An empirical check on best linear unbiased prediction genetic evaluation using pig field recording data

The theory and use of best linear unbiased prediction in genetic evaluation are well developed. However, there has been little empirical checking of its efficacy in practice. The objective here was to use a large body of Canadian pig performance records to check on the predicted benefits of BLUP in genetic evaluation. Phenotype records were available on fat depth and on days to 100 kg on some 65 000 progeny born in 1994 and 1995 from parents evaluated before 1994. Rank correlations between parent and progeny in data were calculated within herd-year-season to avoid effects due to differences in these factors. Computer simulation studies were also run to check on the predicted results. The simulation results confirmed the expectations on the higher correlation of mid-parental EBV than of mid-parental phenotype with progeny genotype and a regression (of progeny phenotype on mid-parental EBV) of unity when all relevant pedigree and performance data were used. In the data analysis, the (rank) correlations with progeny phenotype were consistently higher (36 and 27%) for mid-parental BLUP genetic evaluation than for mid-parental phenotypes, confirming the superiority of the BLUP evaluations over phenotypes. However, the regression of progeny phenotype on mid-parent BLUP EBV was usually less than the predicted value of unity. Simulation results suggest that either the base population heritability was lower than that used in the evaluation or that the information used was incomplete. Key words: Best linear unbiased prediction, EBV, pigs, performance, selection