Developmental Increases in the Reliability of Cognitive Assessment Bias G x Age Estimates

The heritability (G) of cognitive ability increases substantially across development. However, previous studies of G x Age interaction—also known as the Wilson Effect—have not controlled for developmental changes in the reliability of cognitive assessment. Unmodeled reliability changes may bias estimates of cognitive growth and G x Age interaction, obscuring the developmental etiology of cognitive ability, but this possibility has not been explored empirically due to a lack of appropriate data and statistical methods. Using an expanded version of the Louisville Twin Study data set in which Ronald Wilson (1983) originally documented G x Age interaction, we first replicated heritability increases between ages 3 months and 15 years. Then, we examined the extent to which unmodeled reliability changes biased estimates of G x Age interaction using continuous time dynamic modeling. Our-best fitting model, which corrected for reliability differences across test battery and age, estimated that heritability increased from .26 to .55. Reliability increased substantially across development, from less than .30 in infancy to approximately .80 at age 15. Models that did not control for developmental reliability increases yielded downwardly biased estimates of cognitive growth and differentiation, G x Age interaction, developmental decreases in shared environmental effects, and increases in non-shared environmental effects. Results indicated that a re-evaluation of previous G x Age estimates may be warranted, and that future studies should control for developmental fluctuations in reliability. Our continuous time dynamic models improved estimation of developmental change in cognitive ability and can be easily adapted to other psychological constructs.