2. Introduction

Best prediction (BP) [VanRadenVanRaden1997,VanRadenVanRaden1998] is a computational method derived from selection index [CameronCameron1997] that allows the prediction of missing yields from measured yields and condensation of test day data into lactation yields and persistencies, and was approved in 2002 for lactation record calculation by the International Committee for Animal Recording [International Committee for Animal RecordingInternational Committee for Animal Recording2006]. It requires only phenotypic covariances, and assumes that herd means and variances are known. Reverse prediction may be used to obtain daily yields from lactation yields and persistencies. Both single and multiple trait prediction are supported. Calculation of lactation records for milk (M), fat (F), protein (P), and somatic cell score (SCS) for cows calving in January 1997 and later using BP began in November 1999, replacing the test interval method and projection factors [Shook, Jensen, and DickinsonShook et al.1980,Shook, Johnson, and DickinsonShook et al.1980] at AIPL.

Norman et al. NormanVRWC1999 showed that the advantages of BP are small compared to the Test Interval Method (TIM) [Sargent, Lytton, and WallSargent et al.1968] for most 305-d lactations but larger for lactations with infrequent testing or missing component samples. Some biases in TIM were eliminated by the introduction of Shook factors [Shook, Johnson, and DickinsonShook et al.1980], but BP proved to be more useful when dealing with many alternative testing plans. It permits more precise estimation of records for SCS because test days are adjusted for stage of lactation. Yield records calculated using BP have slightly lower standard deviations (SD) because BP regresses estimates toward the herd average.

The Canadian system for estimating 305-d records, multiple trait prediction (MTP) [Schaeffer and JamrozikSchaeffer and Jamrozik1996], is similar to BP in many respects. In a recent validation study Quist et al. QuistLHLMK2007 compared estimates of actual daily and 305-d yields from MTP with data collected from on-farm milk meters and reported that concordance correlations increase with DIM. They also reported that MTP overestimates actual 305-d yield, which is consistent with results from BP.

The original program for computing BP, AIPLDCR, was limited to 305-d lactations, required recompilation to change parameters, used only simple linear interpolation for calculating standard curves, and could not provide BP for individual days of lactation. A new version of the program, BESTPRED, has a number of advantages over the original routines:

Lactations of any length can be modeled.
Lactation-to-date, 305-d, 365-d, and projected yields are provided.
Persistencies of milk, fat, protein, and SCS are now comparable on within, rather than across, lactation groups.
The autoregressive function used to model correlations among test day yields has been updated.
Program options are set in a parameter file and read at runtime.
Diagnostic plots are available for all traits.
BP of individual daily yields, test day yields, and standard curves are now output.
The program accepts input in, and provides output in, either pounds or kilograms for M, F, and P.

This document describes how to obtain, install, and run the BESTPRED v 2 software package for calculating yields and persistencies using best prediction. It also provides a mathematical overview of BESTPRED, references to the schenftific literature, and notes on computational details and troubleshooting.

See About this document... for information on suggesting changes.