Causation in contemporary analytical philosophy

Max Kistler

To cite this version:

Max Kistler. Causation in contemporary analytical philosophy. Costantino Esposito, Pasquale Porro.
La Causalità-La Causalité-Die Kausalität-Causality, Brepols, pp.635-668, 2002. �ijn_00000404�

HAL Id: ijn_00000404

https://hal.science/ijn_00000404
Submitted on 14 Oct 2003

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 635

Max Kistler

Causation in contemporary
analytical philosophy

At the beginning of the 20th century, Bertrand Russell, in his famous paper On
the Notion of Cause, pleaded for the «complete extrusion [of the word ‘cause’]
from the philosophical vocabulary»1. He gave three arguments for the thesis that
«the reason that physics has ceased to look for causes is that, in fact, there are
no such things» and that the whole conceptual apparatus of causation is «a rel-
ic of a bygone age, surviving, like the monarchy, only because it is erroneously
supposed to do no harm»2.
Russell’s first argument is that the idea, inherent in the concept of causation,
of an intrinsic difference between an active cause and a passive effect, is an-
thropomorphic. This makes the concept inadequate for the analysis of natural
causal relations, with the exception of human actions. «Nothing can operate ex-
cept a volition. The belief that causes “operate” results from assimilating them,
consciously or unconsciously, to volitions»3.
His second argument will prove particularly important in what follows. The
concept of causation is associated with the “principle of causality” according to
which causality is the source of regularities in nature. (For Hume, it even con-
sists fundamentally in regularities.) The principle of causality says that the same
cause always has the same effect. Now, it is essential to the concept of causation
to be applicable to macroscopic events such as car accidents, avalanches, or
deaths. The formidable complexity of such events contrasts with the simplicity

1 B. RUSSELL, On the Notion of Cause [1912], in B. RUSSELL, Mysticism and Logic [1917], Routledge,

London 1986, 173.

2 RUSSELL, On the Notion…, 173.

3 RUSSELL, On the Notion…, 183. Russell continues to condemn this anthropomorphic aspect of the

common notion of causation, when he undertakes, two years later, to rehabilitate the concepts of causa-
tion and of causal law. «Historically, the notion of cause has been bound up with that of human volition.
[...] The cause is supposed to be “active”, the effect “passive”. [...] But all such ideas, as applied to
physics, are mere anthropomorphic superstitions» (B. RUSSELL, Our Knowledge of the External World
[1914], Routledge, London 1993, 227).

«Quaestio», 2 (2002), 635-668

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 636

636 Max Kistler

of the conditions that are the object of the laws of physics. According to Russell,
the requirements of applicability to complex macroscopic events and of follow-
ing the strict regularity of a law of nature, are incompatible, although both are
central to causation. The more the description of a complex event involves suf-
ficient detail and completeness to allow the application of physical laws, the less
it is possible that the event will ever recur. However, without recurrence, the idea
of regularity is unintelligible4. This blocks the possibility to justify the scientif-
ic legitimacy of the concept of causation by its link to lawful regularity via the
principle of causality, which might have appeared promising, given that modern
science has substituted the search for laws for the search for causes.
Russell’s second argument establishes that there are no regularities at the
level of concrete events. His third argument goes a step further. It says that in
20th century physics, there are not any laws of succession at all, which might
have played the role of grounding causation. The progress of physics consists in
replacing them by laws of coexistence or by equivalences or functional depen-
dencies expressed in equations. This thesis about the form of laws in mature sci-
ence explains according to Russell the failure of the project to make the concept
of causation respectable by reducing it to nomological regularity. According to
the nomological conception of causation5, two events c and e are linked as cause
and effect, if and only if c and e are instances of types of events, such that there
is a law that all events of type c are followed by an event of type e. However, ac-
cording to Russell, the functional laws of contemporary physics are incapable of
playing this role. We can reconstruct Russell’s argument as a reductio. The con-
cept of causation has two essential aspects. First, the cause is temporally sepa-
rate from the effect and precedes it. Second, there is an asymmetry between
cause and effect in that the former determines or produces the latter, but not vice
versa, and this is intrinsic, i.e. not relative to a description of these events. The
reduction of causation to regular succession according to law of nature can only
use functional laws of coexistence, for mature physics has no other laws. How-
ever, this eliminates the two characteristic traits of causation, for the terms of
functional laws are not separate in time, and there is no intrinsic asymmetry be-
tween them. Hence the reduction of causation to nomological regularity fails.
Given this conclusion, the role of Russell’s second argument is to give an addi-

4 See RUSSELL, On the Notion…, 181. Nancy Cartwright has more recently taken up this observation

of Russell’s, to draw the opposite conclusion from Russell’s own. «I hold just the reverse of Russell’s view.
I am in favour of causes and opposed to laws». N. CARTWRIGHT, How the Laws of Physics Lie, Clarendon
Press, Oxford, 1983, 74.
5 Kant has offered a transcendental justification of the grounding of causation in lawful regularity. Re-

cently, Davidson (D. DAVIDSON, Laws and cause, «Dialectica», 49 (1995), 263-279) has undertaken to re-
formulate a transcendental justification which is Kantian in spirit.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 637

Causation in contemporary analytical philosophy 637

tional reason for why there can be no laws of succession between complex events.
It is important for Russell’s demonstration of the bankruptcy of the concept of
causation, to show that the absence of such laws is no historically contingent fact,
but has a conceptual ground, namely the incommensurability of the terms of
quantitative laws (a restricted number of measurable quantities) and the terms
of causal relations (concrete events with an infinite number of properties).

1. The absorption of causation in the model of deductive-

nomological explanation

Russell’s arguments have not succeeded in convincing the philosophical com-

munity of the need to eliminate the concept of causation from a scientifically in-
formed world view. Many continue to think that the concept of causation is nec-
essary to understand not only our interaction with the environments of everyday
life, but also science.
Nancy Cartwright accepts Russell’s thesis that «causal principles cannot be
derived from the causally symmetric laws of association»6, but she objects to his
thesis that «laws of association are all the laws there are»7. Her reason is the fol-
lowing: We are capable of intervening in the world. This capacity presupposes
that those interventions abide by nomological constraints, or laws of nature, that
we are able to know, at least approximately. Moreover, science allows us to de-
velop and refine our «effective strategies»8 of intervention, and above all to ex-
plain why they work.
Hilary Putnam has also drawn attention to the indispensability of the concept
of causation for the analysis of scientific experimental and explanatory practice,
and not only in sciences like biology that have not completely accomplished the
quantitative formulation of all their laws. We need it for physics too, when it
comes to understanding the interaction of concrete and complex systems:

«When we come to sciences less fundamental than fundamental particle physics, say
sociology or history, or even evolutionary biology, or even to chemistry or solid-state
physics, then, of course, we find that causation-as-bringing-about is invoked con-
stantly. [...] Even in fundamental physics such talk becomes indispensable when we
apply the physics to actual systems that are (of course) not the whole universe-regard-
ed-as-a-closed-system»9.

6 CARTWRIGHT, How the Laws…, 21.

7 CARTWRIGHT, How the Laws…, 21.
8 CARTWRIGHT, How the Laws…, 21.

9 H. PUTNAM, Is the Causal Structure of the Physical Itself Something Physical?, in H. PUTNAM, Real-
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 638

638 Max Kistler

Even Russell himself, only two years after his radical critique of the concept
of cause, partly rehabilitated it10, however, after cutting out the anthropomorphic
aspects of the common sense notion of cause. Science provides no grounding for
the existence in nature of a compulsion or necessity, in which the cause produces
the effect. The idea of such a necessity has its only origin in the experience of
our own voluntary acts or acts of volition11. We must also refrain from thinking
of the cause as being “active” and of the effect as being “passive”; furthermore,
we must eliminate from the concept of causation «everything that differentiates
between past and future»12. However, says Russell, we need the notion of cau-
sation in order to understand the evolution of physical systems that are «less than
the whole state of the universe»13, which is of course all we ever try to do in prac-
tice. According to the position articulated by Russell in 1914, we can know «that
certain constant relations are observed to hold […] between events of specified
kinds with given intervals of time between them»14, and he now thinks that such
constant relations between events can be considered as «causal laws»15.
The reintegration of the notion of causation within the conceptual apparatus
of the philosophy of science has found its canonical expression in the assimila-
tion of causation to the relation of explanation, conceived according to the so-
called ‘deductive-nomological’ model. This conception of causation belongs to
the core of the doctrine of logical empiricism. As such, it was the standard view
on the nature of causation from the 1930’s to the 1970’s when criticism and al-
ternative proposals began to proliferate. I shall present this conception in some
detail because all later alternative conceptions are motivated by the attempt to

ism with a Human Face, J. CONANT (ed. by), Harvard University Press, Cambridge, MA 1990, 86; italics
Putnam’s.
Since then, Putnam has changed his mind about the objective reality of causation. He now regards it
as an effect of perspective that makes its appearance relatively to our interest of searching explanations,
which is governed by a pragmatic logic. See H. PUTNAM, Renewing Philosophy, Harvard University Press,
Cambridge, MA 1992, 47ff.
10 Russell maintains his thesis that «in a sufficiently advanced science, the word ‘cause’ will not oc-

cur in any statement of invariable laws». (RUSSELL, Our Knowledge…, 223). However, he now acknowl-
edges that «there is, however, a somewhat rough and loose use of the word ‘cause’ which may be preserved.
The approximate uniformities which lead to its pre-scientific employment may turn out to be true in all
but very rare and exceptional circumstances, perhaps in all circumstances that actually occur. In such
cases, it is convenient to be able to speak of the antecedent event as the “cause” and the subsequent event
as the “effect”» (RUSSELL, Our Knowledge…, 223).
11 Cfr. RUSSELL, Our Knowledge…, 229.

12 RUSSELL, Our Knowledge…, 227.

13 RUSSELL, Our Knowledge…, 230.

14 RUSSELL, Our Knowledge…, 230.

15 RUSSELL, Our Knowledge…, 230.

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 639

Causation in contemporary analytical philosophy 639

overcome the difficulties encountered by the logical empiricist’s absorption of

the notion of causation in the deductive-nomological model of explanation16.
According to the model of scientific explanation put forward by Hempel17 and
developed by Hempel and Oppenheim18, an explanation is a deductive argu-
ment. Its conclusion is called the “explanandum” – that which needs to be ex-
plained –, its premises constitute the “explanans” – that which explains. Part of
the premises consists in «statements of antecedent conditions» (often called
“initial conditions”), and part in “General Laws”, whereas the conclusion con-
tains a «description of the empirical phenomenon to be explained»19. An ade-
quate explanation must satisfy the following requirements: the explanandum
must be a logical consequence of the explanans, for only the logical necessita-
tion of the conclusion of a valid argument by its premises can sufficiently
“ground” (our belief in the truth of) the explanandum. Furthermore, the ex-
planans must contain an essential reference to general laws. In other words, an
explanation is not adequate if it is possible to deduce the explanandum from the
particular premises alone. However, Hempel and Oppenheim explicitly ac-
knowledge the legitimacy of purely general explanations, in which the explanans
contains only general laws and which have a law as their explanandum. This is
meant to allow integrating in the model the derivation of laws from other, more
general laws, such as the derivation of the general form20 of Kepler’s laws from
Newton’s laws. The deductive-nomological (from now on “D-N”) model is only
intended to cover the explanation of empirical phenomena. Therefore, it requires
the explanans to have empirical content, in other words it must be capable of be-
ing tested by observation or experimentation. Finally, an explanation can only
be correct if all statements in the explanans are true. This implies that we have
to abandon the aim to know with certainty whether an explanation is correct. The
problem of induction21, i.e. the fact that inductive arguments are never logical-
ly valid, condemns us to construct explanations with hypothetical laws whose
truth will for logical reasons remain forever uncertain.

