Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Inference to the Best Explanation: Is It Really Different from Mill's Methods?

Author(s): Steven Rappaport


Source: Philosophy of Science, Vol. 63, No. 1, (Mar., 1996), pp. 65-80
Published by: The University of Chicago Press on behalf of the Philosophy of Science
Association
Stable URL: http://www.jstor.org/stable/188226
Accessed: 30/07/2008 02:25

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=ucpress.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the
scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that
promotes the discovery and use of these resources. For more information about JSTOR, please contact support@jstor.org.

http://www.jstor.org
INFERENCE TO THE BEST EXPLANATION: IS IT REALLY
DIFFERENT FROM MILL'S METHODS?*

STEVEN RAPPAPORTtt
Department of Philosophy
De Anza College

Peter Lipton has attempted to flesh out a model of Inference to the Best Ex-
planation (IBE) by clarifying explanation in terms of a causal model. But Lipton's
account of explanation makes an adequate explanation depend on a principle
which is virtually identical to Mill's Method of Difference. This has the result of
collapsing IBE on Lipton's account of it into causal inference as conceived by the
Causal-Inference model of induction. According to this model, many of our in-
ductions are inferences from effects to their probable causes, and Mill's Methods
are canons to guide such inferences. Thus, Lipton's account of IBE fails to rep-
resent an advance over the already familiar Causal-Inference Model of induction.

In his attractive book, Inference to the Best Explanation (1993), Peter Lip-
ton offers a model of inference to the best explanation (IBE) as a descrip-
tion of many of the inductive inferences we make both in science and
everyday life. Lipton (1993, 22), initially tells us that "we infer what would,
if true, be the best explanation of our evidence." This is a fairly standard
initial description of IBE given by its proponents. For instance, according
to Thagard (1978, 77), ". . . inference to the best explanation consists in
accepting a hypothesis on the grounds that it provides a better explanation
of the evidence than is provided by alternative hypotheses." It seems true,
if rather uninformative, to say that many of our inductive inferences fit
this description. Lipton tries to reduce significantly the air of emptiness
attaching to the initial account of IBE largely by unpacking the notion of
explanation in terms of a causal model of explanation. But this maneuver
threatens to collapse Lipton's account of IBE into what may be called "the
Causal-Inference account of induction." This is characterized by Lipton
himself as follows:
A fourth account of induction, the last I will consider in this section,
focuses on causal inference. It is a striking fact about our inductive
practice, both lay and scientific, that so many of our inferences depend
on inferring from effects to their probable causes..... The best known
account of causal inference is John Stuart Mill's discussion of the
*Received January 1995; revised April 1995.
tI want to thank an anonymous referee for useful comments on this paper.
tSend requests for reprints to the author, Department of Philosophy, De Anza College,
Cupertino, CA 95014.

Philosophy of Science, 63 (March 1996) pp. 65-80. 0031-8248/96/6301-0005$2.00


Copyright 1996 by the Philosophy of Science Association. All rights reserved.

65
66 STEVEN RAPPAPORT

"methods of experimental inquiry." The critical methods are the


Method of Agreement and the Method of Difference. (1993, 20)

So the Causal-Inference model says that many of our inductive inferences


in science and everyday life are from effects to their likely causes. And
Mill's Methods, especially the Method of Agreement and the Method of
Difference, are canons to guide such inferences. It will be argued here that
Lipton's model of IBE is little different from the Causal-Inference view,
and indeed fails to represent any real advance over Mill's Methods. This
conclusion has some important implications for IBE which will be brought
out in the concluding remarks.
Lipton attempts to put some flesh on the bare bones of the model of
IBE by appealing to a pair of distinctions, to wit, the best actual expla-
nation versus the best potential explanation, and the likeliest versus the
loveliest explanation. As Lipton (1993, 59) notes, the model of IBE must
be seen as affirming that we infer the best potential, not the best actual,
explanation. For IBE is supposed to be a mode of inductive inference, and
even correct inductions sometimes lead to false conclusions. But if we
always infer the best actual explanation, the conclusions of our inferences
to the best explanation would always be true for a hypothesis to convey
an actual explanation of a phenomenon implies that the hypothesis is true.
To avoid misunderstanding, it is not that the conclusion of a correct or
warranted IBE is "Hypothesis H is the best potential explanation." As
Lipton (1993, 60) says, "According to Inference to the Best Explanation,
then, we do not infer the best actual explanation; rather we infer that the
best of the available potential explanations is an actual explanation." So,
the conclusion of a correct IBE is of the form "H is an actual explana-
tion.'" It is just that to be a correct or warranted inference, this conclusion
does not have to be true-again, correct inductions sometimes lead to false
conclusions. Lipton says relatively little about potential explanation when
initially introducing the notion.2 He does say this:
'A distinction is often made between inference and argument. Inference is a psychological
process which regularly involves reaching one belief on the basis of other beliefs. But an
argument is a group of statements one of which is the conclusion with the rest being the
premises. Lipton usually talks as if his model of IBE characterizes a class of inductive infer-
ences. Lipton wants his model of IBE to say something about the generation of hypotheses
and not just be an account of confirmation or inductive support. But he (1993, 62) also
regards his model as describing a class of strong or correct inductive arguments. Viewing
IBE as a form of argument, one could describe it as follows: The observed data are such and
such; Hypothesis H is the best explanation of these data; so, (probably) H is true. (This way
of representing Lipton's IBE as a form of argument is found in Achinstein 1992, 354.) Lipton
often seems to regard the conclusion of IBE as of the form "H is an actual explanation,"
rather than "H is true," though, of course the former entails the latter.
2Borrowing from Hempel, Lipton (1993, 60) does say that intuitively a potential expla-
nation is something that meets all the conditions of an actual explanation except perhaps
that of truth.
INFERENCE TO THE BEST EXPLANATION 67

