Association for Public Policy Analysis & Management

Intergenerational transmission of socioeconomic status has long been of interest to researchers and policy makers, with the U.S. having been referred to as “the land of opportunity”. With datasets now reaching timespans that can include three generations, researchers are now exploring whether grandparents matter for their grandchildren’s socioeconomic outcomes, conditional on that of the parents. In other words, in a regression of child’s outcomes on that of the parents and that of the grandparents, what is the coefficient for grandparents?

Early theoretical work by Becker & Tomes (1979) suggested that one would estimate a small negative grandparental coefficient in a multigenerational regression including parents and grandparents, reflecting faster than geometric decline in status associations over generations. However, several recent multigenerational studies are yielding small positive grandparental coefficient estimates, which suggests slower than geometric decay.

In this paper, we re-examine biases that arise from life cycle profiles and measurement error in income, econometric issues with a rich history in the intergenerational literature, but which have distinct implications for the mutigenerational regression setting. We show that, theoretically, a small positive grandparent estimate could be an artifact of measurement error, even if the true parameter were zero or slightly negative.

Empirically, we illustrate the implications of these biases by applying now customary time-averaging approach as well as an instrumental variables approach to income data for all individuals born in Norway between 1974 and 1978 along with that for their fathers and grandfathers. Although we do indeed find evidence that measurement issues in fathers’ income can lead to upward bias in the coefficient for grandfathers’ income, we also find that a small positive grandparent coefficient remains after attempting to eliminate (or at least mitigate) these biases.