The Heritability of Cognitive Ability Across the Lifespan: A Meta-Analysis of Twins Studies

Abstract

We conduct a meta-analysis of twin studies of the proportions of variance in cognitive ability that can be attributed to variation in genetic endowment and variation in shared-environment (h2 and c2 respectively). We use kernel median regression to estimate how these variance shares change over the life course and to estimate confidence intervals on the conditional medians at each age. We use median regression to deal with the censoring of variance shares at 0 and 1. This is particularly important for c2 estimates from the teenage years through adulthood where estimates of exactly zero are frequently reported. We use kernel regression to allow for general non-linearity in the changes in the variance components over time. An exhaustive literature survey found 116 studies – many more than the number used in any previous review or meta-analysis.

Our results confirm previous studies that find a declining role for shared-environment and an increasing role for differences in genetic endowment in explaining differences in cognitive ability as people age. Unlike past studies (Bergen et al. 2007) we find that the pattern of change is substantially non-linear with the increase in h2 and the decline in c2 coming almost entirely before the age of 12 and stabilizing thereafter until some unknown point in adulthood. Confidence intervals confirm that these changes with age are reliable. We can find no evidence that shared-environment explains any variance of cognitive ability in adults within races or ethnic groups.

Our findings for those under 20 and those over 20 are difficult to reconcile. For those from 15 to 20 years of age we find that shared-environment explains a stable and statistically significant share of the variance in cognitive ability (20%). On the other hand, our results for adults rule out a value of c2 greater than .15.

We suggest two possible explanations for this. The paucity of studies of those over 20 may make it impossible for us to identify another decline that starts later in life. Another possibility is that c2 estimates in adult samples are biased downward due to homogeneity of the self-selected subjects. Attempts to find evidence that would allow us to distinguish between these possibilities were inconclusive.

Keywords: Intelligence, Behavior Genetics, Meta-Analysis