Multiple and Single Trait Analyses for Estimating Genetic Parameters in Simulated Populations Under Selection

Abstract

Records were simulated for two traits for a one-way model with random sire groups. Sire effects and residuals were generated from bivariate normal distributions with heritabilities of .30 for each trait. Genetic correlations simulated were .15, .45, and .75 while residual correlations were .15, .35, and .55. Genetic values for 50 sires and residuals corresponding to 100 daughters per sire were simulated. Initially, all daughters were assigned two records. However, if a daughter's first record placed her in the bottom 0, 20, 40, or 60% of the population her second record was not included in the corresponding analysis. An iterative restricted maximum likelihood procedure was used to estimate variance and covariance components by two multiple trait and one single trait algorithm. One multiple trait method required an assumption of zero residual covariance, but the other did not.After 30 replications, results were compared by selection intensities and genetic and residual correlations. With no selection, all estimation methods produced estimates within one standard error of the parameters. Under selection, the multiple trait algorithm requiring no assumptions about residual covariance was superior to the other two in producing accurate estimates, especially with heavy culling and with high underlying correlations of residuals or sires.