16 On the D-N model of explanation and its relation to causation, see M. KISTLER, Causalité et lois de

la nature, Vrin, Paris 1999, chap. 5, and A. BARBEROUSSE, M. KISTLER et P. LUDWIG, La philosophie des sci-
ences au XXe siècle, Flammarion, Paris 2000, chap. 5.
17 C.G. HEMPEL [1942], The Function of General Laws in History, repr. in C.G. HEMPEL, Aspects of Sci-

entific Explanation, The Free Press, New York 1965, 231-243.

18 C.G. HEMPEL and P. OPPENHEIM, Studies in the Logic of Explanation, repr. in HEMPEL, Aspects…, 245-

290.
19 HEMPEL and OPPENHEIM, Studies…, 249.

20 Kepler’s laws mention the sun, which is an individual object. Hence, these laws cannot be derived

without particular premises, such as a statement of the sun’s mass. However, the general laws describing
the relative movement of two masses can be derived from general premises only.
21 See BARBEROUSSE, KISTLER, LUDWIG, La philosophie…, chap. 2.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 640

640 Max Kistler

According to Hempel and Oppenheim, any explanation that satisfies the re-
quirements of the D-N model is ipso facto a causal explanation. Hence, it is pos-
sible to use the model to analyse the causal relation: the conjunction of the ini-
tial conditions is taken to be a description of the cause, and the event described
by the explanandum is taken to be its effect22. Insofar as the laws appearing in
the explanans are explicitly considered to be empirical regularities, the assimi-
lation of the cause-effect relation to the explanatory relation in the D-N model
leads to a version of the regularity theory of causation. To that extent, the account
still has to face Russell’s second and third arguments sketched above. Interest-
ingly, Hempel later acknowledged, to a certain extent, the inadequacy in prin-
ciple of the D-N model to the cause-effect relation, and this for several reasons.
He acknowledges (in line with the conclusion of Russell’s second argument)
that, as long as the cause does not comprise the state of the entire universe at an
instant, it is never strictly possible to deduce with logical necessity the descrip-
tion of the state of a limited system, from the description of a different state of
the system and laws.

«If the system is not isolated, i.e., if relevant outside influences act upon the system
during the period of time from the initial state invoked to the state to be explained,
then the particular circumstances that must be stated in the explanans include also
those outside influences»23.

Later, Hempel24 even more explicitly acknowledged that the deduction of par-
ticular events (whose extent is narrower than the entire universe) is never strict-
ly possible, for «if the lawlike premises of deductive-nomological explanations
are not to be patently false, we must explicitly recognize the explanations’ aux-
iliary reliance on qualifying clauses, commonly tacit, which govern such premis-
es»25, clauses that Hempel calls «provisos», and that others call «ceteris paribus»
clauses26.

22 «The type of explanation which has been considered here so far is often referred to as causal ex-

planation. If E describes a particular event, then the antecedent circumstances described in the sentences
C1, C2, ..., Ck may be said jointly to “cause” that event, in the sense that there are certain empirical reg-
ularities, expressed by the laws L1, L2, ..., Lr, which imply that whenever conditions of the kind indicated
by C1, C2, ... Ck occur, an event of the kind described in E will take place» (HEMPEL and OPPENHEIM, Stud-
ies…, 250). In short, «by virtue of thus presupposing general laws which connect “cause” and “ef-
fect”, causal explanation conforms to the D-N model» (HEMPEL, Aspects…, 348).
23 HEMPEL, Aspects…, 352.

24 See C. G. HEMPEL, Provisos: A Problem Concerning the Inferential Function of Scientific Theories,

in A. GRÜNBAUM and W. C. SALMON (ed. by), The Limitations of Deductivism, University of California Press,
Berkeley 1988, 19-36.
25 A. GRÜNBAUM, Preface, in GRÜNBAUM and SALMON (eds.), The Limitations…, x.

26 On this problem, see KISTLER, Causalité…, chap. 3.

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 641

Causation in contemporary analytical philosophy 641

The second reason Hempel gives for the inadequacy of the D-N model as a
means to analyze the cause-effect relation (and which is reminiscent of Russell’s
third argument) concerns «laws of coexistence»27. Hempel admits that the mere
deducibility of the explanandum from initial conditions and general laws is no
sufficient reason to consider those conditions as the cause of the event described
by the explanandum. He gives the following counterexample to this thesis. Take
the explanation of the period of an (ideal) pendulum by its length. This expla-
nation uses the law relating the period T of an ideal pendulum to its length L: T=
–—
2π √ L/g (where g is the acceleration of free fall at the surface of the Earth). It
links two properties28 the pendulum possesses at the same time; therefore the
events or facts corresponding to the possession of these properties cannot stand
in the relation of cause to effect, which requires temporal separation29. More
generally, Hempel admits that there are a certain number of types of explanation
that satisfy the criteria of the D-N model, but which cannot be interpreted as
causal explanations: besides explanations using laws of coexistence, this is the
case where the explanandum is not a particular event but a law. In the deduc-
tion of (the general form of) Kepler’s laws from Newton’s, no particular event is
named that could be interpreted as cause or effect.
Finally, Hempel notes that there are explanations satisfying all the require-
ments of the D-N model, but where the “initial” conditions contain a reference
to a state of the system that is posterior to the state described in the explanan-
dum. One example is the explanation of the refraction of a light ray crossing the
interface separating two media with different refractive index. The explanans
mentions, besides Fermat’s principle according to which the path of the ray is
such as to minimize its overall duration, the points of departure and arrival of
the ray, to deduce the point of its passage of the interface. Here, the arrival men-
tioned in the premises is posterior to the passage of the interface mentioned in
the conclusion. Hempel notes that the «uneasiness»30 produced by explanations
of this kind stems from the apparent impossibility of interpreting them as causal
explanations, for it seems a priori excluded that refractions and similar types of
physical processes constitute cases of backwards causation where the effect pre-

27HEMPEL, Aspects…, 352.

28To be able to attribute its period to the pendulum at a given instant, one must consider it as «a dis-
positional characteristic» (HEMPEL, Aspects…, 352).
29 R. CUMMINS, The Nature of Psychological Explanation, MIT Press, Cambridge, MA 1983, 3 uses

this example to refute more generally the identification of scientific explanation with the discovery of a
“covering law”. Hempel (HEMPEL, Aspects…, 352 ff.) notes, concerning this example, that intuitively, the
explanation of the pendulum’s period by its length seems more acceptable than the explanation of its
length by its period. However, none of the logical requirements of D-N explanation allows to justify this
intuitive preference, the law linking these two variables being a symmetrical law of coexistence.
30 HEMPEL, Aspects…, 353.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 642

642 Max Kistler

cedes the cause31. As a result, Hempel despairs of finding general conditions,

which it would be sufficient to impose on D-N explanation, in order to yield a
model of the causal relation.
Many expositions of the doctrine called the “deductive-nomological theory of
causation”, by the leading philosophers of the logical empiricist tradition take
fewer precautions than Hempel, and identify the cause-effect relation directly
with the relation between the initial conditions and the explanandum in a cor-
rect D-N explanation. According to Carnap,

«What is meant when it is said that event B is caused by event A? It is that there are
certain laws in nature from which event B can be logically deduced when they are com-
bined with the full description of event A»32.

Popper, though a severe critic of logical empiricism in many regards, also iden-
tifies causal explanation with D-N explanation, without any further constraints:

«To give a causal explanation of an event means to deduce a statement which describes
it, using as premises of the deduction one or more universal laws, together with certain
singular statements, the initial conditions»33.

As we have seen, Hempel gives a number of reasons arguing against this D-N
account of causation. The counterexamples can be divided in two categories34:
the first contains D-N explanations in which the relation between explanans and
explanandum cannot be taken to be causal because it is not a relation between

31It is controversial whether there is backwards causation in quantum mechanics. Cfr. H. REICHEN-
BACH, The Direction of Time [1956], University of California Press, Berkeley 1991, and P. DOWE, Back-
wards Causation and the Direction of Causal Processes, «Mind», 105 (1996), 1-22.
32 R. CARNAP, Philosophical Foundations of Physics, Basic Books, New York 1966.

33 K.R. POPPER, Logik der Forschung [1934], J.C.B. Mohr, Tübingen 199410, 31; engl. transl. The Log-

ic of Scientific Discovery, London, Routledge 19925, 59, italics Popper’s. There are many variants of this
doctrine. Stegmüller expresses the idea semantically, in terms of the truth conditions of causal statements.
«Wo ein individueller Kausalzusammenhang festgestellt wird, da wird behauptet, dass eine spezielle
Ereignisfolge unter ein (hypothetisch angenommenes) Naturgesetz subsumiert werden könne» (W.
STEGMÜLLER, Probleme und Resultate der Wissenschaftstheorie und Analytischen Philosophie, Vol. I.: Erk-
lärung, Begründung, Kausalität [1969], Springer, Berlin 1983, 512-513). Davidson defends the idea in
terms of our justification for believing particular causal statements. «It does not follow that we must be
able to dredge up a law if we know a singular causal statement to be true; all that follows is that we know
there must be a covering law. And very often, I think, our justification for accepting a singular causal state-
ment is that we have reason to believe an appropriate causal law exists, though we do not know what it
is»: D. DAVIDSON, Causal Relations [1967], in D. DAVIDSON, Essays on Actions and Events, Clarendon Press,
Oxford 1980, 160.
34 Such counter-examples have been collected by P. W. HUMPHREYS, The Causes, Some of the Causes,

and Nothing But the Causes, in Minnesota Studies in the Philosophy of Science, vol XII: Scientific Expla-
nation, University of Minnesota Press, Minneapolis 1989, 283-306, 300-301, and W. SALMON, Four
Decades of Scientific Explanation, University of Minnesota Press, Minneapolis 1990, 46-50.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 643

Causation in contemporary analytical philosophy 643

spatio-temporally separated particulars. The second contains explanations con-

cerning causal processes, but where neither can the initial conditions be identi-
fied with the cause nor the conclusion with the effect. The following belong to
the first category.
a) The explanation of general laws, such as Kepler’s laws from Newton’s laws.
b) Logical and mathematical explanations, proceeding by deduction from ax-
ioms or theorems, which are not interpreted or which have abstract models as in-
terpretations.
c) Explanations by laws of coexistence, such as the explanation of the period
by the length of the pendulum.
d) Analytical or compositional35 analysis, in particular functional analysis:
an important type of scientific explanation does not focus on the occurrence of
an event or on the acquisition of a property by an object, but rather on the na-
ture of a property. For example, statistical physics explains what the property of
an object O being at a certain temperature consists in, by indicating the under-
lying nature of that property: it identifies the mean kinetic energy of O’s compo-
nents as the microscopic reduction base of that property. There is no causal re-
lation between the properties of the components and their interactions on the one
hand, and the reduced property of the whole object on the other, if only because
the instantiations of the reduced and reducing properties are not separated in
space and time. An important type of such non-causal explanations concerns the
nature of dispositional properties, as when the solubility of sugar in water is ex-
plained by properties of the sugar molecules and their interactions with water
molecules.
e) Some explanations satisfy the purely formal requirements of the D-N mod-
el, but use a trivial generalization in the place of a genuine law of nature. Thus,
one can “explain” the fact that John has not become pregnant, by deducing it
from the fact that he has taken contraceptive pills for the whole year, together
with the “law” that no man who regularly takes contraceptive pills will become
pregnant. There is of course neither a link of relevance nor of causation between
men’s taking birth-control pills and their avoiding pregnancy.
The second category of counterexamples to the D-N theory of causation con-
sists of cases where the explanation is about a causal process, but where never-
theless the initial conditions do not cause the explanandum.
a) Explanations using extremal principles, such as the explanation of the path
of a refracted light ray mentioned above.

