The negative effects of population stratification can sometimes be avoided at the study design level (by matching controls to cases on potentially important confounders that mark population structure) or the data analysis level (by adjusting the results for these confounders – see later in this protocol series). It should be noted that matching is only essential when the frequency of the confounder shows such a marked difference between cases and controls that it cannot be adjusted for in the analysis, or in situations where the confounder cannot be accurately measured. ‘Overmatching’ on unnecessary variables will actually reduce power, since all matching variables will need to be taken account of in the analysis 22. Population stratification is minimised when controls are matched to cases on ethnicity (or when cases and controls are restricted to a particular ethnic group), often ascertained through self-description 23, although the extent to which this can avoid stratification depends on the population under study and the differences in disease prevalences and allele frequencies across populations 24-27. Further matching on sex can reduce population stratification in situations where
