Families as Natural Experiments

Abstract

In this article, the author argues that data on siblings provide a way to account for the impact of unmeasured, omitted variables on relashionships of interest. This is possible because families form a sort of natural experiment. Family members are likely to have many shared experiences, as well as a common genetic heritage, but relationships between variables defined as differences between family members cannot be attributed to these shared family characteristics. Although fixed- and random-effects models are discussed as one means to make use of information on siblings from the same family, the author proposes a latent-variable structural equation approach to the problem. This model provides estimates of both within-and between-family relationships, and it accounts for the impact of measurement error.