Analyse problems, not data

The last fifty years or so have seen a transformational change in statistical methodology, from a discrete set of specific methods to a single, integrated paradigm. An early example is the seminal paper by Nelder and Wedderburn (1972) that introduced the unifying concept of the generalised linear model for independently replicated data. Later computational advances have stimulated a comparable unification for modelling data with various kinds of dependence, for example in time and/or in space. I argue that this transformation should encourage statistical scientists to change their focus from analysing data to solving problems. I give an example from an ongoing study of the acquisition of natural immunity to leptospirosis among slum-dwellers in northern Brazil.