are
thus covered by IRLs. What happens at fertilization is covered
by laws of chance of a different sort: interaction in basic groups
is no longer relevant, and even the basic groups themselves
seem to lose their importance. A statistical discussion of, say,
haemophilia and its carriers will deal with frequencies in large
populations rather than the proportions in individual families
which, since the families are fairly small, will often differ widely
from the theoretical probabilities. Again, physicists do not tell
us that the decay rate of radium depends upon interaction (in
even the loosest sense) between adjacent atoms: no basic group
is relevant here. But if basic groups go, over what can a
statistical law be universally quantified? How are we to re-
concile its statistical character with the need for repetition, for
distributiveness, to distinguish a law of working from a mere
statistical fact about the universe?
 STATISTICAL LAWS 239
The obvious answer which I have been withholding so long
is that a statistical law assigns the corresponding probability or
chance to each individual that falls under it, that it has the form
'Every A has an x per cent chance of becoming B (or of pro-
ducing a B)\ It is this assignment of a measured chance of the
change in question to each individual instance of the relevant
assemblage of conditions that allows a law to be at once uni-
versally distributive, as a law of working, it seems, should be,
and yet merely statistical in its outcome. Bernoulli's Law of
Large Numbers ensures that if each instance of A has an x per
cent chance of becoming B, there is a high probability that the
frequency, in any large collection of As, of becoming B will be
near to x per cent.6
But this answer, though obvious, is paradoxical in the light
of what has been said earlier. If we are to reduce the prob-
ability of individual outcome events to the probabilification of
the corresponding statements by the conjunction of descrip-
tions of antecedent conditions with statements of statistical laws,
we cannot without circularity take those laws to assign a prob-
ability of such an outcome to individual antecedent situations.
The probability interpretation of a statistical law requires that
the probability thus assigned should be some genuinely intrinsic
feature of each individual occurrence of the antecedent condi-
tions; in other words, an objective chance or propensity. But I
have argued elsewhere against the recognition of any such
entities,7 and it would clearly be inconsistent to introduce them
while keeping even as close as I have kept to the Humean
reduction of necessity to universality. On the other hand, the
rejection of intrinsic, not logically relational, objective single
case probabilities leaves us nothing to use as the predicate in
the proposed formulation of a statistical law. Of course, this
proposed formulation will do as a rough indication (it is not
as rough as the alternative 'x per cent of As are B'), and in
conjunction with Bernoulli's Law and Bayes's Theorems it in-
dicates how such law-statements are to be applied and tested;
but it will not do as a formulation of a law by which the world
actually works, as a literal description of what goes on. If there
6
I have examined the interpretation and application of this law in Truth,
Probability, and Paradox, pp. 204-13.
' Ibid., pp. 133-45, '79-87. and 232-5.
240 STATISTICAL LAWS
are no such probabilities, a law statement that assigns them to
all situations of a certain sort cannot be literally true. So we
still have on our hands the problem of saying just what a
statistical law of working could be.
The only way out of this difficulty is that indicated by the
theory of limiting frequency. We know what it is for, say,
tosses of a coin to approach a certain limiting frequency for
heads: in short runs the frequency may diverge widely from
this limit, but most longer runs approach it more closely, and
greatly divergent longer runs are rare. If there is a statistical
law of working about, say, the decay of radium atoms, the state
of affairs must be similar to this. If radium atoms are considered
in one or other possible series, for example a series determined
by increasing distance at a time t± from some arbitrarily chosen
point, their decaying or not decaying within some period of
time may exhibit the sort of pattern that makes a series look
as if it were a random collective approaching a certain limiting
frequency. Its doing this, I suggest, would constitute the hold-
ing of a statistical law of working. This statement may well
seem crude. But this is deliberate. It is all too easy in this area
to sacrifice reality to precision. We are tempted to say that in a
series that has a limiting frequency for some feature there is a
low probability for long runs that diverge widely from that limit-
ing frequency. But in the formulation of a law which is intended
to provide a reductive analysis of objective probability this
'probability for long runs' cannot remain unanalysed; it in
turn will have to be analysed in terms of limiting frequency;
and then we shall not be able to avoid a vicious infinite regress
if such a probability turns up in the account of what a series
with a limiting frequency is. Again, it is plausible to interpret
probability statements about individual objects, such as 'This
penny, tossed in the normal way, has an x per cent chance of
falling heads', as assertions of limiting frequency in a merely
hypothetical series, as being about what would happen if this
penny as it now is were tossed indefinitely.8 But no such hypo-
thetical statement can be an adequate formulation of a fully
objective law of working. Of course, we can and must frame
hypotheses to the effect that such and such a law holds. There
is no objection to laws, statistical or otherwise, being hypo-
8
Cf. Truth, Probability, and Paradox, pp. 177-8, 231-2.
 STATISTICAL LAWS 241
thetical in this sense. All laws of nature will be the objects of
such hypotheses and will be known only in so far as such hypo-
theses are confirmed or corroborated. But it would be quite
another matter, and quite inadmissible, to include a statement
about what would happen i f . , . within the formulation of a
(potentially) objective law,
A statistical law, then, must consist in some outcome's having
some limiting frequency in certain actual series or sequences of
instances of the assemblage of antecedent conditions. Its dis-
tributive character must lie not in any universal assignment of a
probability of that outcome to each such instance, nor even in
any universal assignment of probabilities of various frequencies of
that outcome to runs of such instances of various finite lengths,
but merely in the fact that long runs of such instances in various
spatio-ternporally determined series usually exhibit frequencies
close to the limiting one.
There is, indeed, another way in which we can interpret
such terms as 'tendency' or 'propensity' and which allows for a
universal assignment of a certain propensity to each antecedent
instance. A loaded die has, at every throw, the same physical
asymmetry which somehow accounts for its landing 6-up more
than one-sixth of the time. We could identify its propensity to
favour 6-up not with the disposition whose manifestation is the
limiting frequency but with the occurrent ground of that dis-
position, the physical asymmetry. In this sense, every throw of
the die contains this propensity, And by contrast an unbiased
die is physically symmetrical; if its propensity to fall equal
numbers of times with each of the six faces up is identified with
this symmetry, we can assign this equality-favouring propensity
to every throw. Similarly, all radium atoms share a nuclear
composition which somehow accounts for their being unstable
and for radium having the half-life that it has, a nuclear com-
position which is different from that of a stable element like
lead or of a radioactive element with a different half-life like
radon. If we again equate their propensity to decay not with the
disposition but with this occurrent ground of the disposition,
we can assign this propensity universally, to every radium atom
at all times up to that at which it actually decays. But it is
very obvious that these universal assignments are not in them-
selves statements of any laws of working. They would become
243 STATISTICAL LAWS
so only if the propensity were interpreted as a disposition rather
than as the ground of a disposition. But if we analyse dis-
positions, as we must, in terms of laws, we cannot without
circularity include the assignment of dispositions in a strict
formulation of those laws. This approach would give us a law
of working as a universally quantified propensity-assignment
only by way of an equivocation about 'propensity'.
This exclusion of chance or probability from the formulation
of statistical laws not only makes that formulation cruder and
clumsier than we might desire but also deprives us of what
looks like an explanatory concept. We feel that if each instance
of the antecedent conditions had the appropriate chance of pro-
ducing such and such an outcome, this would explain why, in
most long runs of such instances, this outcome occurs with a
frequency fairly close to that chance. But since the supposed
chance can be introduced only as an unknown somewhat which,
by obeying the standard probability calculus, issues in a much
higher chance for corresponding frequencies in long runs, and
hence in a high chance that most long runs of instances of the
antecedent conditions will exhibit the corresponding frequency
of the outcome, where this last 'chance' is taken to justify a
corresponding degree of belief, the supposed explanation is
spurious. The explanatory item is a mere 'that which would
account for' what it is supposed to explain. There is, indeed,
no objection to talk about 'powers' or (more old-fashioned)
Virtues' if these terms are taken as place-holders for genuinely
explanatory items still to be discovered;9 but a 'chance' or
'propensity' assigned to each individual instance of the ante-
cedent conditions cannot serve as such a place-holder: nothing
could ever fulfil the promise it offers. Such items as the nuclear
composition of atoms or the physical symmetry or asymmetry of
a die will not fill this bill; they will not explain, in the way
promised by the term 'chance', the appearance of an approach
to a limiting frequency which is the observed fact in question.
This last claim may be disputed. In the case of the die we
can at least sketch an explanation. There is a considerable
range of variation between possible ways of placing the die in
the dice-box and throwing it onto the table. Any one such way
leads deterministieally, by strict laws of sufficient causation, to
• Of. Truth, Probability, and Paradox, p. 143.
 STATISTICAL LAWS 843
a certain result, a certain face's being uppermost when the die
comes to rest. If the die is symmetrical, equal portions of this
range of possible variation will thus lead to the results I -up,
2-up, and so on. But if the die is asymmetrical, say if it is
loaded to favour 6-up, a larger portion of this range of possible
variation will lead, deterministically, to 6-up than to any other
result. Add to all this the vital further fact that in a certain long
run of throws of the die the actual ways of placing the die in
the dice-box and throwing it were scattered fairly evenly over
the whole range of possible ways, and we have an explanation
why in this run the six faces came up about an equal number
of times each (if the die was symmetrical) or unequally, with
6-up the most frequent (if it was loaded).
But this explanation depends on the conjunction of physical
symmetry or asymmetry with deterministic laws of working
and with a brute fact of collocation, the actual scatter of throw-
ing methods used in this run.
A similar explanation could be given, or at least sketched, of
the frequencies that result from the genetic lottery. For example,
a heterozygous parent plant produces male and female gametes
carrying the allelomorphs A and a in equal numbers. In a
particular process of self-pollination, these gametes are shuffled
around in such a way that about half the A females are fertilized
by A males and about half by a males, and the a females like-
wise. In this individual process, the frequencies among the
offspring of close to one quarter AA, one half Aa, and one quarter
aa result from the equal numbers of A and a gametes (itself
the product of an IRL) conjoined with deterministic laws of
working and a fact of collocation, the actual shuffling that
occurred.
If we now go on from a particular process of self-pollination
in one plant to what regularly happens in the self-pollination of
plants that are heterozygous with respect to this pair of allelo-
morphs, the main problem is what to say about these shufflings.
It is a mathematical fact that if there are m male and n female
gametes (m>«), and half of each are A and half a, then of the
possible unions resulting from all possible shufflings exactly one
quarter will be AA, and so on. But what sort of fact is it that
actual self-pollination processes incorporate shufflings which are
pretty fairly representative of the range of mathematically
244 STATISTICAL LAWS
possible shufflings ? Comparison with other examples of shuffling,
say of large numbers of grains of sand of different colours placed
in a tray that is then mechanically agitated, shows that the
characteristic apparently random results of shuffling need not,
and presumably in general do not, depend upon any indetermin-
ism in the pure laws of working, but simply on the conjunction
of facts of collocation—the exact shape of the tray, the exact
initial arrangement and shapes of the variously coloured grains,
and so on—with deterministic laws. That actual shufflings are
pretty fairly representative of the range of mathematically
possible shufflings is then, in terms of the distinctions drawn
in Chapter 8, a mixed law. Various applications of this and
analogous principles have the consequence that certain kinds
of set-up regularly produce outcomes which are apparently
random in order and arrangement but which seem to approach
definite limiting frequencies for the different possible kinds of
outcome. Some such set-ups—dice-throwing, card-shuffling,
the stirring up of marbles in a lottery-barrel, and so on—are
typically used as gambling devices. These in general involve
fairly close approximations to equality between certain aspects
of the set-up—a die is meant to be physically symmetrical with
respect to all six faces, the cards in a pack are equal in size and
surface texture, and so on—and consequently also in the result-
ing limiting frequencies for alternative outcomes. But exactly
analogous explanations underlie the unequal limiting frequencies
generated by asymmetrical dice and unequal set-ups in general.
We may, then, give the name of gambling device laws, or
GDLs, to all those laws each of which consists in a certain kind
of set-up regularly generating outcomes in apparently random
order with limiting frequencies for alternative outcomes, where
this can in principle be explained in the sort of way indicated
by our examples. GDLs, then, are mixed laws; they are deriva-
tive from deterministic pure laws of working in conjunction
with certain relevant and fairly permanent features of the set-
up—the symmetry or the exact asymmetry of the die, the
equality or inequality of the numbers of A and a gametes, and
so on—and with other more obscure facts of collocation, almost
impossible to describe in detail, which by the complexity and
scattering of their relevant aspects help to introduce the appear-
ance of randomness into the results. Consequently, even if we
 STATISTIC3AL LAWS 245
state a GDL in a hypothetical form, putting into the ante-
cedent explicit descriptions of all the fairly permanent relevant
features of the set-up, we cannot thereby 'purify' it, and turn
the mixed law into a pure law of working, as we could (in
principle) with the mixed laws discussed in Chapter 8, What
I have called the more obscure facts of collocation keep it
impure. If their obscurity could be overcome and we could, for
any particular process or run of trials, include in the ante-
cedent of a hypothetieally formulated law a description of all
the relevant detail, then we should have a derivative pure law
of working, but a deterministic one, which this particular pro-
cess or run would exactly instantiate. But if we put into the
antecedent only a rough general description of these obscure
contributory facts of collocation, our law will still be a statistical,
not a deterministic, one, assigning only limiting frequencies to
series of the antecedent conditions, but also it will remain a
mixed law, dependent on the unstated and obscure facts of
collocation, not a pure law of working.
GDLs are different from IRLs, although, as we have seen,
laws of both kinds may play a part in the same complex pheno-
menon, covering, for example, different phases of a genetic
process. IRLs rest upon patterns of interaction of some kind
within basic groups and may issue in exact proportions within
each basic group (and therefore, automatically, in any collec-
tion of such basic groups); GDLs rest rather on an input of
complex, scattered, detail in relevant collocations, and issue
only in limiting frequencies. But both are only mixed laws, and
in both cases the pure laws of working from which they are
partly derived may be, and in all the ordinary cases presumably
are, deterministic laws, laws of sufficient causation.
Consequently, we can reaffirm our thesis that such items as
the symmetry or asymmetry of a die will not explain in the
way required the appearance of an approach to a limiting fre-
quency of this or that outcome. We can concede that it helps
to explain this, but only in conjunction with some obscure facts
of collocation. What is even more serious is that such explana-
tions dispense with statistical laws of working, resolving what
appear to be such into mixed laws derived from the conjunction
of collocations with deterministic laws of working. Any such
explanation, then, is very far indeed from fulfilling the promise
246 STATISTICAL LAWS
offered by the assignment of an objective chance or propensity
to each instance of the antecedent conditions. We can then, in
the light of this discussion, reaffirm the thesis that whereas talk
about 'powers' or 'virtues' can be interpreted, harmlessly, as
merely foreshadowing the discovery of genuinely explanatory
items, the analogous talk about 'chances' or 'propensities' can-
not; at any rate not without departing from the claim that
what these terms explain, or foreshadow an explanation of, are
statistical laws of working,
What our discussion brings out, then, is the need to dis-
tinguish such statistical laws of working (if there are any), which
we may call SLWs, from both IRLs and GDLs. Whether there
are any SLWs is a question that we cannot answer with con-
fidence. The orthodox view of quantum mechanics asserts that
there are, and indeed that what look like macroscopic deter-
ministic laws are derivative from SLWs at the micro level, and
are therefore (in general) themselves only SLWs that approach
determinism as a limit. But a heretical view of quantum
mechanics in effect suggests that the quantum laws themselves
are what I have called GDLs rather than SLWs. I cannot
adjudicate this controversy; I have been concerned merely
to bring out what SLWs, if there are any, would involve. The
concept of an SLW can be developed coherently and without
conflict with our empiricist, semi-Humean, treatment of natural
necessity. And yet when it is clarified and distinguished, some-
thing rather mysterious is left. There is, I har?e argued, an
unavoidable crudity of formulation: an SLW must find its dis-
tributiveness simply in the fact that certain actual series appear
to approach limiting frequencies and that most long runs in
such series do not diverge far from the limiting frequencies:
we cannot make this formulation neater by introducing universal
assignments of chances or probabilities without abandoning the
claim that such a law-statement is an objective description of
how things work. Also, there is an inevitable lack of explanation:
we cannot go behind the brute fact of actual approaches to
limiting frequency with random distribution without sliding
from SLWs to GDLs and hence to deterministic laws of working.
There may, indeed, seem to be a gap in our argument here.
I have said only that the laws of working from which GDLs
are partly derived may be and in ordinary cases presumably are
 STATISTICAL LAWS 247
deterministic, not that they must be so. There is no thing in theory
to preclude the derivation of a GDL from the conjunction of
collocations (both the relatively permanent features and the
obscure, scattered, details) with pure laws of working some of
which are SLWs. True, but that only shifts the problem. If
what we initially took to be an SLW is explained as being, after
all, a GDL which itself is dependent partly on an SLW, then
it is this new SLW that both calls for and resists explanation.
We cannot, however, make very much of this, since our
empirical counterpart to Kneale's necessity gave even deter-
ministic laws only a partial intelligibility even in principle: even
a deterministic interaction requires, inescapably, a law of work-
ing which does not consist just in the regular persistence of
anything. That any interaction should go one way rather than
another, however regularly it does so and however many items
or quantities are conserved through the transaction, remains in
part a purely contingent fact which no penetration into under-
lying features, no transcendent hypotheses, could ever render
completely intelligible. Yet it remains true that a statistical law
of working resists intelligibility still more. On the other hand,
it is worth repeating that statistical laws of working make it
particularly easy to feed in our criterion for the direction of
causation, and that those suggested for the central quantum
mechanical case clearly provide for a recognizable kind of
structural continuity.
 IO

Extensionality—Two Kinds
of Cause

OF what kind are the entities that stand in causal relations?

