Working retail

Mon 03 Dec 2007 11:26 AM

I've been thinking about the distinction between retail and wholesale arguments in philosophy of science. A retail argument is about a specific theory, specific kinds of entity, or a specific practice. A wholesale argument promises a conclusion about all or most of science. Wholesale arguments are often stated in more modest terms; for example, the conclusion might be about all the theories in mature sciences. However, such modest authors typically slip back into talking about science simpliciter over the course of an essay.

Suppose one studied 1000 specific cases, and the retail arguments led to a realist victory in 900 of them and an antirealist victory in 100. One might then be tempted to say that most of the other cases will turn out realist as well. Roughly, we should expect 90% of cases to be victories for the realist.* This is what Arthur Fine calls piecemeal realism. It generates a wholesale argument by generalizing over many retail arguments.

I am suspicious of wholesale arguments, but that suspicion is threatened here. If retail arguments can be successful, then we might consider them for a great many different cases. And then we might generalize.

However, the generalization would only be justified if (a) the cases involved formed a homogenous reference class and (b) the cases studied were representative.** Neither of these assumptions is likely to be true.

a. Positing entities serves different functions, and scientific posits are of different kinds. Even if belief is merited in 90% of the posits studied, there is no reason to expect that belief will be merited in 90% of the posits in mature sciences generally. The factors that make the difference between realism and antirealism are more fine-grained than that.

b. Philosophers, historians, and sociologists of science select cases to study based on their own background, on what research material is available, and on what they hope to establish in their enquiry. One might make out general trends based on cases studies, but one will not be able to draw any precise statistical conclusions.

If most retail arguments did favor realism, then of course we would form the expectation of the next retail argument that it would favor realism. But we will never be secure enough in this expectation that we could do without the retail argument entirely and rest on a probability that realism wins in the unexplored cases.

* Nothing that I say here requires that realism be the 9/10th victor. Everything should still hold mutatis mutandis if the preponderance of cases went antirealist.

** One also needs to assume that scientific cases comprise a well-defined sample space. There might be reasons to be dubious of this, too.