Problems in the family: Controlling for age, period or cohort in sibling comparison designs

The sibling comparison design is increasingly used across the social sciences as a means of controlling for observed and unobserved confounding, but this design is not without its methodological challenges. One challenge is to control for age, period or cohort when studying exposures that are measured in units of time. In such cases, variation between siblings in the exposure is typically collinear with variation in age, period or cohort – which makes controlling for these factors highly problematic. We address this challenge by showing how it is possible to control for cohort (or age or period) and obtain unbiased estimates of the effect of an exposure that is measured in units of time, albeit given certain assumptions. Using a simulation study that compares a series of estimators, we show that bias can be minimised by including a group who are discordant on age, period or cohort but concordant on the exposure.