When we decide which explanation to infer, we often start from a


group of plausible candidates, and then consider which of these is the
best, rather than selecting directly from the vast pool of possible ex-
planations. In later chapters, we will see further reasons for preferring
this restricted notion of potential explanation. (1993, 61)
So, a potential explanation of a phenomenon is a member of a pool of
plausible explanations of that phenomenon.
The distinction between loveliness and likeliness of explanations is im-
portant for Lipton. About this distinction he says the following:
It is important to distinguish two senses in which something may be
the best of competing potential explanations. We may characterize it
as the explanation that is most warranted: the 'likeliest' explanation.
On the other hand we may characterize the best explanation as the
one which would, if correct, be the most explanatory or provide the
most understanding: the 'loveliest' explanation. The criteria of likeli-
ness and loveliness may well pick out the same explanation, but they
are clearly different sorts of standards. Likeliness speaks of truth;
loveliness of potential understanding. (1993, 61)
Lipton thinks that an interesting version of IBE will not represent it as of
the following form:
The observed data are such and such.
Hypothesis H is the likeliest of the competing potential explanations
of these data.
So, (probably) H is true.
For, given the reliance on the notion of likeliness in the second premise,
an air of triviality hangs about this form of inference. Instead, Lipton
(1993, 63) says that IBE is inference to the loveliest of the competing po-
tential explanations. In other words, an interesting version of IBE is of
this form:
The observed data are such and such.
Hypothesis H is the loveliest of the competing potential explanations
of these data.
So, (probably) H is true.
The problem confronting this version of IBE is saying something useful
about what makes a (potential) explanation the loveliest among the con-
testants.
Lipton seems to adopt two rather disparate answers to this problem.
But one of them is only marginally connected with his version of IBE,
while the other is central. At times Lipton (1993, 71) says or implies that
the loveliness of an explanation is determined by the familiar, if hard to
68 STEVEN RAPPAPORT

define, virtues of simplicity, unification, and theoretical elegance. But these


three virtues of hypotheses do not really get developed and integrated into
Lipton's account of IBE. Instead, something very much like Mill's Method
of Difference is seen by Lipton as a key determinant of loveliness, and
worked into his model of IBE. He says:

According to Mill, we find the cause of a fact in some prior difference


between a case where the fact occurs and an otherwise similar case
where it does not. Mill's central mechanism for inferring the likeliest
cause is almost the same as the mechanism of causal triangulation
which helps determine the loveliest explanation. (1993, 77)
We need to connect Lipton's mechanism of causal triangulation, or as he
also call it, the "Difference Condition," to IBE. This will help lay the
groundwork for showing that Lipton's model of IBE collapses into the
Causal-Inference model.
Lipton adopts a causal model of explanation according to which to
explain why a particular event or phenomenon occurs is to cite a cause of
it, and to explain a causal regularity (e.g., decapitation causes death in
humans) is to specify the mechanism linking cause and effect.3 This is an
attractive approach to explanation which is largely correct. In addition,
Lipton adopts the position, familiar from the work of van Fraassen and
Garfinkel, that frequently the explanandum is a contrastive phenomenon
consisting of a fact and a foil. We explain why Jones rather than Smith
got paresis, with Jones getting paresis being the fact and Smith not getting
paresis being the foil. But what makes for an adequate explanation of a
contrastive phenomenon? The following passage contains the crux of Lip-
ton's answer:

Accordingly, I propose that, for the causal explanation of events, ex-


planatory contrasts select causes by means of what I will call the 'Dif-
ference Condition'. To explain why P rather than Q, we must cite a
causal difference between P and not-Q, consisting of a cause of P and
the absence of a correspondingevent in the case of not-Q. (1993, 43)