Between what sorts of item does causing go on? In Chapter 2
this question was brushed aside with the interim suggestions
that although causes and effects would be spoken of as events,
'event* would be used just as a general term to stand in for such
items as 'the hammer's striking the chestnut*, and that in con-
structing conditional analyses of singular causal statements we
could go back from such nominalizations to the corresponding
clauses or their negations: 'The chestnut would not have be-
come flatter if the hammer had not struck it.' Similarly in
Chapter 3 regularities were formulated in terms of types of
event or situation. But these ways of speaking slurred over a
cluster of issues which deserve attention.1
At least two questions have been raised. Are causes events or
facts? And are causal statements extensional? Both are con-
troversial. Philosophers have long been inclined to speak of one
event causing another. The double assassination at Sarajevo was
surely an event, and it was at least the triggering cause of the
First World War, Standing conditions and states of affairs are
admitted to be also causally relevant, to be perhaps predisposing
causes, but they can be accommodated, under a tolerant inter-
pretation of the term, as static events, as Ducasse's unchanges.2
1
They are discussed, for example, by Donald Davidson, 'Causal Relations',
Journal of Philosophy, Ixiv (1967), 691-703; Zeno Vendler, 'Causal Relations', ibid.,
pp. 704-13, also in Vendler's earlier 'Effects, Results, and Consequences' in Analytical
Philosophy, ed. R. J. Butler; G. E. M. Anscombe, 'Causality and Extensionality',
Journal of Philosophy, Ixvi (1969), 152-9; and Adam Morton, 'Extensional and
Non-truth-functional Contexts', ibid., pp. 159-64.
1
Though Ducasse seems to distinguish a standing condition from an unchange:
cf. Chapter 5.
 EXTENSIONALITY 249
Yet it is also tempting to say 'The fact that there was a
short circuit caused it to be the case that there was a fire' or that
Caesar's having crossed the Rubicon caused the Civil War;
these go together because the phrase 'Caesar's having crossed
the Rubicon' is an imperfect nominal, and the word 'fact* goes
with the imperfect nominals whereas the word 'event' goes with
the perfect nominals,3 Again, the extensionality of at least some
causal statements seems intuitively obvious. We can surely sub-
stitute co-referring expressions in singular causal statements: if
the assassination at Sarajevo in 1914 caused the First World
War, so did the assassination in the capital of Bosnia in 1914,
and so did Gabriel Prinzip's best-known action; if Caesar's
crossing of the Rubicon caused the Roman civil war, so did
Caius Julius's crossing of the boundary of Italy; and so on. On
the other hand, even if we agree that there was an international
crisis because President de Gaulle made a speech we may
hesitate to say that there was a crisis because the man with the
biggest nose in France made a speech;4 moreover, there is a
logical argument which is believed to show that any context
which is extensional must also be truth-functional, which causal
contexts certainly are not.5 In favour of another aspect of
extensionality it might be argued that statements of causal
regularity relate types of event or situation, and must neces-
sarily retain their truth-value when coextensive predicate terms
are substituted for one another. An excessively simple regularity
will serve to illustrate this point: if every event of type A is also
of type B and vice versa, then 'Every event of type A is followed
contiguously by an event of type C" will retain its truth-value
when 'B' is substituted for A'. But against this it might be
argued that of the coextensive features A and B one might be
causally relevant to C and the other not; for example, someone
who held a double aspect theory of mind and brain might say
that the neurophysiological aspect has effects which cannot
properly be ascribed to the mental aspect of the same events;
he could not deny the corresponding regularities, but he might
reply that this shows only that there is more to causing than
regular sequence.
3
Vendler, 'Causal Relations'; this is discussed below.
* Anscombe, op. cit.
1
Davidson, op. cit.; Anscombe, op. cit.
250 EXTENSIONALITY
Since the logical argument mentioned above has tended to
dominate recent discussions, we may consider first whether it is
sound and what, if anything, it shows. It purports to show that
if a context is extensional it is truth-functional, so that it can
be applied to prove that since causal contexts are not truth-
functional they cannot be extensional either. However, 'exten-
sionaP is ambiguous. It may mean that the substitution of
co-referring singular terms does not affect truth-value, or that the
substitution of coextensive predicates does not affect truth-value.
Let us use Quine's term 'referentially transparent' (or, briefly,
just 'transparent') for the former meaning, and reserve 'exten-
sional' for the latter. The argument we are to consider is then
concerned with transparency, not extensionality: it purports to
show that if a context is transparent it is truth-functional. It
runs as follows.
Let lp' represent any sentence, and 'F(/>)! any sentence con-
taining the sentence represented by '/>'. That is, 'F(...)' is a
sentence-forming operator on sentences, or briefly a context.
Suppose that 'F(...)' is such that logical equivalents are inter-
changeable within it salva veritate. Suppose further that 1F(...)'
is referentially transparent, so that co-referring expressions are
interchangeable in 'F(p)' salva veritate provided that they are
so in '/>' itself. Then consider x(x = x &/>)> that is, the class of
xs such that both (x is identical with x) and p. If p is true, this
will be the class of xs such that x = x, but if p is false it will be
the empty class. Consequently *x(x = x & p) — x(x — x)'—the
statement that {the class of xs such that both (x is identical
with #) and p} is identical with {the class of xs such that (x is
identical with x}}—will be logically equivalent to 'p', since,
whatever 'p' may be, it is necessarily true when 'p' is true and
false when 'p' is false. Moreover, if 'q' has the same truth-value
as 'p', x(x = x & p) will be the same class as x(x — x & q),
since each will be x(x = x) if 'p' and 'q' are both true, and each
will be the empty class if y and 'q' are both false. We then have
the following steps:
1 (i) p+-*g —assumption
2 (2) F(p) —assumption
2 (3) F(x(x — x Sap) = x(x — x))—from (2), by substitu-
tion of logical equivalents in 'F(...)'
i, 2 (4) F(x(x — x & q) = x(x = x))—from (3), by substi-
 EXTENSIONALITY 351
<
tution of co-referring expressions in F(.,.)', in the
light of (i) and the above remarks.
i, a (5) F(q)—from (4), by substitution of logical equi-
valents in 'F(...)'
i (6) F(p) -*• F(q}~~-from (2), (5), by conditional proof,
i (7) F(q) -> F(p)—by a similar series of steps to (a)-(6).
i (8) F(p)*->F(q)—from (6) and (7).
(9) (P++9) -+(F(p)++(q))—from (1), (8), by con-
ditional proof.
Now (9) says that the context 'F(...)' is truth-functional, and
this has been proved from the suppositions that '/*"(„.)' is such
as to allow substitution within it of logical equivalents and of
co-referring expressions. That is, if 'F(...)' is both transparent
and allows substitution of logical equivalents, it is truth-
functional.6
This argument has been widely accepted and respected, but
is it sound? As Professor Ansconibe says, someone might com-
plain about its artificiality, but that is beside the point. The
most suspect move is that from (2) to (3), resting on the alleged
logical equivalence of'/)' and 'x(x = x & p} — x(x = x)'—with
of course, the reverse move from (4) to (5). The point of this
move is that it replaces the proposition '/>', which may be about
any subject whatever, with one which asserts the identity of
two classes, and therefore allows the substitution at (4) of
another expression '£(x = x & q)' which refers to the same
class as does '#(# = x &. p)\ Cummins and Gottlieb say roundly
that this step is fallacious. If the set abstraction operator is
denned contextually, so that
x(Fx) = x(Gx) == df(x)(Fx^Gx)
we can no longer move validly from (3) to (4), since the apparent
class-identity of (3) has been reduced by this definition to a
statement that two predicates are coextensive, and our initial
assumption was only that 'F(...)' was transparent, not that
6
The argument is credited originally to Frege and variants have been used by
Godel ('Russell's Mathematical Logic' in The Philosophy of Bertrand Russell, ed.
P, A. Schilpp, p. 129, n. 5) and Quine ('Three Grades of Modal Involvement'
(1953), reprinted in The Ways of Paradox, pp. 156-74) as well as by Davidson and
Anscombe in the papers referred to above, p, 248, n. i, It is discussed by Morton
in the paper referred to above and by Robert Cummins and Dale Gottlieb in
"On an Argument for Truth-functionality', American Philosophical Quarterly, ix
(1972), 265-9.
asa EXTENSIONALITY
it was extensional in the sense of allowing interchange of
coextensive predicates. If, alternatively, the set abstraction
operator makes definite descriptions out of predicates—that is,
if we take sets seriously as individual objects to which reference
can be made—then '/»' and 'x(x = x Scp) = x(x = x)' are not
logically equivalent, for the latter implies the existence of the
class of xs such that x = x, whereas the former does not. It
might be objected, however, that the class of xs such that
x = x—the universal set—is guaranteed by logic alone to exist
and to be non-empty, and therefore that the explicit commit-
ment to it in 'x(x = x & p) = x(x = x)' does not destroy the
latter's logical equivalence to '/>'. As Anscombe says, what we
need is the description of a class x(Gx & />) where 'G' is a re-
spectable class-forming predicate such that x(Gx) logically can-
not be empty. There is room for dispute here.7 The argument
could, then, be blocked by a narrow interpretation of the term
'logical equivalents' in the assumption that '/*"(...)' is such that
logical equivalents are interchangeable within it salva veritate, or
vindicated by a broader interpretation of this term. It is hard
to see how anything of importance could turn upon this choice.
Still, the conclusion of this argument is paradoxical. It seems
intuitively obvious that some causal statements are transparent
without being truth-functional, even if others are referentially
opaque, and we must be suspicious of an argument that pur-
ports to exclude the former possibility. In 'The fact that Smith
fainted caused it to be the case that Jones ran for help' truth-
value will be preserved by any substitution of co-referring
expressions for 'Smith' or for 'Jones' and by any substitution of
7
Cummins and Gottlieb discuss Quine's version of the argument, which intro-
duces as equivalent to 'p' the formula 'X (x = $ & p) = £ (x = <£)' where '</>' is a
name for the empty set. They rebut an objection that the existence of X (x = </>)
is guaranteed by first-order theory with identity because (3*) (x = c) is a theorem,
and also point out that the proof fails if the schematic letter 'c' is used throughout
instead of '<j>', but they do not consider the more serious objection that the existence
of the empty class $ itself, and hence of the class of which this is the only member,
is specifically guaranteed by logic. This would be the analogue of the claim men-
tioned above, that the existence and non-emptiness of the class A (*==*) are
logically guaranteed. Anscombe points out that the argument could be constructed
in terms of numbers rather than classes: the number of numbers n such that (n is
an even prime, and p) will be i if'/)' is true and o if 'p' is false, so that 'the number
of numbers n such that (n is an even prime and p) is i' will be logically equivalent
to '/>'. But this would still be open to the objection that the former statement
implies the existence of the number i whereas 'p' does not.
 EXTENSIONALITY 253
predicate-expressions with the same sense for Tainted' or 'ran
for help', and equally by any substitution of logical equivalents
for the sentences 'Smith fainted* and 'Jones ran for help', since
these will naturally preserve the reference and the sense of each
sentence as a whole—for example, 'Smith either fainted or both
walked and didn't walk* and "Jones didn't not run for help'.8
What is peculiar about the argument we are examining is that
the sense of 'p' is lost at step (4),
This has come about not by any substitution of co-referring
expressions in 'p' itself, nor by any substitution of logical equi-
valents for 'p' itself, but by the substitution of what may be
considered as a logical equivalent of 'p' followed by the sub-
stitution of a co-referring expression within that equivalent. In
constructing the argument, one construes the assumptions about
'F(...)' as meaning not merely that each kind of substitution is
allowed on its own, but that substitutions of one kind are
allowed within the products of substitutions of the other kind.
If the assumptions apply to the context 'F(...)' as such, and
hence to any statement of the form 'F(p}\ then we must so
construe them. But there might be a particular sentence 'p'—
or a whole special class of such sentences—such that when it
was inserted into the context 'F(...)' each kind of substitution
was allowed separately but successive substitutions of the two
kinds were not allowed. A rnodiifkation of the story of Puss in
Boots may illustrate this. The giant might well have said to the
cat, 'Certainly you may eat any mice you find in this castle, and,
to amuse and impress you I shall be happy to change myself
temporarily into an animal of any sort you care to name', but
if he had had his wits about him he would have added, 'But
this does not mean that after I have changed myself into a
mouse to oblige you, you may then eat any mice you find in
this castle*. And we should show a like caution when we realize
that successive substitutions of the two kinds can produce a loss
or change of sense which no substitution of either kind could
produce on its own in such a sentence as ordinarily enters, as a
clause, into causal statements of the sort we are discussing.
We can concede that if the context 'F(...)' is such that substi-
tutions of both kinds in the contained sentence always pre-
serve the truth-value of the whole statement, then it must be
* Cumminii and Gottlieb, op. cit., p. 269.
854 EXTENSIONALITY
truth-functional—allowing, that is, the broad interpretation of
logical equivalence which escapes the criticism of Cummins
and Gottlieb—and yet defend the intuitive claim that a par-
ticular sentence 'F(p)' may be such that replacements either of
'p' itself by logical equivalents or of singular terms in 'p* by
co-referring expressions do not alter the truth-value of F(p)\
but that replacement of 'p' by a sentence of the same truth-value
may alter the truth-value of 'F(p)'. The point is that when 'p'
is replaced by its logical equivalent at step (3), a singular term
'£(x = x &. p)* is introduced in a position different from that
occupied by, say, 'Smith' in our last example (which is typical
of the position occupied by names in ordinary causal sentences,
the ones which intuitively look transparent); and it may well
be that when this new sentence is embedded in the context
'F(...)', a singular term in this new position is not replaceable
salva veritate by a co-referring one.
These considerations draw the claws of the argument, to
which excessive deference has been shown. We can therefore
consider the substantive questions directly, free from the threat
that any transparent context for sentences will have to be
merely truth-functional.
Those who have deferred to the argument have moved in
different directions. Anscombe 'finds it harmless to say that
causal statements are intensionaP, and to say that this applies
not only to those of the form 'p because q\ but also to those of
the form 'A brought it about that^'—whether 'A' designates an
event or a substance—and indeed to all in which a proposition
'p' is embedded in a context to yield a statement lF(p}\ But
her reason for welcoming this conclusion is merely that there
is a way of construing a clause containing a definite description
which makes it important which of a number of co-referring
definite descriptions is used. If we read 'There was a crisis
because the man with the biggest nose in France made a
speech' in such a way as to assert that there was a crisis because
his was the biggest nose, it is false, whereas 'There was a crisis
because the French president made a speech', read in the same
way, is true. But there is also a way of reading both sentences,
giving, Anscombe suggests, the definite description a different
scope, which gives the two sentences the same truth-value so
long as the two descriptions refer to the same man. But this
 EXTENSIONALITY 255
reading, which makes the occurrence of the definite descrip-
tions transparent, escapes the conclusion of the logical argument
by taking these descriptions out of the embedded sentence and
prefixing them to <F(...)' as a whole, presumably allowing the
embedded sentence to be intensional: 'Concerning the man
with the biggest nose; there was a crisis because he. . . .**
We can link this suggestion with what was said above. What
I referred to as the position occupied by 'Smith' in our example
—and normally by names in ordinary causal sentences—is one
such that a definite description there would naturally be read
as having large scope, as being prefixed to 'F(p)' as a whole,
whereas the position occupied by 'x(x — x &/>)' in (3) is such
that a definite description there has small scope.
Quine's versions of the argument have been criticized on
these grounds by Richard Sharvy (following Arthur Smullyan):
'definite descriptions given large scope will behave properly
with respect to substitutivity of identity' but 'there is no
justification for performing such substitutions for or of definite
descriptions which are given smaller scope'.10 Thus 'F(...)'—
which may be, as in our examples, a causal context, or a modal
one, or a belief one, and so on—may well be transparent with
respect to definite descriptions with large scope but not with
respect to definite descriptions with smaller scope (and names
can be treated correspondingly). The moves from (2) to (3)
and from (4) to (5) will be valid only if the descriptions of the
classes x(x = x & p) and x(x — x & q) are taken as having
small scope, as being strictly inside the context *F(,,.y and as
governed by 'F'; but the move from (3) to (4) would be valid
only if these descriptions of classes had large scope. We can
concede that if *F(...)' were such as to allow substitutivity of
identity for all names and definite descriptions it would have
to be truth-functional, as the argument shows: but it could be,
and some causal contexts clearly are, such as to allow substitu-
tions of co-referring terms which function as definite descrip-
tions with large scope only, without being truth-functional.
• Of. Anscombe, op. cit., p, 156.
10
'Truth-Functionality and Referential Opacity', Philosophical Studies, xxi
(1970), 5-9, referring to Arthur Smullyan, 'Modality and Description', Journal of
Symbolic Logic, xiii (1948), 31-7. Similar points are made and applied to other,
but related, Quinean arguments in an unpublished article by C. J. F. Williams,
who brought Sharvy's treatment to my attention.
356 EXTENSIONALITY
Davidson, on the other hand, thinks that the argument's
conclusion is to be evaded by denying that the logical form of
a statement like 'The short circuit caused the fire' is given by
'The fact that there was a short circuit caused it to be the case
that there was a fire', and by saying that its true logical form is
that of a two-place relation whose terms are events, where an
event is something complex and concrete with features beyond
those that we use to describe it. The event which is Flora's
drying of herself is also Flora's drying of herself with a coarse
towel on the beach at noon: 'we must distinguish firmly between
causes and the features we hit on for describing them'.11 It is
a consequence of this treatment that we can be said to have
specified the whole cause of an event even when we have not
wholly specified it. The event which is a certain man's eating
of a particular dish, which, as Mill remarked, we should be apt
to call the cause of his death, is, from this point of view, the
whole cause, since as a concrete occurrence it is an eating of
food containing just whatever poison it did contain by a man
with just such a bodily constitution in just such a state of
health, and so on. The event qua concrete occurrence includes
everything that was relevant to the production of the effect and
much more besides. Yet Davidson stops short of including
all the circumstances in the event: since Caesar's death was
the death of a man with more wounds than Brutus inflicted,
Brutus's stab, even if it was sufficient for Caesar's dying, was
not the whole cause of Caesar's death. Although Brutus's stab-
bing of Caesar was a stabbing of him shortly after the other
conspirators had stabbed him, Davidson does not take the con-
crete event, Brutus's stabbing of Caesar, to include these other
stabbings as parts of itself, but says that it 'was literally (spatio-
temporally) only part of the cause'.12 And this is just as well:
if we included the circumstances in the event we should have
to conclude, in a far more damaging sense than Hume intended,
that anything may cause anything. The unfortunate Cinna the
poet might have been rightly blamed for Caesar's death, since
his writing of some bad verses was, no doubt, a writing-of-bad-
verses - at - a - time - when - not - far - away - the - conspirators - were-
stabbing-Caesar.
" Davidson, op. cit,, p. 697.
" Ibid., p. 699.
 EXTENSIONALITY 257
We can identify events, in this concrete occurrence sense,
simply by their spatio-temporal regions, well enough to allow
one event but not another roughly contemporaneous event to
be the cause of a certain effect. We then have a view of causing
very similar to that of Ducasse;13 a cause leads on to or pro-
duces an effect by being followed contiguously by it, and by
being the only change (or unchange) upon which the effect so
follows. I think we can and sometimes do think and speak of
such concrete occurrences as causes. We can plausibly say, of
such a cause-event, that if in the circumstances it had not
occurred the effect-event would not have occurred either: a
possible world constructed from the actual world by excising
the whole concrete occurrence and then letting things run on
in accordance with the actual laws of working will not contain
the effect. (Presumably we should have to insert some fairly
harmless, neutral occurrence in the place of the one excised: if
we cut out Flora's drying of herself, we may put in her simply
lying on the beach for that period of time.) Such a counter-
factual conditional will be particularly plausible if we know
some law or set of laws of working by which it would be sus-
tained; but even where we do not know such a law, we may
believe that there is some such law.14
We can, then, and sometimes do, take causes (and effects)
as concrete occurrences and causing as a two-place relation be-
tween such events. But it is far from clear that this is the best
treatment, and it is certainly not the only possible one. Even
if we deferred, unnecessarily, to the logical argument discussed
above, we could keep 'The fact that. . . caused it to be the case
that . . . " as an operator on propositions or sentences in
Anscombe's way, by taking these contexts as intensional and
by interpreting the transparent occurrences of definite descrip-
tions within them as having large scope, that is, as being pre-
fixed to the whole sentence rather than being governed by this
" Cf. Chapter 5.
14
Davidson (op. cit., p. 702) says that "Ducasse is right that singular causal
statements entail no law; Hume is right that they entail there is a law'. I have
argued against the latter claim in Chapter 3, and the arguments of Chapter 2
would show that even if a law were so entailed it would not need to be one of
Hume's form, making the cause an event of some type that is sufficient for events
of the same type as the effect. We may, but need not, believe that there is a law
which makes the cause necessary in the circumstances for the effect: the other half
of the regularity is still more dispensable.
858 EXTENSIONALITY
causal operator. It seems a disadvantage of concrete occurrence
causes that they will nearly always include irrelevant com-
ponents. In Mill's example, this man's eating of this dish was
also, perhaps, an eating of something with a silver fork by a
man with a red moustache, and neither of these features may
have had any bearing at all on his subsequent death. And we
saw, in Chapter 5, how Ducasse was forced to include the air
waves from the song of a canary in the cause of a particular
breaking of a particular window pane. We are, sometimes at
least, interested in a more selective, more discriminating rela-
tion than can hold between concrete occurrences. I spoke in
Chapter 3 of the progressive localization of a cause. This is an
important kind of advance in knowledge, but it would be
severely restricted if only concrete occurrences were recognized
as causes. Davidson himself gives examples of sentences in which
particular features of events, or facts rather than events, are
mentioned as causes: 'The slowness with which controls were
applied caused the rapidity with which the inflation developed';
'The collapse was caused, not by the fact that the bolt gave way,
but by the fact that it gave way so suddenly and unexpectedly.'
Davidson suggests that these are 'rudimentary causal explana-
tions', that 'Explanations typically relate statements, not events',
and that the 'caused' of these sentences is different from that of
'straightforward singular causal statements' and 'is best ex-
pressed by the words "causally explains" ',15
We need not dispute about which sort of statement is the
more straightforward: what matters is that there is room, and
need, for statements of both kinds. And if the logical argument
had any force, it would clearly raise a difficulty for the context
'The fact that. . . causally explains the fact that. . .', since it is
a context for sentences, it is obviously not truth-functional, and
yet singular terms (for example, 'the bolt' in Davidson's sen-
tence above) can occur transparently in such an explanation
sentence.
Zeno Vendler argues from linguistic evidence that it is facts,
rather than events, that count as causes. His thesis is that if
noun phrases denoting events are represented by 'e' (with sub-
scripts if necessary) and those denoting facts by'/', our sentences
have the three distinct forms:
« Op. cit., pp. 702-3.
 EXTENSIONALITY 359
el is the effect of ez.
fi is the result of/2.
jfis the cause of e.
In other words, effects and what they are effects of are events
(that is, what I have been calling concrete occurrences), results
and what they are the results of are facts, but a fact is said to
be the cause of an event.
Vendler's linguistic argument centres upon the distinction
between perfect nominals and imperfect norninals: 'the noun
phrases typically associated with the "fact"-group turn out to
be imperfectly nominalized sentences, whereas those appro-
priate to the "event"-groupare exclusively perfectly nominalized
sentences'. The difference is that 'In the imperfect nominal the
verb keeps some of its verb-like features: it retains the verb-
object intact; tenses, modals and adverbs may be present; and
the whole structure is subject to negation', whereas 'The verb
in the perfect nominal sheds these verb-like features and behaves
like a noun: it may take relative clauses, adjectives, articles, and
prepositions'.16 Thus 'that he sang the song' and 'his having sung
the song' are imperfect nominals, 'his singing of the song' and
'the beautiful singing of the song we heard' are perfect nominals.
To me, one part of Vendler's thesis, namely that imperfect
nominals can stand very naturally as subjects of 'caused' and
as occupants of the second gap in 'The cause o f . . . was . . .'
is entirely convincing. But I cannot accept the other part, that
perfect nominals cannot play these roles. We can accept 'Caesar's
having crossed the Rubicon caused the war' and 'The cause of
the Roman civil war was Caesar's having crossed the Rubicon';
but can we not accept also 'Caesar's crossing of the Rubicon
caused the war' and 'The cause of the war was Caesar's crossing
of the Rubicon' ? However, this is a purely linguistic issue, with
which I am not much concerned. Even if Vendler were right,
we could say 'Caesar's crossing of the Rubicon had-as-one-of-
its-effects the Roman civil war' or we could coin a term, per-
haps borrowing Mill's 'producing cause', for an event of which
some other event is the effect, and say 'The producing cause
of the war was Caesar's crossing of the Rubicon'. However the
point is to be linguistically accommodated, we want both
events and facts as antecedents in (different) causal relations.
16
Vendler, op. eit, p. 707.
a6o EXTENSIONALITY
But there is a further issue. According to Vendler, facts are
referentially transparent. Oedipus's having married Jocasta is
the same fact as his having married his mother, and if the latter
caused the tragedy, so did the former. Yet the latter description
helps to explain the tragedy in a way that the former does not.
What is referentially transparent may be, for that very reason,
explanatorily opaque. But, as Davidson suggests, it is particu-
larly when we are concerned with explanation that we take
facts rather than events as causes. Vendler's transparent facts
may be what the linguistic evidence picks out as the most
favoured category for causes, and may none the less be only
half way to a thoroughly explanatory sort of cause.
Let us work with the notion of a causally relevant feature. That
Brutus's dagger had a sharp point was, presumably, causally
relevant to Caesar's death, that it had an ornamented hilt pre-
sumably was not. That the bolt in Davidson's example gave
way suddenly was causally relevant to the collapse of the
bridge, that it gave way noisily was not. We can then form the
concept of a minimally complete causal account, one which mentions
all the features that were actually causally relevant to some
event. This is, of course, an ideal: we do not in general expect
to be able to give such an account. If determinism is not pre-
supposed, we should not assume that the complete set of rele-
vant features was in the strong, counterfactual, sense sufficient
for the event, but if the account is causal rather than probabilistic
we can assume that each relevant feature was necessary in the
circumstances for that event. Now these relevant features will
all be general; what is vital is not this bolt's giving way but the
giving way of a bolt at this place; what mattered about the
assassination at Sarajevo was not that it was committed by
Prinzip but that it was committed by a Bosnian who had
received arms and instruction in Serbia.
The same drive towards explanation, however, will lead us
to take also what is caused not as an event but as a fact, as
what Vendler calls a result: war's breaking out in August 1914
rather than the outbreak of war, or Caesar's dying rather than
Caesar's death. For it will be such facts only that can be ex-
plained rather than a concrete occurrence in all its infinite
detail. But again we cannot stop at a referentially transparent
fact. Oedipus's having certain obvious and conscious motives
 EXTENSIONALITY a6i
causally explains his marrying the one woman who had a cer-
tain combination of physical and social attractions; to explain
his marrying his mother we must either add the fact that his
mother was the unique possessor of these attractions in the
relevant spatio-temporal region, or substitute, in Freud's man-
ner, some unconscious motives for the conscious ones.
But in a minimally complete causal account, not everything
dissolves into generality. It is Caesar's being stabbed in such and
such a way, conjoined with his having just such a physical
make-up, that caused, or causally explains, Caesar's dying. It is
such and such features of one or more things of such and such
kinds being together at this place and time that causally explains
the fact that an event of such and such a sort occurred near this
place at a time slightly later than this. It is still a singular causal
account that is being given, and at least one singular term will
be included in the description alike of the cause and of the
result—the same term in both—holding together what are
otherwise general, existentially quantified statements. With
respect to this term, then, the whole account will be referentially
transparent, but this term can be considered either as a pure
proper name or as a definite description with large scope, pre-
fixed to the account as a whole: 'Concerning Caesar,'—or
'Concerning the bald Roman dictator,'—'his being stabbed '
Everything causally relevant will have been mentioned else-
where. This singular term, which as it were stands outside the
clauses which state the cause and the result and enters into
them (in the shape of pronouns that refer to it) only to connect
them with one another, is a much purified version of the
causal field introduced in Chapter 2.
The typical form of a minimally complete causal account will
then be:
Vs being B and there being an x such that x is C and x has
relation R to a and . . . caused there to be a z such that z
is D and z has relation S to a.'
What is here said to cause something may be called an
explanatory cause. An explanatory cause is a fact, but of an un-
usually pure sort. Everything causally relevant about it is
expressed by general terms and existential quantifications; the
individual that this fact is about, the only item with respect to
a6a EXTENSIONALITY
whose description the expression that states this fact is referen-
tially transparent, is causally irrelevant, everything that is
causally relevant about this individual having been covered
by explicit monadic, dyadic, or polyadic predicates. And the
result of which this is the cause is a fact of an equally pure
sort.
A minimally complete causal account is explanatory in at
least two ways. The conjunction of features in the cause will
be related to that in the result by some regularity which makes
the former necessary and perhaps also sufficient for the latter.
Also, since the first conjunction includes all the features that
are causally relevant to the latter, this regularity will be what
we called in Chapter 8 a pure law of working, and as such will
be at least in part a pattern of persistence, that is, there will
be some qualitative or structural continuity uniting the former
set of features to the latter.
Pushed to this extreme, facts as causes seem to have every
advantage over events. Why, then, do we bother to recognize
producing causes as well as explanatory causes? The reason
lies, as so often, in the extent of our ignorance. A minimally
complete account is an ideal which we can hardly ever reach.
But we can use this concept to elucidate some of the peculiarities
of the kinds of causal statements we actually make.
Let us start with the following approximation to a minimally
complete account: 'There being an explosion here just after
midnight was caused by'—or, rather, according to Vendler,
'was a result of—'someone's pressing at midnight a button
which was connected to a pair of contacts which were connected
by electrical conductors to a charged battery and a detonator
which was in contact with a quantity of TNT which was here5.
Apart from 'here' and 'at midnight' all the components of the
explanatory cause are relevant general features. If we change
'a button' to 'the button', yielding a definite description, it
makes little difference, though no doubt the uniqueness of the
button satisfying the general description was not causally rele-
vant. But other changes are more serious.
Suppose we replace 'someone' by 'Tom' and 'a button' by
'the button on the red box'. We have now added, let us suppose,
true but irrelevant bits of description, and if we still read it as
purporting to be a minimally complete account, and hence as
 EXTENSIONALITY 263
asserting that it was causally relevant that Tom pressed the
button and that it was (the only one) on the red box, the state-
ment must be counted false. But we can read these items as
giving some gratuitous but true and harmless further informa-
tion, that is, in Anscombe's way as 'Concerning Tom and con-
cerning the red box, there being an explosion . . . was a result
of his pressing . . . a button on it which . . .'; and this will still
be true. Again we may leave out part of the explanatory
cause: 'There being an explosion here just after midnight was
a result of someone's pressing a button.' This is still true if it
claims that some button's being pressed was causally relevant,
though not if it claims that this was the whole explanatory
cause, that the pressing of a button as such brought about the
explosion.
Alternatively, we may replace part of the explanatory cause
by what we previously introduced as an irrelevant definite
description: 'There being an explosion here just after midnight
was a result of someone's pressing at midnight the button on
the red box.' Now the explosion was not a result of the button's
being on the red box, but it was a result of features which the
button on the red box uniquely possessed; the phrase 'the (one)
on the red box' is a kind of substitute for the unstated relevant
features. Taken in this way, the statement remains true, and
although it is hardly more relevantly informative than the last
variant it is more helpful in indicating where further relevant
information is to be sought. We should still construe it as start-
ing 'Concerning the button on the red box, there being an
explosion here just after midnight was a result of someone's
pressing it at midnight' but as going on 'and of its (uniquely)
possessing certain other qualities and relations'. Moreover,
someone might use this statement not perversely, suppressing
parts of a known minimally complete account, but because it
expressed all he knew. Observing the red box, the button on it,
someone's pressing it and, immediately afterwards, the explo-
sion, and employing the method of difference either explicitly
or implicitly, he might well arrive at just that mixture of
knowledge and ignorance which this statement as we have con-
strued it would express. Since this is how the phrase 'the button
on the red box' is functioning, its occurrence is referentially
transparent: any other unique description of this object would
364 EXTENSIONALITY
do the same job equally well, and we can explain this logically
by saying that this description is prefixed to the whole state-
ment, not governed by the intensional operator 'was a result
of. We might then be tempted to argue that in the original
minimally complete account with 'a button' changed to 'the
button', the whole phrase 'the button which was connected . . .
which was here' also occurs transparently, since it can be re-
placed salva veritate by any other correctly identifying phrase,
such as 'the button on the red box'. But this would be wrong.
Truth is preserved through this replacement only by a thorough
reconstrual of the sentence, by moving the description from the
place where it is governed by 'was a result of to a place where
it is not. The sentences which, as Vendler says, present referenti-
ally transparent facts as causes, or as what have results, are
typically of the form illustrated by 'There being an explosion
here just after midnight was a result of someone's pressing at
midnight the button on the red box'; but an in principle
superior, but often unattainable, account has the minimally
complete form which is intensional throughout.
Finally, we may say 'The explosion here was caused by
Tom's midnight movement'. Tom's midnight movement is an
event, not a fact, and what we now have is a producing, not
an explanatory, cause. All the causally relevant features are
now concealed under a blanket phrase. But again it is easy to
tell a story in which someone's actual knowledge of the incident
would be accurately expressed by this sentence: the observer
sees Tom enter a darkened room, hears him moving about
and immediately afterwards hears, and sees, the explosion. Of
course, an event description may include features that were
causally relevant to the effect, but a statement which says that
one event was the effect of another, or was caused by another,
makes no claim that the features used to identify the cause-
event were causally relevant. There is nothing false or even
misleading about 'The international crisis was an effect of the
speech made by the man with the biggest nose in France'.
Ignorance is the main reason for taking referentially trans-
parent facts as causes rather than giving minimally complete
accounts with only existentially quantified expressions to intro-
duce causes, but of course the advantages of brevity or of say-
ing just as much as is needed for the purpose in hand often
 EXTENSIONALITY 965
justifies our saying less than we know. Similar reasons will
often explain our taking concrete events as causes rather than
facts. We may know that a certain event caused another with-
out knowing what features of the former event were relevant,
and therefore without being able to specify any facts as causes.
And since less knowledge is needed to pick out an event as a
cause, the knowledge that is needed is more easily acquired.
One event's causing another is observable whereas the identi-
fication of a fact as a cause requires some theory, some assump-
tions, or some comparison of cases with one another. As we
saw in Chapter 5, Ducasse identifies causing with a relation
between concrete events with spatio-temporal boundaries, and
thereby makes causing observable. It is one concrete event that
in the most obvious sense leads on to or produces another.
But it is general features that enter into regularities and con-
tinuities.
We need then, to recognize both kinds of cause, producing
causes and explanatory causes, events and facts, and at the
same time to distinguish them, in order to understand what we
think and say about causal relations. Two problems touched
upon earlier may be further clarified with the help of this dis-
tinction. One is that of quantitative over-determination: the
hammer-blow that flattened the chestnut was harder than it
needed to be to produce this result, and this seemed to make
questionable the judgement that this blow was necessary in the
circumstances for the flattening.17 We can now say that the
concrete event, this hammer-blow, was the producing cause
of the chestnut's change of shape, it was what actually led on to
this change; but the explanatory cause was the fact that there
was here a blow of at least such-and-such a momentum. Each
can be said, in a slightly different sense, to have been necessary
and sufficient in the circumstances for the result; our earlier
difficulty arose from the fact that the concrete event was not
necessary in the sense in which the fact was; the event could
have been a bit different without altering the result, whereas
the fact as stated could not. But this is only a further refinement
of the contrasts drawn earlier between plugging in or leaving
17
Cf. Chapters 2 and 6. The corresponding distinction, between facts and events
as results or effects, has already been used in Chapter 2 (pp. 46-7) to clear up a
similar problem about alternative over-determination.
266 EXTENSIONALITY

