ACCOUNTING FOR COVARIANCES AMONG TEST DAY MILK YIELDS IN DAIRY COWS

Abstract

Daily milk weights from 1006 lactations on 775 Holstein-Friesian cows in 42 herds and monthly test-day weights from 102 540 lactations on 73 717 cows in 17 481 herd-year-seasons were used to study the influence of covariances among milk weighings within a lactation on three models for describing the shape of the lactation curve for individual cows. The models included a gamma function, an inverse quadratic polynomial function, and a regression model of yields on day in lactation (linear and quadratic) and on log of 305 divided by day in lactation (linear and quadratic). For each model, several variance-covariance matrices of the observation vector were used. Models were compared on the basis of squared deviations of predicted versus actual milk weights and on the correlation between predicted and actual weights through the lactation averaged over cows. Better predictions were observed when covariances among test-day yields were ignored while models could be ranked regression model, gamma function, and inverse quadratic polynomial function in order of best to worst. Heritability estimates for the parameters of the various models and for 305-d milk yield ranged from 0.11 to 0.30. Genetic correlations were estimated and predictions of correlated responses in 305-d yield from selecting on various combinations of parameters from each method were computed. The best combination of parameters of the gamma function gave a relative efficiency of 74.7% as compared to selection for 305-d yield alone. Key words: Lactation curves, covariances, Holsteins