So, from among the antecedent conditions or causal histories of both P


and not-Q-note that we are trying to explain why P rather than Q4-we
must find an event or condition that is in the causal history of P but not
that of not-Q. This is Lipton's Difference Condition. For instance, we
might explain why Jones rather than Smith got paresis by citing the fact
that Jones has syphilis while Smith does not. Jones having syphilis is
3Lipton (1993, 33-34) says that the causal model is not complete, as there are non-causal
explanations.
4Lipton (1993, 39-40) argues that this is not tantamount to trying to explain why P and
separately trying to explain not-Q.
INFERENCE TO THE BEST EXPLANATION 69

among the antecedent conditions of Jones having paresis but Smith having
syphilis is not among the antecedent conditions of Smith not having pa-
resis. Lipton (1993, 47) acknowledges that while the Difference Condition
specifies a necessary condition for an adequate explanation of why P rather
than Q, it does not supply a generally sufficient condition. A major reason
for this is that not all differences in the causal histories of the fact and foil
are explanatory in a given context. To use one of Lipton's own examples,
suppose we ask why Able rather than Baker got the teaching position in
the philosophy department. Able had a strong letter from Quine while
Baker did not, yet this difference does not supply the explanation. For
Baker would not have gotten the job even with a strong letter from Quine,
as Baker's area of specialization did not fit the department's needs while
Able's did.5
Lipton's account of explanation, and especially the Difference Condi-
tion, instills some genuine content into the idea of an explanation being
the loveliest among the competing potential explanations of the data. Sup-
pose we have three hypotheses H1, H2, and H3 that are competing po-
tential explanations of data D. Imagine that only H2 turns out to meet
the Difference Condition, i.e., only H2 specifies a circumstance in the
causal history of the fact but not that of the foil. On Lipton's account,
this makes H2 the loveliest of the three contestants. And Lipton's model
of IBE says that we would infer that (probably) H2 is true or an actual
explanation.
However, in light of Lipton's account of loveliness, it is hard to see how
his model of IBE differs from the Causal-Inference view of induction.
Recall that on Lipton's model of IBE, the IBE argument schema is this:
The observed data are such and such.
Hypothesis H is the loveliest of the competing potential explanations
of these data.
So, (probably) H is true.
An advocate of the Causal-Inferenceview could easily say that this schema
does indeed characterize a mode of correct inductive argument, and one
which is widely used in science and everyday life. Moreover, the Causal-
Inference theorist adds, Mill's Methods, including the Method of Differ-
ence, are means for picking out the loveliest among competing potential
explanations of the data.6 Now Lipton takes his Difference Condition to
5Lipton does not supply any general principle with which to supplement the Difference
Condition. He (1993, 47) simply says, "The considerations that govern selection from among
the causally relevant differences are numerous and diverse." And he goes on to cite a few
such considerations.
6But it might be said that: (A) Mill himself cannot then be a advocate of the Causal-
Inference view of induction as described here; for (B) the IBE argument schema appears to
characterize what Mill (1970, 322-323) calls the "hypothetical method;" and (C) Mill rejects
70 STEVEN RAPPAPORT

pick out the loveliest explanation. But given its very close resemblance to
Mill's Method of Difference, Lipton's model of IBE does not seem to be
an account of induction separate from the Causal-Inference account. Lip-
ton is sensitive to this issue. He says:

What made this articulation possible is the close analogy between


contrastive explanation and the Method of Difference. The analogy
is so close that some readers, convinced that Inference to the Best
Explanation really is different from hypothetico-deductivism, may
now wonder whether it is not just a catchy name for Mill's Method
itself, rather than a distinctive account. This raises a third challenge
to Inference to the Best Explanation. Even if the account is more than
Inference to the Likeliest Cause, and improves on the hypothetico-
deductive model, is it really any better than Mill's Methods? (1993,
112-113)

Lipton tries to meet the challenge he describes here; but, it will be argued,
he is not successful.
Lipton's effort to distance his model of IBE from Mill's Methods and
the Causal-Inference model of induction consists in specifying two alleged
defects of the Method of Difference that do not attach to IBE on Lipton's
account of it. Here is Lipton's description of one of the putative defects:
The Method of Difference has two serious limitations not shared by
Inference to the Best Explanation. The first is that it does not account
for inferred differences. The Method of Difference sanctions the in-
ference that the only difference between the antecedents of a case
where the effect occurs and one where it does not marks a cause of
the effect. Here the constrastive evidence is not evidence for the ex-
istence of the prior difference, but only for its causal role. The method
says nothing about the discovery of differences, only about the infer-
ence from sole difference to cause. So it does not describe the workings
of the many contrastive inferences where the existence of the differ-
ence must be inferred, either because it is unobservable or because it
is observable but not observed. Many of these cases are naturally

