ARS | Publication request: DETECTION AND ADJUSTMENT OF ABNORMAL TEST-DAY YIELDS

Submitted to: Journal of Dairy Science
Publication Type: Abstract Only
Publication Acceptance Date: April 15, 2002
Publication Date: N/A

Technical Abstract: Test-day milk, fat, and protein yields were investigated to develop a method of detecting abnormal values to replace the current use of the sick code to exclude test-day yields. In this investigation, both abnormally low and high yields were detected. Yields were considered abnormal if they fell outside the range of .6 to 1.5 x the predicted yield. This range was selected to classify the high and low 1% of the distribution as abnormal. To calculate the predicted yield, first a slope was estimated based on days in milk and test-day yield. The estimated slope was applied to the previous yield. If a test was detected as abnormal it was checked against a predicted yield which was a linear interpolation of the last previous normal test and the following test. The herd average was used when there were < 3 tests and to determine an acceptable range for component percentages. Applying this method to over 93 million test-day yields from calvings in 1997 or later identified 1.8% of the milk, 3.4% of the fat, and 1.9% of the protein yields as abnormal. The higher percentage for fat reflects its greater variability. Lactation yields were calculated after replacing abnormal yields with a floor or ceiling set to .6 or 1.5 of the predicted value. For 561,063 milk lactations with one or more abnormal tests and a following lactation the correlation between consecutive lactations increased from .692 to .693. For 951,387 fat lactations the correlation increased from .653 to .660 and for 488,653 protein lactations, the increase was from .686 to .694. These adjustments apply a consistent standard to detection of outlier test-day yields and improve the correlation with subsequent lactation yields.