35 Cummins distinguishes explanations of these types from causal explanation by subsumption under

a causal law. Furthermore, Cummins distinguishes the compositional analysis of a system from the analy-
sis of a property. He calls the latter “functional analysis” (CUMMINS, The Nature of Psychological …, 15)
if the property is dispositional, otherwise “property analysis”.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 644

644 Max Kistler

b) The D-N model allows the explanation of the cause by the effect. To take
a famous example36, a tower projects a shadow. The (non-causal) explanation of
the height of the tower by the length of the shadow satisfies the requirements of
the model just as well as the (causal) explanation of the length of the shadow by
the height of the tower. However, only in the second case can the initial condi-
tions be interpreted as a cause of the explanandum.
c) The D-N model allows the explanation of an event by another, where both
are effects of a common cause, although neither is a cause of the other. The ex-
planation of the storm by the fall of the barometer is legitimate insofar as the link
between these events is nomological, rather than accidental. However, the ini-
tial conditions do not indicate a cause of the explanandum; rather, both are ef-
fects of a common cause, the fall of air pressure.
d) Henry Kyburg37 has noted that the D-N model makes the following expla-
nation legitimate: a pinch of salt dissolves in warm water. This can be explained
by the fact that the salt has been hexed and that by necessity, all samples of
hexed salt dissolve in warm water. In this case, there really is a causal relation,
but the presence, among the initial conditions, of a piece of information without
relevance for the effect, makes it impossible to identify the whole set of those
initial conditions with the cause.
e) Michael Scriven38 has discovered a counterexample similar to the preced-
ing one in its structure, but which poses are more pressing problem. It raises the
so-called ‘preemption problem’, which arises in cases of potential causal overde-
termination. A bridge collapses as a result of being overloaded. A bomb that
would have been sufficient to destroy the bridge explodes just after it has col-
lapsed. The explanation of the collapse by the explosion meets the D-N model’s
requirements without indicating the cause of that event. This is expressed by
saying that the efficacy of the bomb has been “preempted” by the overload. Ac-
cording to Hempel39, it is possible to rescue the model from refutation by this
case, by taking into account the process leading to the destruction of the bridge,
as it lawfully evolves in time. On closer inspection, the law that allows deducing
the destruction of the bridge from the explosion of the bomb governs the propa-
gation of the shock wave, from the space-time location of the explosion and with
finite speed. However, in the case described by Scriven, the D-N explanation

36 Cfr. S. BROMBERGER, Why-Questions [1966], in S. BROMBERGER, On What We Know We Don't Know,

University of Chicago Press, Chicago and CSLI, Stanford 1992, 75-100, 83.
37 H.E. KYBURG JR., Comment, «Philosophy of Science», 32 (1965), 147-151.

38 M. SCRIVEN, Explanation, Prediction, and Laws, in H. FEIGL and G. MAXWELL (ed. by), Minnesota

Studies in the Philosophy of Science, III, University of Minnesota Press, Minneapolis 1962, 170-230.
39 HEMPEL, Aspects…, 420.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 645

Causation in contemporary analytical philosophy 645

will turn out not to be correct for at the moment of the shock wave’s arrival at the
location of the bridge, there already is no more bridge there to be destroyed.
These counterexamples do not only show that the deductive-nomological the-
ory of causation, the contemporary version of Hume’s regularity theory, is inad-
equate. They also set the agenda for the search of an alternative account40. The
four major theories of causation, which are to be discussed here, will be evalu-
ated first of all by their capacity to cope with those problem cases. We shall con-
sider in turn the counterfactual, the interventionist, the probabilistic and the
transference account.

2. The counterfactual theory of causation

David Lewis has proposed returning to Hume’s second definition of cause ac-
cording to which «we may define a cause to be an object followed by another
[…] where, if the first object had not been, the second had never existed»41.
Lewis’ proposal consists in reducing causation between two events to a finite
chain of intermediate events, each link of which causally depends on the pre-
ceding42. The notion of causal dependence between events is then analysed with
the help of the counterfactual dependence between the propositions saying that
those events have taken place. According to the counterfactual theory of cau-
sation, event e depends causally on event c if the following two counterfactu-
als are true: if c had occurred, e would have occurred; if c had not occurred, e
would not have occurred43. The distinction between causal dependence and the
causal relation itself is introduced in order to guarantee the transitivity of the
causal relation, for causal dependence is not transitive44. A situation is con-

40 Lewis motivates his elaboration of a counterfactual theory of causation by the enumeration of prob-

lems encountered by the deductive-nomological account. «I have no proof that regularity theories are be-
yond repair, nor any space to review the repairs that have been tried. Suffice it to say that the prospects
look dark. I think it is time to give up and try something else». (D. LEWIS, Philosophical Papers, vol. II,
Oxford University Press, New York 1986, 160).
41 D. HUME, An Enquiry Concerning Human Understanding, in Enquiries Concerning Human Under-

standing and Concerning the Principles of Morals, L.A. Selby-Bigge and P.H. Nidditch (ed. by), Claren-
don Press, Oxford 19753, 76.
42 Precedence is here understood logically. It is determined by the order of the members of the chain,

without prejudging the temporal precedence of the cause with respect to the effect. Indeed, Lewis gives
an independent argument against the possibility of backwards causation, where the effect is earlier in time
than the cause. I shall present and criticize that argument below.
43 The first counterfactual carries the weight in cases where the events c and e do not occur, whereas

the second carries the weight in cases in which they do occur. Cfr. LEWIS, Philosophical Papers…, 167.
44 «Causation must always be transitive; causal dependence may not be so»: LEWIS, Philosophical Pa-

pers…, 167. Among the defenders of the counterfactual approach, some have criticized Lewis’ move to
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 646

646 Max Kistler

ceivable where the premises of the following argument are true but its conclu-
sion false45:
If Otto had gone to the party, then Anna would have gone.
If Anna had gone, then Waldo would have gone.
Therefore, if Otto had gone, then Waldo would have gone.
The conclusion is false if Waldo is Otto’s unlucky rival for Anna’s affections,
and does all in his power to avoid his rival. The non-transitivity of counterfactu-
al dependence can be explained by the fact that their evaluation implicitly pre-
supposes a concrete situation, and that the situation presupposed by the con-
clusion differs from that presupposed in the premises46.
Lewis’s proposal still faces some of the counterexamples to the deductive-
nomological account.
a) There seem to be counterfactuals expressing dependence of the cause on
the effect. Here is such an apparently true “backtracking” counterfactual, i.e. a
counterfactual whose antecedent bears on an event that happens later than the
event described by the consequent. If there had not been a solar eclipse on 11
August 1999, the sun, moon and earth would have had different positions or dif-
ferent speeds on 10 August 1999. Lewis calls this «the problem of effects»47.
b) Similarly, two effects of a common cause seem to depend counterfactually
on each other. Given that Smith and Jones watch all (and only) soccer matches
broadcast by TV XY, it seems that if Smith had seen the match on day D, then
Jones would have seen it (because XY would have shown it), and if Smith had
not seen it, Jones would not have seen it either (because XY wouldn’t have shown
it). Lewis calls this «the problem of epiphenomena»48.
c) Finally, the counterfactual analysis seems to provide an incorrect result in
cases of preemption and causal overdetermination. If a prisoner is executed by
two bullets A and B shot by two soldiers at the same time, each of which would
alone have been sufficient for death, we have a case of causal overdetermina-

guarantee the transitivity of causation by stipulation. As Geert Keil points out (G. KEIL, Handeln und
Verursachen, Vittorio Klostermann, Frankfurt am Main 2000, 297), within a theory aiming at reducing cau-
sation to counterfactual dependence, it seems plausible to accept the fact that counterfactual dependence
is not transitive, and to conclude that causation is not transitive either. Ned Hall (N. HALL, Causation and
the Price of Transitivity, «Journal of Philosophy», 97 (2000), 198-222) shows that if one accepts “nega-
tive events”, such as absences and omissions among causes and effects, then the thesis that counterfac-
tual dependence is sufficient for causation enters in conflict with the thesis of the transitivity of causa-
tion.
45 The example is from David Lewis: D. LEWIS, Counterfactuals, Blackwell, Oxford 1973, 32.

46 Cfr. KEIL, Handeln…, 295-296.

47 LEWIS, Philosophical Papers…, 170.

48 LEWIS, Philosophical Papers…, 170.

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 647

Causation in contemporary analytical philosophy 647

tion, or, in Lewis’ words, of «redundant causation»49. Intuitively, each bullet

seems to be a cause of death. However, it is not true that if, say, bullet B had not
been shot, the prisoner would not have died. Or, to take the case of preemption
mentioned earlier, if the bridge hadn’t been overloaded, it would have collapsed
nevertheless. Hence, the counterfactual analysis denies, wrongly, that the over-
load is, in the situation sketched above, the cause of the bridge’s collapsing50.
The counterfactual theory inherits, it seems to me, some of the difficulties of
the nomological theory because it tacitly relies on the same foundations51. (This
is a controversial thesis that I will try to justify below). Let us follow Lewis52 (and
Stalnaker53) in their conception of the truth-conditions of counterfactual condi-
tionals, in terms of possible worlds. Let us assume that A is false in the actual
world (or, equivalently, that event a does not occur in the actual world: A is the
proposition that a occurs54). On which condition is the counterfactual “if it were
that A, then it would be that C” (equivalent to: if a had occurred, then c would
have occurred») true? According to Lewis, it is true in the actual world “if and

49LEWIS, Philosophical Papers…, 193.