the hypothetical method as a distinct mode of induction capable of establishing the (prob-
able) truth of conclusions. (See Skorupski 1991, 197-202 for a discussion of Mill on the
hypothetical method.) However, (B) and (C) do not justify (A). The IBE argument schema,
supplementedwith Mill's Methods to identify lovely explanations, is not a mode of induction
that is distinct from Mill's Methods. And so the IBE argument schema plus Mill's Methods
is not the hypothetical method to which Mill objects. For that hypothetical method is infer-
ring the likely truth of a hypothesis merely on the basis of the hypothesis accommodating
observed data. The IBE argument schema plus Mill's Methods sanctions the inference to the
likely truth of a hypothesis which specifies a cause as identified by Mill's Methods. And Mill
would hardly reject the legitimacy of such an inference.
INFERENCE TO THE BEST EXPLANATION 71

described by Inference to the Best Explanation, since the difference is


inferred precisely because it would explain the contrast. (1993, 114)

As Mill's Method of Difference is usually presented, it does sanction in-


ference to a cause when the antecedent circumstances from which the cause
is selected are already known to exist. One examines a positive instance (a
case in which the effect whose cause is sought is present) and a negative
instance (a case in which the effect is absent). Assuming one finds just one
(relevant) antecedent circumstance which is present in the positive instance
but absent in the negative instance, that circumstance is inferred as the
cause. As Lipton says, it is this circumstance'scausal role, not its existence,
that is inferred. The existence of the antecedent circumstance has already
been established prior to the use of the Method of Difference.
But Lipton has not established any distance here between his model of
IBE and the Causal-Inference account of induction. The Causal-Inference
account does not say that Mill's Method of Difference is the only means
for picking out the loveliest explanation. The account advocates the use
of all of Mill's Methods. And one of the methods, to wit, the Method of
Residues, does allow inference to the existence of a cause, and not just to
the causal role of a condition already known to exist. The Method of
Residues can be characterized as follows:

If (a) antecedent conditions C , C2, ... Cn and Cn + 1 are the likeliest


candidates for the complex cause of a total phenomenon P, and (b) it
is known that part of P is the effect of C1, C2, . . . Cn, then infer that
the residue of P (the other part of P) is the effect of Cn + 1.

It is important to note that the antecedent condition Cn + 1 may not be


known to exist prior to the application of the Method of Residues. Clause
(a) just represents C1, C2, . . . Cn and Cn + 1 as the likeliest candidates
for the cause of P, and this does not mean that they are the cause, or that
they even exist. The likeliest candidate for the cause of a person's death
may be a heart attack-the person had a previous heart attack, his father
died of a heart attack, etc. Yet it could turn out that the person died from
being poisoned, and no heart attack occurred. However, clause (b) entails
that C1, C2, . . . Cn are known to exist already. Should clauses (a) and
(b) both hold, then the Method of Residues sanctions the inference to the
conclusion that Cn + 1 is the cause of the residue of P. Of course, this
conclusion entails that Cn + 1 exists. And so if Cn + 1's existence was not
already known, then the application of the Method of Residues establishes
its (probable) existence. Treatments of Mill's Methods in elementary logic
textbooks sometimes contain a famous example of the use of the Method
of Residues to establish the existence as well as the causal role of an an-
tecedent circumstance: Leverrier's discovery of the planet Neptune in
72 STEVEN RAPPAPORT

