A Science paper claiming lifespan heritability is ~50% actually models a hypothetical world with zero extrinsic mortality, not the real world where it is 23-35%.
Key Takeaways
Real-world twin studies (Danish, Swedish cohorts) put lifespan heritability at 23-35%; the paper’s 46-57% figure comes from simulated data with extrinsic mortality zeroed out.
The paper builds a Gompertz-style death-probability model, fits a Gaussian over genetic parameters to match historical data, then simulates identical and fraternal twins under varied extrinsic mortality.
Heritability rising when non-genetic noise is removed is mathematically trivial; the paper’s value is quantifying how much it rises and providing a simulation framework for more recent cohorts.
Twin datasets used for baseline estimates include people born as early as 1870, so lower modern extrinsic mortality likely means true current heritability is already higher than 23-35%.
Science’s editorial practice of stripping math from quantitative papers made the methods nearly unverifiable from the main text; key modeling choices appear only in a vague appendix.
Hacker News Comment Review
Commenters converged on a core semantic point: heritability is h = Var(genetic)/Var(phenotypic), a population-and-environment-contingent statistic, not a fixed biological constant, making cross-study comparisons structurally ambiguous.
There is tension between the statistical genetics definition and lay intuition; one commenter noted human bipedalism has near-zero heritability by the standard formula because phenotypic variance is near zero, illustrating how the metric misleads.
Skepticism about practical intervention implications: reducing harmful environmental factors (lead exposure) reliably moves outcomes, but boosting positive inputs has shown near-zero effect, which some read as indirect evidence that residual variance is mostly genetic.
Notable Comments
@ndr: flags the bipedalism edge case to show h = Var(genetic)/Var(phenotypic) produces counterintuitive results when phenotypic variance collapses.
@svnt: “heritability is hopelessly overloaded” – argues current terminology makes coherent gene-environment discussion impossible.
