Parameter estimates from the DeFries and Fulker [(DF) Behav. Genet. 15
:462-473, 1985] regression method can be greater than unity or less th
an zero. This occurs when the monozygotic correlation is greater than
twice the dizygotic correlation. Sensible values can be obtained in th
ese cases by fitting a constrained DF model that estimates genetic and
nonshared environmental variance components only. In this article I d
emonstrate that the original Df model yields positively biased heritab
ility estimates and negatively biased estimates of shared environmenta
lity when data are significantly influenced by genetic nonadditivity.
The magnitude of the bias is algebraically expressed. I then describe
a simple regression equation that provides unbiased estimates of the s
tandardized additive and dominance genetic variance components. Result
s of a study of 6 million twin pairs from the Monte Carlo Twin Registr
y demonstrate that the DF additive and dominance genetic parameter est
imates are virtually identical to those obtained by maximum-likelihood
procedures. Finally, I derive the expectations for the constrained DF
model and show that the genetic parameter estimates from this model a
re negatively biased estimates of broad-sense heritability.