1845.7 Prior to 1845, it was known that the orbit of Uranus calculated
from the gravitational effect of the Sun and the interior planets disagreed
somewhat with the observed orbit of Uranus. Knowing that just part of
the observed orbit of Uranus is due to the gravitational effect of the Sun
and the interior planets, Leverrierinferred that the residue of the phenom-
enon-the other part of Uranus' orbit that is not the result of the Sun and
the interior planets-is due to the presence of a planet beyond Uranus that
was causing the perturbation in its orbit. Of course, in late 1846 Neptune,
the planet whose existence and causal role Leverrier had inferred, was
observed very close to the place that Leverrier had predicted.
So Lipton may be correct in thinking that Mill's Method of Difference
only allows the inference to the causal role of an antecedent condition
already known to exist, whereas IBE on Lipton's account of it sanctions
inference to the very existence of a cause. But it does not follow that the
Causal-Inference model represents our inductions as limited to establish-
ing the causal role of antecedent circumstances whose existence has to be
independently determined. The Causal-Inference model advocates using
all of Mill's Methods, and not just the Method of Difference. Or in other
words, all of Mill's Methods, including the Method of Residues, are means
to identifying the loveliest explanation. And the Method of Residues does
sanction the inference to a cause whose existence has not already been
established.8
Lipton thinks that Mill's Method of Difference has a second defect
7For instance, see Copi and Cohen 1994, 500-501. Copi and Cohen's general, schematic
description of the Method of Residues appears to assume that the existence of the circum-
stance which is inferred as the cause of the residue of the phenomenon is known prior to the
application of the Method of Residues. But this assumption does not hold for their own
example of Leverrier'sdiscovery of Neptune to illustrate the use of the Method of Residues.
It is worth noting that Lipton (1993, 96) acknowledges that Leverrier's discovery of the
existence of Neptune is an application of the Method of Residues.
8Mill (1970, 260) says that the Method of Residues is a modification of the Method of
Difference. Mill's reason for taking the Method of Residues as an elaboration of the Method
of Difference seems to be this: the Method of Residues, like the Method of Difference,
involves both positive and negative instances. The positive instance is the total phenomenon
P together with the total cause Cl, C2, . . . Cn and Cn + 1. The negative instance is a case
in which the residue of P is absent as is Cn + 1. (Mill says that the negative instance is
inferred or deduced, as we do not observe a case of the residue of P and Cn + 1 both being
absent.) But Mill's position here does not seem entirely correct. We can think of P together
with Cl, C2, . . . Cn and Cn + 1 as including the residue of P as well as Cn + 1. So, we
have a positive instance as the Method of Difference requires. And we can regard a case in
which the residue of P as well as Cn + 1 are absent as the negative instance. The Method
of Difference would then sanction inferring Cn + 1 as the cause of the residue of P. But the
problem is that to apply the Method of Difference here we must already know that, were
the residue of P absent, Cn + 1 would also be absent. And this is what is implied by the
conclusion that Cn + 1 is the cause of the residue of P, a conclusion inferred by the Method
of Residues at least sometimes before we have the negative instance required to apply the
Method of Difference. So, the Method of Residues does not seem to be a modification of
the Method of Difference, if that means that applying the former is always reducible to an
application of the latter.
INFERENCE TO THE BEST EXPLANATION 73

which makes his model of IBE superior to, and therefore different from,
Mill's Methods and the Causal-Inference view of induction. The alleged
defect is described by Lipton as follows:
The second weakness of the Method of Difference is that it does not
account for the way we select from among the multiple differences.
Although Mill's strict statement of the Method of Difference sanc-
tions an inference only when we know that there is a sole difference
in the histories of fact and foil, Mill recognizes that this is an ideali-
zation. However similar the fact and foil, there will always be more
than one difference between their antecedents. Some of these will be
causally relevant, but others not. The problem of multiple differences
is the problem of making this discrimination. (1993, 115)

Lipton goes on to assert that Mill himself proposes a pair of inadequate


solutions to the problem of multiple differences.
Imagine that I see some leaves on the ground and they are not moving.
A wind comes up and the leaves move. I infer that the wind caused the
leaves to move. This is an application of the Method of Difference, with
the leaves moving being the positive instance and the leaves being still the
negative instance. As Lipton notes, Mill (1970, 256) sometimes states the
Method of Difference so that it would seem to sanction the inference to
the wind being the cause only if

(1) the sole difference between the antecedents of the negative and
positive instances is the wind blowing.
But of course, (1) would no doubt be false. For many events would prob-
ably occur as the wind came up-a person sneezes on a certain street
corner in Singapore, a certain canary starts singing, etc.-which were not
among the antecedents of the negative instance. But Mill does not intend
that the Method of Difference applies to our example of the leaves only
if (1) were true. He says:
The two instances which are to be compared [in using the Method of
Difference] must be exactly similar in all circumstances except the one
which we are attempting to investigate: they must be in the relation
of A B C and B C, or of a b c and b c. It is true that this similarity
of circumstances needs not extend to such as are already known to be
immaterial to the result. And in the case of most phenomena we learn
at once, from the commonest experience, that most of the co-existent
phenomena of the universe may be either present or absent without
affecting the given phenomenon; or, if present, are present indiffer-
ently when the phenomenon does not happen and when it does. (1970,
256)
74 STEVEN RAPPAPORT

Thus Mill intends that the application of the Method of Difference to our
leaves example requires not (1) above, but instead

