I develop an alternative to standard semantic theories of subjunctive or “counterfactual”
conditionals using “causal models” of roughly the sort investigated by Judea Pearl (1999, 2000)
and Spirtes et al. (2000).
The currently standard approach, due primarily to Robert Stalnaker
(1968, 1970) and David Lewis (1973, 1986b), appeals to relations of “similarity” of possible
On this view, ‘If A had been, then C would have been’ is true iff C is true in all possible
worlds in which A is true and which are otherwise “similar” to actuality.
Lewis suggests that similarity of possible worlds is a sort of global resemblance,
something like the overall similarity of cities, faces, and philosophies (1973 94-5).
worlds can be specified without reference to causal relations, natural necessities, or other
The idea is that we can then use counterfactuals to provide reductive
analyses of causation, dispositions, and other causal concepts.
I argue that this approach gets the order of analysis backward: a proper understanding of
counterfactuals requires reference to causal relations and practically nothing else.
idea that I urge is nicely expressed by Igal Kvart (1986 44):
In considering the contrary-to-fact assumption, we consider this change, and only this
change, and contemplate the possible effects of this change
against the background of the
actual course of events, and only these effects.
Consequently, in considering this change,
we are not prompted to question those actual events which have nothing to do with it.
In evaluating ‘If A had been, then C would have been’, we start with a causal network of actual
events. If A is actually false, we introduce a minimal break
in that causal network that allows A
to be true.
Then we trace out the causal consequences of that break, holding fixed other matters
which are not influenced by the initial break or its consequences.
‘If A had been, then C would
have been’ is true iff C is true in such situations.
I will use “causal models” roughly like those of
Judea Pearl (1999, 2000) and Spirtes et al. (2000) to represent causal networks and “minimal
breaks” in them.
§1 presents Lewis’s theory of counterfactuals and raises some counterexamples.