Splitters and poolers

The world can be divided into those who split and those who pool. For the former, the devil is in the detail. There is no point in talking about the effect of a given treatment on patients in general. Patients do not arrive ‘in general’ at the surgery door, they arrive in particular and it is personal treatment they seek (or at least, this is generally the case). For the latter, what is applicable to one is applicable to one and all. The world is awash with chance and contingency. Only by repeated study and by averaging can we hope to make any sense of it at all.

On the whole, physicians are splitters and statisticians are poolers. Being a statistician I can make that sort of statement. If I were a physician, I would have difficulty in saying anything about it at all.

In fact, intermediate positions are possible, and certain statistical approaches, most notably the Bayesian one, address this explicitly. The Bayesian approach would be to use one's prior belief that apparently similar things often behave in fairly similar ways. It is a matter of reasonable belief, for example, that the effect of a treatment is unlikely to be very different when applied to Roman Catholics then when applied to Anglicans.