50Lewis (LEWIS, Philosophical Papers…, 205-207) proposes to justify the intuition that the overload
is the cause of the collapse although it does not counterfactually depend on it, by pointing out that the col-
lapse is nevertheless quasi-dependent on the overload. The relation between the overload and the collapse
is causal because it is intrinsically similar to other relations between overloads and collapses, in which
no preempting cause is present. In those other situations, there is counterfactual dependence; hence (by
definition of this concept) there is quasi-dependence in the situation in which the preempting cause is
present. Causation being an intrinsic relation depending only on the process linking cause and effect, the
relation is causal even in the situation in which the preempting cause is present. Hence, quasi-depen-
dence is sufficient for causation, even in the absence of dependence. However, there are cases of pre-
emption, in which the effect is also quasi-dependent on the preempted cause. Schaffer (J. SCHAFFER,
Trumping Preemption, «Journal of Philosophy», 97 (2000), 165-181) shows that Lewis’ theory of quasi-
dependence systematically fails in cases of “trumping preemption”, where the preempted process is never
interrupted and goes to completion, just as the preempting process. In such cases, the theory of quasi-de-
pendence yields the wrong result that the preempted cause is a cause because the effect is quasi-depen-
dent on it. Schaffer (SCHAFFER, Trumping Preemption, 175) reports the following example, attributing it to
Bas van Fraassen and Ned Hall: The major and the sergeant, both before the corporal, shout «charge!» at
the same time and the corporal decides to charge. The higher officer’s order “trumps” the lower officer’s;
hence the major’s order, not the sergeant’s, is the cause of the corporal’s decision. However, it is at least
conceivable that the competing process from the sergeant’s order to the corporal’s decision not be inter-
rupted. In that case, the decision is quasi-dependent on the sergeant’s order just as it is on the major’s.
Lewis (D. LEWIS, Causation as Influence, «Journal of Philosophy», 97 (2000), 182-197, 185) himself ac-
knowledges that situations of this kind refute his theory of quasi-dependence.
51 However, these problems do not arise in exactly the same way. In a case of preemption, e.g., the

nomological theory yields the false result that the preempted event is a cause, whereas the counterfactu-
al theory yields the false result that the real (preempting) cause is not a cause, because the effect would
have occurred even if it had been absent. (See LEWIS, Philosophical Papers…, 200).
52 D. LEWIS, Counterfactuals, Blackwell, Oxford 1973.

53 R. STALNAKER, A Theory of Conditionals, in N. RESCHER (ed. by), Studies in Logical Theory, Black-

well, Oxford 1968.

54 See LEWIS, Philosophical Papers…, 166.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 648

648 Max Kistler

only if some (accessible55) world where both A and C are true is more similar to
our actual world, overall, than is any world where A is true but C is false”56.
There is no universal algorithm allowing us to evaluate the comparative similar-
ity of any pair of possible worlds. However, one can establish a hierarchy of cri-
teria, in decreasing order of importance. What is, according to Lewis, responsi-
ble for the largest dissimilarity between a world w0 and a world w1, is that w1 con-
tains «big widespread, diverse violations of law»57 with respect to the laws of w0.
In other words, in order to find the nearest world where the antecedent A of the
counterfactual is true, one must eliminate all worlds containing major violations
of the laws of w0. According to Lewis’ second most important criterion, the most
similar worlds are those with a «perfect match» of all facts existing in the largest
spatio-temporal region of w0. The second least importance is given to the crite-
rion according to which worlds are closer if they contain fewer «small localized
simple violations of law»58. In other words, w1 is closer to w0 than w2, if w1 con-
tains fewer events which appear as “miracles” from the point of view of the laws
of w0 than w2. The least weight is given to similarity with respect to particular
matters of fact.
Lewis tries to justify his thesis that backtracking counterfactuals are never
true on the basis of the application of these criteria of similarity between possi-
ble worlds59. This thesis turns out to be crucial for the defence of the counter-
factual theory of causation, because many (perhaps all) counterexamples it faces
get their plausibility (which Lewis takes to be only apparent) from a (sometimes
implicit) reliance on backtracking counterfactual dependence. The first coun-
terexample that raises the “problem of effects” (a above) is explicitly based on

55 The concept of accessibility of one possible world from another allows us to construct different con-

cepts of possibility and necessity. In particular, physical necessity can be distinguished from logical ne-
cessity, on the basis of the distinction between possible worlds in which our actual laws of nature hold and
worlds in which they do not. The former are accessible from the actual world with respect to physical pos-
sibility: a proposition is physically necessary if and only if it is true in all physically accessible possible
worlds; a proposition is physically possible if and only if there exists a physically accessible world in
which it is true. Cfr. LEWIS, Counterfactuals, 4-5.
56 LEWIS, Philosophical Papers…, 41.

57 LEWIS, Philosophical Papers…, 47.

58 LEWIS, Philosophical Papers…, 48.

59 The counterfactual theory might be tempted, like the nomological theory, to avoid the problem of

having to judge falsely that effects are also causes of their own causes (the “problem of effects”), simply
by stipulating that the cause always precedes its effects in time. It seems, however, preferable to avoid
this easy solution, for three reasons: cfr. LEWIS, Philosophical Papers…, 170. 1) Such a stipulation solves
neither the problem of epiphenomena nor the problem of preemption. 2) It prejudges a priori that back-
wards causation is impossible, whereas the controversial issue of the existence of backwards causation
seems to be empirical and should therefore be decided on scientific grounds. 3) It makes a causal theory
of time a priori inconceivable.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 649

Causation in contemporary analytical philosophy 649

a backtracking counterfactual. The counterexample of the common cause (b

above), the proposition that if Smith hadn’t seen the match, then Jones hadn’t
seen it either, is grounded on the backtracking counterfactual that if Smith had-
n’t seen it, then TV station XY wouldn’t have broadcast it60.
Let us consider the following example61. In world w0, during the cold war and
Nixon’s presidency of the USA, there is a red button that can trigger, with per-
fect reliability, a nuclear holocaust. We can imagine that w0 is our actual world.
In w0, Nixon does not push the button. According to Lewis, the counterfactual
“if Nixon had pushed the button, there would have been a nuclear holocaust” is
true, whereas the backtracking counterfactual “if there had been a nuclear holo-
caust, then N. would have had pushed the button” is false. The reason why the
former is true is that the closest world where N. pushes the button is, according
to Lewis, a world that exactly resembles our own (w0), until the moment imme-
diately preceding N.’s fatal act, but in which «a tiny miracle takes place» at pre-
cisely that instant, such that «the deterministic laws of w0 are violated at w1 in
some simple, localized, inconspicuous way»62. W1 differs from w0 only by «a few
extra neurons [that] fire in some corner of Nixon’s brain»63. Among the worlds
meeting that requirement, the nearest to w0 is that which contains no other mir-
acles or any other violations of the laws of nature. The consequent of the “for-
ward” counterfactual is therefore true in w1, because our laws inevitably deter-
mine the occurrence of the nuclear holocaust, once the button is pushed.
What about the truth value of the backtracking counterfactual that “if there
had been a nuclear holocaust, then N. would have had pushed the button”? Ac-
cording to Lewis, it is false, because the closest world to w0 in which its an-
tecedent is true, i.e. in which there is a nuclear holocaust, is a world in which
that holocaust is caused by a tiny miracle immediately preceding it. In effect,
Lewis shows that worlds in which the miracle is located after the holocaust, are
necessarily more distant from w0 than those where the miracle is located before
it. A miracle in w1 occurring immediately after the holocaust maximizes the sim-
ilarity with w0, as far as its final phase starting with the holocaust is concerned.
However, such a miracle would not at all be «small, localized, inconspicuous»64,
because a very large miracle is needed to “wipe out” the large number of effects

60 Lewis shows (LEWIS, Philosophical Papers…, 200-201) that at least certain cases of preemption are

only counterexamples to the counterfactual theory if there are true backtracking counterfactuals.
61 LEWIS, Philosophical Papers…, 43ff.

62 LEWIS, Philosophical Papers…, 44. On miracles, see also LEWIS, Counterfactuals, 75.

63 LEWIS, Philosophical Papers…, 44.

64 LEWIS, Philosophical Papers…, 44.

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 650

650 Max Kistler

spreading out from a single cause65. This asymmetry stems from the existence of
irreversible processes, which is, according to a plausible hypothesis, also at the
origin of the asymmetries of causation and time66.
However, contrary to what Lewis asserts, the irreversibility of physical
processes is no obstacle to the possibility of reasoning backwards in time. Lewis’
observation that the miracles that must be postulated to allow a perfect match
between w0 and a time span of a possible world in which the antecedent of a
counterfactual is true, are smaller if they are located earlier than the occurrence
of the antecedent, rather than later, obliges us only to look for the nearest possi-
ble world, among the worlds where the miracle takes place before the holocaust
described in the antecedent. However, this does not yet completely determine
the temporal location of that miracle: Lewis thinks that it must be located im-
mediately before the event described in the antecedent, i.e. the nuclear holo-
caust; this would indeed let the counterfactual come out false.

«The proper solution to both problems [the problems of effects and of epiphenomena;
M.K.], I think, is flatly to deny the counterfactuals that cause the trouble. If e had been
absent, it is not that c would have been absent. [...] Rather, c would have occurred just
as it did but would have failed to cause e»67.

In our case, Lewis’ verdict would be: if e (the holocaust) had been present, it is
not that c (the event where N. pushed the button) would have been present;
rather, c would still have been absent but e would have been produced by a mir-
acle immediately before e.
Lewis’ analysis of such cases presupposes that the counterfactual can be
evaluated on the basis of a possible world in which a miracle occurs within the
stretch of time between the events described by the antecedent and the consequent
of the counterfactual. However, this destroys the capacity of that world to ground
our counterfactual judgment. The possibility of reasoning in a counterfactual
way presupposes that we consider that the laws of the actual world are valid with-
out exception in the world on the basis of which we evaluate the counterfactual
and in which its antecedent is true, within the stretch of time between the events
described by the antecedent and the consequent.
Indeed, on what can we ground our intuitive judgments of the truth values of
counterfactuals, if not on the assumption that the counterfactual world in which

65 «The big miracle required for perfect reconvergence consists of a multitude of little miracles, spread

out and diverse»: LEWIS, Philosophical Papers…, 47.

