Design choices for observational studies of the effect of exposure on disease incidence