(2) the sole difference between the relevant antecedent circumstances


of the negative and positive instances is the wind blowing.
And by "relevant antecedent circumstances" is meant antecedent circum-
stances which are epistemicallypossible causes of the phenomenon whose
cause we seek. (To say that it is epistemically possible that C is a cause of
E, is to say that we do not know, or justifiably believe, that C is not a
cause of E.)
So, part of Mill's own approach to the problem of multiple differences
is specifying that the Method of Difference sanctions the inference to the
causal role of the only difference between relevant antecedent conditions
of the negative and positive instances of the phenomenon whose cause is
sought. The Method of Difference allows us to ignore antecedent circum-
stances that are not relevant in the sense described above. But Lipton
thinks that this is inadequate. The reason he (1993, 195-196) gives for this
assessment is that Mill's approach fails to tell us how to determine whether
or not an antecedent condition is relevant. But this claim of Lipton's is
not really true. In the passage from A System of Logic quoted above, Mill
notes that the "commonest experience" enables us to reject as epistemi-
cally possible causes of a phenomenon most of the antecedent circum-
stances. Strictly, it is not experience or observation alone, but experience
plus principles like
(3) if an antecedent condition C is present in the absence of a phe-
nomenon P, then C is not a cause of P
which allows us to judge as irrelevant legions of antecedent circumstances
of a phenomenon whose cause we seek.9 In our leaves being blown by the
wind example, a person may start singing as the wind comes up and the
leaves start to move. But we have observed situations in which someone
started singing and nearby leaves failed to move. This sort of observation,
coupled with principle (3), warrants ruling out the singing as a cause of
9It seems clear that we do often use (3) to rule out antecedent conditions as epistemically
possible causes in particular contexts in which we seek a cause. Often philosophers think of
causes as sufficient conditions, necessary conditions, or INUS conditions. (3) is true if a
cause is thought of as a sufficient condition. And (3) appears to be true if a cause is regarded
as an INUS condition. Condition C is an INUS condition provided that C is an indispensable
part of a sufficient condition that is not also a necessary condition. If an antecedent condition
C1 is present in the absence of a phenomenon P, then C1 is not an INUS condition. For if
it were, then another condition C2 of which C1 is an indispensable part would have to be
present, and C2 would be a sufficient condition for P. But C2 is not sufficient for P, as C2
is present while P is absent. Finally, (3) is not correct if a cause is thought of as a necessary
condition. A necessary condition for a phenomenon P can of course be present in the absence
of P.
INFERENCE TO THE BEST EXPLANATION 75

the leaves moving. So, an advocate of the Causal-Inference model of in-


duction can partly handle the problem of multiple differences by stating
the Method of Difference in terms of relevant antecedent circumstances.
And, contrary to Lipton, she can say, as does Mill himself, something
useful about how to determine whether or not an antecedent circumstance
is relevant.
Appealing to the idea of relevant antecedent circumstances is only part
of Mill's approach to the problem of multiple differences. As Lipton him-
self says:

Second, while passive observations will seldom satisfy the requirement


of the method [i.e. the Method of Difference], Mill claims that care-
fully controlled experiments where a precise change is introduced into
a system will often leave that change as the only possible material
difference between the situation before the change and the situation
afterwards. (1993, 115)
Mill does suggest that the problem of multiple differences can also be dealt
with by using artificial or controlled experiments. Suppose that we delib-
erately produce in a situation a single circumstance C, and then observe
an occurrence of a phenomenon P whose cause we are seeking. The neg-
ative instance is the situation prior to C being produced, P being absent
in this situation; while the positive instance is the situation after C is pro-
duced and P occurs. The controlled experiment has provided us with a
single difference between the relevant antecedent circumstances of the pos-
itive and negative instances of P. The Method of Difference would allow
us to infer C as the cause of P.
Lipton regards as inadequate Mill's attempt to deal with the problem
of multiple differences by appeal to controlled experiments. To back up
this evaluation Lipton (1993, 116) says that ". . . Mill is too sanguine
about the powers of experimental technique to eliminate all but one pos-
sible relevant difference." But this is very weak. Why is Mill too sanguine
about the use of controlled experiments? Controlled experiments have in
fact often been successfully used to find causes, which suggests that Mill
is not unreasonably sanguine about the ability of controlled experiments
to establish as a cause a single antecedent circumstance of the phenomenon
of interest. And Mill does not have to think that controlled experiments
always succeed in order to see such experiments as a means of dealing
with the problem of multiple differences.
Recall that Lipton thinks that he can distance his model of IBE from
Mill's Methods and the Causal-Inference view of induction by showing
that an advocate of the Causal-Inference view cannot handle in an ade-
quate way the problem of multiple differences. But Lipton fails to persua-
sively support his negative assessment of what Mill says in answer to this
76 STEVEN RAPPAPORT