66 See REICHENBACH, The Direction of Time...

67 LEWIS, Philosophical Papers…, 170.

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 651

Causation in contemporary analytical philosophy 651

the antecedent is true, is governed by the same laws as the actual world? Tradi-
tionally68, one of the main functions a knowledge of laws is thought to have is to
allow us to judge which course events would have taken in counterfactual cir-
cumstances. Our capacity to make such judgments is necessary in order to make
decisions: we evaluate the consequences of each possible action we might
choose. We can only hypothesize these consequences on the condition of as-
suming that the laws of nature do not change after such a possible act. It seems
therefore necessary to ground the evaluation of counterfactuals on the require-
ment that the possible world to be considered for that evaluation must be clos-
est to ours with respect to the facts existing at the time of the antecedent, and
must obey the same laws as the actual world during the temporal interval be-
tween the antecedent and the consequent. The requirement of holding the actu-
al laws constant at least for that time span, forces us to situate the miracle before
the consequent (the event where N. pushes the button). In this case, there are no
principled obstacles any more to backwards counterfactual reasoning. The coun-
terfactual is true if the consequent is true in the world that is closest to the ac-
tual world with respect to the facts at the time of the antecedent (posterior in
time) and in which, during the interval between antecedent and consequent, the
laws are the same as the actual laws69. Even if not all laws applying to a given
situation are reversible, it is sufficient for the truth of a given backwards coun-
terfactual that the consequent event c is, in the circumstances, necessary for the
antecedent event e. The backwards counterfactual about Nixon is true if, in the
circumstances, other possible causes that could have triggered the holocaust
without the intervention of a miracle are situated in possible worlds more distant
from the actual world than the world where N. triggers the holocaust by pushing
the button.
To the extent that this argument refutes Lewis’ thesis that no backwards coun-
terfactual is true, his defence of the counterfactual theory of causation, threat-
ened by the counterexamples mentioned above, is deprived of its main concep-
tual resource. If backwards counterfactuals can be true, the counterfactual the-
ory is defeated by the problems of effects and of epiphenomena.
So far, I have agreed to reason in terms of “miraculous” worlds. As Lewis is

68 Cfr. W. KNEALE, Probability and Induction, Oxford University Press, Oxford 1949; N. GOODMAN,

Fact, Fiction and Forecast [1955], Bobbs-Merrill, Indianapolis 19733; E. NAGEL, The Structure of Science,
Routledge and Kegan Paul, London 1961; F. JACKSON, A Causal Theory of Conterfactuals, «Australasian
Journal of Philosophy», 55 (1977), 3-21; D.M. ARMSTRONG, What is a Law of Nature?, Cambridge Uni-
versity Press, Cambridge 1983; D.W. ZIMMERMANN, Shoemaker's Argument for his Theory of Properties,
«Facta philosophica», 2 (2000), 271-290.
69 See JACKSON, A Causal Theory…, 9ff.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 652

652 Max Kistler

careful to point out70, miraculous worlds are not inconsistent: they do not con-
tain any violations of their own laws. Rather, they appear miraculous from the
point of view of the actual world. Their laws (those of w1, which are strictly fol-
lowed in w1) differ from ours in such a way as to permit that the course of events
differs from the actual course only in a tiny portion of space-time but nowhere
else. However, the conception of such a world seems to contain a contradiction.
By definition, a law is a constraint determining a global regularity71; hence, even
a tiny difference between two possible worlds as to the laws valid in them, must
have as a consequence a vast divergence in facts and particular events through-
out all past and future. Furthermore, any change with respect to a particular fact,
even as tiny as a few extra neurons firing in N.’s brain, requires the violation of
several laws, beginning with the laws of conservation. The counterfactual analy-
sis faces the following dilemma.
Either the worlds that are candidates for being the basis of evaluation of a
counterfactual because they are very similar to ours are coherent worlds in the
sense that their laws are strictly valid. But then these laws must be the same as
our actual laws, otherwise their difference from actuality would necessarily be
vast with respect to particular facts, and we would have no means to make judg-
ments about what happens in them because we cannot reason on the basis of laws
that differ from ours in an unknown way.
Or else these worlds are truly miraculous. However, a contradictory world
whose laws are no laws (because they are violated) cannot serve as a basis for
the evaluation of a counterfactual.
It is possible to avoid the consideration of miraculous possible worlds, by
grounding the evaluation of counterfactuals in possible worlds that strictly share
the actual laws but that lack the counterpart of a part of the history of the actu-
al world72. To evaluate a “forward” counterfactual where the antecedent con-
cerns an event at t, one considers a possible world containing the counterpart of
a temporal part of the actual world, which extends from t to infinity in future, but
which does not contain any counterpart of the actual past, before t. This solution
requires rejecting Lewis’ realist conception which has it that all possible worlds
are as real as the actual world. It presupposes conceiving them as devices for
reasoning. As such they must be internally consistent but need not be possible
complete worlds. A complex world beginning at t without having a past is a pos-
sible world in this sense, although it could not possibly be a complete world.

LEWIS, Philosophical Papers…, 44-5.

70

According to regularity theories of laws, a law simply is a global regularity.

71

72 See M. KISTLER, L'identité des propriétés et la nécessité des lois de la nature, «Cahiers de philoso-

phie de l'Université de Caen, Numéro spécial: Le réalisme des universaux», 38-39 (2002).
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 653

Causation in contemporary analytical philosophy 653

3. The interventionist or manipulability analysis of causation

As we have seen, the asymmetry of causation raises important difficulties for

both the nomological and the counterfactual approach to causation. The inter-
ventionist analysis73 tries to overcome them by grounding that asymmetry in the
asymmetry between the past, which we cannot influence, and the future, which
presents itself to us agents as an «open ‘space’ of possibilities»74. If A is the oc-
currence of an avalanche, and B and C are the collapses of two houses caused
by A, on what ground can we reject the following problematic counterfactuals:
the first, which raises the “problem of effects”, explicitly expresses backwards
reasoning: “if B had not occurred, then A would not have occurred”. The sec-
ond, which raises the “problem of epiphenomena”, also presupposes, although
only implicitly, backwards reasoning: “if B had not occurred, then C would not
have occurred”. The interventionist analysis proposes judging their truth value
on the basis of a possible world differing from ours simply by a human interven-
tion, which plays the role of Lewis’ miracle. The worlds we compare in order to
evaluate counterfactuals share our actual laws, but differ with respect to a free
action. In the case of the example, the minimal intervention that prevents B con-
sists in directly protecting the house in B. But if this had been done, A as well
as C would still have occurred, which is what grounds our judgment that the two
controversial counterfactuals are false75.
More generally, the interventionist or manipulability account of causation76
reduces the existence of a causal relation between two distinct events A and B
to the fact that «bringing about the occurrence of A would be an effective means
by which a free agent could bring about the occurrence of B»77. However, this
proposal encounters two major problems78. First, it is anthropocentric, in that it
postulates a conceptual dependency between causation and our human capaci-
ty to act. Not content to note that our actions are causes, it holds that generally,

73 D. GASKING, Causation and Recipes, «Mind», 64 (1955), 479-487; G.H. VON WRIGHT, Explanation

and Understanding, Cornell University Press, Ithaca, N.Y. 1971; CARTWRIGHT, How the Laws…; H. PRICE,
Agency and Causal Asymmetry, «Mind», 101 (1992), 501-520; P. MENZIES AND H. PRICE, Causation as a
Secondary Quality, «British J. Phil. of Science», 44 (1993), 187-203; KEIL, Handeln…
74 KEIL, Handeln…, 440.

75 This is Keil’s way of reasoning. Cfr. KEIL, Handeln…, 442.

76 Menzies and Price call it the «agency theory» (MENZIES AND PRICE, Causation as a Secondary…,

187).
77 MENZIES AND PRICE, Causation as a Secondary…, 187.

78 Hausman mentions three difficulties: «The manipulability theory is narrowly circular, unjustifiably

anthropomorphic, and it makes causal inferences on the basis of passive observation needlessly prob-
lematic»: D. HAUSMAN, Causation and Experimentation, «American Philosophical Quarterly», 23 (1986),
143-154, 145.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 654

654 Max Kistler

any event c is a cause of another event e if and only if a human agent could ob-
tain e by bringing about c. This approach is condemned to consider causal rela-
tions taking place without the reach of possible human actions to be causal on-
ly by analogy. It is in principle impossible for a human agent to bring about the
explosion of a supernova. Hence, the interventionist analysis doesn’t apply to
that process and cannot explain, at least not directly, why it is causal. Further-
more, decisions to consider one relation or process to be causal because analo-
gous, or similar enough, to an action, and another relation or process not to be
causal because not similar enough, seem arbitrary. Hence, the theory seems to
have no clear application to causal relations taking place without the reach of
possible human actions.
The second problem stems from the circularity of the account. The interven-
tionist account uses our capacity to act as a conceptual tool to analyse causation
in general. However, when we act, we cause events. Hence, the concept of ac-
tion presupposes the concept of causation. The circularity can be avoided if one
explicitly takes the concept of action to be primitive and unanalysable, in par-
ticular to be incapable of an analysis in causal terms79. However, in that case we
don’t seem to have any analysis of the concept of causation at all, but only the
thesis that a certain species of causal interactions, namely human actions, are
paradigmatic and can serve as a model for the whole genus. In the absence of a
direct analysis of what characterizes that genus, i.e. causation itself, the theory
leaves us in the dark as to what, among the properties of human actions, can be
generalized to all causal relations.
This critique is not meant to deny the fact, well established by von Wright80,
that all knowledge of objective causal relations must rely on experimental inter-
ventions81. Our capacity to intervene in the natural course of events is our only
epistemic handle on causation. In von Wright’s own words,

«it is established that there is a causal connection between p and q when we have sat-
isfied ourselves that, by manipulating the one factor, we can achieve or bring it about
that the other is, or is not, there. We usually satisfy ourselves as to this by making ex-
periments»82.

79 The dependence of the concept of intervention on the concept of causation makes it impossible to

use the interventionist theory in the causal theory of action. According to the latter, actions differ from
other bodily movements by their causal origin in an appropriate mental event. However, it is obviously
circular to analyse both action in terms of causation, and causation in terms of action. Cfr. KEIL, Han-
deln…, chap. III, 3.
80 VON WRIGHT, Explanation…

81 These methods of investigation of causal relations are the object of John Stuart Mill’s (J.S. MILL, A

System of Logic, Ratiocinative and Inductive [1843], Parker, London 1848) famous classification.
82 VON WRIGHT, Explanation…, 72; italics von Wright’s.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 655

Causation in contemporary analytical philosophy 655

However, we can acknowledge this important epistemic fact about our knowl-
edge of causation, without obscuring the fact that natural causal relations (which
are not actions) do not, for their objective existence, depend on human actions.

4. The probabilistic theory of causation

The fundamental thesis of the probabilistic approach to causation is that A is

cause of B if and only if the occurrence of A raises the chance of the occurrence
of B. The theory has a twofold origin. It can be considered as a weakened ver-
sion of the deductive-nomological theory of causation, which is obtained by sub-
stituting probabilistic or statistical laws for the deterministic laws in the ex-
planans83. However, the theory can also be seen to be motivated by the theory of
rational choice84. Here is the link. It is rational for an agent to do A in order to
obtain B if and only if the agent believes that A causes B. According to the nor-
mative theory of choice, it is rational to do A in order to obtain B if and only if
the agent believes that P(B|A) > P(B|¬A). If both equivalences are correct, one
can conclude that A causes B if and only if P(B|A) > P(B|¬A)85.
The major problem with this approach concerns its application to the analy-
sis of singular causation, i.e. to causal relations between particular pairs of
events. To take an example well known in the debate on probabilistic causation,
let us assume that the chance of developing lung cancer is higher among smok-
ers than non-smokers, in the sense that the frequency of cancer is higher in the
population of smokers than in that of non-smokers. The probabilistic theory of
causation, in its simplest version, concludes that smoking is a cause of cancer.
However, the statistical correlation between two populations is not sufficient to
justify a causal judgment in any individual case86. It seems perfectly compati-

83 Frederick Suppes introduces the theory in this way. Cfr. F. SUPPES, A Probabilistic Theory of Causal-

ity, North Holland, Amsterdam 1970.

