Accounting for scale effects in genetic evaluation of New Zealand dairy cattle : a thesis presented in partial fulfilment for the degree of Master of Agricultural Science in Animal Science at Massey University

Loading...
Thumbnail Image
Date
1994
DOI
Open Access Location
Journal Title
Journal ISSN
Volume Title
Publisher
Massey University
Rights
The Author
Abstract
It is well known that variation in lactation yields tends to increase with average production. Failure to account for this scale effect may cause overestimation of genetic merit for sires with a majority of daughters in high-variation herds and vice-versa. The current system of sire evaluation in New Zealand overcomes this problem by expressing daughters performance as a proportion of contemporary average performance. The objectives of this study were to quantify the magnitude of scaling (heterogeneous variance), and to identify methods to stabilise the variance of milkfat yields for use in the genetic evaluation system of dairy cattle through best linear unbiased prediction (BLUP) using an animal model across breeds. Lactation records of dairy cows calving between 1986 and 1989 were obtained from the Livestock Improvement Corporation of the New Zealand Dairy Board. There were milkfat yields from 2,004,854 lactations in 83,805 contemporary groups (herdyear- age; HYA). The data were divided into three equal-sized subsets based on HYA mean; these being (kg milkfat ± sd) High (H), 172 ± 28; Medium (M), 152 ± 26; and Low (L), 139 ± 25. The methods investigated for the accounting of scaling were: adjustment by the HYA sd (SD-adjustment); scaling by the HYA mean (MEAN-correction); and natural logarithmic transformation (LOG-transformation) of milkfat yield. The overall correlation between HY A means and HYA sd's was 0.44. This value was reduced to 0.31 in SD-adjusted, -0.27 in the MEAN-corrected and -0.24 in the LOG-transformed data. Ideally, the transformed data should exhibit independence between the mean and standard deviation. Breeding values of sires were separately estimated from each data subset using a mixed model. Product-moment and rank correlations between breeding values for sires estimated from the independent subsets and with variable minimum number of daughters were in the overall comparisons (L-M, L-H and M-H) lower than expected correlations, reflecting inaccuracies in sire evaluation when scaling is ignored. Product-moment and rank correlations were similar for SD-adjustment and MEAN-correction, but LOGtransformation reduced the calculated correlations in the L-M, L-H and M-H comparisons. Estimates of the genetic correlations between production in pairs of environments were obtained from the ratio of observed to expected correlations. These estimates ranged from 0.82 to 1.01 for the linear yields. Estimates of genetic correlations were similar for SD-adjusted and MEAN-corrected data, but for LOG-transformed data these were reduced, especially in the L-H comparison which ranged from 0.77 to 0.87. Results confirm the problem of scaling on genetic evaluation of New Zealand dairy cattle. MEAN-correction and LOG-transformation methods are not appropriate because they tend to overcorrect the scaling problem. SD-adjustment is not satisfactory but seems to be more appropriate than no adjustment. An alternative method is proposed based on a Bayesian approach, which takes into account any relationship between variance and mean.
Description
Keywords
Dairy cattle, Cattle-breeding, Genetics, New Zealand, Scaling effect
Citation