Hd. Norman et al., Comparison of test interval and best prediction methods for estimation of lactation yield from monthly, am-pm, and trimonthly testing, J DAIRY SCI, 82(2), 1999, pp. 438-444
A method with best prediction properties that condenses information from al
l test days into measures of lactation yield and persistency has been propo
sed as a possible replacement for the test interval method and projection f
actors. The proposed method uses previously established correlations betwee
n individual test days and includes inversion of a matrix for each lactatio
n. Milk weights that were representative of monthly, a.m.-p.m., and trimont
hly test plans were examined to compare the accuracy of best prediction and
test interval methods for estimating lactation yield. Individual milk weig
hts or daily yields of 658 Canadian cows in 17 herds were selected to corre
spond to test intervals for 100,000 US cows. For a.m.-p.m. testing, the ini
tial milk weight that was credited was selected randomly from the a.m. or p
.m. milking and was alternated thereafter. Trimonthly credits were from one
of the first three designated test day weights, selected randomly, and eac
h third designated test weight thereafter. Correlations between 305-d actua
l lactation yield and lactation estimates by the test interval method were
0.97, 0.96, and 0.93 for monthly, a.m.-p.m., and trimonthly testing, respec
tively. Corresponding correlations for the best prediction method were 0.97
, 0.97, and 0.93. Standard deviations of differences between estimated and
305-d actual yields for monthly, a.m.-p.m., and trimonthly testing were 373
, 400, and 546 kg, respectively, for best prediction regressed on herd mean
, which was a reduction in estimation error of 4, 6, and 10% over the test
interval method. The advantage of best prediction was moderate if two milk
weights were recorded monthly and was larger if testing was less frequent.
Advantages also were found for fat and protein yields estimated by multitra
it best prediction for records with reduced component sampling.