Best prediction of yields for long lactations

J.B. Cole1, D.J. Null, and P.M. VanRaden

Animal Improvement Programs Laboratory, Agricultural Research Service, USDA, Beltsville, MD 20705-2350

2009 J. Dairy Sci. (?)
© American Dairy Science Association, 2009. All rights reserved.
Individuals may download, store, or print single copies solely for personal use.
Do not share personal accounts or passwords for the purposes of disseminating this article.


Lactation records of any reasonable length now can be processed with the selection index method known as best prediction (BP). Previous prediction programs were limited to the 305-d standard used since 1935. Best prediction was implemented in 1998 to calculate lactation records in USDA genetic evaluations, replacing the test interval method used since 1969 to calculate lactation records. Best prediction is more complex but also more accurate, particularly when testing is less frequent. Programs were reorganized to output better graphics, give users simpler access to options, and provide additional output, such as BP of daily yields. Test-day data for 6 breeds were extracted from the national dairy database, and lactation lengths were required to be ≥500-d (Ayrshires and Milking Shorthorns) or ≥ 800-d (all others). Average yield and SD at any day in milk (DIM) were estimated by fitting 3-parameter Wood's curves (milk, fat, protein) and 4-parameter exponential functions (somatic cell score) to means and SD of 15- (≤ 300 DIM) and 30-d (> 300 DIM) intervals. Correlations among test day yields were estimated using an autoregressive matrix to account for biological changes and an identity matrix to model daily measurement error. Autoregressive parameters (r) were estimated separately for first- (r = 0.998) and later-parities (r = 0.995). These r were slightly larger than previous estimates due to the inclusion of the identity matrix. Correlations between traits were modified so that correlations between somatic cell score and other traits may be non-zero. The new lactation curves and correlation functions were validated by extracting test day data from the national database, estimating 305-d yields using the original and new programs, and correlating those results. Daily BP of yield were validated using daily milk weights from on-farm meters in university research herds. Correlations ranged from 0.900 to 0.988 for 305-d milk yield. High correlations ranged from 0.844 to 0.988 for daily yields, although correlations were as low as 0.015 on day 1 of lactation, which may be due to calving-related disorders that are not accounted for by BP. Correlations between 305-d yield calculated using 50-d intervals from 50 to 250 DIM and 305-yield calculated using all test days to 500 DIM increased as test day data accumulated. Many cows can protably produce for >305 DIM, and the revised program provides a flexible tool to model these records.

(Key words: best prediction, milk yield, long lactations)