out, from a possible world, the hammer-blow as a whole and

altering its momentum, and between neolithic and functional
concepts of causation.
The other problem is the legal difficulty in the case of the
uncertificated officer, Sinon.18 Sinon's navigation of his ship
on the date in question was part of the cause of the collision;
that is, this bit of navigating as a concrete event (or perhaps
performance) was a partial producing cause of the collision.
Among the features of this concrete event were that it was
navigation by an uncertificated officer, that it was none the less
navigation by an experienced and in general competent officer,
and that it was, however, negligent. The very reasonable find-
ing which the judge had some difficulty in formulating can be
put in our terms as follows: the fact that the navigation was
negligent was part of the explanatory cause of the collision, but
the fact that it was by an uncertificated officer was not; whereas
if the officer who navigated negligently had been incompetent
as well as uncertificated, then the fact that he was uncerti-
ficated would have been (indirectly, by way of the fact that he
was incompetent) a part of the explanatory cause. Speaking in
terms of Vendler's referentially transparent facts, we can say
that the collision was a result of Sinon's navigating negligently,
and also, therefore, of an uncertificated officer's navigating
negligently, but we must remember that the phrase 'an un-
certificated officer' occurs here transparently, that is, it is
really prefixed to the whole account and not governed by the
intensional operator 'was a result of.
The distinction between these two kinds of cause is, then,
clear and useful. But there are still very close relations between
them. That Sinon navigated negligently is a fact; for that very
reason negligence in navigation was a feature of the concrete
event or performance, Sinon's navigation on that date; some-
one's navigating negligently is a type of event, and Sinon's
performance was an instance of that type just because the fact
mentioned is a fact. These close systematic connections between
events, their features, event-types, and facts make it legitimate
to move freely and without excessive caution between them in
the way that I have done in earlier discussions, in particular
in Chapter 3. For some purposes we need to attend to these
11
Chapter 5, quoted from Hart and Honor£, Causation in the Law, pp. 112-13.
 EXTENSIONALITY 367
distinctions, but for most of our investigation of causal regu-
larities and our ways of discovering them we do not. The
presence on some occasion of one of the 'factors' or 'conditions'
discussed in Chapter 3 would be a fact: but it would also be a
concrete event's having a certain feature and being an instance
of a certain type of event.
In the light of this distinction, we may also take up again
the question raised in Chapter 3 whether causal statements are
implicitly general. If a singular causal statement about events
were implicitly general, its analysis would include the claim
that the producing cause and the effect could be described in
such a way that the existence of one was related to that of the
other by some causal law.19 But such a statement does not entail
this, but only that the one event led on to the other and that
if the one had not occurred, in the circumstances, the other
would not, and while this counterfactual might be backed by
laws, we saw in Chapters 2 and 3 that it does not need to be
backed even by the claim that there is a law that would back
it. But what about a singular explanatory cause statement? It
is explicitly general in the sense that it makes essential reference
to certain general features, but does it implicitly claim that
there are laws connecting the general features mentioned on
the cause side with those on the result side ? Does 'The fact that
Sinon navigated negligently caused his ship to collide with
another' entail that there is some law in which 'x navigates
negligently' and Vs ship collides with another ship' are con-
stituent elements? I have little doubt that there is such a law:
there are circumstances the conjunction of which with negligent
navigation will always lead to a collision, but the conjunction
of which with careful and skilful navigation would lead to no
" Cf. Davidson, op. cit., p. 699.1 have omitted his suggestion that the law would
make cause and effect both necessary and sufficient for one another, which could
not be defended for concrete events even if we accepted thoroughly deterministic
laws. Though a cause-event may be necessary for its effect, it is so only in the sense
that a possible world from which it was excised would not contain the effect-
event; all that a law could demonstrate as being necessary for the effect is that an
event of a certain sort should have occurred, not that this one in all its concrete
detail should have done so. And the same applies to sufficiency: this concrete
cause-event may be shown by laws to be sufficient for the existence of an effect
with such and such features, but, even if strict determinism holds no isolable
event, but only the previous condition of the whole neighbourhood, would be
sufficient for the effect in all its concrete detail.
a68 EXTENSIONALITY
collision: that is, there is a law which makes negligent naviga-
tion an inus condition of collisions. Such a law, if known,
could be used to back up the explanatory cause statement. But
I do not think that that statement entails that there is such a
law. All it entails is that the causal feature mentioned was
necessary in the circumstances for the result feature mentioned,
and this claim is to be analysed into the appropriate counter-
factual conditional along with an assertion of causal priority.
Finally, we may return to the question raised earlier whether
causal statements are extensional in the sense of allowing the
substitution of coextensive predicates. Statements about pro-
ducing causes will be extensional, since in them predicates are
used only to identify concrete occurrences, and if one of a pair
of coextensive predicates helps to do this, the other will serve
equally well. But this is not true of explanatory cause state-
ments. On a pure regularity theory it would be true: coextensive
predicates are interchangeable salva veritate in statements of
regularity. But the sorts of structural and qualitative continuity
suggested in Chapter 8 could link one, but not the other, of
a pair of coextensive features with some result. Even if the
features A and B were coextensive, A's being regularly followed
by C might be a pure law of working, but B's being followed by
C a mixed law—if B's being coextensive with A were a fact of
collocation or partly derivative from collocations. And even if
the coextensiveness of A and B were the result of pure laws of
working, each being produced by some common cause D in
accordance with such laws, we could have A's being followed
by C as a pure law, and while B's being followed by C would
then be a derivative pure law, the direction of causation
(between B and D), as the diagram shows, would be such as
to prevent B's causing C in virtue of the relationships shown

Diagram (vii)
 EXTENSIONALITY 269
(though of course there could also be an additional relationship
that would need an arrow from B to C to represent it).
The considerations we have dealt with under the headings of
causal priority and the necessity of causes, when added to a
regularity theory, may prevent coextensive features from being
interchangeable in causal laws. I have just argued that explana-
tory cause statements do not entail regularities or that there
are regularities; still less do they reduce to them. Our use of
such statements involves some employment of the notions of
causal priority and the necessity of causes which I 'have tried
to clarify in Chapters 7 and 8, Explanatory cause statements,
therefore, are not extensional In the way that pure regularity
statements would be.
 II
Teleology

ARE there final causes as well as efficient causes? Or can

teleology be somehow reduced to efficient (or 'physical* or
'mechanical' or 'Humean') causation? When this question is
discussed, it is commonly assumed that the efficient causation
with which teleology may be contrasted or to which it may be
reduced is essentially a matter of laws or regularities, and that
these laws make causes sufficient for their effects, that these
causes are necessitating causes. Little account, if any, is taken
of what I have called the direction of causation or of what I
have discussed in Chapter 8 under the heading of the necessity
of causes. If I have been right in earlier chapters in rejecting
or qualifying this cluster of assumptions, it is very likely that
an inadequate concept of efficient causation will have produced
false contrasts and incorrect solutions of the problem of relating
final to efficient causes. I have already argued in Chapter 5
that Hart and Honore* concede too much to Mill with regard
to physical causation, and as a result take as distinctively
characteristic of interpersonal causation features which are in
fact common to it and physical causation. Much the same may
have happened in the discussion of final causes.
Let me then sum up the main points that I have suggested
and have tried, with varying degrees of conclusiveness, to
establish about efficient causation. First, the conceptual and
epistemological priority of singular causal statements to gen-
eralizations (Chapters 1,2). Secondly, an initial analysis of the
meaning of those singular statements in terms of conditionals,
especially those which claim that a cause (whether event or
fact) is necessary in the circumstances for its effect or result,
with a further analysis of those conditional statements which
shows that they may, or may not, be backed by belief in
 TELEOLOGY 271
regularities (Chapter a). Thirdly, an account of the form of
causal regularities, typically highly complex and very incom-
pletely known, with a demonstration that this form is closely
related to the standard basic methods of observation and
reasoning through which we discover and can up to a point
establish such regularities, these observations in turn being
similar to those which give rise to even the most elementary
and least sophisticated counterfactual conditional thinking
involved in causal judgements (Chapter 3). Fourthly, a demon-
stration that laws of functional dependence introduce a causal
concept which is in some ways new, but which has such close
connections with and formal similarities to the 'neolithic' con-
cept that the former is best regarded as a development and
refinement of the latter rather than a rival or complete replace-
ment for it; and with this a plea for two other extensions of the
causal concept to cover successive phases of persisting things
and self-maintaining processes (Chapter 6). Fifthly, I argued
that we have a concept of the direction of causation as some-
thing intrinsic to causal processes, which, though discovered
initially in our active and apparently free interventions, is not
incurably anthropocentric, nor is it simply derived from the
asymmetry of time; I attempted to define this causal priority
in terms of a contrast between fixity and unfixity, and to relate
it both to the dispersal of order and to the direction of explana-
tion (Chapter 7). Sixthly, in considering the necessity of causes,
I accepted Hume's argument that there can be no logically
necessary connection between causes and effects either as they
are known or as they are in the objects, and that there is no
necessity2 j which would support a deductively valid a priori
inference from cause to effect or vice versa; I argued that the
apparent power of causal laws to sustain counterfactual condi-
tionals was to be explained without assuming that such laws
differ in form or content from accidental generalizations; but
I tried to show that an important distinction can be drawn
between pure laws of working, collocations, and mixtures of the
two, and speculated that processes in accordance with pure laws
of working would display certain kinds of continuity or per-
sistence (Chapter 8). Seventhly, although I rejected Kantian
and transcendental arguments not only for the principle that
all alterations take place in accordance with rules but also for
272 TELEOLOGY
the more modest claim that there must be some regularity and
some persistence, I noted that instantiations of the latter pre-
sumption are required for our ordinary basic methods of detect-
ing causal relationships, and that they could be defended as an
avoidance of arbitrarily selective scepticism and perhaps also
more directly (Chapters 4 and 8). Eighthly, I showed how
statistical laws of working might fit into this general theory
(Chapter 9). And ninthly I argued that this theory left room
for the explanation of common-sense and legal restrictions on
the use of causal terms, and that puzzles about both extension-
ality and the observability of causation could be cleared up by
recognizing that both events and facts can be causes, in slightly
different but closely related senses (Chapters 5, 8, and 10).
It is against the background of such a richer concept of
efficient causation that questions about the distinctness and
irreducibility of final causation should be considered. A rough
general definition of teleology or final causation—I shall take
these two terms to be synonymous—is that an account or
explanation is teleological if it makes some essential use of the
notion that something—perhaps an event, or a state, or a fact—
is an end or goal to which something else is, or is seen as, a
means. But exactly what it would be for something to be an
end or a goal, or again a means, remains to be clarified, and
different teleological concepts result when these terms are
given various more exact meanings.
One elementary case is where we see something (a type or
an individual) as necessary for something else. The repeated
contractions of the heart are necessary for the continued circu-
lation of the blood in this animal body or in some large class
of animal bodies; the defeat of the Persians at Salamis was
(perhaps) necessary for the flowering of Athenian culture in the
age of Pericles. But this relationship, where one thing is neces-
sary (in the circumstances) for another is, as we have seen, the
very one which is central in our ordinary concept of efficient
causation: once we have seen that efficient causes need not be
sufficient (even in the circumstances) for their effects or results
we must admit that we have as yet no distinction at all between
final and efficient causes. But there is a difference if we see the
result as good, from our own point of view or from some point
of view which we for the moment endorse or play along with.
 TELEOLOGY 273
If we regard the art and literature of fifth-century Athens as a
good thing, then we may see the Greek victory at Salamis,
because it was necessary for this good thing, as fitting satis-
factorily into a certain pattern, and in a much-extended sense
of that elastic word 'explain', we may feel that its place in that
pattern explains that victory. But of course this does not in
itself (without, say, the postulation of some divine or immanent
historical purpose) explain the occurrence of that victory, it does
not explain why the Greeks got the better of the Persians.
Similarly, once we see the circulation of the blood as good for
the individual animal, we see the fact that repeated contrac-
tions of the heart are necessary in the circumstances for that
circulation as giving a certain point to those contractions: we
in a sense explain them just by showing that they have this
'function'; but in this case it is possible to add two further
sorts of explanation: if (in a certain large class of animals) the
circulation of the blood is in turn necessary for continued life,
the fact that heart-contractions are necessary for circulation
helps to explain why all living animals (of that class) have
repeatedly contracting hearts; and it may also (in a way that
I shall examine later) figure in an evolutionary explanation of
how it comes about that all such animals have hearts which
contract regularly. More simply, we may just (as von Wright
puts it)l take the occurrence of the result for granted, and in
relation to this we shall see something's being necessary for that
result as explaining that something. But this relation itself, the
antecedent's being necessary in the circumstances for the result
(whether in an individual case or generally) is, for all we have
so far found, just that of efficient causation: all that makes it
distinctively ideological is either a subjective or a relative point
of view—we take the result for granted, or see it as good, or
note that it is good for the animal in question—or the incor-
poration of this relation into some wider account—purposive or
evolutionary or referring simply to conditions for individual
survival—which explains either why the antecedent came about
or came into existence or simply why it does exist in all members
1
Explanation and Understanding, p. 58. von Wright calls such examples as that of
the heart quasi-teleological, but examples like our Salamis one quasi-causal, because
he thinks that there are distinctively ideological steps in the chain that connects,
say, Salamis with the Parthenon.
374 TELEOLOGY
of a certain class. It is in such further accounts, then, that
distinctively teleological features must be sought, not in the
mere relation of something's being necessary for a certain result.
On the other hand, we must reject the view of, for example,
Moritz Schlick that there is, necessarily and trivially, no real
difference between final and efficient causes, that 'it makes no
difference whether we say "the past determines the future" or
"the future determines the past" '.a In Schlick's thought, this
is simply a consequence of a pure functional dependence view
of causation, that is, a refined pure regularity view, and more
generally of an extreme positivism. Given his assumptions,
there is no room for the sort of causal priority I have tried to
elucidate in Chapter 7, or indeed for anything except one or
other sort of regularity or functional dependence and inferences
and calculations based on these, which of course may move in
either direction. But if I am right there is a distinctive order in
which events come to be fixed, and an explanation by way of
efficient causes retraces this path of fixing, and (as we saw) this
is why we can explain the co-occurrence of collateral effects by
reference to their common cause, but not the co-occurrence of
joint or over-determining causes by reference to their common
effect. In ordinary simple cases it would be simply wrong, not
a trivially equivalent alternative, to explain the occurrence of a
cause-event (or cause-fact) by reference to its effect or result.
But of course if there were backward causation, then indeed
the later event would explain the earlier one(s), and (as we
saw with Scriven's precognition experiment) would explain
the co-occurrence of a set of earlier events. It would be through
such a later event that the earlier ones became fixed. Now if
there were such a case, we could call the later event an end or
goal, simply in virtue of its relative temporal position, rather
than (as we have been doing) a time-reversed cause. And then
this would be in a very clear sense a case of teleology or final
causation: the 'end* would indeed be responsible for the coming
about of the 'means', and any adequate explanation of these
earlier-occurring items would have to refer to this end. Of
course, it would be a trivial, merely verbal, choice whether we
called this final causation or time-reversed efficient causation;
* 'Causality in Everyday Life and in Recent Science', in Fcigl and Sellars,
Readings in Philosophical Analysis, pp. 515-33, especially p. 527.
 TELEOLOGY 275
but it would be radically and intrinsically different from
ordinary, forward, efficient causation, and it might well be
described as the purest and simplest kind of teleology. The only
trouble is that such cases seem never to occur. This possibility
has, nevertheless, some importance in that opponents of teleo-
logy are inclined to assume that this would be the only sort of
teleology that was really worthy of the name, and to conclude
that objections to it are fatal to teleology in general. Thus
Charles Taylor quotes Hull as saying that 'In its extreme form
teleology is the name of the belief that the terminal stage of certain
environmental-organismic cycles somehow is at the same time
one of the antecedent determining conditions which bring the
behaviour cycle about.'3 Defenders of teleology, however, do
not adopt this 'extreme form'. It would be characteristic of this
extreme form that the end or goal must actually occur or be
realized; it must, in our terminology, be fixed first in order to
be able both to fix and to explain the earlier items: only an
already concrete remoter future could drag the immediate
future into existence. But it is not in this way that ends or goals
figure in plausible ideological accounts: in them a goal can play
its part even if it fails to be realized at all in the individual case,
and a fortiori even if, though it is realized, its occurrence is not
fixed in advance.
There are several distinguishable kinds of example to be
considered. First, there is conscious purposive human action;
it is obviously from this that we first acquire the notions of end
or goal, of a means to an end, and so on. Secondly, there is the
goal-seeking action or behaviour of the higher but non-verbal
animals, which are unable to report to us any conscious pur-
poses that they may have, but whose behaviour itself invites
purposive description; a stock example is a carnivore stalking
and catching its prey. Thirdly, there is the apparent teleology
of goal-seeking and feedback mechanisms, homing rockets,
thermostats, and the like. Fourthly, there is the functional
appropriateness of organs and their working seen in the light
of actual or possible evolutionary explanations. I shall discuss
all of these, but in what may seem to be a less natural order,
taking first the purposive behaviour of non-verbal animals.
J
Taylor, The Explanation of Behaviour, p. 16, quoting G. L, Hull, Principles of
Behaviour, p. 26.
a?6 TELEOLOGY
The mere fact that such animals are unable to tell us that
they have purposes does not, indeed, show that they do not
have them, and it is possible that their thinking and acting are
closely analogous to those of human beings: if so, they will be
covered by the analysis of conscious purposive human behaviour
which is still to be given. But for the present I shall ignore this
possibility, and consider only what we can say on the basis of
rather unimaginative observation of animal behaviour.
The stock example4 is of a carnivorous animal in a certain
state—hungry, with eyes, sense of smell, muscles, and so on in
good working order—all of which I sum up as 'this A'; an.
environment E which includes an appropriate victim; a
sequence of behaviour /?—looking, smelling, stalking, leaping,
and so on—and a goal G—the predator's catching and eating
its prey. This J's performing B in E can be said in some sense
to lead to G; we need not specify whether it is necessary for 6?
or sufficient for G or both or whether it merely favours G, con-
siderably increases the likelihood of G. The description and
explanation that we naturally give is that this A does B in E
because B (in E) leads to G; and this is a teleological account,
it refers essentially to the goal G. Now that B (in whatever
precise sense) leads to G is a fact of efficient causation: this kind
of teleology at any rate presupposes and involves efficient
causation. But the controversial question is whether the teleo-
logical account is somehow equivalent or reducible to an effi-
cient causal one. If we construe the 'because' as a reference to a
law, we are led to formulate a teleological law, 'Whenever an
animal of sort A is in a situation such that B would lead to G,
it does /?, but when it is in a situation such that B would not
lead to G, it does not do B\ It can then be argued5 that all
(and only) situations which satisfy the teleological description
of being such that B would lead to G must also satisfy some
intrinsic description, say 7', and hence that the teleological law
can be replaced by the law 'Whenever an animal of sort A is
in a situation of kind / it does /?, but when it is in a situation
of a kind other than /it does not do B\ This clearly has (if we
add, as we can, the right direction of causation) the form of
4
Cf. Taylor, op. cit., pp. 9-17.
* Gf. D. Noble, 'Charles Taylor on Teleological Explanation', Analysis, xxvii
(1966-7), 96-103.
 TELEOLOGY 277
a law of efficient causation, and it could be used to back up an
ordinary causal explanation about an individual case, 'This A
did B in E because E was of kind /'. It must be conceded that
there will always be some intrinsic description '/' which is
coextensive with the teleological description 'such that B would
lead to G'.6 But I have argued (at the end of Chapter 10) that
coextensive predicates need not be interchangeable in explana-
tory cause statements as they are in pure regularity statements.
It would therefore in principle be possible that this A should
have done B in E because B (in E} would lead to G but not
because E was of kind /, although it was, and although all and
only situations in which B would lead to G are, of kind /. But
how can this formal possibility be realized? We have deliber-
ately excluded any appeal to the notion that the animal sees E
as being such that B would lead to G: this would be relevant to
an account in terms of conscious purpose, which we are for the
moment leaving aside. Again, there might be an evolutionary
story to be told, that this A has an instinctive tendency to do
B in situations of kind /because those members of some ancestral
species which happened to have a stronger tendency than
others to do B in /-situations achieved G more often than those
others, survived when those others did not, and passed on this
tendency to their descendants including this A. In this story
mention of B's leading to G cannot be eliminated in favour of
the coextensive predicate-term '/'„ But this is a further explana-
tion, it involves another species of teleology to which we have
still to turn: is there anything irreducibly teleological in the
behaviour of this A? Well, there might have been some process
of 'reinforcement' in the individual history of this A; it might
have acquired a stronger tendency to do B in /-situations as a
result of often getting food when it did so but remaining
hungry when it did not. This explanation too refers essentially
to B's having led to G: it is this feature of/-situations, not their
being / as such, which has produced the reinforcement. But
this too is a further explanation and we can still ask: Is there
anything irreducibly teleological in the behaviour of this A as
it is now ? If we exclude conscious purpose, further explanations
in terms of evolution and reinforcement, and backward causa-
tion, there seems to be no way in which the fact that E is such
6
Cf. Taylor, 'Teleological Explanation—a Reply to Denis Noble', ibid., pp. 141 -3.
278 TELEOLOGY

