The most accurate confidence intervals for estimates of heritability are based on Fisher’s Z-transformation. Using these methods, Loehlin and Nichol’s (1976) analysis is confirmed, viz., that it is pointless to estimate heritabilities from the classical twin method unless one is prepared to recruit upwards of 800 twin pairs for study. Even then the assumptions of that method are so improbable as to leave reasonable doubt about the true value of H. Estimates of H from the correlation of monozygotic twins reared apart (MZA twins), on the other hand, is remarkably more stable than that obtained by comparing MZ with DZ correlations. Moreover, estimates of H based on the MZA design rest upon more reasonable (and often testable) assumptions.
