Restriction is an alternative to matching, which is used to prevent confounding in a study during the stage of the study design . Restriction consists in limiting the entrance to the study on a restricted number of subjects, on the basis of a possible confounder. For example, if we think that gender is a confounder, we may enroll in our study only males. In this way we will be able to study the association, for example, on the risk of cancer X and drinking whisky, without the need to account for gender as a potential confounder.
The disadvantage of using restriction is that we will not be able to quantify the absolute effect of drinking whisky and cancer X in the population, but our estimate will just represent the association in the male group.
It is also possible to perferm a "restricted analysis", by conducting an analysis in one specific stratum, once a study has been conducted with a unmatched or matched design. This could mean to enroll males and female in the study and then to perform a sub-analysis only with one gender.
Obviously the consequence is that the subsequent results would only apply to that stratum of the population and could not be generalised. Sample size would also have to be calculated accordingly in order to maintain a sufficient power in the study.
1. Rothman KJ; Epidemiology: an introduction. Oxford University Press 2002, p.108-111.