84 In this respect, it can be considered as a variant of the manipulability theory, discussed in the pre-

ceding section.
85 Papineau (D. PAPINEAU, Pure, Mixed, and Spurious Probabilites and Their Significance for a Re-

ductionist Theory of Caustion, in Minnesota Studies in the Philosophy of Science, vol XII: Scientific Ex-
planation, ed. by P. Kitcher, University of Minnesota Press, Minneapolis 1989, 307-356, 321 and 328)
and Mellor (D.H. MELLOR, The Facts of Causation, Routledge, London and New York 1995, chap. 7) pre-
sent this argument as a major motivation for the probabilistic theory. It is certainly not conclusive, in that
it only allows one to draw the conclusion that the beliefs are equivalent: it shows that it is equivalent to be-
lieve that A causes B and to believe that P(B|A) > P(B|-A). It is of course invalid to conclude from this,
without further argument, that the objects of these beliefs are equivalent too.
86 «Given the conceptual independence of token-level causal facts from type-level causal facts, it

should not be surprising that what is true at the type level is […] conceptually independent of what is true
of the token level, and that token-level causation cannot be straightforwardly understood in terms of type-
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 656

656 Max Kistler

ble with our assumption that John’s smoking does not cause cancer in him, ei-
ther because he doesn’t get cancer at all or because his cancer is causally due
to some other factor, e.g. to the inhalation of asbestos dust. Hence, higher chance
of occurrence of an event of type B, within a population that also satisfies A, in
comparison with the chance of B in a population that does not satisfy A, is not
sufficient for concluding that the occurrence of event a (of type A) causes event
b (of type B), even if both have occurred. The fact that A raises the odds of B is
not a necessary condition for a’s causing b either, for an individual event a (of
type A) can well cause the individual event b (of type B) while in the general
case, the occurrence of events of type A lowers the probability of occurrence of
an event of type B. This is illustrated in the following case87: Moriarty, Watson,
and an unstable rock are at the edge of a cliff. Holmes stands at the foot of the
cliff. Watson must leave, but he knows that if he doesn’t act before leaving, Mo-
riarty will push the rock over the edge of the cliff, with the intention of killing
Holmes. The only thing Watson can do in order to try to save Holmes, is to push
the rock himself hard enough for it to fall down beyond Holmes. Watson man-
ages to get hold of the rock and pushes it as hard as he can. Unfortunately, and
improbably, he fails and Holmes is killed. The cause of Holmes’ death is an event
of a type which diminishes the chance of his death. The event of pushing the rock
«had a tendency to prevent»88 Holmes’ death, but has nevertheless caused it.
The probabilistic theory seems ill suited for the analysis of causal relations
between particular events for two other reasons. The terms of probabilistic cor-
relations are abstract entities, such as properties, factors, types of events or prop-
erties of events, e.g. smoking or developing cancer. These entities are situated
neither in space nor in time. They cannot therefore be causally related strictly
speaking, to the extent that causality links entities situated in space and time,
generally conceived of as events. This suggests that the probabilistic theory
might really be a theory not so much of causation, but of explanation or of the
nomological link between properties, in other words, of laws of nature. In the D-
N model of explanation, it could play both of these roles.
The second reason for the inadequacy of the probabilistic theory for the

level causal relations»: E. EELLS, Probabilistic Causality, Cambridge University Press, Cambridge 1991,
16.
87 The example is due to I.J. GOOD, A Causal Calculus I-II [1961-62], in I.J. GOOD, Good Thinking,

University of Minnesota Press, Minneapolis 1983. Other examples can be found in E. EELLS and E. SOBER,
Probabilistic Causality and the Question of Transitivity, «Phil. of Science», 50 (1983), 35-57; EELLS, Prob-
abilistic Causality, 281ff.; M. HEIDELBERGER, Kausalität: Eine Problemübersicht, «Neue Hefte für Philoso-
phie», 32-33 (1992), 130-153, 151; D. EDGINGTON, Mellor on Chance and Causation, «British Journal for
the Philisophy of Science», 48 (1997), 411-433, 420.
88 GOOD, A Causal Calculus…, 216-217.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 657

Causation in contemporary analytical philosophy 657

analysis of causation between spatio-temporal events is this. The statistical cor-

relations that constitute the point of departure of the probabilistic analysis are
often spurious, in the sense that the correlation exists between two (types of) ef-
fects of a common (type of) cause, but where these effects are not linked as cause
and effect. Hence, the theory needs to be completed by the following require-
ment: the fact that A raises the probability of B justifies belief in the existence
of a causal relation between (events of type) A and B, only if there is no third fac-
tor that is the real cause of both. Such a factor is called a ‘screening factor’. The
correlation between barometer fallings and storms is spurious; the theory pre-
dicts that it is not causal because the fall in air pressure constitutes a screening
factor: Within a set of situations in which air pressure is constant, the correla-
tion between barometer fallings and storms disappears. However, this strategy
cannot be generalised. Insofar as the sets, within which the conditional proba-
bilities are evaluated, contain more than one element, it is always possible to
partition the set in such a way that the conditional probabilities give a different
result, while of course such a change in the calculation of the probability does-
n’t affect the reality of the causal relations. This fact is known as ‘Simpson’s para-
dox’89. As an illustration, take the famous case of the alleged sexist discrimina-
tion in the admissions policy of the University of Berkeley. Cartwright90 cites a
study showing that the frequency of admission at Berkeley is higher among male
than among female applicants. However, the existence of this statistical correla-
tion does not allow inferring the existence of a “causal”91 relation between be-
ing a male applicant and the fact of being favoured in the selection. For it can
be shown that there is another partition of the set of candidates that inverts the
result, if the relevant conditional probabilities are evaluated separately within
the sets of this partition: The partition with respect to the different departments
selecting the candidates. It has turned out that the correlation between the gen-
der of the candidate and her admission is not grounded in the existence of a
“causal” relation between these two factors, but on a common “causal” factor:
men have the tendency to candidate in departments in which it is easier to be
accepted, while women candidate more often in the more selective departments.
To simplify, let us consider a case in which there are only two departments and
where the relevant frequencies are these92:

89 After E.H. SIMPSON, The Interpretation of Interaction in Contingency Tables, «J. of the Royal Statis-

tical Society», Ser. B, 13 (1951), 238-241. Cfr. EELLS, Probabilistic Causality, chap. 2.
90 CARTWRIGHT, How the Laws…, 37.

91 As I have already noted, calling abstract entities, such as properties or factors, ‘causes’, seems to

be a category mistake. With respect to these entities, one should rather speak of nomological dependen-
cy, which is indeed a domain where the probabilistic theory can fruitfully be deployed.
92 The numbers are taken from EELLS, Probabilistic Causality, 63.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 658

658 Max Kistler

Department 1 Department 2 Total

Male 81/90 2/10 83/100

Female 9/10 18/90 27/100

In this particular situation, global evaluation of the probabilities – without dis-

tinguishing between the two departments – yields the result that the chance to
be accepted is more than three times higher for male than for female candidates.
However, the correlation between the gender of the candidates and their admis-
sion disappears if the probabilities are evaluated within the subsets obtained by
partitioning the set of candidates according to departments. The first department
accepts 90% of the male but also 90% of the female candidates, whereas the sec-
ond department accepts only 20% of men but also only 20% of female candi-
dates. Eells93 shows that the existence of a global positive correlation between
being male and being accepted is not only compatible with the absence of such
a positive correlation (under a different partition), but also with the existence of
a negative correlation (i.e. the fact that the departments in fact admit women
more often than men) in the case of an evaluation by departments. It can also
happen94 that the evaluation on the basis of a partition in departments shows that
there really is discrimination against women, but that it is smaller than the glob-
al evaluation would have made it appear. In short, it is impossible to draw any
conclusive causal information whatsoever from an analysis of conditional prob-
abilities calculated in an inhomogeneous set. To the extent that it is possible to
introduce new partitions by paying attention to new factors, it can always hap-
pen that the conditional probabilities change radically. The fact that judgments
of statistical correlation always remain relative to the choice of partition, and that
the direction of correlation may depend on that choice, makes them inappropri-
ate to ground absolute causal judgments.
The only possible way out seems to be to require, as Cartwright does, that the
probabilities be evaluated within absolutely homogeneous sets, so that no
causally relevant factor is left hidden from the analysis95. However, this solution
tends to deprive the theory of its probabilistic elements, and to make it equiva-

93EELLS, Probabilistic Causality, 66f.

94See EELLS, Probabilistic Causality, 68.
95 According to Cartwright, conditional probabilities must be evaluated «holding fixed other causal

factors» (CARTWRIGHT, How the Laws…, 28), and one is only justified in inferring the existence of a causal
relation on the basis of a rise in probability, if the set of situations over which the probabilities are taken
are «homogeneous with respect to all other causal factors» (CARTWRIGHT, How the Laws…, 25). However,
when it comes to considering real events, there are always other causal factors; no pair of concrete situa-
tions is homogeneous with respect to all causal factors.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 659

Causation in contemporary analytical philosophy 659

lent to the nomological theory. For the case of deterministic causal relations, the
relation of the conditional probability that B, given that A, reduces to the nom-
ic relation according to which B nomically depends on A. As Eells shows, in or-
der to evaluate correctly the dependence of B on A, all situations must be ex-
cluded in which there exists a correlation between A and B but where A is not
causally related to B, and all situations in which the correlation between A and
B is enhanced or diminished by factors influencing B, but which are not caused
by A. Hence, to judge whether A causes B, one evaluates the probability of B,
given that A, in all populations that are of types A and B but are perfectly ho-
mogeneous with respect to all potentially interfering factors. This undermines
the ambition of the probabilistic theory to deliver an analysis of the causal rela-
tion, for the following reason.
In a deterministic world, if all potentially interfering external factors are held
constant (let us assume that they are systematically absent), the probability that
Y, given that X, is 1 if X causes Y, and 0 if X does not cause Y. The intermedi-
ate values are the result of the variability of the external interfering factors,
which are here excluded. This undermines the fundamental idea of the proba-
bilistic theory of causation. The intermediate probabilities do not measure X’s
impact on Y (which can only exist or not exist) but the probability of interfer-
ences by external factors. However, that probability is independent of the rela-
tion between X and Y itself. Apart from the case of irreducibly indeterministic
processes, the content of the attribution of intermediary probabilities does there-
fore not bear on the causal relation itself, but on our ignorance of the interfering
factors. Therefore, the probabilistic theory does not give a new analysis of the
causal relation itself: with respect to objective causation, the probabilistic theo-
ry reduces to the nomological theory.
The probabilistic theory draws its plausibility from the analogy between the
evaluation of conditional probabilities and the evaluation of a series of experi-
ences, following Mill’s methods. It is practically impossible to exclude all inter-
fering factors, and the probability that B, given an experience where A has been
observed and where the interfering factors have been eliminated to the extent
that this is practically possible, will be intermediary. But the reason does not lie
with the relation between A and B itself.
Eells96 proposes a probabilistic theory of particular causal relations, which
tries to reduce what I shall call ‘causal responsibility’ of a fact for another fact,
to a certain evolution of the probability of the effect in the course of time, be-
ginning with the time of occurrence of the cause, depending on this cause and

96 EELLS, Probabilistic Causality, chap. 6.

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 660

660 Max Kistler

on the external interfering factors. However, the probabilistic theory of particu-

lar causal relations encounters the same difficulties as the probabilistic theory
of the causal dependence between factors.
In the case of particular causal relations, the probability of the effect given
the cause is conceived as the limit of the frequencies of the effect in an infinite
series of populations in which the cause is present. This series contains an infi-
nite number of copies of the situation in its totality, with all interfering factors
included. However, in a deterministic universe, to the extent that the situations
in this series do not differ at all with respect to all factors that affect directly or
indirectly the effect, if fact X is causally responsible for fact Y, then the proba-
bility that Y, given X, is 1. In this case, the resulting analysis of the relation of
causal responsibility is not probabilistic any more. The probabilistic theory of
causation, at least in its most elaborate form given by Eells, fails to solve the
problem Russell’s second argument raised against the regularity theory of cau-
sation. The attribution of probabilities below 1 is only justified to the extent that
part of the context is ignored.