problem. Indeed, what Mill says in answer to the problem of multiple


differences may not be the last word on the subject, but Mill's remarks
about relevance and controlled experiments convey something true and
useful about how to remove multiple differences so that the Method of
Difference may be used to infer a cause. Lipton has failed to distance his
model of IBE from the Causal-Inference view of induction on account of
the latter's handling of the problem of multiple differences, since Lipton
has not shown the inadequacy of what Mill or an advocate of the Causal-
Inference view has to say about multiple differences. But it will be useful
to buttress the discussion with a consideration of how Lipton's own model
of IBE approaches the problem of multiple differences.
Lipton thinks that his model of IBE does a better job of dealing with
multiple differences than does Mill. Lipton (1993, 116) says that his model
of IBE resolves the problem of multiple differences by sanctioning the
inference to a difference between the causal histories of fact and foil that
provides the loveliest explanation of the evidence. But, as Lipton is aware,
this is hardly useful without an account of what makes an explanation the
loveliest. Lipton takes his account of contrastive explanation-which was
summarized above-to partly fill the need for an answer to "What makes
for a good or lovely explanation?" But recall that Lipton's Difference
Condition plays a key role in his account of the adequacy of an expla-
nation. And by Lipton's own admission, his Difference Condition is vir-
tually the same as Mill's Method of Difference. Thus it is hard to see how
Lipton's model of IBE is an advance over Mill's Methods in handling the
problem of multiple differences. Indeed, from what has been said so far,
any difficulty Mill's Method of Difference has in handling the problem in
question will simply be inherited by Lipton's model of IBE.
Lipton (1993, pp. 117-122) does make some additional, if rather
sketchy, remarks about the multiple differences problem. He emphasizes
the role of backgroundbeliefs of ours both in coming up with a short list
of circumstances which might be a cause of the phenomenon of interest,
as well as in selecting a cause of the phenomenon from the short list. The
flavor of Lipton's remarks is well-conveyed by the following example he
gives:
When my computer did not come on in the morning, I at first inferred
that there was a blown fuse, since this would explain the contrast
between its behavior on other days and its lack of behavior today.
Then I notice another difference: my computer was unplugged. Given
my background knowledge about electricity, I had little difficulty se-
lecting which difference was the more likely cause. Even if the fuse
had blown, this would not explain why my computer was not running.
(1993, 118)
INFERENCE TO THE BEST EXPLANATION 77

But recall that Mill himself notes that "the commonest experience" aids
us in coming up with a short-list of antecedent circumstances-a list of
relevant antecedent circumstances-which are possible causes of the phe-
nomenon we seek to explain. It is not clear to what extent Mill's "com-
monest experience" and Lipton's background beliefs differ. The common-
est experience furnishes us with all sorts of background beliefs about what
does not cause leaves to move, computers not to work, etc. In his remarks
about background beliefs, Lipton does not appear to be making any sig-
nificant advance over what Mill or an advocate of the Causal-Inference
view of induction does or could say about the multiple differences prob-
lem.
The Causal-Inference model of induction says that many of our induc-
tive inferences are from effects to their likely causes. And Mill's Methods
are canons to guide such inferences. Or alternatively, the Causal-Inference
model affirms that many of our inductive arguments are of the following
form:
The observed data are such and such.
Hypothesis H is the loveliest of the competing potential explanations
of these data.
So, (probably) H is true.
If true, the premises of arguments fitting this schema warrant acceptance
of their conclusions, i.e., arguments fitting the schema are correct or war-
ranted inferences. And Mill's Methods are the means for identifying the
hypothesis which is the loveliest explanation of the data. That is, at least
part of the answer to the question "What makes for a lovely explanation?"
is that it is a causal explanation which has been reached by applying Mill's
Methods. Lipton's model of IBE also says that many of our inductions fit
the above argument schema, and the model of IBE regards such inferences
as warranted. And Lipton's causal account of explanation, which relies
on the Difference Condition, is supposed to at least partly answer the
question "What makes for a lovely explanation?" But since Lipton's Dif-
ference Condition, so central to his contrastive account of explanation, is
virtually indistinguishable from Mill's Method of Difference, there is little
distinction between Lipton's model of IBE and the Causal-Inference view
of induction. Or so it has been argued here.
Moreover, Lipton makes an effort to separate his model of IBE from
the Causal-Inference view and Mill's Methods by arguing both that (a) his
model of IBE, unlike the Causal-Inference view, allows for the inference
to the existence of a circumstance that is a cause of the phenomenon under
investigation, and (b) the Causal-Inference view's reliance on Mill's Meth-
ods makes it incapable of handling the problem of multiple differences.
But it has been suggested here that Lipton's claim (a) is false, as it over-
78 STEVEN RAPPAPORT

