Restricted maximum likelihood (REML) estimation of genetic parameters for type traits in Normande cattle breed

Abstract

Data on type traits (29 variables) from progeny of 437 Normande sampling bulls (16 356 daughters), connected through the progeny of 37 service bulls (14 170 daughters) were analysed in order to estimate genetic parameters (i.e. heritabilities and genetic correlations). In addition to the usual Henderson's Method 3, several methods for calculating REML estimates were used: Fisher's method of scoring, Newton-Raphson, EM (expectation-maximization) and accelerated EM. In terms of computational requirements, accelerated EM seemed to be the best for implementing REML. Differences between Henderson 3 and REML estimates and their sampling errors were very small. Possible reasons for this are discussed. Values of genetic parameters for type in the Normande breed were similar to their counterparts in Friesians or Holsteins. Furthermore, there was no need to use such a large number of variables for selection purposes. New synthetic variables could be constructed from a principal component analysis applied to the matrix of genetic correlation coefficients, thus allowing a substantial reduction of the useful variables and an improvement of their heritability values. The underlying idea was to define independent selection objectives on the standardized genetic scale.