David Lewis on indicative and counterfactual
conditionals
Robert J. Fogelin
Citing Ernest Adams as its source, in Counterfactuals David Lewis presents
the following argument concerning the relationship between indicative and
counterfactual conditionals:
The first conditional below is probably true, but the second may very
well be false. ...
[1] If Oswald did not kill Kennedy, then someone else did.
[2] If Oswald had not killed Kennedy, then someone else would
have.
Therefore there really are two different sorts of conditional; not a
single conditional that can appear as indicative or as counterfactual
depending on the speaker s opinion about the truth of the antecedent.
(p. 3)1
It may seem hard to imagine an argument more decisive than this, but, in
fact, without considerable further argument not provided by Lewis, it does
not establish what it is intended to establish.
The argument is directed against the view, briefly alluded to, that indic-
ative and counterfactual conditionals do not differ in their structure as
conditionals, but differ only in giving different indications concerning the
truth value of the antecedent, i.e., indicative conditionals express no
commitment concerning the truth value of the antecedent whereas coun-
terfactuals express a commitment to the falsehood of the antecedent.2
Lewis holds that this view cannot be correct because it is possible to find
pairs of conditionals which are otherwise the same except that one is indic-
ative, the other counterfactual, where the indicative conditional is
probably true, but the counterfactual conditional may well be false. If that
is right  and Lewis s specimen pair of conditionals is intended to show that
it is right  then, as Lewis concludes, there must be more to the difference
1
Though Lewis is right in attributing this style of counterexample to Adams, this
particular example does not appear in the source Lewis cites. See: Ernest Adams,
 Subjunctive and indicative conditionals , Foundations of Language 6 (1970): 89 94,
and David K. Lewis,Counterfactuals. Cambridge: Harvard University Press, 1973.
2
In  Inferential Constructions (American Philosophical Quarterly 4, 1966: 15 27), I
defended a view more or less along the lines of the one that Lewis is attacking. I think
that there may be difficulties with the view expressed in that paper, but not, as I shall
try to show, the difficulty that Lewis raises against positions of this kind.
Analysis 58.4, October 1998, pp. 286 89. © Robert J. Fogelin
indicative and counterfactual conditionals 287
between indicative and counterfactual conditionals than an indication of
truth value commitments concerning the antecedent.
Clearly, Lewis s argument rests on the assumption that the sole salient
difference between [1] and [2] relevant to the differing probability assess-
ments is mood shift. Lewis does not defend this pivotal assumption,
presumably because it strikes him as obviously true. After all, except for
mood shift, the words stay the same. Against this, I suggest that a reason-
able case  indeed, a stronger case  can be made for an alternative account
of these disparities in probabilistic assessment: they depend on changing
contextual settings in which we imagine these conditionals being
employed. That is, we are inclined to give [1] and [2] contrasting probabi-
listic assessments because we are inclined to imagine them being used in
contrasting contexts. This is shown by the fact that the contrast in proba-
bilistic assignments disappears when we hold contexts constant.
We can begin by looking at the specimen counterfactual [2].
[2] If Oswald had not killed Kennedy, then someone else would
have.
We can imagine this sentence being uttered in a context where, knowing
that Oswald had shot Kennedy, people are reflecting on the possibility that
others were involved in a conspiracy to kill Kennedy. In such a setting, we
can imagine someone asserting [2] because he believes, and would be
understood to believe:
[2c] If Oswald had not killed Kennedy, then a fellow conspirator
would have stepped in and killed him.
I have labeled this conditional [2c] because it makes an explicit reference
to a conspirator stepping in  something that [2] itself does not do. This
person asserts [2] because he accepts [2c].