that B would lead to G, rather than the fact that E is of kind /,

could be causally relevant. This A (as it is now, including its
instincts and its acquired or developed characteristics) must
surely be reacting to some occurrent intrinsic features of E, it
must be instantiating some law (perhaps only an approximate
regularity) that As do B in situations of kind /, where their
being / is the causally relevant feature. But, as Taylor shows,7
the teleologist has one further resource. This A displays appro-
priately different behaviour in different situations. Where E is
such that BI would lead to G, it does B^ where E is such that
B% would lead to G, it does £2, and so on. And though each
performance on its own could be explained by a law that As
do Bl in situations of kind Ilt or that A do J92 in situations of
kind /2, and so on, we find no common feature of all the
In-Bn correlations (n — 1 , 2 , . . . ) until we note that Bn is what
in In leads to G. Doing in various Es whatever in each E would
lead to G is the common feature of this A's diverse performances.
(Of course, this will hold only over some range of variation
in E; there will be limits to our predator's versatility.) This
teleological law is superior to the mere conjunction of a set of
efficient causal laws (As in /t do B1 and -4s in /2 do B2 and so on)
both because it explains them all by subsuming them under a
more general theory and because it enables us to predict, as
the finite set (^4s in In do Bn for n = 1,2, ... r) does not, that
if in a situation of some further kind Jr+l some performance
Br+l would lead to G, then in an /,+l-situation an A will do
Bf+l.
It is clear that our knowledge and understanding of a par-
ticular thing—say this A—or things of a certain kind—say As—
can be such that the most unitary general account that we can
give, and the one with the greatest predictive value, is a teleo-
logical one. But it is misleading to say that this teleological
account is therefore more basic than its efficient causation
rivals. The fundamental laws of working, the ways in which
the Bn events actually come about, are here still laws of efficient
causation and involve no reference to G—that is, as long as we
attend only to what happens now, and do not go back to
further explanatory accounts concerning reinforcement or evolu-
tion. We have so far, then, found a place for indispensably
7
In the Analysis article referred to in footnote 6, p. 377.
 TELEOLOGY 279
ideological accounts within our knowledge and understanding
of, and predictions about, certain kinds of behaviour; but we
have not yet found processes which are objectively, in them-
selves, irreducibly ideological.
This sort of example does, however, refute a move sometimes
made by anti-teleologists who suggest that any causal law what-
ever can be recast, by mere verbal juggling, in a teleological
form, and therefore that this form reveals no important dis-
tinction: 'do not basic causal laws in effect say that in any
situation, things of a certain kind . . . will do whatever is
required in order that they remain in certain relations one to
another, or in order that they continue to move in a certain
direction of change?'8 A Newtonian solar system does indeed
do whatever is required to keep constant the total sum of
potential and kinetic energy, to keep each body's acceleration
directly proportional to the resultant force acting on it and
inversely proportional to its mass, the force being a certain
function of all the masses and distances in the system; but such
keeping constant and keeping proportional are not results causally
brought about by what the system initially does as G is causally
brought out by Bn. What the solar system persistently 'achieves'
is a certain mathematical form which is realized in different
concrete arrangements of the positions and velocities of the
various bodies; but this is quite different from the convergence
of the different sorts of behaviour, B^ in /1} JBZ in 7a, and so on,
towards a single type of goal-event G. The different actions
that As initially perform in these different kinds of situation do
not, as we have seen, lend themselves to any intrinsic common
description: we can reach a common description of ^x in Ilt
Bz in /a, and so on only by saying that each is what there leads
to the remoter result G; but the various moves that the solar
system initially makes from various prior positions are open
to an intrinsic common description (though of an essentially
mathematical sort) and do not call for any common remoter
result to unify them.
The very fact that what unifi.es an A's different patterns of
behaviour is teleological, whereas its fundamental laws of work-
ing are non-teleological, creates a problem: why does it do, in
8
T. L. S. Sprigge, 'Final Causes', Aristotelian Society Supplementary Volume xlv
(1971), 149-70, csp. p. 161.
a8o TELEOLOGY
each kind of situation, whatever would lead to G? It must have
one inbuilt pattern of response to 7l5 another inbuilt pattern
of response to 72, and so on: but how does it come about that
it has this odd collection of inbuilt features? That each response
when triggered by the appropriate situation should lead to the
same goal G is, on the face of it, an extraordinary coincidence
which demands some further explanation in a way that the
regular conformity of a solar system to Newtonian laws does
not. This is the element of truth in the theological Argument
from Design: there are kinds of things, notably the behaviour
of various organisms, which exhibit 'marks of design', co-
existences of features which taken in themselves would be
remarkable coincidences, and which call for some further
explanation. But, in principle, several different kinds of ex-
planation are possible. Our 'coincidence' could be resolved by
reference to the conscious purposes of a very skilful and powerful
designer and maker, or (up to a point) by reference to reinforce-
ment, or by reference to an evolutionary process; or an initial
explanation of the 'coincidence' might be given in terms of the
animal's own conscious purpose to achieve G. But something is
called for to resolve the tension between the teleological
common description and the mere conjunction of efficient
causal laws.
Although it is not true that any causal law whatever can be
trivially recast in a teleological form, some special kinds of
mechanism do invite teleological description of a sort analogous
to that which we have been examining. The governor on a
steam engine sometimes opens and sometimes closes a valve in
such ways as to bring the engine speed near to a certain value,
and a homing rocket changes its direction in such ways as to
bring it eventually into contact with an elusive target, though
either of these is far less versatile than a hungry wolf. These
teleological descriptions are literally true of such devices: they
are in no way metaphorical or animistic or anthropomorphic.
They are, of course, quite compatible with detailed mechanical
descriptions in terms of purely efficient causation, and are
indeed deducible from some such descriptions. On the other
hand, it is quite possible that the teleological description should
be the best one that we know, providing more general explana-
tions and more powerful predictions than any available causal
 TELEOLOGY 281
account. We may find out by a few experiments that the governor
does something that keeps the engine speed fairly constant
without knowing how it does this; and after rockets of this kind
have had one or two spectacular successes in reaching targets
that have done a fair amount of zigzagging about we may pre-
dict that no movements of the target, however erratic, will
elude them, and yet not know at all how this comes about.
Once again our knowledge and understanding of something
may be distinctively and irreducibly teleological, a teleological
account may be the best one available to us, and yet the basic
laws of working will be efficient causal laws. And here again
the teleological aspect calls for some further explanation, and
the more versatile the device, the more clearly it calls for this,
because the greater would the coincidence be if there were no
unitary explanation for its achieving the same result by diverse
means in diverse conditions. But with these devices the required
explanation is at hand in the conscious purpose, ingenuity, and
skill of their makers.
In fact, we know in outline how such devices work; by feed-
back. Some (perhaps random or externally caused) change
brings the device in some sense nearer to its goal. The control
mechanism reacts to this increased 'nearness' by producing a
further change in the same direction. But if some change takes
the device further from its goal, the control mechanism reacts
to this decreased 'nearness' by producing a change in the
opposite direction. The evolutionary explanation of teleo-
logical behaviour in organisms has a rough formal analogy to
this. Some differential genetic characteristic X^ tends to pro-
duce behaviour J5t in situations of type /jj if B± in 7X tends to
lead to (7, where G is in turn something that increases the
organism's chance of having descendants, the organisms that
have X{ will have more descendants than otherwise similar ones
that do not, and if X^ tends to be inherited in some way there
will, as time goes on, be proportionately more members of the
species that do B1 in Iz; but, conversely, if Xz tends to produce
behaviour Bs in situations of type Ie, and Bs in Ie tends to pre-
vent G, there will in time be fewer members of the species that
have Xz and so do Bz in Iz: natural selection constitutes a kind
of feedback. Taking the species as a whole as the system,
various moves that actually bring an earlier phase of the system
282 TELEOLOGY
'nearer' to G are fostered, while any move that actually takes
it further from G is attenuated; that is why later phases of the
system tend to display in various situations whatever behaviour
favours G. In this story, as I mentioned earlier, the reference to
G is uneliminable: actual attainings of G by doing J8t in I± and
actual failures to attain G as a result of doing Bz in Iz are part
of the causal process by which X1 has been multiplied and Xe
made rare. It is the fact that situations of kind 7j are such that
an A's doing B{ in them, leads to G, not the fact that they are
/! (though these features are, by hypothesis, coextensive), con-
joined with the fact that As possessing Xl tend to do Bl in Ilt
that has fostered X1, and conversely it is the fact that Iz situa-
tions are such that an A*s doing Bz in them tends to prevent G,
not the fact that they are Iz, that has made Xa rare. But although
the (type-event) G that is the goal in the teleological description
of the behaviour of this At the product of the evolutionary
development, thus plays an uneliminable part in the evolu-
tionary explanation of this developed behaviour, it is not as a
goal, not as something for the sake of which something else is
done, that it figures in that story, but simply as a result that
has come about from certain causes on some occasions and that
has been prevented from coming about by other factors on other
occasions, and that, when it occurs, has certain further effects.
We thus have the slightly paradoxical but not, I hope, really
obscure consequence that although the evolutionary explana-
tion of what, taken on its own, invites a teleological description
makes essential reference to the G which is the goal in that
description, that evolutionary story is not itself a teleological
one. The steps it narrates both are and are described as pro-
cesses of efficient causation only.
I have here discussed an evolutionary explanation of the
goal-seeking behaviour of an animal. A closely analogous
account could be given of the functional appropriateness of
organs in organisms of all sorts. The presence and the operation
of each organ will correspond to our Blt 7?2, and so on. The
co-operation and mutual support of the various organs and
their operations (in a healthy organism) correspond to the
convergence of B^ in 71} B2 in 72, and so on to the single type-
event G] it is these that make a teleological description appro-
priate and (since the detailed workings themselves involve only
 TELEOLOGY 383
efficient causation) call for some further explanation. And again
it is an evolutionary story that provides this explanation,
making essential mention of some goal G—the harmony of
organic functions which promotes the survival and reproduction
of organisms—and yet itself involving only processes of effi-
cient causation.
It would be labouring the obvious to show that the same
applies to a psychological story about reinforcement—though
such a story gives only an incomplete explanation, since it
needs a further evolutionary story to explain why the organisms
in question are capable of having tendencies reinforced.
It is worth stressing that when we say of this A that it does
BI in /! because Bl in I± leads to G, and Bz in 72 because Bz
in /2 leads to G, and in general Bn in In because Bn in In leads
to G—in other words, that in each situation it does whatever
leads to G because it leads to G—we are speaking correctly
about the extended evolutionary process, whereas if we were
speaking only about the short-term response of this A as it now
is to its environment, we should say that it does B± in /x because
this environment is intrinsically of kind /19 Bz in 72 because it is
intrinsically of kind /2, and so on.. (In each case we are speaking
of fact-causes, explanatory causes, as these were distinguished
in Chapter 10, and we could hardly make the present point
without the help of this or some equivalent distinction.) The
distinctively teleological description of this A's behaviour, that
in each situation it does whatever leads to G, can therefore be
appropriately used in two very different states of knowledge.
As we saw, it can be used as the most general and most pre-
dictively powerful account available to us after a fairly super-
ficial observation of this A's behaviour in several different
situations, and while we as yet know nothing of the detailed
causal processes by which this A now responds to its environ-
ment, let alone any evolutionary process by which it has come
into existence with these tendencies and capacities. In this use
it is not an explanatory cause statement, nor a producing cause
statement either: it is what we may call a surface teleological
account. But this same description can be used in the knowledge
of and with reference to at least the sketch of an evolutionary
story: it is then serving as a condensation of an explanatory
cause statement, that this A does (for all n) Bn in In and
384 TELEOLOGY
so tends to achieve G because some ancestral proto-^s did Bn
in 7n and so achieved G and so survived and reproduced their
kind better than other proto-^s which did something else in In
and so failed to achieve G. It is only because the same type-
event G is mentioned twice in this explanatory cause statement,
as the general result of this A'5 present (versatile) behaviour and
as what some proto-^s achieved while others did not, and also
because the same set of type-events Blt B2, and in general Bn,
in situations of types I1, I2, and in general In, are also mentioned
twice, as what this A now does and as what some proto-
As did, that the condensation of this explanatory cause state-
ment takes the teleological form: this A does in each situation
whatever leads to G because it leads to G. There is, in these
respects, an essential logical connection between the descrip-
tion of the evolutionary explanatory cause and the description
of the behaviour which it explains. Yet no one can doubt that
this is a statement of efficient causation of a thoroughly standard
kind, relating logically distinct occurrences and discovered only
empirically—facts which should warn us against concluding
hastily in other cases that any logical connection between ante-
cedents and consequents is an obstacle to ordinary efficient
causal relations.
In all the kinds of case mentioned so far an initially teleo-
logical description has been underpinned and replaced by a
story about efficient causation. But, it might be said, it is
logically possible that there should be something which called
for an ultimately teleological description. There might be an
animal—or, for that matter, an inanimate system—which was
found to do in a wide range of Es whatever in each E would
lead to G, so that it was inductively reasonable to predict that
it would do so even in further, unobserved, kinds of E, and yet
these diverse performances (Bl in /1} Bz in 72, etc.) might not
lend themselves to any intrinsic common description, the only
distinctive common feature of these performances might be
their leading to G, and there might be no further explanation
to be found either in terms of the animal's—or the system's—
having been skilfully designed and manufactured by a maker
whose purpose was that it should achieve G, or in terms of
evolution or reinforcement or anything of that sort. Even a
backward causation explanation might be excluded by the
 TELEOLOGY 285
occasional failure of G to be actually realized, even where the
animal had done the right B for the E as it then was. That is,
the animal's—or the system's—behaviour might invite teleo-
logical description for the reasons indicated by Taylor, and yet
this might be left as ultimate by the non-realization of any
efficiently caused underpinning.
It must be conceded that this is a logical possibility; but
there is no evidence that anything of this sort actually occurs.
If it did occur, it would be very natural to suppose that this
animal—or system—had a conscious purpose of achieving G,
and knew or rightly believed that each Bn in 7n would lead to
G, and that this explained its diverse G-ward performances.
That is, this case would then be assimilated to that of conscious
purposive action which we have still to consider. But if this
explanation also were excluded we should be faced with a very
puzzling situation indeed. The sheer unexplainable coincidence
of such ultimate non-conscious teleology, of the animal's—or
the system's—doing in each E whatever would lead to G but
for no reason whatever in which the type-event G figured any-
where, would be far more surprising than, say, mere indetermin-
ism (as in the random differences in the outcome of relevantly
identical situations covered by quantum mechanics). So it is
just as well that it seems not to occur.
So far, then, we have found several uses for distinctively
teleological accounts and explanations, but no objective pro-
cesses which are in themselves teleological in a way that either
excludes their being instances of efficient causation or makes
their teleological aspect objectively more basic than their
efficient aspect. It remains to be considered whether this holds
good when we come to what; must be the hard core of final
causation—as well as being the original source of our teleo-
logical concepts—our own experience of conscious purposive
human action.
The central and crucial case may be represented by the same
letters as before: A—now a human agent—does B in situation
E in order to bring about G. But since we are now concerned
with conscious action, not with automatic responses to the
environment, it matters less of what kind the situation E is
than of what kind A sees it as being; so let us say that A sees E
as being a situation of kind /. Then if A does B in order to
286 TELEOLOGY
bring about G, he must believe that his doing B in this /-situa-
tion will lead to G. As before, 'lead to' is deliberately imprecise:
A may believe that his doing B is necessary in the circumstances
for G, or sufficient in the circumstances, or both, or merely that
it will increase the likelihood of G's coming about: there will
be examples of all these sorts, but the differences between them
do not affect the main issues. Equally it does not matter
whether A's belief is dispositional or occurrent, confident or
uncertain, true or false, or even whether it is knowledge rather
than belief. But what A believes is that there is some relation of
efficient causation between his doing B in E and the occurrence
of G. He need not, of course, be making any explanatory causal
judgement, he need not claim to know what it is about E, or B,
that will make the doing of B in E lead to G; rather he must see
his doing of B in E as some sort of producing cause of G.
Purposive action, then, involves in this respect not efficient
causation itself but a belief about efficient causation. But, of
course, more than this belief is required: A also desires or wants
or intends to bring about G.
Now the natural and traditional9 view is that A's believing
this and his wanting (or desiring or intending) that are joint
efficient causes of his doing B', that is, they are factors which
together constitute the cause or at least a partial cause of his
doing B. Moreover, on this view they would be, in our termino-
logy, explanatory causes. It is causally relevant that A's conative
state should be wanting to bring about G, and that his cognitive
state should be believing that B would lead to G, with the same
description of the goal G in both. And it is causally relevant
that what he believes to be some sort of efficient cause of G is
the same B that, when he finally acts, he sees himself as doing.
If A turned on the burner in order to boil the kettle, what is
causally relevant about his wanting is that it was his wanting
this kettle to boil, what is causally relevant about his believing
is that it was his believing that his turning on of this burner
would lead to this kettle's boiling, and it is causally relevant
that the subject of this belief is his turning on of this burner,
while the performance of which all this is the explanatory cause
is seen by A as his turning on of this burner. If, for example,
9
Gf. D. Davidson, 'Actions, Reasons, and Causes', Journal of Philosophy, lx
(1963), 685-700.
 TELEOLOGY 287
A saw his ultimate action simply as his rotating that white
knob ninety degrees anticlockwise, this action could not be
caused in the proposed way by his wanting to boil the kettle
and his believing that turning on this burner would make the
kettle boil. And equally, even if he saw his action as his turning
on of the burner, it could not be caused in the proposed way
if what he wanted was that the kettle should boil while what
he believed was that his turning on of the burner would lead
to the production of a quantity of water vapour. We need the
descriptive links between his action as he sees it, the content
of his belief, and the (intentional) object of his desire.
But these descriptive links which seem to be required for
such an explanatory cause statement have been seen as fatal to
the traditional view, as showing that there is no efficient causal
relation here at all Efficient (or 'Humean') causation requires
logically independent causes and effects, whereas the supposed
'cause* and 'effect' in any voluntary action are logically con-
nected. Thus Melden, arguing indeed against a rather simpler
theory of 'acts of volition', says: 'Let the interior event which
we call "the act of volition" be mental or physical (which it is
will make no difference at all), it must be logically distinct
from the alleged effect—this surely is one lesson we can derive
from a reading of Hume's discussion of causation. Yet nothing
can be an act of volition that is riot logically connected with
that which is willed—the act of willing is intelligible only as
the act of willing whatever it is that is willed,'10 This logical
connection argument has been widely accepted; and yet, as it
stands, it is little more than a bad pun on the phrases 'logically
distinct' and 'logically connected'. What we can learn from
Hume's discussion is that cause and effect must be logically
distinct occurrences (or 'existences'), that it must be logically
possible that either should occur while the other does not. But
the supposed act of volition would need to be connected with
its alleged effect (in a successful voluntary performance) only
by the appearance of some true description of that effect in
the content of the act of volition. Acts of volition would be
identified (at least partly) by their content; if A voluntarily
does B the act of volition theory is committed to saying that
what happens is truly describable as A's doing B, and that this
10
A. I. Melden, Free Action, p. 53,
288 TELEOLOGY
same description, ^4's doing B, is the content of A's act of
volition, it is as his doing of B that he 'wills' it. But of course
this logical connection' would not prevent the act of volition
from being a logically distinct occurrence from the doing of S,
it would not make it logically impossible that either should
occur without the other. There is no more logical connection
here than there is between someone's hoping that Leeds will
win the Cup Final and Leeds's winning the Gup Final, and
though these two may not be causally connected, it is no lack
of logical distinctness that prevents this.
But this is liable to be confused with a different and trickier
point. On the act of volition theory, A'& voluntary raising of his
arm consists of his willing that his arm should rise and its rising
as a result: the voluntary action as a whole consists of an act of
will, a bodily movement, and an efficient causal connection
between them. But it is tempting to say rather that what A wills
is that he should raise his arm, and that the effect of his willing
is that he raises his arm. If we say this, we are taking the whole
voluntary action as the 'effect'—and also a description of it as
the content of the act of will. That is, on the act of volition
theory, we have now incorporated the act of volition in the
'effect', and that 'effect' is indeed no longer a distinct occurrence
from the act of volition, not because a description of the 'effect'
occurs as the content of that volition, but because the volition
has been included in the 'effect'. But we can, and should, pre-
sent the theory coherently by keeping them apart. Analogously,
the movement of a cricketer's bat cannot cause a drive, for the
supposed effect, the drive, includes that movement. None the
less, part of the movement of the bat (leaving out the follow
through) is a partial cause of the movement of the ball, and the
whole process of bat-movement temporally overlapping with
ball-movement and causally related to it is what is called a
drive.
Such arguments as Melden's have been criticized on such
grounds as these,11 and the force of these criticisms is recognized
by, for example, von Wright: 'The logical dependence of the
specific character of the will on the nature of its object is fully
11
e.g. by Davidson in the article mentioned in footnote 9, p. 286 and by F. M.
Stoutland in 'The Logical Connection Argument', American Philosophical Quarterly
Monograph iv (1970), 117-30.
 TELEOLOGY 289
compatible with the logical independence of the occurrence of an
act of will of this character from the realization of the object.'12
It is worth stressing that in order to bring out the latter logical
independence it is not necessary to say—though it may be
true—that the agent's willing or Intending or desiring is open
to some other description which is independent of the nature
of its object. And, as I have insisted, it is the description which
is 'logically dependent on the nature of its object' which pre-
sents the causally relevant feature of this willing, the one which
must be mentioned in an explanatory causal statement.
But though von Wright accepts these criticisms, he still
endorses a version of the logical connection argument. His
thesis is that if it is asked 'how . . . one ascertains (verifies)
whether an agent has a certain intention, "wills" a certain
thing—and also how one finds out whether his behavior is of a
kind which his intention or will is supposed to cause' it will
turn out that one cannot answer the one question without also
answering the other, and hence that 'the intention or will
cannot be a (humean) cause of his behavior', because 'The
facts which one tries to establish [are] not logically independent
of one another 5 . 13 There is, he holds, a schema of practical
inference (or practical syllogism) which is that of a teleological
explanation turned upside down. If the teleological explanation
is 'A does B in order to bring about G' the practical inference
is (roughly):
A intends to bring about G.
A considers that he cannot bring about G unless he does B.
Therefore A sets himself to do B,1*
And, von Wright argues, the practical inference, when properly
formulated, is logically binding. Several precautions are needed.
A will not set himself to do B if he thinks (or knows) that he
cannot do B. So von Wright takes it that the first premiss
'contains, in a concealed form, that the agent thinks he knows
how to bring about' G,IS Also, what A intends and considers
at one time cannot, in the required sense, be logically connected
12
Explanation and Understanding, p. 94.
" Ibid., pp. 94-5 and 115-16.
14
Ibid., p. 96. I have altered the letters to agree with those used above.
15
Ibid., p. 103.
ago TELEOLOGY
with his setting himself to do something at a later time: the
logical connection argument cannot apply to previous inten-
tions but only, if at all, to concurrent intentions. So the practical
inference schema, in its final—and, von Wright argues, logic-
ally binding—form, becomes:
From now on A intends to bring about G at time t,
From now on A considers that, unless he does B no later than
at time t, he cannot bring about G at t.
Therefore, no later than when he thinks time t has arrived,
A sets himself to do B, unless he forgets about the time or is
prevented.16
Now von Wright argues that in order to verify that A sets
himself to do £, it is not enough to observe that he did B in the
sense that the result came about through A's movements: even
if this happens, we must ascertain that A's behaviour 'is inten-
tional under the description "doing B" *. Also, we may need to
verify that A set himself to do B but failed to do 5. So (in either
case) to verify the conclusion we have to verify either the pre-
misses of this practical inference or those of some other similar
inference which would show the behaviour to be intentional
under the description 'doing B'. He also argues, in a Witt-
gensteinian vein, that one can verify the premiss that A has
such and such an intention (even if A is oneself!) only by
recognizing intentionality in A's behaviour, and 'Behavior gets
its intentional character from being seen by the agent himself or
by an outside observer in a wider perspective, from being set in
a context of aims and cognitions', as happens 'when we construe
a practical inference to match it'.17
Non-Wittgensteinians might suspect that if we accepted both
von Wright's arguments we should be caught in a vicious circle
which made it impossible to 'verify', or to have any good reason
for believing, either the premisses or the conclusion of any
practical inference whatever. But let that pass. There are
several more important comments to be made.
16
Explanation and Understanding, p, 107. I am here concerned only with von
Wright's use of this 'practical syllogism'; but there is another and I think more
interesting sort of practical syllogism which is not, like this, merely descriptive
of* an agent but systematizes an agent's own rational deliberation, by which he
may arrive at a decision to act. I have examined this briefly in a discussion on
'Practical versus Theoretical Reason* in Practical Reason, ed. S. Korner.
" Explanation and Understanding, pp. 114-15.
 TELEOLOGY 391
First, the possibility of verifying the conclusion by verifying
the premisses of some other practical inference is a red herring.
Except in special cases where an action is intentionally over-
determined—where the agent is killing two birds with one
stone—there will be just one ideological explanation of an
action and so, when it is turned upside down, just one relevant
practical inference. In such a case von Wright's first argument
seems to show that the conclusion of this practical inference
can be verified only by verifying the premisses of this practical
inference.
But, secondly, would this show that what the conclusion
describes is not as an occurrence logically distinct from what the
premisses describe? Surely not. What the conclusion describes,
A's setting himself to do B, has been taken to include A's
intending to do B, But even if A's only reason for intending to
do B is that he intends to bring about G and sees doing B as
necessary for G, his having this reason and his intending to do
B can still be distinct occurrences. Further, if his setting him-
self to do B only includes, and is not exhausted by, his (now)
intending to do B (right away), it would for that reason be
partly a distinct occurrence from his having the reason specified
by the premisses, even if his intending to do B were not so
distinct. The obstacles to independent verification mentioned
in this first argument have no tendency to show that the occur-
rence described by the conclusion is included in those described
by the premisses.
A third and still more important point is this. The tendency
of von Wright's second argument is indeed to suggest that these
occurrences are not distinct, because the 'occurrence' described
by the premisses is just that described by the conclusion seen
in a certain way. Of course this view is supported by well-
known difficulties about inner mental states or events, diffi-
culties which, I believe, can be met, though I cannot discuss
them here. But whatever we make of these difficulties, we
cannot consistently take von Wright's view about concurrent
intentions if we admit that there are also literally preformed
intentions, intentions which exist before any behaviour mani-
fests them. For these intendings must be, as occurrences,
logically distinct from any related actions: it cannot then be
impossible that exactly similar items should continue to exist
aga TELEOLOGY
and so be concurrent with, but logically distinct from, the
actions that fulfil them. No problems about independent veri-
fication can undermine this possibility of independent existence.
We could evade this argument only by interpreting the talk
about previous or preformed intentions as fictitious in a high
degree.
I have discussed von Wright's account because it seems to
me to make the best case that can be made for even a partial
acceptance of the logical connection argument; but it seems
to me that it fails completely.
But suppose that we were to waive all these objections, and
concede von Wright's conclusion, what would follow? The
proposed practical inference, as a logically valid schema, would
hold within and for a certain systematic way of interpreting the
actions in question. The descriptions 1A intends to bring about
G', c^4 considers that he cannot bring about G unless he does /?',
and 'A sets himself to do B'—or rather their more exact counter-
parts in the final version—are now being so read that the first
two entail the third: anything that counts as falsifying the third
automatically falsifies at least one of the others. We have a
network of interlocking descriptions which must be applied or
withheld as a whole. Then since the teleological explanation is,
as von Wright says, just this schema turned upside down, it too
is only a network of interlocking descriptions: we have to read
*A does B in order to bring about G' as short for '(A does B]
and (A believes that he cannot bring about G without doing B}
and (A intends to bring about G)' with such senses that the
first of these conjuncts is entailed by the two others together,
with a hint that this entailment holds. And this is a conceptualist
view of teleology with a vengeance: we have nothing but a set
of interlocking concepts, with no shadow of an account or
explanation of how A came to do B, Once we have a genuine
logical connection between the action on the one hand and the
belief and intention on the other, as opposed to the spurious
logical connection that Melden offered—the mere 'dependence
of the specific character of the will on the nature of its object'—
we do indeed exclude efficient or Humean causation, but we
also abandon all claim to be explaining how things come about.
We cannot now speak of processes of final causation. Nor are
there, in this sense, any teleological laws of working that would
 TELEOLOGY 293
be alternatives or rivals to those of efficient causation. A teleo-
logical account, thus construed,, simply leaves the question of
how things come about off to one side, and so leaves it wide
open.
Although this is a possible and consistent view, it is an
extremely unnatural one—so unnatural that it is almost im-
possible to adhere to it for long. Our natural way of speaking
of conscious purposive human action treats intendings, desir-
ings, believings, and so on as elements in accounts of how
actions come about. It seems that von Wright himself so treats
them when, in the next chapter, he discusses the course of
historical events. But if we revert to this natural treatment, we
cannot consistently read into it the logical connection barrier
to efficient causation. If intendings, and so on, are so construed
as to be permissible elements in any genetic account, they can
figure there, as I suggested earlier, as factors that help to make
up (explanatory) efficient causes of people's doing what they
consciously do. Only an indefensible confusion of two separ-
ately defensible interpretations of teleological descriptions of
human action produces the appearance of a special kind of
causation by which actions come about but which includes
features that prevent it from being a species of efficient causa-
tion.
We can, if we like, accept von Wright's conceptual teleology:
I do not think that either the ordinary meaning of the termino-
logy of intentions or any epistemological difficulties about
mental events force us to do so. Alternatively we can interpret
teleological descriptions of human action (and perhaps some
animal action also) as explanatory causal accounts. But then
if A does B because he intends (or wants or desires) to bring
about G and believes that his doing B would lead to G, there
is not only ^4's belief about efficient causation between his
doing B and the occurrence of G, there is also an objective
relation of efficient causation between ^4's intending (or wanting
or desiring) this, his believing that, and his doing of B.
A view which is almost equivalent to von Wright's (and to
which von Wright refers with approval) is taken by Stoutland.18
After arguing, correctly, that the weak type of logical connec-
tion mentioned by, for example, Melden does not exclude a
18
In the article referred to in footnote n, p. 288.
894 TELEOLOGY

