Ch. Mallinckrodt et al., APPROXIMATE CONFIDENCE-INTERVALS FOR HERITABILITY FROM METHOD-R ESTIMATES, Journal of animal science, 75(8), 1997, pp. 2041-2046
Method R estimates of heritability (h(2)) and associated confidence in
tervals (CI) were obtained from simulated data using a single trait, d
irect effects, full animal model, with 50% subsampling. Five hundred d
ata sets were simulated for each of five levels of h(2) (.10, .20, .30
, .40, and .50) and two types of pedigree structure (random pedigree s
tructure [N = 2,000] that varied over simulations, or the pedigree str
ucture from a real data set [N = 2,644] that was constant for all simu
lations). The first 10, 20, and all 50 h(2) estimates were used to obt
ain 80, 90, 95, and 99% CI for each data set. The variance of h(2) est
imates within data sets approximated the sampling variance of the h(2)
estimates. The Box-Cox transformation was used to normalize the distr
ibution of estimates from each data set. Confidence intervals were com
puted on the transformed scale as CI = mu +/- (T x sigma), where mu an
d sigma = the mean and SD of the N transformed h(2) estimates, respect
ively, and T = the critical value from the T distribution for a 1 - al
pha CI, with df = N - 1. Upper and lower CI bounds were converted back
to the original scale by reversing the transformation. The percentage
s of CI containing the true h(2) value, pooled across all levels of h(
2), types of pedigree, and number of estimates used to obtain CI, for
80, 90, 95, and 99% CI were 81.14, 90.96, 95.27, and 98.76%, respectiv
ely. These results suggested that Method R h(2) estimates can be used
to obtain reliable CI.