ABSTRACT
The ideal implementation of the scienti®c method is to develop a hypothesis and then design the perfect study to refute it (Popper, 1983). However, in the ®eld of lifecourse epidemiology, it is rare to have the resources or time to design a study that can collect relevant data from scratch. The majority of data are obtained opportunistically, using pre-existing databases. It is therefore usually necessary to adopt a more pragmatic approach, in which a suitable dataset for answering a particular question is sought, perhaps with an ideal hypothesis in mind, and the exact hypothesis to be tested is decided once the limitations of the data are known.