contingent connection such as causation is now commonly

taken to be, Stoutland says that there is also a strong type of
logical connection between an intention and its fulfilment which
does exclude a contingent connection. Intending that p makes
it 'a priori necessary that p occur if normal conditions obtain',
and in general 'To attribute a goal to a system, in the sense
involved in a teleological theory, is to say that it is a priori
necessary that the system realize the goal if normal conditions
obtain, that is, if nothing interferes'.19
Now this last remark is far too weak distinctively to charac-
terize a teleological account. If I call something a reflector I
am saying that it is a priori necessary that it reflects light if
normal conditions obtain, and similar comments apply to the
calling of something a conductor, a cure for chilblains, or
indeed a cause of such-and-such in the inus-condition sense,
and while such a way of speaking describes an antecedent in
terms of a consequent and (given some positive valuing of the
consequent) develops easily into at least a functional descrip-
tion, it is obviously quite compatible with saying that the ante-
cedent causally brings about that consequent.
There are some hints20 that Stoutland regards a system as
having a goal if it tends by some internal process to achieve
some result, external conditions needing only not to interfere,
but not if external conditions play some more positive role in
bringing about- the result. But even if any distinction of this
sort could be drawn firmly, it would be no more than a distinc-
tion between 'transeunt' and 'immanent' causation, that is
between what is ordinarily called causation and the sort of
self-maintaining process for whose inclusion, in an extended
notion of causation, I have argued in Chapter 6. There is
nothing intrinsically non-contingent about such an 'immanent'
relation.
Stoutland must then fall back on a conceptual relation be-
tween intention and fulfilment. 'It is not simply contingent that
my intention to visit my friend should result in my visiting him;
my visiting him, given normal conditions, is a criterion of my
intending to visit him. That is part of our concept of intention.'21
But this, with a sense of ^criterion' strong enough to exclude a
19 20
Explanation and Understanding, pp. 126, 128. Ibid,, p. 127.
« Ibid., p. 128.
 TELEOLOGY 395
contingent (causal) relation, is equivalent to von Wright's con-
ceptual teleology: the intending, as intending, is not now a
distinct occurrence from the visiting, and does not explain how
the visiting came about, any more than the infant Edward
Heath's being a future Prime Minister explains his eventually
becoming Prime Minister. If it were claimed (though, I think,
neither Stoutland nor von Wright does claim this) that an
intention to do p is a previous existent which in itself, and not
just in being described as an intention to dop, logically requires
that if circumstances are normal p will be done, we should have
to reply that there can be no such animal. Hume's arguments
exclude this from reality, not merely from our concept of
efficient causation.
We can, then, repeat part of our interim conclusion: though
we have found several uses for distinctively teleological accounts
and explanations, we have found no objective processes which
are in themselves teleological in a way that precludes their
being instances of efficient causation. On the other hand, con-
scious purposive action which calls for the kind of explanatory
causal account suggested does seem to involve a distinct species
of efficient causation. This might be challenged by someone
who adopted a physicalist or epiphenomenalist view. While I
arn sceptical about such theories, it seems to me to be more
important to insist that even if some such theory were adopted,
an adequate explanatory causal account in its terms would have
somehow to reproduce the characteristic teleological form,
keeping intact the descriptive links between the repeated occur-
rences of our terms '£' and 'G*. Some physicalist counterpart
would have to be found for the agent's seeing his action as of a
certain kind, for his intending a certain event under a certain
description,, and for his believing that a certain kind of action
would lead to a certain kind of outcome. And if this were done,
even the physicalist account would be teleological in that some
representation of a B-G causal sequence would figure essentially
in the explanatory cause of what we now describe as A's doing
B. It might be that this teleological sort of account could be
reduced to a more basic causal account which eliminated what
I am here calling the teleological form; but there is no obvious
reason why this should be possible,, Even if it were possible, an
adequate explanatory account would involve some repetitions
ag6 TELEOLOGY
of descriptive terms, just because the pure laws of working
would, as I have argued, involve some element of continuity
and persistence. In any case, teleology as a species presents no
challenge to efficient causation, of the richer sort for which I
have argued, as the genus that covers all actual laws of working
and perhaps all processes by which things come about.
It is, then, causation, in the sense of what satisfies the wider
and richer concept for which I have argued, that holds the
universe together, that makes it more than just a collection of
four-dimensional scenery. Moritz Schlick, like others in the
Humean tradition, was very critical of the notion of a causal
tie: 'After the scientist has successfully filled up all the gaps in
his causal chains by continually interpolating new events, the
philosopher wants to go on with this pleasant game after all
the gaps are filled. So he invents a kind of glue and assures us
that in reality it is only his glue that holds the events together
at all. But we can never find the glue; there is no room for it,
as the world is already completely filled by events which leave
no chinks between them.'22 But this criticism does not touch
the view for which I have argued. Causation is not something
between events in a spatio-temporal sense, but is rather the
way in which they follow one another. It involves regularities,
universal or statistical, in particular what I have distinguished
as pure laws of working, but it is not exhausted by them; it
includes also the spatio-temporal continuity stressed by
Ducasse, the qualitative or structural continuity, or partial
persistence, which I have sorted out from Kneale's more
rationalist concepts, and the features which constitute the
direction of causation. When Hume said that the principles of
association are to us the cement of the universe,23 he meant that
they are the links that connect the ideas of unobserved things
with impressions, that give rise to the non-rational inferences
which we naturally make. Inference is not as central in causa-
tion as Hume thought, but the causal inferences which we
undoubtedly make have sometimes more authority than he
allowed. When we get things right, our causal inferences retrace
or anticipate the sequences by which the universe creates itself.
11
'Causality in Everyday Life and Recent Science', in Readings in Philosophical
Analysis, ed. H. Fcigl and W. Sellars, pp. 515-33, at p. 522.
23
Abstract, p, 32.
 APPENDIX

