K.R. Koots, J.P. Gibson, C. Smith and J.W. Wilton
Anim. Breed. Abst. (1994) 62:309-338
Abstract
The number of published heritability estimates for some beef traits is sufficient to conduct analyses to determine factors affecting the estimates, and to make recommendations regarding appropriate mean values. A literature search found a total of 1656 heritabilities for 70 traits from 287 studies covering 45 years (1946 to 1991). There were many heritability estimates for growth traits, but far fewer for the economically important traits relating to reproduction and carcass merit. Weighted least-squares analyses were performed with up to seven factors fitted for each trait [breed, country data origin (field or experimental), feeding management, estimation method, sex, and decade of data collection]. Breed and country were identified as significant effects in many cases. Contrary to widespread belief, heritabilities derived from field data and experimental data did not differ for growth traits. Heritabilities estimated using animal models did not differ from those by other methods of estimation. For most traits, males had lower heritabilities than females. Heritability estimates were also affected by the mean and phenotypic standard deviation of the population in which they were derived. In general, the observed standard deviation among estimates was twice the predicted theoretical standard deviation. In a covariance analysis across studies, heritability estimates for many traits were positively correlated with those of other traits. There was evidence that for traits with low heritability, average heritabilities were substantially overestimated, and so considerably less response to selection for such traits is expected than is currently predicted. The weighted mean heritabilities presented here are recommended for use when reliable estimates are not available within a population. Where estimates within a population are available, these should be combined with the weighted literature average. A simple rule would be to weight each, the local estimate and the pooled mean, by the inverse of their squared standard error.