looks the role the Method of Residues can play in sanctioning the inference
to a cause whose existence has not already been established. And it has
also been argued that Lipton fails to establish claim (b); and moreover,
his own efforts to deal with the problem of multiple differences add little
to what an advocate of the Causal-Inference model does or could say
about the problem. In sum, Lipton's model of IBE does collapse into the
Causal-Inference view.
Proponents of IBE have made various claims on its behalf. Two of the
more exciting ones are (C1) induction is IBE, and (C2) the reasoning by
which scientific theories are justified is in all cases IBE. But the collapse
of Lipton's model of IBE into the Causal-Inference view of induction, has
negative implications for (C1) and (C2). In particular, neither claim is
defensible when IBE is understood in the manner characterized by
Lipton's model. So if Lipton's model of IBE is adopted, then neither (C1)
nor (C2) can be accepted.
(Cl) is perhaps particularly associated with Gilbert Harman.10Behind
(Cl) is a distinction between a kind of induction being basic and a form
of induction being non-basic or derivative. A non-basic form of induction
is one that is a special case of another sort of induction, while a basic
variety of induction is one that cannot be seen as a special case of another
sort of induction. For instance, in a well-known article, Harman (1989)
argues that enumerative induction is a special case of IBE, which he re-
gards as the basic form of induction. Of course, this makes enumerative
induction non-basic or derivative.1'(Cl) is a compressed way of expressing
the view that IBE is the sole basic kind of induction, with any other variety
of induction being a special case of IBE. But (Cl) would not be true were
IBE understood in accordance with Lipton's model. Consider the form of
induction-call it "statistical generalization"-in which a generalization
about a population is inferred from information about a sample drawn
from that population. Under suitable conditions, statistical generaliza-
tions are correct or warranted inductive inferences. But they are not a
special case of IBE as characterized by Lipton's model. For IBE as Lipton
views it has as a conclusion a hypothesis which furnishes a causal expla-
nation of the data, whereas at least often the conclusions of statistical
generalizations do not even purport to afford causal explanations of their
'OHarman(1973, 140) says, "We are led to construe induction as inference to the best
explanation, or more precisely as inference to the best of competing explanatory statements."
It is important to note that Harman (1973, 130) says that IBE is not always inference to a
conclusion causally explaining the data. So Harman would presumably not view Lipton's
account of IBE as adequately describing all cases of this form of induction.
'Richard Fumerton (1980) argues that at least on one account of IBE, it is just a special
case of enumerative induction. Thus for Fumerton, IBE is a derivative mode of induction.
Fumerton does not consider the possibility that IBE collapses into causal inference as de-
scribed by the Causal-Inference Model. But then Lipton's version of IBE was not available
as the time Fumerton's article was written.
INFERENCE TO THE BEST EXPLANATION 79

supporting data. Thus, one implication of the conclusion that Lipton's


model of IBE collapses into the Causal-Inferenceview of induction, is that
(Cl) would have to be abandoned if Lipton's model is adopted as an
account of IBE. This would detract considerably from the appeal of Lip-
ton's model for those advocates of IBE who see it as the basic mode of
induction.
Philosophers of science have long sought a comprehensive account of
the confirmation of scientific theories. Hypothetico-deductivism and Baye-
sian confirmation theory are two prominent efforts to supply such an
account. But some philosophers, dissatisfied with hypothetico-deductiv-
ism as well as Bayesian confirmation theory, have seen IBE as affording
a comprehensive account of theory confirmation in science. For example,
Thagard says:
However, I shall show that actual cases of scientific reasoning exhibit
a set of criteria for evaluating explanatory theories. Besides filling in
a crucial gap in Harman's account of inference to the best explanation,
the criteria furnish a comprehensive account of the justification of
scientific theories. I shall argue that this account has many advantages
over the hypothetico-deductive model of theory confirmation. (1978,
76)
This passage is an expression of (C2) above. But it is difficult to see how
(C2) could be maintained were Lipton's model of IBE accepted. Again,
as argued here, Lipton's model of IBE is little different than the Causal-
Inference view of induction. Thus, if Lipton's account of IBE is adopted,
then (C2) would have to be seen as tantamount to the assertion that the
Causal-Inference view-which relies on Mill's Methods to identify
causes-is a comprehensive account of theory confirmation in science. But
this assertion is not true. Confirmation of scientific theories does not al-
ways proceed by way of induction as described by the Causal-Inference
model. In justifying a theory, scientists simply do not always appeal to a
premise affirming that the theory is the best causal explanation of the data
as determined by Mill's Methods. Since the Causal-Inference view is not
adequate as a comprehensive account of confirmation, an implication of
the conclusion argued for in this article is that acceptance of Lipton's
model of IBE would require relinquishing the claim that IBE is the form
of reasoning by which scientific theories are confirmed.

REFERENCES
Achinstein, P. (1992), "Inference to the Best Explanation: Or, Who Won the Mill-Whevell
Debate?", Studies in the History and Philosophy of Science 23:349-364.
Copi, I. and Cohen, C. (1994), Introductionto Logic, Ninth Edition. New York: Macmillan.
Frumerton, R. (1980), "Induction and Reasoning to the Best Explanation", Philosophy of
Science 47.'589-600.
80 STEVEN RAPPAPORT

Harman, G. (1973), Thought.Princeton: Princeton University Press.


. (1989), "The Inference to the Best Explanation", reprinted in B. Brody and R.
Grandy (eds.), Readings in the Philosophy of Science. Englewood Cliffs, NJ: Prentice-
Hall, pp. 323-328.
Lipton, P. (1993), Inference to the Best Explanation. London: Routledge.
Mill, J. S. (1970), A System of Logic. London: Longman.
Skorupski, J. (1991), John Stuart Mill. London: Routledge.
Thagard, P. (1978), "The Best Explanation: Criteria for Theory Choice", The Journal of
Philosophy 75:76-92.

You might also like