Eliminative Methods of Induction

(Mill's Methods of Induction)

i. Introduction
J O H N S T U A R T M I L L , inhis System of Logic1 set forth and discussed
five methods of experimental inquiry, calling them the method of
agreement, the method of difference, the joint method of agreement
and difference, the method of residues, and the method of con-
comitant variation. He maintained that these are the methods by
which we both discover and demon!! trate causal relationships, and
that they are of fundamental importance in scientific investigation.
In calling them eliminative methods Mill drew a rather forced
analogy with the elimination of terms in an algebraic equation. But
we can use this name in a different sense: all these methods work
by eliminating rival candidates for the role of cause.
The inductive character of these methods may well be questioned.
W. E. Johnson called them demonstrative methods of induction,2
and they can be set out as valid forms of deductive argument: they
involve no characteristically inductive steps, and no principles of
confirmation or corroboration are relevant to them. But in each
case the conclusion is a generalization wider than the observation
which helps to establish it; these are examples of ampliative induction.
The general nature of these methods may be illustrated by
examples of the two simplest, those of agreement and of difference.
Mill's canon for the method of agreement runs: 'If two or more
instances of the phenomenon under investigation have only one
circumstance in common, the circumstance in which alone all the
instances agree is the cause (or effect) of the given phenomenon.'
For example, if a number of people who are suffering from a certain
disease have all gone for a considerable time without fresh fruit or
vegetables, but have in other respects had quite different diets, have
lived in different conditions, have different hereditary backgrounds,
and so on, so that the lack of fresh fruit and vegetables is the only
1 2
Book III, Ghs. 8-10. Logic, Part II, Gh. 10.
298 ELIMINATIVE INDUCTION
feature common to all of them, then we can conclude that the lack
of fresh fruit and vegetables is the cause of this particular disease.
Mill's canon for the method of difference runs: ( If an instance in
which the phenomenon under investigation occurs, and an instance
in which it does not occur, have every circumstance in common
save one, that one occurring in the former; the circumstance in
which alone the two instances differ, is the effect, or the cause, or
an indispensable part of the cause, of the phenomenon.' For example,
if two exactly similar pieces of iron are heated in a charcoal-burning
furnace and hammered into shape in exactly similar ways, except
that the first is dipped into water after the final heating while the
second is not, and the first is found to be harder than the second,
then the dipping into water while it is hot is the cause of such extra
hardness—or at least an essential part of the cause, for the hammer-
ing, the charcoal fire, and so on may also be needed. For all this
experiment shows, merely dipping iron while hot into water might
not increase its hardness.
The method of agreement, then, picks out as the cause the one
common feature in a number of otherwise different cases where the
effect occurs; the method of difference picks out as the cause the
one respect in which a case where the effect occurs differs from an
otherwise exactly similar case where the effect does not occur. But
in both the conclusion is intended to say more than that this was
the cause of that effect in this instance (or this group of instances).
The conclusion in our first example is that this particular disease is
always produced by a lack of fresh fruit and vegetables, and in our
second example that dipping iron which has been heated and
hammered in a particular way into water while it is hot always
hardens it.
There are many weaknesses in Mill's description of these methods,
but there is no need to criticize his account in detail. The interesting
questions are whether there are any valid demonstrative methods of
this sort, and if so whether any of them, or any approximations to
any of them, have a place in either scientific or commonsense
inquiry. Several reconstructions of the methods have been offered;
the most thorough treatment I know of is that of von Wright, but
I find it somewhat unclear.3 I shall, therefore, attempt another
reconstruction.
In giving a formal account of the reasoning involved in these
methods, I shall use an old-fashioned sort of logic, traditional
(roughly Aristotelian) logic but with complex (conjunctive and
J
G. H. von Wright, A Treatise on Induction and Probability. Earlier accounts arc
those of Johnson (referred to above, p. 297) and C. D. Broad, *The Principles of
Demonstrative Induction', Mind, xxxix (1930), 303-17 and 426-39.
 ELIMINATIVE INDUCTION 399
disjunctive) terms. The letters 'A', '6', and so on stand for kinds of
event or situation, or, what comes to the same thing, for features the
possession of which makes an event or situation one of this or that
kind. These conditions (Mill's 'circumstances') can therefore be pre-
sent or absent on particular occasions (in particular instances}. To
say that all A are B would be to say that whenever feature A is
present, so is feature B. To say that A is necessary for B is to say
that whenever B is present, A is present, and to say that A is suffi-
cient for B is to say that whenever A is present, B is present also.
A conjunctive feature AB is present whenever and only when both
A and B are present. A disjunctive feature (A or B) is present when-
ever at least one of A and B is present. A negative feature not-A
(written A) is present whenever and only when A is absent. The
following argument forms are obviously valid:
(i) All A are C
Therefore, All AB are C
(ii) All A are BC
Therefore, All A are B (and All A are C)
(iii) All (A or B) are C
Therefore, All A are C (and All B are C)
(iv) All A are B
Therefore, All A are (B or C)
(v) All A are C
and All B are C
Therefore, All (A or B} are C.
In expounding the reasoning implicit in these methods, I shall
also use the letters *X\ 1T\ "£' as variables whose instances are the
features or kinds of occurrence represented by *A\ (B', and so on,
and I shall take the liberty of quantifying with respect to them; for
example, to say that for some X, X is necessary for B will be to say
that some feature occurs whenever B occurs; this statement would
be true if, say, all B are C. There is nothing unsound in these pro-
cedures, and I hope nothing obscure. The slight unfamiliar!ty of
these techniques is, I believe, more than compensated for by the
fact that for this particular task they are more economical than the
obvious alternatives.4
To avoid unnecessary complications, let us assume that the
4
The classic exposition of this extension of traditional logic to allow for complex
terms is in J. N. Keynes's Formal Logic (4th edition, 1906), Appendix G, pp. 468-
538. It was much discussed by John Anderson in lectures at Sydney University.
For our present purposes we can take the universal propositions as not having
existential import, and thus avoid the need for restrictions on the valid argument
forms.
300 ELIMINATIVE INDUCTION
conclusion reached by any application of one of these methods
(other than that of concomitant variation which we shall leave aside
for the present) is to have the form 'Such-and-such is a cause of such-
and-such a kind of event or phenomenon', where a 'cause' will, in
general, be both necessary and sufficient for the phenomenon,
though in some variants of the methods it will be taken as necessary
only, or as sufficient only. Let us assume also that we can distinguish
in some way causes from effects, and think of the methods only as
identifying causes. (Mill, as the canons quoted show, mixes this
task up with that of identifying effects, and so far as their logical
form is concerned the methods could be applied to'any problem
about conditions that are necessary for something, or sufficient,
or both, but their most interesting applications, and the ones
for which the required assumptions are most plausible, are to the
identification of causes.) However, the cause will be such, and will
be necessary or sufficient or both, in relation to some field, that is,
some set of background conditions: our question is, say, 'What
causes this disease in human beings living in ordinary conditions,
breathing air, and so on?' or, 'What causes the greater-than-
ordinary hardness in iron in ordinary circumstances and at ordinary
temperatures?' To say that A is necessary for B in the field F will
be to say that whenever B occurs in (or in relation to) something
that satisfies the conditions summed up as F, A occurs there also—
which will be expressed accurately enough by 'All FB are A'—and
so on. I shall use 'P* to represent the 'phenomenon' whose 'cause' is
being sought, and I shall call an occasion on which P is present a
positive instance, and one on which it is absent a negative instance. The
observation that supports the conclusion will be an observation of
the presence or absence of various conditions, each of which might
be causally relevant, in one or more (positive or negative) instances.
But since, as I said, the conclusion regularly goes beyond the
observation, and yet each method is supposed to be demonstrative,
that is, deductively valid, the conclusion must be drawn not from
the observation alone, but from it in conjunction with an assumption.
This pattern, assumption and observation together entailing a conclusion,
is characteristic of all these methods. And there are obvious pro-
portional relations between the three items. The less rigorous the
assumption, the stronger the observation needs to be if we are to get
the same, or perhaps even any, conclusion. With the same observa-
tion, a less rigorous assumption will yield a weaker conclusion, if
any. And so on.
But what sort of assumption is required, and what is it for it to
be more or less rigorous? Since we are to arrive at the conclusion that
 ELIMINATIVE INDUCTION 301
a certain condition is, in the sense indicated above, a cause of the
phenomenon, and to do so by eliminating rivals, we must assume
at the start that there is some condition which, in relation to the
field, is necessary and sufficient (or which is necessary, or which is
sufficient) for this phenomenon, and that it is to be found somewhere
within a range of conditions that is restricted in some way.
For a formal exposition, it is easiest to take the assumption as
indicating some set (not necessarily finite) of possibly relevant
causal features (Mill's 'circumstances' or 'antecedents'). Initially I
shall speak in terms of a list of such possible causes (p-cs), but, as I
shall show, we can in the end dispense with any such list. A p-c,
since it is possibly causally relevant in relation to the field in
question, must—like the phenomenon itself—be something that is
sometimes present and sometimes absent within that field: it must
not be one of the conditions that together constitute the field.
But are we to assume that a p-c acts singly, if it acts at all? If the
p-cs are A, Bt C, etc., are we to assume that the cause of P in F
will be either A by itself or B by itself, and so on? Or are we to
allow that it might be a conjunction, say AC, so that P occurs in F
when and only when both A and C are present? Are we to allow
that the actual (necessary and sufficient) cause might be a dis-
junction, say (B or /)), so that .P occurs in F whenever B occurs,
and whenever D occurs, but only when at least one of these occurs?
Are we to allow that our p-cs may include counteracting causes, so
that the actual cause of P in Fmay be, say, the absence of C (that is,
not-C, or C), or perhaps BC, so that P occurs in F when and only
when B is present and C is absent at: the same time?
There are in fact valid methods with assumptions of different
kinds, from the most rigorous, which requires that the actual cause
should be just one of the p-cs by itself, through those which pro-
gressively admit negations, conjunctions, and disjunctions of p-cs
and combinations of these, to the least rigorous, which says merely
that the actual cause is built up out of some of the p-cs in some way.
There are in fact eight possible kinds of assumption, namely that
the actual cause is:
1. one of the p-cs.
2. either one of the p-cs or a negation of one.
3. either a p-c or a conjunction of p-cs.
4. either a p-c or a disjunction of p-cs.
5. a p-c, or the negation of a p-c, or a conjunction each of whose
members is a p-c or the negation of a p-c.
6. a p-c, or the negation of a p-c, or a disjunction each of whose
members is a p-c or the negation of a p-c.
303 ELIMINATIVE INDUCTION
7. a p-c, or a conjunction of p-cs, or a disjunction each of whose
members is a p-c or a conjunction of p-cs.
8. a p-c; or the negation of a p-c; or a conjunction each of whose
members is a p-c or the negation of one; or a disjunction each
of whose members is a p-c, or the negation of one, or a con-
junction each of whose members is a p-c or the negation of one.
Analogy with the use of disjunctive normal form in the preposi-
tional calculus makes it easy to show that any condition made up
in any way by negation, conjunction, and disjunction from a set
of p-cs will be equivalent to some condition allowed by this eighth
kind of assumption, which is therefore the least rigorous kind of
assumption possible. The form of the observation determines whether
a method is a variant of the method of agreement, or difference, and
so on. But since each form of observation may be combined with
various kinds of assumption, there will be not just one method of
agreement, but a series of variants using assumptions of different
kinds, and similarly a series of variants of the method of difference,
and so on. To classify all these variants we may use a decimal
numbering, letting the figure before the decimal point (from i to 8)
indicate the kind of assumption, and the first figure after the decim'al
point the form of observation, thus:
1. a variant of the method of agreement.
2. a variant of the method of difference.
3. a variant of the joint method (interpreted as an 'indirect method
of difference').
4. a new but related method.
Further figures in the second place after the decimal point will be
used for further subdivisions.
A complete survey would take up too much space, but some of the
main possibilities will be mentioned.

2. Agreement and Difference—Simple Variants

Positive method of agreement
Let us begin with an assumption of the first kind, that there is
some necessary and sufficient condition £ for P in F—that is, for
some £ all FP are £ and all F% are P—and £ is identical with one
of the p-cs A, B, C, D, E. We obtain a variant of the method of
agreement (1.12) by combining this assumption with the following
observation: a set of one or more positive instances such that one
p-c, say A, is present in each, but for every other p-c there is an
instance from which that p-c is absent. This yields the conclusion
that A is necessary and sufficient for P in F.
 ELIMINATIVE INDUCTION 303
For example, the observation might be:
A B C D E
/,. p a p . a
I2 P P a a
{
where p' indicates that the p-c is present, 'a' that it is absent, and
a dot that it may be either present: or absent without affecting the
argument. 7X and /2 are positive instances: /x shows that neither B
nor E is necessary for P in F, and hence that neither can be the £
that satisfies the assumption, similarly Iz shows that neither C nor
D can be this ^; only A can be, and therefore must be, this ^; that
is, A is both necessary and sufficient for P in F.
Since this reasoning eliminates candidates solely on the ground
that the observation shows them not to be necessary, there is an-
other variant (i.i i) which assumes only that there is some necessary
condition % for P in F and (with the same observation) concludes
only that A is necessary for P in F.

Negative method of agreement

Besides this positive method of agreement in which candidates
are eliminated as not being necessary because they are absent from
positive instances, there are corresponding variants of a negative
method of agreement in which they are eliminated as not sufficient
because they are present in negative instances. The required observa-
tion consists of one or more negative instances such that one p-c,
say A} is absent from each instance, but for every other p-c there is
an instance in which it is present. For example:
A B C D E
N a P
N a P P
N a P
If the assumption was that one of the p-cs is sufficient for P in F,
this observation would show (1.13) that A is sufficient, while if the
assumption was that one of the p-cs is both necessary and sufficient,
the same observation yields the conclusion (1.14) that A is both
necessary and sufficient,

Method of difference
The simplest variant of this method (1.2) combines the assump-
tion that one of the p-cs is both necessary and sufficient for P in F
with this observation: a positive instance /j and a negative instance
304 ELIMINATIVE INDUCTION
NI such that of the p-cs present in I± one, say A, is absent from JV1}
but the rest are present in JVj. For example:
A B C D E
A p p p a .
JVi a p p . p
Here D is eliminated because it is absent from /j and hence not
necessary, and B, C, and E because they are present in JV^ and hence
not sufficient. Only A therefore can be, and so must be, the £ that
is both necessary and sufficient for P in F. Note that with an assump-
tion of this first kind it would not matter if, say, E were absent
from /! and/or D were present in jVj: the presence of the actual
cause A in It but not in JV^ need not be the only relevant difference
between the instances. But this would matter if we went on to an
assumption of the second kind. We may also remark that the method of
difference, unlike some variants of the method of agreement, requires
the assumption that there is some condition that is both necessary
and sufficient for P in F> As we shall see with variants 4.2 and 8.2,
the conclusion may not fully specify the resulting necessary and
sufficient condition, and the factor picked out as (in another sense)
the cause is guaranteed only to be an inus condition or better; but the
assumption needed is that something is both necessary and sufficient.

Joint method of agreement and difference

Mill's exposition of this method mixes up what we may call a
double method of agreement, that is, the use of the positive and negative
methods of agreement together, with what Mill himself calls an
indirect method of difference, in which the jobs done by the single
positive instance and the single negative instance in the method of
difference are shared between a set of positive and/or a set of
negative instances. The latter is the more interesting and distinctive.
Its simplest variant (1.3) combines the assumption that one of the
p-cs is both necessary and sufficient for P in F with this observation:
a set S^ of positive instances and a set Sn of negative instances such
that one of the p-cs, say A, is present throughout Si and absent
throughout Sn, but each of the other p-cs is either absent from at
least one positive instance or present in at least one negative
instance. For example:
A B C D E
ofA P P P P a
M/2 p p p a p
^ tNi a p a a a
n
\N2 a a p a a
 ELIMINATIVE INDUCTION 305
This assumption and this observation together entail that A is neces-
sary and sufficient for P in F. As the example shows, none of the
other p-cs, B, C, D, E, could be so, given such an observation:
yet St by itself does not yield this conclusion by the positive method
of agreement, nor Sn by the negative method, nor does any pair of
positive and negative instances from those shown yield this con-
clusion by the method of difference.

3. Agreement and Difference—Complex Variants

When we go on to an assumption of the second kind, allowing
that the actual cause may be the negation of a p-c, we need slightly
stronger observations. Thus for variants of the positive method of
agreement (2.11 and 2.12) we need this: two or more positive
instances such that one p-c (or a negation of a p-c) say A> is present
in each instance, but for every other p-c there is an instance in
which it is present and an instance from which it is absent. The
observation described for i.n and 1.12 above would be too weak:
for example, if D were absent from both /x and /2 as we there allowed,
73 would not be eliminated, and we could not conclude that A was
the actual cause.
For the corresponding variant of the method of difference (2.2)
we need this: a positive instance I{ and a negative instance JV^ such
that one p-c (or a negation of a p-c), say A, is present in /t and
absent from JV13 but each other p-c is either present in both 7X and
jVj or absent from both. For example:
A B C D E
A p p a a p
JV^ a p a a p
Since B is present in JV15 B is not sufficient for P in F, but since B
is present in Il} B is not necessary for P in F. Since C is absent from
/!, C is not necessary for P in F, but since C is absent from JVl3 ?? is
not sufficient for P in F. Thus neither B nor C, nor either of their
negations, can be both necessary and sufficient for P in F; D and E
are ruled out similarly, so, given the assumption that some p-c or
negation of a p-c is both necessary and sufficient, A must be so.
This is the classic difference observation described by Mill, in which
the only possibly relevant difference between the instances is the
presence in /t of the factor identified as the actual cause; but we
need this, rather than the weaker observation of 1.2, only when we
allow that the actual cause may be the negation of a p-c.
The joint method needs, along with this weaker assumption, a
similarly strengthened observation: each of the p-cs other than A
306 ELIMINATIVE INDUCTION
must be either present in both a positive instance and a negative
instance or absent from both a positive instance and a negative
instance; this variant (2.3) then still yields the conclusion that A is
both necessary and sufficient for P in F.
We consider next an assumption of the third kind, that the actual
cause is either a p-c or a conjunction of p-cs. This latter possibility
seems to be at least part of the complication Mill described as an
intermixture of effects. This possibility does not affect the positive
method of agreement, since if a conjunction is necessary, each of its
conjuncts is necessary, and candidates can therefore be eliminated
as before. But since the conjuncts in a sufficient (or necessary and
sufficient) condition may not be severally sufficient, the negative
method of agreement is seriously affected. The observation described
and exemplified above for 1.13 and 1,14 would now leave it open
that, say, BC rather than A was the required sufficient (or necessary
and sufficient) condition, for if C were absent from JV1} B from JV2,
and either from JV3, then BC as a whole might still be sufficient.
We can, indeed, still use this method with a much stronger observa-
tion (3.14), namely a single negative instance JVj from which one
p-c, say A, is absent, but every other p-c is present. This will show
(with our present assumption) that no p-c other than A, and no
conjunction of p-cs that does not contain A, is sufficient for P in F.
But even this does not show that the actual cause is A itself, but
merely that it is either A or a conjunction in which A is a conjunct.
(We may symbolize this by saying that the cause is (A. . .), where
the dots indicate that other conjuncts may form part of the cause,
and the dots are underlined, while A is not, to indicate that A must
appear in the formula for the actual cause, but that other conjuncts
may or may not appear.)
The corresponding variant of the method of difference (3.2) needs
only the same observation as 1.2; but it, too, yields only the weaker
conclusion that (A. . .) is necessary and sufficient for P in F (but
therefore that A itself is necessary, but perhaps not sufficient). For
while in the example given for 1.2 above B, C, D, and E singly are
still eliminated as they were in 1.2, and any conjunctions such as BC
which, being present in Jlt might be necessary, are eliminated because
they are also present in JV\ and hence are not sufficient, a conjunc-
tion such as AB, which contains A, is both present in /x and absent
from JV1} and therefore might be both necessary and sufficient. Thus
this assumption and this observation show only that A is, as Mill
put it, 'the cause, or an indispensable part of the cause'. The full
cause is represented by the formula '(^i_n)'> provided that only p-cs
which are present in 7j can replace the dots.
 ELIMINATIVE INDUCTION 307
The corresponding variant of the joint method (3.3) needs a
single negative instance instead of the set 6"n, for the same reason as
in 3.14, and the cause is identified only as (A. . .).
With an assumption of the fourth kind, that the actual cause is
either a p-c or a disjunction of p-cs, the negative method of agree-
ment (4.13 and 4.14) works as in 1,13 and 1.14, since if a disjunction
is sufficient, each of its disjuncts is so. It is now the positive method
of agreement that suffers. For with the observation given for 1.12
above, the necessary and sufficient condition might be, say, (B or
C); for this disjunction is present in both I1 and I2, though neither
of its disjuncts is present in both. Thus the observation of 1.12 would
leave the result quite undecided. We'need (for 4.12) a much stronger
observation, namely a single positive instance in which one p-c,
say A, is present but from which every other p-c is absent, but even
this shows only that the cause is—with the same interpretation of
the symbols as above—(A or . . .). This assumption (that the cause
may be a disjunction of p-cs) allows the possibility of what Mill
called A plurality of 'causes', each disjunct being a 'cause' in the sense
of being a sufficient condition. What we have just noted is the well-
known point that this possibility undermines the method of agreement.
The method of difference, on the other hand, still survives and
still needs only the observation of 1.2; this eliminates all p-cs other
than A) and all disjunctions that do not contain A, either as being
not sufficient because they are present in JV"t or as not necessary
because they are absent from 1^ The only disjunctions not elimin-
ated are those that are present in /x but absent from JV15 and these
must contain A. Thus this observation, with the present assumption,
still shows that (A or . . .) is necessary and sufficient for P in F;
that is, the actual cause is either A itself or a disjunction one of whose
disjuncts is A and the others are p-cs absent from JVj. Hence A
itself, the differential feature, is sufficient for P in F but may not be
necessary.
The joint method with this assumption (4.3) needs a single positive
instance instead of the set Sit but can still use a set of negative
instances, and it too identifies the cause as (A or . . .).
As the assumptions are relaxed further, the method of agreement
needs stronger and stronger observations. For example in 6.12, a
variant of the positive method with the assumption that there is a
necessary and sufficient condition which may be a p-c, or a negation
of one, or a disjunction of p-cs and/or negations of p-cs, the observa-
tion needed is this: a set St of positive instances such that one p-c,
say A, is present in each, but that for every other possible combina-
tion of the other p-cs and their negations there is an instance in
308 E L I M I N A T I V E INDUCTION
which this combination is present (that is, if there are n other p-cs,
we need 2n different instances). This observation will eliminate every
disjunction that does not contain A (showing it not to be necessary)
and will thus show that (A or . . .) is necessary and sufficient, and
hence that A itself is sufficient, for P in F. A corresponding variant
of the negative method of agreement (5.14) shows that (A. . .) is
necessary and sufficient, and hence A itself necessary, for P in F—
a curious reversal of roles, since in the simplest variants the positive
method of agreement identified a necessary condition and the
negative one a sufficient condition.
The method of difference, however, continues to need only the
observation prescribed for i.2 3 if negations are not admitted, or, if
negations are admitted, that prescribed for 2.2. But the conclusions
become progressively weaker, that is, the cause is less and less com-
pletely specified. By far the most important variant of this method—
indeed it is the most important of all those that deal in agreement
or disagreement or any combination of them—is 8.2. This has an
assumption of the eighth kind, in effect that for some £, £ is neces-
sary and sufficient for P in Fy and % is a condition represented by
some formula in disjunctive normal form all of whose constituents
are taken from the p-cs. With the observation of 2.2, this yields the
conclusion that (A. . . or . . .) is necessary and sufficient for P in F.
For every condition built up in any way from p-cs other than A
will either be present in both /j and JV19 and so not sufficient (because
it is present in N1), or absent from both I1 and N1, and so not
necessary (because it is absent from / x ); also any condition in normal
form in which A occurs only negated will similarly be absent from
/j if it is absent from JV1} so it will either be absent from both, and
hence ruled out as not necessary, or it will be present in JV1} and
hence ruled out as not sufficient. Consequently £ must be identical
with some condition in disjunctive normal form in which A occurs
unnegated, that is, with something covered by the expression
'(A . .or. . .)'. Since each disjunct in such a necessary and sufficient
condition is itself sufficient, this observation, in which the presence
of A in 7j is the only possibly relevant difference between /t and JV1}
shows even with the least rigorous kind of assumption that A is at
least a necessary part of a sufficient condition for P in F (this suffi-
cient condition being (A. . .)), that is, that A is an inus condition
of P in F, or better. (It is this method 8.2 that has been discussed in
Chapter 3; as was said there, the conclusion can also be expressed
thus: For some X and for some ¥ (which may, however, be null),
all F (AX or ¥} are P, and all FP are (AX or T}.}
 ELIMINATIVE INDUCTION 309
The joint method, as an indirect method of difference, ceases to
work once we allow both conjunctions and disjunctions of p-cs as
candidates for the role of actual cause. But what we called the
double method of agreement, which in the simplest variants involved
redundancy, the positive and negative halves merely, as Mill says,
corroborating one another, comes into its own with our eighth,
least rigorous, kind of assumption. In 8.12, as in 6.12, if there are
n p-cs other than A, the set of 2n positive instances with A present
in each but with the other p-cs present and absent in all possible
combinations will show that (A or ,. . .) is necessary and sufficient
and hence that A is sufficient. Similarly in 8,14, as in 5.14, the
corresponding set of 2n negative instances with A absent from each
will show that (A. . .) is necessary and sufficient and hence that A is
necessary. Putting the two observations together, we could conclude
that A is both necessary and sufficient for P in F.
A new method, similar in principle, can be stated as follows (8.4):
if there are n p-cs in all, and we observe 2n instances (positive or
negative) which cover all possible combinations of p-cs and their
negations, then (assuming that some condition somehow built up
out of these p-cs is both necessary and sufficient) the disjunction of
all the conjunctions found in positive instances is both necessary and
sufficient for P in F, For example, if there are only three p-cs, A, B,
and C, and we have the observation set out in the following table:
P A B C
a P P P
P P P a
P p a p
a p a a
a a p p
p a p a
a a a p
a a a a
then (ABC or ABC or ABC) is necessary and sufficient for P in F.
For if these are the only possibly relevant factors, each combination
of p-cs and their negations along with which P occurs at least once
must be sufficient for /*, and each such combination in whose pres-
ence P is absent must be non-sufficient for P; but the disjunction of
all the sufficient conditions must be both necessary and sufficient,
on the assumption that some condition is so.
We find thus that while we must recognize very different variants
of these methods according to the different kinds of assumptions
used, and while the reasoning which validates the simplest variants
3io ELIMINATIVE INDUCTION
fails when it is allowed that the actual cause may be constituted by
negations, conjunctions and disjunctions of p-cs combined in various
ways, nevertheless there are valid demonstrative methods which use
even the least rigorous kind of assumption, that is} which assume
only that there is some necessary and sufficient condition for P in F,
made up in some way from a certain restricted set of p-cs. But with
an assumption of this kind we must be content either to extract (by
8.2) a very incomplete conclusion from the classic difference observa-
tion or to get more complete conclusions (by 8.12, 8.14, the com-
bination of these two, or 8.4) only from a large number of diverse
instances in which the p-cs are present or absent in systematically
varied ways.
There are two very important extensions of these methods, which
consist in the relaxing of restrictions which have been imposed for
the sake of clarity in exposition. First, since in every case the
demonstration proceeds by eliminating certain candidates, it makes
no difference if what survives, what is not eliminated, is not a single
p-c but a cluster of p-cs which in the observed instances are always
present or absent together: the conclusion in each case will be as
stated above but with a symbol for the cluster replacing 1A\ For
example, if in 2,2 we have, say, both A and B present in /L and
both A and B absent from N1, but each other p-c either present in
both or absent from both, it follows that the cluster (A, B) is the
cause in the sense that the actual cause lies somewhere within this
cluster. Given that this- duster will either be present as a whole or absent
as a whole, its presence is necessary and sufficient for P in F, A similar
observation in 8.2 would show that, subject to the same proviso,
this cluster is an inus condition of P in F, or better, and hence that
either A or B is an inus condition, but perhaps each is an inus con-
dition and perhaps their conjunction (AB) is so. Secondly, in order
to eliminate candidates it is not really necessary to list them first.
The observation required for 2.2 or 8.2, for instance, is only that a
certain p-c (or, as we have just seen, a certain cluster of p-cs) should
be present in /t but absent from JV^ while in every other respect
that might be causally relevant these two instances are alike. Or for
8.12 we need only observe that A is present throughout a set of
diverse positive instances while all the other factors that might be
causally relevant occur, and fail to occur, in all possible combina-
tions in these instances. To be sure, we could check this conclusively
only if we had somehow counted and listed the possibly relevant
factors, but without this we could have a mass of evidence that at
least seemed to approximate to the required observation.
 ELIMINATIVE INDUCTION 311

