The Wilson Effect: The Increase in Heritability of IQ With Age

Narrative synthesis of ~10 named twin/adoption datasets spanning ages 5 to 80+: Boomsma 2002 (8 Dutch cross-sectional twin studies), Haworth 2010 (6 pooled twin cohorts), 5 MZA studies, Texas Adoption Project (~300 families, 14 kinships), Teasdale & Owen Danish military siblings/half-siblings, Reynolds 2005 Swedish twins at 65, unrelated-reared-together literature (13 childhood + 5 adult studies)

Twin/adoption samples from childhood to old age, drawn almost exclusively from Western industrial democracies; the author explicitly limits generalization to that domain and notes few participants were raised in real poverty or by illiterate parents.

Narrative synthesis, NOT a formal meta-analysis: key datasets are plotted against a non-linear age axis in a single figure (Fig 2) with a hand-fitted polynomial through the Dutch estimates. No pooling, no weighting, no confidence intervals. The headline adult c2 = 0.10 is one subtraction, not a fitted estimate: adult MZ-apart r = 0.74 (mean of five studies) subtracted from an assumed adult MZ-together r = 0.84 (p.925). Estimates are of BROAD heritability (additive + dominance + assortative-mating variance), so they are not comparable to narrow-sense or SNP heritabilities.

The correct statement of the ceiling on family-environment effects: shared environment for IQ falls to about 0.10 in adulthood — NOT to zero. That residual leaves room for roughly 4-5 IQ points of shared-environment influence, which is almost exactly what full adoption delivers (Kendler 2015: +4.41). 'Shared environment goes to zero' is a common overstatement and this database does not use it. Two caveats the abstract hides and full text supplies. (1) The 0.10 is the top of the paper's own adult range, obtained by one subtraction (MZT 0.84 - MZA 0.74); every adult estimate from a design WITHOUT twins in the same paper is 0.00-0.04, and the Dutch reference series sits at 0.00 from age 12 through age 50. The honest reading is "adult c2 is somewhere between 0.00 and 0.10, and the point estimate depends entirely on which design you trust" — which still refutes "zero" as a certainty but does not establish 0.10 as a measured quantity. (2) The 0.80 asymptote is not permanent: heritability of g declines in old age (about 80% at 65 to 60% at 82; 0.62 at age 80), partly through attrition.

This is the genetically-sensitive design itself.

The standard reference for the Wilson effect. Extends Haworth et al. (2010) into adulthood. The paper's own robustness argument is constructive replication: adoption and unrelated-reared-together designs containing no twins reproduce the twin-study trajectory, and the pattern holds across the Netherlands, UK, US, Australia, Norway, Denmark and Sweden (p.927).