The need for theory

hmmm, that doesn't rhyme quite as well. Ben Bolker brought the following quote from Efron and Tibshirani (1986; "Bootstrap methods for standard errors ...") to my attention:

An important theme of what follows is the substitution of computing power for theoretical analysis. This is not an argument against theory, of course, only against unnecessary theory.

I've often thought of the need for theory as falling along a continuum of 1/n, so when your sample size is small you need strong theory to make predictions, and when large you can get away with less theory. In either case it helps if your theory is well tested in other cases, or you risk making predictions that are completely bogus.

Wise decisions and predictions

Daniel Sarewitz is a leader in the Science-Policy interface area, and last year he had this to say in an opinion piece in Nature last year:

If wise decisions depended on accurate predictions, then in most areas of human endeavour wise decisions would be impossible. Indeed, predictions may even be an impediment to wisdom. They can narrow the view of the future, drawing attention to some conditions, events and timescales at the expense of others, thereby narrowing response options and flexibility as well.

Would “projections” also lead to the same trap? According to Kevin Trenbarth, the difference is that a projection makes no effort to start from the actual initial state of the system, and so all that can be evaluated is the change from the assumed initial state. As a result, there is no expectation on the part of the “projector” that the projection will actually come to pass. In contrast, a prediction is made in the expectation that the future will look similar to the prediction, although as far as I can tell, the same tools are used for both. Intriguingly this is yet a third way to define the difference between a projection and a prediction. Either way, I think predictions and projections run the risks described by Sarewitz.