4. The Method of Residues

We have, in the wide range of variants so far indicated, covered
only three of Mill's five methods. His method of residues needs only
brief treatment: it can be interpreted as a variant of the method of
difference in which the negative instance is not observed but con-
structed on the basis of already known causal laws.
Suppose, for example, that a positive instance /x has been observed
as follows:
A B C D E
Ii P P a P a
Then if we had, to combine with this, a negative instance in which
B and D were present and from which At C, and E were absent, we
could infer (according to the kind of assumption made) either by
2.2 that A was necessary and sufficient for P in F, or by 8.2 that
(A. . . or. . .) was so} and so on. But if previous inductive inquiries
(of whatever sort) can be taken to have established laws from which
it follows that given ABCDE in the field F, P would not result, there
is no need to observe JV^; we already know all that an observation of
JVj could tell us, and so one of the above-mentioned conclusions
follows from I± alone along with the appropriate assumption.
Again, if the effect or phenomenon in which we are interested
can be measured, we can reason as follows. Suppose that we observe
a positive instance /1} with the factors present and absent as above,
in which there occurs a quantity A;X of the effect in question, and
suppose that our previously-established laws enable us to calculate
that given ABCDE in F there would be a quantity #2 of this effect;
then we can regard the difference (xl— xz) as the phenomenon P
which is present in 7t but absent from (the calculated) JV^. With an
assumption of the first, second, fourth, or sixth kind—that is, any
assumption which does not allow conjunctive terms in the cause—
we could conclude that the cause of P in this instance 1^ was A alone,
and hence that A is sufficient for the differential quantity (#1—-#2)
inF.
To make an assumption of any of these four kinds is to assume
that effects of whatever factors are actually relevant are merely
additive, and that is why we can conclude that the extra factor in Ilt
namely A, itself produces the extra effect (xl—x2). But with an
assumption of the third, fifth, seventh, or eighth kind, which allows
conjunctive causes and hence Mill's intermixture of effects, we could
conclude only that a (sufficient) cause of (xl—#2) was (A. . .). Given
the other factors that were present in both /j and JVlf A was
3ia E L I M I N A T I V E INDUCTION
sufficient for this differential effect; but it does not follow that
A is sufficient for this in relation to F as a whole.
Though Mill does not say so, it is obvious that such a use of
constructed, calculated, instances is in principle possible with all the
methods, not only with the method of difference in the way outlined
here.

5. Methods of Concomitant Variation

Whereas Mill called this just one method, there is in fact a system
of concomitant variation methods mirroring the various presence-
and-absence methods we have been studying. These too will be
forms of ampliative induction: we shall argue from a covariation
observed in some cases or over some limited period to a general rule
of covariation that covers unobserved instances as well. These
methods, then, work with a concept of cause which makes the (full)
cause of some quantitative phenomenon that on which the magnitude
of this phenomenon functionally depends: causation is now the con-
verse of dependent covariation. As I have argued in Chapter 6, this
functional dependence concept is best regarded as a development
and refinement of 'neolithic' concepts of causation defined in terms
of necessity and (perhaps) sufficiency, that is, in terms of conditional
presences and absences of features. To let it count as causation,
functional dependence requires, of course, the addition of the rela-
tion of causal priority. This indeed plays some part in the formal
methods that we are studying in so far as it is used in deciding what
are possibly relevant causal factors (that is, what can count as p-cs)
and in interpreting a covariation as a directed causal relation, but
we can take these points for granted and leave them aside in most
of the analysis.
The typical form of a functional causal regularity will then be
that the magnitude of P in the field F is always such-and-such a
function of the magnitude of the factors, say, A, B} C, and D, which
we can write:
Pf=f(A,B,C,D).
Where such a regularity holds, we may call the whole right hand
side of this equation, that is, the set of actually relevant factors
together with the function/1 which determines how they are relevant,
the full cause of P in F, while we may call each of the actually relevant
factors a partial cause. The full cause is that on which the magnitude
of P in F wholly depends; a partial cause is something on whose
magnitude the magnitude of P in F partly depends.
A thorough investigation of such a functional dependence would
 E L I M I N A T I V E INDUCTION 313
involve two tasks, the identifying of the various partial causes and
the determination of the function f. Only the first of these two tasks
can be performed by concomitant variation methods analogous to
those already surveyed.
There are concomitant variation analogues of both the method of
agreement and the method of difference; that is, there are ways of
arguing to a dependence of P on, say, A both from the observation
of cases where P and A remain constant while other possibly relevant
factors vary and from the observation of cases where P and A vary
while other possibly relevant factors remain constant.
As before, we need an assumption as well as an observation, but
we have a choice between different kinds of assumption. The more
rigorous kind (corresponding to the second kind of assumption
above) would be that in F the magnitude of P wholly depends in
some way on the magnitude of X, where X is identical with just one
of the p-cs. The less rigorous and in general more plausible kind
(corresponding to the eighth kind of assumption above) would be
that in F the magnitude of P wholly depends in some way on the
magnitudes of one or more factors X, X', X", etc., where each of
these is identical with one of the p-cs.
The simpler covariance analogue of the method of difference com-
bines the more rigorous assumption with this observation: over some
period or over some range of instances in F, P has varied while A has
varied but all the other p-cs have remained constant. None of the
other p-cs, then, can be identical with the X in the assumption, so
A must be so. That is, the conclusion is that the magnitude of P in F
depends wholly in some way not jet determined on that of A.
The more complex covariance difference analogue combines this
same observation with the less rigorous kind of assumption above.
The observation does not now show that some p-c other than A,
say 2?, is not a partial cause. But it does show that the magnitude of
P in F cannot depend wholly, in any way, on any set of factors that
does not include A, for every function of every such set has remained
constant while P has varied. This ensures, therefore, that A is a
partial cause, but leaves it open whether the full cause is simply of
the form/ (A) or whether there are other partial causes as well. This
observation and this assumption, therefore, show that a full cause of
P in Fhf(A, • • •). Repeated applications of this method, with other
factors being varied one at a time, could fill in further partial causes,
but could not close the list.
The simple covariance analogue of the method of agreement com-
bines our more rigorous kind of assumption above with this observa-
tion: over some period or range of instances in F, P has remained
3H ELIMINATIVE INDUCTION
constant while A has remained constant while every other p-c has
varied. We want to argue here that since B, say, has varied while P
has not, B cannot be identical with the assumed Xon whose magnitude
that of P depends. But this does not follow. It might be that .P's
magnitude varied with and in sole dependence upon the magnitude
of B, and yet that P was not responsive to every change in the mag-
nitude of B: there might be flat stretches, plateaux, in the curve for
PF =f(B}) with values for B plotted on the #-axis and values for PF
on the j>-axis. So to eliminate B we must either strengthen the
assumption by adding that P is responsive to every change in the
magnitude of X, or strengthen the observation to include every
possible degree of variation of every p-c other than A, or strengthen
both assumption and observation together, so that we can say that
every p-c has varied to a degree to which, if this p-c were our X,
P would be responsive. Only in some such way could we validly
reach the conclusion that the magnitude of P in F depends in some
way on that of A alone.
A complex covariance analogue of the method of agreement'en-
counters much the same problem as complex variants of the original
method. Even if P has remained constant while all the p-cs other
than A have varied, this does not, with our less rigorous assumption,
exclude the possibility that PF —f(B, (7), say. For it might be that
the actual variations of B and C and the actual function/ were such
that the effects of these variations cancelled out:/might be such
as to allow the actual changes in B and C to compensate for one
another. It will not help now to strengthen the assumption to include
the claim that P is responsive to all variations in the X$. But will it
do if we greatly strengthen the observation, to claim that all the
p-cs other than A have varied independently (and hence in all
possible combinations) over the whole possible range or magnitude
of each? Even this claim—which is in any case so extreme that we
could not hope to do more than approximate to its fulfilment—
would yield only the conclusion that the magnitude of P in F does
not depend wholly on p-cs other than A, that is, that A is at least a
partial cause. It does not, as we might be tempted to suppose, show
that all p-cs other than A are irrelevant, and that P depends only
upon A. For it might be that at some values of A changes in the values
of all the other p-cs were ineffectual, and yet that at other values of
A some variations in some of the other p-cs made a difference.
Consequently, if with an assumption of our less rigorous kind we are
to- be able to close the list of partial causes, nothing less than an
analogue of 8.4 will do. That is, we must be able to say that A, as
well as all the other p-cs, has joined in this mutually independent
 ELIMINATIVE I N D U C T I O N 315
variation over the whole possible range of each. And then if the
value for P has remained constant for each constant value of A, we
can conclude that P's magnitude depends on that of A alone.
As I have said, these concomitant variation methods only identify
partial causes of the functional dependence sort, and may in an
extreme case tend to close the list of partial causes: they do not in
themselves help with the task of determining the function /. But
there is another device, stiH within the range of eliminative induc-
tion methods, which at least makes a start on this task. As we saw
when discussing the method of residues, we can regard a quantitative
difference in some magnitude, something of the form (x^~x2), as
the effect or phenomenon whose cause is identified by some applica-
tion of the method of difference. Various observations interpreted in
this way will ascribe such-and-such a change in the quantity of P to
such-and-such a change in the magnitude of, say, A at certain
(temporarily fixed) values of the other p-cs, and will so at least
impose constraints on the function /. That is, even when we are
dealing with functional dependences there is an important place
for applications of the neolithic presence-and-absence method of
difference.

6. Uses and Applications of these Methods

I have so far been concerned to show only that there are many
demonstratively valid methods of these sorts. But in reaching a more
exact formulation of them for this purpose I have incidentally re-
moved some of the more obvious objections to the view that such
methods can be and are applied in practice. Thus the introduction
of the notion of a field gives these methods the more modest task of
finding the cause of a phenomenon only in relation to some field,
rather than that of finding a condition which is absolutely necessary
and sufficient. By contrasting the p-cs with the field we have freed
the user of the method of agreement from having to make the
implausible claim that his instances have only one 'circumstance' in
common. He has merely to claim that they have only one of the
p-cs in common, while conceding that whatever features constitute
the field or are constant throughout the field will belong to all the
instances, and that there may be other common features too, but
ones that the user has initially judged not to be possibly relevant to
this phenomenon.
Similarly, the user of the method of difference has only to claim
that no possibly relevant feature other than the one he is picking out
as (part of) the cause is present in 1^ but not in JVlt Also, we have
taken explicit account of the ways in which the possibilities of
316 ELIMINATIVE INDUCTION
counteracting causes, a plurality of causes, an intermixture of effects,
and so on, need to be allowed for, and we have seen how valid
conclusions can still be drawn in the face of these complications,
provided that we note explicitly the incompleteness of the con-
clusions we can now draw or the much greater strength of the
observations required for complete conclusions (for example in 8,4).
By making explicit the need for assumptions we have abandoned
any pretence that these methods in themselves solve or remove the
problem of induction. If the requisite observations can be made, the
ultimate justification for any conclusion reached by one of these
methods will depend on the justification for the assumption used,
and since this proposition is general in form, any reliance that we
place on it or on its consequences will have to be backed by some
other kind of inductive reasoning or confirmation or corroboration.
The eliminative methods cannot be the whole of the logic of scientific
discovery. But it does not follow that they cannot be any part of this.
It is, I think, instructive to see just how far 'deduction from the
phenomena1 (in Newton's phrase) can go, in the light of some not
implausible assumptions.
It is worth stressing the precise form of assumption that our
analysis has shown to be required. It is not a general uniformity of
nature, a universal determinism, but merely that this particular
phenomenon, in this particular field, should have some cause. But
we have also found that whereas our ordinary concept of cause
requires a cause only to be necessary in the circumstances, the assump-
tion needed for almost all our methods is that something (though this
will usually not be the factor that is eventually picked out as, in a
popular sense, the cause) should be both necessary and sufficient for
this phenomenon in this field. We do have to assume what might be
called particular determinism. We must reject the view of Mill that
there is some one 'law of causality* to which every application of
these methods appeals. But at the other extreme we should stop
short of the view of Wittgenstein in the Tractatus that this "law" dis-
solves into the mere logical form of laws in general. Specific deter-
ministic assumptions play a part that makes them in at least one
way prior to the laws discovered with their help.
What may seem more of a problem is that we have to assume a
somehow restricted range of possibly relevant factors, p-cs. But, as
we shall see, some inquiries are conducted against a background of
general knowledge that supplies such a restriction; in others we
simply assume that the only items that are possible causes for some
result, some change, are other perceptible changes in the spatial
neighbourhood and not long before. And what counts as near
 E L I M I N A T I V E INDUCTION 317
enough in space and time, and as being perceptible, can itself be a
matter of working hypotheses. If a conclusion demonstrated with
the help of some particular limitation of p-cs is subsequently falsified,
that limitation can and must be relaxed: we must look further afield
or try other ways of detecting changes that we did not at first per-
ceive. It must be emphasized that though the eliminative methods
are themselves demonstrative, there is no hostility whatever between
the use of them and the use of hypotheses whether as working
assumptions or otherwise. The demonstrative methods need not be
associated with any theory that the method of inquiry as a whole is
watertight, or simply aggregative, or mechanical.
The eliminative methods are in fact constantly in use in the sense
that innumerable procedures of search, discovery, and checking
have implicit within them the forms of reasoning that I have tried
to formulate and make precise. I have argued in Chapter 3 that all
our basic causal knowledge is founded on what are in effect applica-
tions of the method of difference, and that this knowledge has, in
consequence, just the form (of elliptical double regularities) that one
would therefore expect it to have. It is by before-and-after contrast
observations that we know that fire burns, that it cooks meat, that
rattlesnakes are poisonous; Becquerel discovered that the radium he
carried in a bottle in his pocket caused a burn by noticing that the
presence of the radium was the only possibly relevant difference
between the time when the inflammation developed and the earlier
time when it did not, or between the part of his skin where the
inflammation appeared and other parts.
But the causal relations revealed in this way need not be so
general as to be of scientific interest. Fault-finding procedures aim
only at discovering why this particular machine does something
wrong, and hence how it can be put right; but they, too, depend on
trying one change after another and seeing what results.
Suppose that a new drug is being tested. It is administered to
some subject, and some change (good or bad) is noticed in the subject
soon afterwards. There is a prima facie case for supposing that the
administration of this drug can cause—that is, is an inus condition
of—that change. But why, if the method of difference is a demonstra-
tive method, is it only a prima facie case? Simply because the
experimenter cannot be sure that the requirements for that method's
observation have been met: some other relevant change may have
occurred at about the same time. But if the experiment is repeated
and we keep on getting the same result, it becomes less and less
likely that on each occasion when the drug was administered, some
one other, unnoticed but relevant, change also occurred. But (as I
3i8 E L I M I N A T I V E INDUCTION
said in Chapter 3) such repetition should not be taken as a resort to
the method of agreement, it is not the repetition of the A-P sequence
as such that we are relying on: we are using the repetition merely
to check whether on each occasion the requirements for the method
of difference were met. If we keep getting the same result, we can
say that it is probable that they were met, that no other relevant
change occurred, and therefore (given the assumption) that the
conclusion demonstrated by the method of difference holds. But
repetition can have another function as well. Even if we are reason-
ably sure that a particular experiment (or set of experiments) has
shown that A is an inus condition of P in F (or better), we may
repeat it under varying conditions in the hope—which may be ful-
filled—of showing that various factors which were present in our
original /x and JV^, and so might be essential conjuncts with A in the
full cause (A. . . or . . .), are not in fact essential: we are reducing
the indeterminacy represented by the first row of dots in this formula.
Such repetition under varying conditions is an application of the
method of agreement, and turns the whole procedure into some-
thing that can be called (in yet another sense) a joint method: it is
in fact an approximation to 8.4.
As I have said in Chapter 3, the controlled experiment, in which
a control case is deliberately set up alongside the experimental case
and made to match it as closely as possible except with regard to
the item under test, is another well-known application of the method
of difference.
It is often supposed, and the examples I have given might en-
courage the belief, that these methods operate only at superficial
levels of science, that they establish causal connections only between
directly perceivable features. But they are subject to no such restric-
tion. A physicist may take himself to be firing deuterons at bismuth
atoms when all that a superficial observer could perceive was his
twiddling knobs and switches on a large and impressive piece of
apparatus; but if the physicist subsequently finds evidence that some
of the former bismuth atoms are now behaving like polonium atoms,
it will be method of difference reasoning that enables him to say
that deuteron bombardment of bismuth can produce polonium. Of
course in such applications the conclusion reached by a method
which is in itself demonstrative is established only subject to the
proviso that the interpretation of the antecedent conditions and of
the result is correct. But once we have detached the use of these
methods from the claim that they achieve certainty, this is quite in
order. It is also very obvious that the eliminative methods can be
used, not only as part of a mere response to experiences that force
 ELIMINATIVE INDUCTION 319
themselves upon us, but also in procedures for testing hypotheses,
for finding answers to questions that we ask,
Mill thought that these were methods both of discovery and of
proof; and so they can be, but not very easily at once. Rather an
observation which appears to conform to the requirements for one
of the methods may suggest that a certain causal relationship holds,
and a causal hypothesis—whether suggested in this way or in some
other—may be tested and perhaps confirmed by a more thorough
survey or by a carefully constructed experiment, the results of each
being interpreted by another application of the appropriate method.
The method of agreement is frequently used in medical contexts-
locating a source of infection, identifying the substance to which
someone has an allergic reaction, deciding what bacillus is responsible
for a certain disease, and so on. It is particularly in such contexts
that a previously developed background of theory supplies the fairly
rigorous kind of assumption needed for the simpler variants of this
method. In a country in which typhoid is rare, we may hope that
if a dozen cases occur within a few days in one town they all result
from a single source of infection, A person may be allergic to just
one substance, and his response to it may not depend much on
other circumstances. But in these applications too the eliminative
method of reasoning may well be combined with asking questions,
framing and testing hypotheses, rather than the passive reception of
information.
The methods of concomitant variation, particularly those that are
counterparts of the method of difference, along with statistical pro-
cedures that can be regarded as elaborations of these methods, are
constantly used when one or more factors are varied while other
possibly relevant factors are held constant, and calculations are
based on the observed results. In practice the two tasks which I
distinguished, the identifying of actual partial causes of a pheno-
menon and the determination of the function f, tend to be mixed
up together; but the separation of these tasks in our analysis may
help to distinguish different degrees of collusiveness with which
different parts of an hypothesis have been confirmed.

7, Criticisms of these Methods

We have already met most of the stock objections to these methods.

