Do educational opportunities skip generations? Evidence from sibling and cousin correlations in the United States

Cousin correlations are often interpreted as evidence that family influence extends beyond two generations. Lundberg (2020), however, showed that such correlations can be fully consistent with a first-order Markov process once measurement error is taken into account, without requiring any direct role of grandparents or extended family. This article provides a direct empirical test of Lundberg’s framework using three-generation data from the U.S. National Longitudinal Surveys (NLSY79 and the Child and Young Adult sample). Unlike prior studies, I estimate all of the key correlations needed for the test: sibling correlations in both the parent and child generations as well as cousin correlations in the child generation. These estimates, together with established reliabilities of educational attainment, yield a model-implied cousin correlation that can be directly compared to the observed value. The results show that cousin similarities in education are fully accounted for by a dynamic first-order process: transmission from grandparents to parents was stronger than from parents to children. Once this variation in transmission strength and measurement error are incorporated, the model reproduces the observed cousin correlations, leaving little scope for any residual extended-family effect. This conclusion also holds in race-specific analyses, despite higher cousin correlations among White respondents than Black respondents. I discuss the implications of these findings for research on multigenerational mobility and the interpretation of extended-family effects.