Theism and Explanation - Appeared-to-Blogly

Successful Theistic Explanations 125
Another possible response would be to point out, as I did in the previous
section, that even a solitary theistic explanation can be tested by invoking
what statisticians call the “null hypothesis.” What do I mean? If an experimenter
is wanting to test the effi cacy of a drug, then the null hypothesis is
that there will be no statistically signifi cant difference in outcome between a
test group and a control group. More precisely, the null hypothesis predicts
that any differences between the two groups will be attributable to chance.
In this situation, there is only one hypothesis being tested—that the drug
will be effective. Can we say that it is being tested comparatively? In a certain
sense, yes. The hypothesis will be corroborated to the extent that the
null hypothesis is disproved. And it will be disproved to the extent that the
result is signifi cantly more likely to have occurred on the hypothesis under
test than by chance. If the result is represented as E, the hypothesis under
test as H, and the null hypothesis as C, then the hypothesis is corroborated
if (to some signifi cant degree)
Pr(E|H) > Pr(E|C).
How can this observation be applied to a proposed theistic explanation?
Well, where a rival proposed explanation does exist—say, Darwin’s theory
as opposed to the creationist account—then we can ask which of these
would render a range of observations more likely. We can test the theistic
hypothesis comparatively, in precisely Sober’s sense. But where there is no
rival potential explanation—where the proposed theistic explanation is the
only one on offer—one could argue that there does exist a default null
hypothesis over against which it can be tested. It is the claim that the fact
in question occurred by chance. 35
But while this seems a legitimate argument, it is not my preferred
response. My preferred response is simply to deny Sober’s conclusion. If
corroboration by the passing of independent tests is a desideratum in any
explanation, then a repeated failure to pass such tests will count against the
explanation in question. Sober’s point is that no one failure would falsify
such an explanation. If the prediction it makes is merely probable, then
the failure of a prediction does not show that the hypothesis is false. This
is surely true. But it need not mean that no solitary proposed explanation
is testable. If we can use it to predict facts other than that which it was
introduced to explain, then it is testable. And even if repeated failures to
pass such tests do not, strictly speaking, falsify the theory, they should at
least arouse our suspicions. I have argued (2.1.3.2) that we do not need to
demonstrate that an explanation is true in order to have reason to accept
it. But by the same token we do not need to prove that an explanation is
false in order to have reason to reject it, or at least to seek an alternative.
If a proposed explanation is not corroborated by independent tests, it does
not necessarily follow that it is false. (To this extent, Sober is correct.) But
it does follow that we have less reason to regard it as a good explanation,
one we are entitled to accept.