One still outstanding is that they take for granted the most important
aspect of inquiry, the detection, isolation, and analysis of the possibly
relevant factors. Mill, of course, does not ignore this: he devotes a
whole book to 'operations subsidiary to induction'. But what most
320 ELIMINATIVE INDUCTION
needs to be stressed is that there is no need for a finally satisfactory
analysis of factors before the eliminative methods can be applied:
as I argued in Chapter 3, we can start using the methods with a
very rough distinction of factors and obtain some correct results,
which can be later refined by what I called the progressive localiza-
tion of a cause. Also, though classification may be subsidiary to
induction, it is not purely preliminary to it. Things are classified
mainly by their causal properties; it will be by using, in effect, the
method of difference that we find that this material regularly does
such-and-such to that other material, and so recognize it, and other
samples that act. similarly, as all specimens of some one kind of
stuff. A mass of elementary eliminative inductions helps to provide
the setting for more sophisticated ones.
Another stock criticism is that causal relations are not the whole,
or even the most important, concern of science. (We have already
rejected, in Chapter 6, the argument that science is not concerned
with causation at all.) We can concede that science has other con-
cerns as well, but this criticism is otherwise to be rebutted by
insisting on a wide concept of causation and by seeing that these
methods assist the discovery of causes in this wide sense. Especially
since entities on all levels are identified largely by dispositional pro-
perties, that is, by their causal powers, the detection of causation by
particular objects as well as the discovery of causal regularities runs
through all processes of discovery.
Thirdly, it is often thought that the use of and reliance on these
methods is somehow a discredited and discreditable alternative to
the hypothetico-deductive method. In fact there is no incom-
patibility, hostility, or even any necessary difference of spirit between
the use of these two procedures. Though the eliminative methods are
in themselves demonstrative, they are naturally and unavoidably
used in contexts which make their conclusions only tentative. Their
assumptions require to be confirmed in ways in which these methods
cannot themselves confirm them. And the Popperian principle that
hypotheses are corroborated by being subjected to severe tests—
that is, ones which, if an hypothesis is false, are likely to show it to
be false—is admirably illustrated by the methods of agreement and
difference. The hypothesis that A is necessary for P in F is severely
tested by looking for instances of P which fall within the field F but
otherwise differ widely in all conceivably relevant ways, and seeing
whether A is present in them all; if the hypothesis stands up to this
test, it is by the method of agreement that we draw the (in practice
still tentative) conclusion. The hypothesis that A is an indispensable
part of a sufficient condition for P in F that was present in /x is
 ELIMINATJVE INDUCTION 321
severely tested by finding or constructing an instance that is as like
/! as possible in all conceivably relevant ways, except that it lacks
A, and seeing whether P occurs there or not. And if it does not, this
counts as the jVj from which, along with Ilt we draw our conclusion
in accordance with the method of difference.
 Additional Notes
{These notes are indicated in the text by an asterisk)

p. 80. I offer a solution in 'A Defence of Induction', in Epistemotogy in

Perspective: Essays in Honour of A, J. Ayer, edited by G. Macdonald
(Macmillan, forthcoming).

p. 143. But Hume was anticipated by Hobbes and (in a slightly different
kind of compatibilism) by various theologians. See 'Compatibilism,
Free Will and God' by A. G. N. Flew in Philosophy, xlviii (1973).

p. 190. The analysis suggested in this paragraph needs revision. See the
Preface to the Paperback Edition, pages viii—xvi.

p. 214. This book has been republished as Laws, Modalities, and Counter-
factuals, with a foreword by Wesley C. Salmon (University of
California Press, 1976).

p. 238. An interesting and important class of IRLs is found by ethologists

investigating 'evolutionary stable strategies' when there is a mix-
ture of strategies in equilibrium with one another or a single
corresponding mixed strategy. See, e.g., R. Dawkins, The Selfish
Gene (Oxford, 1976), esp. Ch. 5.
 Bibliography
Including only works mentioned in the text, and showing the editions referred to
or quoted

ANDERSON, J. Studies in Empirical Philosophy, Sydney, 1962.

ANSCOMBE, G. E. M. Causality and Determination. Cambridge, 1971.
'Causality and Extensionality.' In Journal of Philosophy, Ixvi (1969),
152-9.
ANSCOMBE, G. E. M., and GEACH, P. T. Three Philosophers. Oxford, 1961.
AVER, A. J. TTie Problem of Knowledge. Harmondsworth, 1956.
Probability and Evidence. London, 1972.
BENNETT, J. F, Kant's Analytic. Cambridge, 1966.
BERKELEY, G. The Principles of Human Knowledge. In The Works of George
Berkeley, ed. A. C. Fraser, Oxford, 1901.
BLACK, M, 'Why Cannot an Effect Precede its Cause?' In Analysis, xvi
(i955~6), 49~58.
Models and Metaphors. Ithaca (N.Y.), 1966.
BROAD, C. D. The Mind and its Place in Nature. London, 1925.
'The Principles of Demonstrative Induction.' In Mind, xxxix (1930),
302-17 and 426-39.
BUNGE, M. Causality. Harvard, 1959.
CHISHOLM, R. M. 'Freedom and Action.' In Freedom and Determinism, ed.
K. Lehrer, New York, 1966,
CHISHOLM, R. M., and TAYLOR, R. 'Making Things to have Happened.'
In Analysis, xx (1959-60), 73-8.
CUMMINS, R.,, and GOTTLIEB, D, 'On an Argument for Truth-functionality."
In American Philosophical Quarterly, ix (1972), 265-9.
D'ALEMBERT, J. Traite de dynamique. Paris, 1921.
DAVENEY, T. F. 'Intentions and Causes.* In Analysis, xxvii (1966-7), 23.
DAVIDSON, D. 'Actions, Reasons, and Causes.' In Journal of Philosophy, IK
(1963), 685-700.
'Causal Relations.* In Journal of Philosophy, Ixiv (1967), 691-703.
DESCARTES, R. Principles of Philosophy. In Descartes: Philosophical Writings, tr,
G. E, M. Anscombe and P. T. Geach, London, 1969.
Rules for the Guidance of our Mental Powers. In Descartes' Philosophical
Writings, tr. N. K, Smith, London, 1952.
DRAY, W. 'Taylor and Chisholm on Making Things to have Happened.*
In Analysis, xx (1959-60), 79-82.
DUCASSE, C. J. Nature, Mind, and Death. La Salle, 1951.
Truth, Knowledge, and Causation. London, 1968.
DUMMETT, M. 'Can an Effect Precede its Cause?1 In Proceedings of the
Aristotelian Society, supplementary volume xxviii (1954), 27-44.
'Bringing about the Past.' In Philosophical Review, Ixxiii (1964), 338-59.
324 BIBLIOGRAPHY
EWING, A. C. Idealism. London, 1934,.
FITZGERALD, E. Rubdiydt of Omar Khayyam, Ed. G. F. Maine, London, 1953.
FLEW, A. 'Can an Effect Precede its Cause?* In Proceedings of the Aristotelian
Society, supplementary volume xxviii (1954), 45-62.
'Effects before their Causes?—Addenda and Corrigenda.' In Analysis,
xvi (1955-6), 104-10.
'Causal Disorder Again,' In Analysis, xvii (1956-7), 81-6.
GALE, R. M. *Why a Cause Cannot be Later than its Effect.' In Review of
Metaphysics, xix (1965), 209-34.
'Professor Ducasse on Determinism.' In Philosophy and Phenomenological
Research, xxii (1961), 92-6.
GARDINER, P. The Nature of Historical Explanation. Oxford, 1952.
GODEL, K. 'Russell's Mathematical Logic.' In The Philosophy of Bertrand
Russell, ed. P. A. Schilpp, Evanston (111.), 1944.
GOROVITZ, S. 'Leaving the Past Alone.* In Philosophical Review, Ixxiii (1964),
360-71.
GRUNBAUM, A. Philosophical Problems of Space and Time. New York, 1963.
'Temporally-asymmetric Principles, etc.* In Philosophy of Science, xxix
(1962), 146-70.
HARRE, H. R. The Principles of Scientific Thinking. London, 1970.
The Philosophies of Science. Oxford, 1972.
HART, H. L. A., and HONORE, A. M. Causation in the Law. Oxford, 1959.
HOME, D, A Treatise of Human Nature. Ed. L. A. Selby-Bigge, Oxford, 1888.
Enquiries. Ed. L. A. Selby-Bigge, 2nd edn., Oxford, 1902.
An Abstract of a Treatise of Human Nature, Ed. J. M. Keynes and P.
Sraffa, Cambridge, 1938.
JOHNSON, W. E. Logic. Cambridge, 1921,
KANT, I. Critique of Pure Reason. Tr. N. K. Smith, London, 1929.
Prolegomena to Any Future Metaphysic, Tr. P. G. Lucas, Manchester,
1953-
KEYNES, J. N. Formal Logic, 4th edn., London, 1906.
KIM, J. 'Causes and Events: Mackie on Causation.' In Journal of Philosophy,
Ixviii (1971), 426"4!•
KNEALE, W. C. Probability and Induction. Oxford, 1949.
'Natural Laws and Contrary to Fact Conditionals.' In Philosophy and
Analysis, ed. M. Macdonald, Oxford, 1954.
LOCKE, J. An Essay Concerning Human Understanding. Ed. A. C. Fraser,
Oxford, 1894.
LYON, A. 'Causality.' In British Journal for the Philosophy of Science, xviii
(1967), 1-20.
MACKIE, J. L. 'Responsibility and Language.' In Australasian Journal of
Philosophy, xxxiii (1955), 143-59-
'Counterfactuals and Causal Laws.' In Analytical Philosophy, ist ser.,
ed. R. J. Butler, Oxford, 1962.
'Causes and Conditions.' In American Philosophical Quarterly, ii (1965),
245-64.
'The Direction of Causation,' In Philosophical Review, Ixxv (1966),
441-66.
 BIBLIOGRAPHY 325
— 'Mill's Methods of Induction.' In Encyclopedia of Philosophy, ed.
P. Edwards, New York, 1967,
Truth, Probability, and Paradox, Oxford, 1973.
Review of The Matter of Chance by D. H. Mellor. In Philosophical
Quarterly, xxiii (1973), 85-7.
'Practical versus Theoretical Reason,' In Practical Reason, ed. S.
Korner, Oxford, 1974.
MCLAUGHLIN, J. A. 'Proximate Cause.' In Harvard Law Review, xxxix
(1925-6), 149 ff.
MARC-WOGAU, K. 'On Historical Explanation.' In Theoria, xxviii (1962),
213-33-
MARTIN, R. 'The Sufficiency Thesis.1 In Philosophical Studies, xxiii (1973),
205-ii.
MELDEN, A. I. Free Action. London, 1961.
MICHOTTE, A. La perception de la causalite. 2e e"d, Louvain, 1954.
MILL, J. S. A System of Logic. 8th edn. (reprinted), London, 1941.
MORTON, A. 'Extensional and Non-truth-functional Contexts.' In Journal
of Philosophy, Ixvi (1969), 159-64.
N AGEL, E. The Structure of Science. London, 1961.
NERLICH, G. C., and SUGHTINO, W, A. 'Popper on Law and Natural
Necessity.' In British Journal for the Philosophy of Science, xviii (1967),
233-5-
NEWTON, I. Philosophiae Naturalis Principia Mathematica, London, 1686.
NOBLE, D. 'Charles Taylor on Teleological Explanation.' Analysis,
xxvii (1966-7), 96-103.
PEARS, D, F. 'The Priority of Causes.' In Analysis, xvii (1956-7), 54-63.
POPPER, K. R. 'The Arrow of Time.' In Nature, clxxvii (1956), 538, also in
Nature, clxxviii (1957), 382, and Nature, clxxix (1958), 1297.
The Logic of Scientific Discovery. 6th impression, London, 1972.
Objective Knowledge. Oxford, 1972.
QUINE, W. V. The Ways of Paradox, New York, 1966.
REICHENBACH, H. JVomological Statements and Admissible Operations. Amster-
dam, 1954.
The Direction of Time. Berkeley, 1956.
ROBINSON, J. A. 'Hume's Two Definitions of "Cause".' In Plume, ed. V. C.
Chappell, London, 1968.
RUSSELL, B. A. W. Mysticism and Logic. New York, 1918.
The Analysis of Mind. London, 1921.
An Inquiry into Meaning and Truth. London, 1940.
'Reply to Criticisms.' In The Philosophy of Bertrand Russell, ed. P. A.
Schilpp. Evanston (111.), 1944.
SCHLIGK, M, 'Causality in Everyday Life and Recent Science.' In Readings
in Philosophical Analysis, ed. H. Feigl and W, Sellars, New York, 1949.
SCRIVEN, M. 'Randomness and the Causal Order.' In Analysis, xvii (1956-
7). 5-9-
— Review of The Structure of Science by E. Nagel. In Review of Metaphysics,
xviii (1964), 403-24.
'The Logic of Cause.' In Theory and Decision, ii (1971)) 49-66.
326 BIBLIOGRAPHY
SHARVY, R. 'Truth-Functionality and Referential Opacity.5 In Philoso-
phical Studies, xxi (1970), 5-9.
SMULLYAN, A. 'Modality and Description.* Journal of Symbolic Logic, xiii
(i94s). 31-7-
SPRIGGE, T. L. S. 'Final Causes.' In Proceedings of the Aristotelian Society,
supplementary volume xlv (1971), I49~7°-
STOUTLAND, F. M. 'The Logical Connection Argument.' In American
Philosophical Quarterly Monograph, iv (1970), 117-30.
STOVE, D. C. 'Hume, Probability, and Induction.5 In Hume, ed. V. C.
Chappellj London, 1968.
Probability and Hume's Inductive Scepticism. Oxford, 1973.
STRAWSON, P. F. The Bounds of Sense. London, 1966.
SUCHTING, W. A. 'Popper's Revised Definition of Natural Necessity.' In
British Journal for the Philosophy of Science, xx (1969), 349-52.
SUPPES, P, A Probabilistic Tlieory of Causality. Amsterdam, 1970.
TAYLOR, C. The Explanation of Behaviour. London, 1964.
'Teleological Explanation—a Reply to Denis Noble.' In Analysis,
xxvii (1966-7), 141-3.
VENDLER, Z. 'Effects, Results, and Consequences.' In Analytical Philosophy,
ist ser., ed. R. J. Butler, Oxford, 1962.
'Causal Relations.' In Journal of Philosophy, Ixiv (1967), 704-13.
VON WRIGHT, G. H, A Treatise on Induction and Probability. London, 1951.
Explanation and Understanding. London, 1971.

Additional Bibliography
DAWKINS, R. The Selfish Gene. Oxford, 1976.
MAC&IE, J. L. 'A Defence of Induction.* In Epistemology in Perspective: Essays
in Honour ofA.J. Ayer, ed. G. Macdonald, London, forthcoming.
Ethics: Inventing Right and Wrong. Harmondsworth, 1977.
'Mind, Brain, and Causation.' In Midwest Studies in Philosophy, iv
(1979). J 9-29-
Problems from Locke. Oxford, 1976.
'Self-Refutation: A Formal Analysis.' In Philosophical Quarterly, xiv
(1964)* 193-203.
'The Transitivity of Counterfactuals and Causation,' In Analysis,
forthcoming.
 Index
Accidental regularities, 83, 196-914 Comte, A., 207
Action, 170-2 Conditionals, counterfactual, 30-4, 53-
Anderson, J., 35 n., 299 n. 4» '33j 140-2, 167, 169, 197-203,
Anscombe, G. E. M., 48, 133 n., 230, 224,229-30
248 n., 249 n., 251-2, 254-5 Contiguity, 19, 135-8, 223
A priori concepts and principles, 91-2 Continuity:
Aquinas, 76 qualitative and structural, 142, 217-
Asymmetry, causal—see Direction of 30. 247
causation spatio-temporal, see Contiguity
Ayer, A. J., 16, 54 n., 162 n., 226 Continuous processes, 155-9, 218-23
Cummins, R., 251, 252 n., 253 n., 254
Bennett, J. F., 91, 93, 94-7, 99, 102,
107-8 d'AJembertjJ., 218-20
Berkeley, G., 95 Daveney, T. F., 25 n.
Black, M., 162 n., 178 n. Davidson, D., 248 n., 249 n., 251 n.,
Broad, G. D., 81 n. 256, 258, 267 n,, 286 n.
Bunge, M., 154, 232 Descartes, R., 12, 209
Determinism, 40-3, 103, n r , 191-2,
Carnap, R., 15 JQ®, 231, 316
Causal field, 35, 63, 145, 261, 300 Direction of causation, 51-3, 56-7,
Causal mechanism, 86, 217, 222-3 85-7> 90. ^9, 160-92
Causa sine qua non, 127-31 Direction of explanation, 164, 184-6,
Causation: 189-90
account of, summarized, 270-2 Direction of time, 162-4, '83-4, 192
backward,161-2,172,173-89, 274-5 Dispersal of order, 183-90, 192
final, see Teleology Dray, W., 162 n.
functional, 143-59, 312-15 Ducasse, C. J., 87, 134-42, 145, 147,
genetic, 86-7, 225-8 156, 229, 231, 248, 257-8, 265, 296
interpersonal, 120-6 Dummett, M. A. E., 162 n.
mnemic, 82
neolithic, 147, 153, 312 Entropy, 187-92
probabilistic, 49-50 Evolution, 277, 281-3
Causes: Ewing, A. C., 86 n.
counteracting, 63 Extensionality, 248-69
explanatory, 261-8, 283-4, 289, 295
fact and event, 31, 46, 130, 248-9, Feedback, 281-3
256-67 FitzGerald, E., 191
plurality of, 61, 307, 316 Fixity, 178-85, 189-91, 225
producing, 259, 264-8 Flew, A., 162 n., i68n.
progressive localization of, 73, 258 Four-dimensional scenery, 23-4 n.,
weighting of, 128-9 226-8, 296
Chisholm, R. M., 162 n., 172 n. Function, 275, 282-3
Coexistence, uniformities of, 204-7,227,
234-5 Gale, R. M.} 138-9, 162 n., 168 n.
Collocations, 206, 210-14, 225, 234, Geach, P. T., 75-6
243~5 Goal-seeking, 275-83
328 INDEX
Godel, K., 251 n. Mach, £., 154
Goodman, N., 49 McLaughlin, J. A., 44-5 n.
Gorovkz, S,, 162 n. Marc-Wogau, K., 44-5 n., 59 n., 62 n,
Gottlieb, D,, 251, 252 n., 353 n., 254 Martin, R., 48
Grunbaum, A., 163 n., 187 n. Melden, A. I., 25 n., 287-8, 292
Mellor, D. H., 234 n.
Harre, H. R., 86 n., 19511. Michotte, A., 223-4, 230
Hart, H. L, A., and Honord, A. M., Mill, J. S., 25, 30, 56, 59 n., 60-3, 68,
44-511., 117-33,136,142. i57>266 n., 7o-5» 79* 81-6, 90, 117-21, 136, 146,
270 148-9, 152, 167, 203-7, 234-5, 297-
Holland, R. F., 133-4, J42» 230 321
Honore, A. M., set Hart, H. L. A., and Mill's Methods of Induction, 25, 59 n.,
Honore, A. M. 68-8t, 107, 135 n., 136, 141, 148-53,
Hull, C. L., 275 297-321
Hume, D., 1-30,54-5, 57, 59-60,76-7, Minimally complete causal account,
86,88-91, 118, I3i, 132-4. 136, HO, 260-5
143, 167, 193-5, 215, 221, 224, 225, Morton, A., 248 n., 251 n.
229,234,287,295-6
Nagel, E., 219-20
Induction, 193-5, 201-2, and see Mill's Necessity, 11-13, 19-23, 26-7, 38-43,
Methods of Induction 86-7, 90,138-40, 193-230
Induction, problem of, 15-16, 195, 316 Nerlich, G. G., aian.
Newton, I., 92, 114, 146, 218-20
Intentions, 285-95
Intermixture of effects, 306,311-12,316 jW.mj, 24
Inus condition, 62, 120, 304, 308, 310 Noble, D., 276 n.
Nomic regularities, 60, 194, 196-228
Nominalization, perfect and imperfect,
Johnson, W. E., 297 31, 46, 130,249,259
Kant, I., 88-116, 161, 196, 225, 231 Overdetermination, 43-7, 164-5, 186,
Keynes, J. M., 15 224, 265 291
Keynes, J. N., 299 n.
Kim, J., 59 n. Pears, D. F,, 162 n.
Kneale, W. C., 86 n., 87, 197-8, 203, Phenomenalism, 215
208-9, 214-17, 228-9, 247, 296 Popper, K. R., 16, 163^, 183, 189,
207, 209-13, 223, 234
Law: Powers, 86
and causation, 117-32 Precognition, 161, 168, 173-6, 178-9,
Laws: 185-6
functional, 143-59, 312-15 Priority, causal (and temporal), see
gambling device, 244 Direction of causation and Direction
interaction resultant, 237-8 of time
Mendelian, 236 Probability, 5, 15-18, 23,49-50, 187-9,
mixed, 210-14, 244-5 232-4,239-42
of working (pure), 235, 239-42, 245- Psychology, 27-8, 54-7, 122-3, 223-4,
7, 292-3, 296 239-42
statistical, 50, 231-47
ideological, 276-80, 295 Quine, W. V., 250, 251 n., 252 n., 355
Legal policy, 127-8
Locke, J., sn., 12, no, 214 Rationalism, i2> 223, 228
Logical connection argument, 25, 287- Realism, 215
95 Referential transparency, 250-5,260-4,
Lyon, A., 30 n, 266
 INDEX 329
Regular succession, 3-4, 36-7, 59-87, Strawson, P. F., 91-104, 107 n,, 109-
160-1 12, 157, 225
Regularities, complex and incompletely Suchting, W. A., 207 n., aofl n., ais n.
known, 60-76 Sufficiency, 38-43, 138-40
Regularity theory of causation, 3-4, Suppes, P., 49-50
ao-i, 23-7, 59-6o, 75-7, 80-7, 154,
160-1, 194-9, 206-7, 214-17, 232-4, Taylor, C., 275-8
270-2 Taylor, R., 162 n.
Reichenbach, H., 163 n., 207, 214 Teleology, 270-96
Reid, T., 81 n. Tendency, 75-6, 241
Reinforcement, 277, 283 Time, measurement of, 113-15
Resemblance between cause and effect, Transcendent hypotheses, 316-17,
24, 158-9,217-24 222-3
Russell, B, A. W., 82, 143-?. I53~4» Transcendental arguments, 91, 96-7,
158,166-7,'90,i9«. i93>a°4 IIO-II
Transcendental idealism, 91-3
Scepticism, 112
Schlick, M., 154, 274, 296 Universals, open and closed, 307
Scriven, M., 44-5 n., 59 n., 162 n.,
168 n., 173 n., 1850. Vendler, Z., 248 n,, 249 n., 258-60,
Sharvy, R., 255 264,266
Slot machines, 40-3, 144 von Wright, G. H., 35 n., 171-3, 182,
Smullyan, A., 255 273 n -> 288-95
Sprigge, T. L. S., 279 n.
Stoutland, F. M., 25 n., 288 n., 293-5 Williams, C. J. F., 255 n.
Stove, D. C., 9, 15, 16 n., 18, 80 Wittgenstein, L., 316