Empirical Bayes Prediction of 305-Day Milk Production

Abstract

A lactation curve described by an algebraic formula can be fitted by regression to the milk weights in the partial record of an individual lactation in progress; however, curves that are obtained in this manner do not provide useful predictions of milk production throughout the remainder of the lactation. This study examined the reasons for this failure and introduced a new empirical Bayes statistical method for fitting Wood's curve that was designed to provide good predictions of future production. The results of a comparison between predictions produced by the new method and predictions from Dairy Herd Improvement Association extension factors were quite favorable to the new method, which has advantages other than greater accuracy; the method does not require preparation of extension factor tables and can be readily adapted to individual herds. Comparisons revealed features of the Dairy Herd Improvement Association predictions that limit their usefulness for some herd management purposes.