5. Transference theories of causation

The general conclusion of the analysis of the alternatives to the deductive-nomo-

logical theory of causation examined so far is that two of these, the counterfac-
tual and the probabilistic theory, are better suited as theories of nomic depen-
dence than as theories of causation. This seems particularly clear in the case of
the probabilistic theory: its object is the analysis of systematic correlations be-
tween different properties (or factors, or between the sets that are their exten-
sions). These correlations can reveal the existence of nomic dependencies ex-
isting between these properties97. However, they do not give any conclusive ev-
idence about the existence of a causal relation between the events exemplifying
these properties. As we have seen in our critical analysis of the deductive-nomo-
logical conception, nomic dependence is not equivalent to causation. Two prop-
erties can be nomically related even where two events exemplifying these prop-
erties are not related as cause and effect. However, I have also tried to show (what
is more controversial) that the theory of causation in terms of counterfactual de-
pendence can also more easily be interpreted as a theory of nomic dependence.
The antecedent of a counterfactual is a proposition asserting the exemplification
of a property (expressed by the predicate of the proposition). In order to deter-

97 On the major contemporary conceptions of the laws of nature, see KISTLER, Causalité…, chap. 2,

and BARBEROUSSE, KISTLER, LUDWIG, La philosophie…, chap. 4.

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 661

Causation in contemporary analytical philosophy 661

mine whether the proposition expressed by the consequent is true in the nearest
possible world in which the antecedent is true, what one has to do is to examine
whether there exists a nomic dependence between the property expressed by the
predicate of the antecedent and the property expressed by the predicate of the
consequent. To answer this question, it is necessary to contemplate possible
worlds sharing our laws of nature.
The fact that the proper object of these theories is nomic dependency between
properties, rather than the causal relations between particular events, is the deep
reason for their failure to avoid counterexamples presented by situations in
which A is nomically linked to B, but where an event a exemplifying A is nev-
ertheless no cause of an event b exemplifying B: either because b is cause of a,
or because a and b are effects of a common cause, or because b has been caused
by a preempting cause c. Hence, the key to the solution to these problems lies
in the search for an account that is no longer at the level of properties but at the
level of the spatio-temporal events that are the relata of particular causal rela-
tions98.
The most promising approach in this respect begins with the following hy-
pothesis. What makes it the case that event a is a cause of event b, is the fact
that something is transferred between these events. It has been proposed99 that
the fundamental characteristic of causal processes is their capacity to transmit
a mark, where a mark is meant to be «the result of an intervention by means of
an irreversible process»100. A paradigmatic case of the introduction of a mark,
which then propagates all along a causal process is the passage of a light ray
through a red filter. The colour is transferred by the ray as long as that process
does not undergo any new interaction. In this respect, causal processes contrast
with pseudo-processes, a clear example of which has been given by Salmon101.
A beacon rotates at the centre of a hollow cylinder and projects a spot of white
light on the internal wall. The world line that consists of the series of illuminat-
ed portions of the wall is a pseudo-process, rather than a causal process, be-
cause, if a red filter is posed on one of the portions of the wall traversed by the
light spot, the spot will be marked red but this mark will not be transmitted along
the pseudo-process which will consist of white light just as it did before that
moment.

98 This is also Ehring’s conclusion. Cfr. D. EHRING, Causation and persistence: A Theory of Causation.

Oxford University Press, New York 1997.

99 See REICHENBACH, The Direction of Time, 198f.; W. SALMON, Scientific Explanation and the Causal

Structure of the World, Princeton University Press, Princeton 1984.

100 REICHENBACH, The Direction of Time, 198. The similarity with the manipulability account is only

apparent. The interactions of transference theory do not necessarily result from human interventions.
101 SALMON, Scientific Explanation…, 141-142.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 662

662 Max Kistler

However, this version of a theory of causation in terms of transference faces

a decisive objection102. The problem is that the definitions of the “ability to
transmit a mark” and of “causal process” form a vicious circle103. The main the-
sis of the theory says that a causal process is a process that is able to transmit a
mark. However, the notion of a mark can only be defined by using the concept
of a causal process. For it is necessary to require that only causal interactions
be able to introduce a mark in a process. A mark is an alteration of a property
due to a local singular interaction. The requirement that this interaction be
causal is necessary, for otherwise the intersection between two pseudo-process-
es could introduce a mark, and this would contradict the very idea of a pseudo-
process.
To escape this circle, Dowe104 has proposed that two events are linked as
cause and effect if and only if they manifest the same amount of a conserved
quantity (from now on “CQ”). Dowe’s account is explicitly opposed to the idea
that such amounts of CQ can be transferred, and cannot therefore be considered
as a version of a transference theory of causation. But it seems natural to use the
concept of amount of a conserved quantity to generalize a traditional variant of
the transference theory according to which causal relations are transmissions of
energy105. In the context of the contemporary conception of conserved quantities
grounded on symmetries, it seems arbitrary to ground causation in the transmis-
sion of energy exclusively, as it would be to ground it in any other particular set
of quantities, such as momentum106, force107, or information108. The idea of gen-
eralizing to any amount of any CQ as the material ground of causal relations can
be justified by reflection on the concept of transference itself. Only what is con-

102 Even the theory’s major advocate, Wesley Salmon, has abandoned it for this reason. Cfr. W. SALMON,

Causality Without Counterfactuals, «Philosophy of Science», 61 (1994), 297-312.

103 Cfr. KITCHER, P. DOWE, Wesley Salmon's Process Theory of Causality and the Conserved Quantity

Theory, «Phil. of Science», 59 (1992), 195-216, 200-201; KISTLER, Causalité…, 89.

104 DOWE, Wesley Salmon's….

105 This idea has been put forward since the discovery of the law of the conservation of energy in the

19th century, in particular by Julius Robert Mayer, and by the school of “energetism”, in particular W.
Ostwald. Cfr. W. KRAJEWSKI, Energetic, informational, and triggering causes, «Erkenntnis», 46 (1997),
193-202. It has more recently been taken up by K. LORENZ, Kants Lehre vom Apriorischen im Lichte gegen-
wärtiger Biologie, «Blätter für deutsche Philosophie», 15 (1941), and in the literature in English by Aron-
son (J. J. ARONSON, The Legacy of Hume's Analysis of Causation, «Studies in the History and Philosophy
of Science», 2 (1971), 135-165; J. J. ARONSON, On the Grammar of ‘Cause’, «Synthese», 22 (1971), 414-
430) and Fair (D. FAIR, Causation and the Flow of Energy, «Erkenntnis», 14 (1979), 219-250).
106 Cfr. FAIR, Causation...

107 Cfr. ARONSON, The Legacy…

108 Cfr. W. KRAJEWSKI, Four Conceptions of Causation, in W. KRAJEWSKI (ed. by), Polish Essays in the

Philosophy of the Natural Sciences, Reidel, Dordrecht 1982.

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 663

Causation in contemporary analytical philosophy 663

served during transmission can be transmitted. This reasoning leads to the pro-
posal to reduce the causal relation between two particular events to the trans-
mission between them of some amount of a CQ109. It is up to science to establish
the explicit list of all the CQ that exist in nature and that can therefore ground
causal relations.
This idea can be elaborated so as to provide a satisfactory account of the sit-
uations that raise insuperable problems for the theories we examined earlier. The
first qualification concerns the asymmetry of causation. The concept of trans-
ference is symmetrical: the individual amount of CQ X is transferred from a to b
if it is present both in event a and in event b. In a possible world in which all
processes are reversible, there is transmission and therefore causation, but
causal relations (and with them, time, if the direction of time is reduced to the
direction of causation) are symmetric. In such a world, there is no intrinsic dif-
ference between the two terms of a causal relation, which makes one the cause
rather than the other. However, in the actual world, most processes are irre-
versible. This irreversibility is at the origin of the asymmetry of both causation
and time. Borrowing from Reichenbach110 the concept of a causal net in which
all events are causally linked to other events, so that no event is causally ab-
solutely isolated, it is possible to consider that insofar as a large majority of the
processes in the net possess the same intrinsic direction, which is the direction
of growing entropy, that direction is the direction of the entire net. In this way, it
is possible to attribute a direction even to intrinsically symmetrical processes,
i.e. to reversible processes.
The problem of effects can then be solved by appealing to the empirical fact
that the majority of actual causal processes have the same intrinsic direction: a
is cause of b because first, an individual amount of a CQ is transmitted between
them and second, because either the transmission process between a and b is
physically asymmetrical and irreversible or it is reversible but part of a network
of causal processes possessing a global direction and imposing it on the trans-
mission process between a and b.
The transference theory overcomes the problems of epiphenomena, preemp-
tion and causal overdetermination rather simply. There is no transmission what-
soever between the events of Jones and Smith watching the same TV program.
Although these events are perfectly correlated qualitatively, special relativity
forbids transmission between them, which would need to have superluminal

109 Cfr. M. KISTLER, Reducing Causality to Transmission, «Erkenntnis», 48 (1998), 1-24; KISTLER,

Causalité…, chap. 1.
110 REICHENBACH, The Direction of Time.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 664

664 Max Kistler

speed111. This provides a simple explanation for the absence of a causal relation
between them. In the case of preemption considered above, the excess weight
has transmitted kinetic energy to the bridge, which is transformed into deforma-
tion energy during the collapse. The explosion of the bomb has transmitted en-
ergy, but not to the bridge, for when the shock wave transmitting this energy ar-
rives, the bridge has already (started to) collapse.
However, transference theory faces its own problems. Sosa and Tooley judge
that «the prospects for a physicalistic reduction of causation do not […] appear
very bright»112, for three reasons. The first two concern only Fair’s version of the
theory113, but not the version sketched above. Sosa and Tooley’s third argument,
however, raises a challenge with respect to clarifying the conceptual status of the
theory. Their first objection to transference theories is that the causal relation
can only be reduced to a type of physical relation if the latter has a direction that
is not itself causally grounded. It is indeed plausible to think that the common
sense concept of transmission is only asymmetrical because its meaning has a
causal and probably even anthropomorphic aspect, its paradigmatic application
being to actions of transference of an object from one place to another. To avoid
both circularity and anthropomorphism, Fair proposes to ground causal asym-
metry on temporal asymmetry. Of two events linked by transmission, the earlier
causes the later. However, we have already noted that first, this strategy makes
the reduction of the asymmetry of time to the asymmetry of causation impossi-
ble and second, that is stipulates for a priori reasons that there is no backwards
causation, whereas this seems to be an empirical issue. We have already indi-
cated the solution to this difficulty. It is to conceive of transference of an amount
of a CQ as the presence of the individual amount in two different events. This
conception is neither implicitly causal nor anthropomorphic. However, it is sym-
metrical. Within this conception, the asymmetries both of causation and of time
(i.e. time’s having a privileged direction) are not a priori consequences of the
concept of causation. Rather, their asymmetry is a contingent property of the
causal relations of the actual world, grounded in the fact that a large majority of
actual causal processes are intrinsically asymmetrical and irreversible: they are
processes with growing entropy114.
The second objection against transference theories that Sosa and Tooley men-

111 The distance between the events of reception is called “space-like”, whereas only “time-like” dis-

tances can be bridged by causal processes, in particular by signals.