There are various ways of explaining the relationship between [2] and
[2c]. The simplest view is that, in the context we are imagining, [2] is just
short for (elliptical for) [2c]. On this approach, asserting [2] just is assert-
ing [2c], though [2] drops out things understood in the conversational
setting. Other accounts are possible as well  one employing presupposi-
tions, for example. But, for our present purposes, there is no need to
explain in detail how [2c] is related to [2]. It is sufficient to note that in the
context we are imagining, in the absence of other relevant considerations,
the probability assigned to [2] is no different than the probability assigned
to [2c]. Those who accept a conspiracy theory might assign [2c] (hence [2])
a reasonably high probability, whereas those like Lewis, deferring to the
findings of the Warren Commission, might say that [2c] (hence [2]) may
very well be false.
The indicative conditional [1] would most naturally be employed in a
288 robert j. fogelin
setting quite different from the one just considered, namely, one where the
identity of Kennedy s assassin is under consideration. In this setting, it is
taken for granted that someone killed Kennedy, and the question is who.
In this setting, someone might assert
[1] If Oswald did not kill Kennedy, then someone else did.
because he assigned a high probability, which he quite reasonably should,
to
[1i] If Oswald wasn t the one who killed Kennedy, then someone else
was.
At least this much should be clear: If Lewis s assignment of contrasting
probabilities to [1] and [2] depends on evaluating them in different
contexts  [2] in a conspiratorial context, [1] in an identificatory context 
then his argument plainly fails. We are thus left with two accounts of the
contrasting probability assignments Lewis s argument rests upon without
being given a reason for choosing between them. In fact, however, a strong
argument can be made against his mood-shift analysis by noting that the
disparities in probability-assignments his argument depends upon do not
occur when the contextual setting is not allowed to vary as it does in the
case given above.
We can first consider what happens when both [1] and [2] are employed
in a context where the identity of Kennedy s assassin is at issue. In that
case, the probabilities for [1] and [2] are respectively determined by the
probabilities of the two following conditionals:
[1i] If Oswald wasn t the one who killed Kennedy, then someone else
was.
[2i] If Oswald had not been the one who killed Kennedy, then some-
one else would have been the one who killed him.
Despite the mood shift, I can think of no reason for assigning different
probabilities to these two conditionals. Here the only difference between
[1i] and [2i] is that [1i] expresses no commitment to the truth or falsehood
of the antecedent whereas [2i] expresses a commitment to the falsehood of
the antecedent  that, of course, being the difference that Lewis claims is
not sufficient to distinguish an indicative conditional from its counterpart
counterfactual conditional.
We get the same result when both [1] and [2] are employed in a context
where the possibility of a conspiracy is salient to the discussion. To make
this vivid, suppose that Kennedy s assassination was part of an elaborate
plot involving a large number of marksmen stationed at various points
along the route of Kennedy s motorcade. Not yet knowing what happened,
the following exchange might take place between two organizers of the
indicative and counterfactual conditionals 289
conspiracy:
A: I wonder whether Oswald was able to kill Kennedy?
B: Don t worry, if Oswald did not kill him, someone else did.
This context generates a new pair of conditionals:
[1c] If Oswald did not kill Kennedy, then someone else stepped in
and did.
[2c] If Oswald had not killed Kennedy, then someone else would
have stepped in and killed him.
As with [1i] and [2i] there seems to be no reason to assign contrasting
probabilities to these two conditionals. So, once again, we fail to get that
disparity in probability assignments upon which Lewis s argument
depends.
It seems then that Lewis s argument fails to establish what it is intended
to establish, that the disparity in probability assignments on which his
argument depends is based solely on mood shift, for a relevant alternative
has not been eliminated: context shift. And the shift in contextual setting
seems to be the better candidate for explaining the disparity he notes, for
this disparity simply disappears when relevant contextual features are held
constant.
Two things in closing: (i) I have concentrated on Lewis s example
because it is simple, initially highly persuasive, and characteristic of a wide
range of similar examples. (ii) The argument presented here does not
depend on adopting any particular account of the underlying logical struc-
tures of either indicative or counterfactual conditionals. It does, however,
show that one widely-accepted reason for thinking that these two kinds of
conditionals must have fundamentally different underlying logical struc-
tures is not persuasive.
Dartmouth College
Hanover, NH 03755-3592, USA
[e-mail]