Statistical strategy: step 1

Before one can select a model one must formulate the research question that the model is being built to address. This formulation — deciding precisely what it is one wants to know — is the first step in using statistical methods. It is the first step in any statistical strategy. This paper contends that often the research question is poorly formulated, so that inappropriate models are built and inappropriate analyses undertaken. To illustrate this three examples are given: explanatory versus pragmatic comparisons in clinical trials, confused definitions of interaction, and two ways to measure relative change. A plea is made that statistics teaching should focus on higher level strategic issues, rather than mathematical manipulation since the latter is now performed by the computer.