112 E. SOSA and M. TOOLEY, Introduction, in E. SOSA and M. TOOLEY (ed. by), Causation, Oxford Uni-

versity Press, Oxford 1993, 4.

113 D. FAIR, Causation and the Flow of Energy, «Erkenntnis», 14 (1979), 219-250.

114 There exist also microphysical processes that are intrinsically asymmetrical, in particular the dis-

integration of K-mesons, also called “kaons”. See DOWE, Wesley Salmon's…, 189.
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 665

Causation in contemporary analytical philosophy 665

tion concerns the causal theory of persistence. There are two major conceptions
of the persistence of macroscopic objects in time. According to “three-dimen-
sionalism” the identity of ordinary objects is determined only in the three spa-
tial dimensions but not in the temporal dimension. Such a three-dimensional ob-
ject, traditionally called a ‘substance’, ‘endures’115 over time without having any
temporal parts, simply by being entirely present at every moment of its existence.
The idea of something being entirely present at many instants of time has its ma-
jor application in the traditional conception of universals, according to which a
universal, say red, is entirely present in every one of its instances, in red objects.
Both substances and universals can be entirely present at more than one place
and time. However, this is inconceivable for four-dimensional entities, such as
events, whose identity is determined according to both space and time. An event
taking place within a space-time region cannot be present at a different space-
time region. Only its parts occupy different space-time regions. According to
“four-dimensionalism” ordinary objects are extended both in space and time,
and have temporal parts. It conceives of the persistence of an object over time
as of “perdurance”, based on a relation between its temporal parts.
For three-dimensionalism, the identity of an object though time, its
«genidentity»116, is a true primitive and irreducible identity. For four-dimen-
sionalism, however, it is no identity in the strict logical sense; it is rather based
on a relation between the different temporal parts of the object, where the rela-
tion may be conceived as causal. However – and this is Sosa and Tooley’s sec-
ond objection against Fair’s version of the transference theory – insofar as the
causal relation is conceived on the model of transference between different ob-
jects, the relation between different temporal parts of one object cannot be con-
ceived as being causal; hence, the four-dimensional (or “perdurance”) view of
persistence through time loses its most plausible foundation. However, the ver-
sion of transference theory sketched above, which takes the terms of causal re-
lations to be events, and not objects, avoids this objection: it can explain the per-
sistence of objects on the basis of a causal relation between the temporal parts
of the object, which are events, thanks to the transmission of amounts of differ-
ent CQ, first of all mass-energy.
The third objection concerns the status of the theory. Transference theory im-
plies a radical change in the very conception of the problem of the analysis of
the concept of causation, insofar as it does not try to provide a purely conceptu-

115 The terms ‘enduring’ and ‘perduring’ (see below) have been introduced by Mark Johnston. Cfr. D.

LEWIS, On the Plurality of Worlds, Blackwell, Oxford 1986.

116 On the notion of genidentity, see M. KISTLER, Le concept de génidentité, in S. LAUGIER (éd. par), Car-

nap et la construction logique du monde, Vrin, Paris 2001, 163-188.

24_Kistler_2.QXD 2-09-2003 11:37 Pagina 666

666 Max Kistler

al and a priori analysis, as the alternative theories do. Rather, its fundamental
hypothesis is that the causal relation is a natural kind of relation whose nature
needs to be discovered a posteriori, by empirical and in principle testable spec-
ulation. In traditional terminology, it aims to discover the “real essence” of the
causal relation117. The discovery of telepathy or other cases of (simultaneous) ac-
tion at a distance would, e.g., constitute an empirical refutation of transference
theory, whereas the alternative conceptions can only be refuted by a priori con-
ceptual arguments. The instantaneous correlations between distant but “entan-
gled” physical systems predicted by quantum mechanics and experimentally ob-
served, do not refute it because they are not causal. They cannot be used to serve
actions at a distance (instantaneously bringing about their effect), such as sig-
nalling or producing changes.
Sosa and Tooley’s thesis that «causation possesses an intrinsic nature, so that
causation must be one and the same relation in all possible worlds»118, can be
accepted, on condition of adding the qualification that causation has this essen-
tial nature in all possible worlds in which it exists. However, the sketched ver-
sion of transference theory no longer conceives this essence, as Sosa and Tooley
do following tradition, as a nominal essence that constitutes the a priori content
of the concept of causation. Rather it conceives it as a real essence to be dis-
covered empirically, a posteriori. This implies that causation does not exist nec-
essarily. There are possible worlds in which no relation belongs to the natural
kind causation, just as some possible worlds do not contain any object belong-
ing to the natural kind gold. However, this thesis does not force us to conclude,
as Sosa and Tooley claim it does, that

«one can appeal to the possibility of worlds that involve causation, but that do not con-
tain the physicalistic relations in question [such as the relation of transference of an
amount of a CQ; M.K.] – or, more radically, that contain no physicalistic states at all
– in order to draw the conclusion that causation cannot, even in this world, be identi-
cal with any physicalistic relation»119.

If causation is a natural kind, such worlds that would refute the conception we
have proposed, do not exist any more than there exist possible worlds contain-
ing gold that does not have the atomic number 79120. In such worlds, there might

117 For the idea of construing causation as a “natural kind” of relation, I am indebted to E.J. Lowe

(personal communication).
118 SOSA and TOOLEY, Introduction, 3.

119 SOSA and TOOLEY, Introduction, 3.

120 Kripke (S.A. KRIPKE, Naming and Necessity [1972], Harvard University Press, Cambridge, MA

1980) has shown that natural kinds, such as gold, possess a real essence, which is however discovered a
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 667

Causation in contemporary analytical philosophy 667

be a different source of correlations, but it would not be causation: it would not

share the real essence of the causal relation of the actual world.

6. Causal responsibility

We have seen that the nomological, counterfactual and probabilistic accounts

are not up to the task of analyzing the singular causal relation between events.
We have also found that they are instead adequate to the analysis of nomologi-
cal dependence, not causation. However, it is possible to acknowledge the use-
fulness of these approaches within the framework of a richer concept of causa-
tion, which takes nomological dependence into account and which deals with a
relation between facts rather than events. If a white billiard ball A moves a red
billiard ball B (by a frontal elastic shock, where B was at rest before the shock
and A and B have equal masses), we have a paradigmatic case of causation. How-
ever, we can characterize this relation more finely, because certain properties of
the cause are responsible for certain properties of the effect, whereas others are
not. The fact that A has momentum M at the moment of its impact on B is causal-
ly responsible for the fact that B has momentum M after the shock. However, the
fact that A is white before the shock is not causally responsible for the fact that
B is red after the shock. The dependence of the effect’s possession of certain
properties on the possession of certain properties by the cause is nomological
and gives rise, first to counterfactual dependence between facts about cause and
effect involving these properties, and second to a relation of probability raising.
If A had not had M before the shock, B would not have had M after the shock.
However, if A had not been white, B would nevertheless have been red. The
probability that B has M at t is greater if it lies on the trajectory of A approach-
ing with M than if B doesn’t lie on it. However, the probability that B is red af-
ter t is the same whether or not A is present.
In this way, the relation of causal responsibility121 allows us to integrate the
contribution of the nomological, counterfactual and probabilistic approaches to

posteriori. There is no possible world in which gold does not have its essential properties, such as its atom-
ic number, 79.
121 Similar relations have been studied by other authors (T. HORGAN, Mental Quausation, in J.E.

TOMBERLIN (ed. by), Philosophical Perspectives 3: Philosophy of Mind and Action Theory, Ridgeview, Atas-
cadero, CA 1989; A. MARRAS, Kim’s Principle of Explanatory Exclusion, «Australasian Journal of Philos-
ophy», 76 (1998), 439-451; B. MCLAUGHLIN, On Davidson's Response to the Charge of Epiphenomenalism,
in J. HEIL and A. MELE (eds.), Mental Causation, Clarendon Press, Oxford 1993, 27-40; J. KIM, The Non-
Reductivist’s Troubles with Mental Causation, in HEIL and MELE, Mental Causation, 189-210) under the
name ‘qua-causation’ or ‘quausation’. Still others consider the relation between facts to be the causal re-
lation itself. «Token causal claims are about the causal role of the actual exemplification of one type for
24_Kistler_2.QXD 2-09-2003 11:37 Pagina 668

668 Max Kistler

causation within the framework of the transference theory. The relation of causal
responsibility of the fact that the cause event (c) has property F for the fact that
the effect event (e) has property G can be written as a two place relation between
the facts Fc and Ge: CR (Fc, Ge).
To a first approximation, the relation of causal responsibility can be analyzed
as a complex relation grounded on the existence of a causal relation between
events c and e122. Building upon a nomological conception of the dependency be-
tween properties123, the relation of causal responsibility CR (Fc, Ge) of fact Fc
about the cause for fact Ge about the effect may be analyzed into a conjunction
C(c,e) ∧ Fc ∧ Ge ∧ L(F,G). C(c, e) means that the events c and e are linked as
cause and effect in virtue of transference; Fc and Ge mean that c and e possess
properties F and G, respectively; L(F, G) means that the exemplification of F by
e depends nomologically on the exemplification of F by c.

7. Conclusion

Contemporary analytic philosophy of science is in the midst of a vigorous debate

on the nature of causation. Each of the main proposals discussed in this chapter
faces important problems. Among the most actively pursued lines of research
aimed at overcoming these problems, are several attempts to combine the in-
sights of these different approaches. For lack of space, I have sketched only one
of them124 in the last section125.

the actual exemplification of another type. [...] 'y's being Y was because of (or token caused by) x's being
X'» (EELLS, Probabilistic Causality, 288; Eells’ italics).
122 I have defended this conception in KISTLER, Causalité…, chap. 5, and M. KISTLER, Causation as

transference and responsibility, in W. SPOHN, M. LEDWIG and M. ESFELD (eds.), Current Issues in Causation,
Mentis, Paderborn 2001, 115-133. In M. KISTLER, Causes as events and facts, «Dialectica», 53 (1999), 25-
46, I defend it against arguments by Mellor (MELLOR, The Facts of Causation) and others, to the effect that
only facts, but not events, can be causally related.
123 Alternatively, this dependency could be conceived according to the models of counterfactual de-

pendency or probability raising.

124 Other proposals include P. DOWE, A Counterfactual Theory of Prevention and “Causation” by Omis-

sion, «Australasian Journal of Philosophy», 79 (2001), 216-226, and J. SCHAFFER, Causes as probability
raisers of processes, «Journal of Philosophy», 98 (2001), 75-92.
125 I am indebted to Richard Carter for improving my English.