British Doctors Study will be similar to the biology of all people, so this intended study population should enable us to draw inferences about many exposures and outcomes that would hold for the general population of the UK (and indeed of the world). If the outcome of interest has not influenced selection into the intended study population, and all of this population is included in the study (i.e., in what we have described as the “actual study population”), then selection bias, and therefore collider bias, should be small. However, for some factors under study, representativeness may matter to a greater degree than is appreciated – for example, if general cognitive ability and childhood socioeconomic position both influence the likelihood of becoming a doctor, and childhood socioeconomic position influences a wide range of health outcomes, bias may still be introduced when attempting to use doctors to draw inferences about the effects, say, of general cognitive ability on disease in a general population. These biases may be small in magnitude, but as we increasingly investigate small associations, their relative importance may be greater than is appreciated.Box 3). In addition, having some baseline data and DNA available on all participants at recruitment into
