Estimates of a trait’s heritability can be used to predict the advance through selective breeding in agriculture and the laboratory where researchers can replicate varieties and locations. These conditions do not apply to human populations, yet considerable attention is still given to high heritability and to small effects of family members growing up together relative to differences within families. This article shows that the conventional partitioning of a trait’s variation produces components that cannot be associated reliably with average differences among varieties and locations (“genotypes” and “environments”), let alone underwrite hypotheses about measurable genetic and environmental influences.
The critique is conceptual and methodological. It begins with agricultural trials in which varieties are raised in multiple replicates in one or more locations. The estimation of heritability and the relative size of shared versus nonshared environmental effects are examined in this ideal situation. Key shortcomings arise from: inattention to the average differences among variety-location combinations (“genotype-environment interaction”); the assumption that, all other things being equal, fraternal twins are half as similar as identical twins; and interpreting unsystematic variation as non-shared environmental effects. It is difficult to remedy the shortcomings in the constrained circumstances of human quantitative genetic research, which centers on genealogically defined varieties (sets of twins or siblings) replicated as few as two times in one or two locations per variety. An adjustment for the first shortcoming takes previously cited human heritability estimates below the fractions of the total variation given by among-location-average variance (“shared environmental effects”) and variety-location-interaction variance. Further remarks are made on the agricultural-human distinction and selective breeding.