Detection and adjustment of abnormal testday yields

1
Detection and adjustment of abnormal test-day
yields
(abstract 125)
G.R. WIGGANS, P.M. VANRADEN, and J.C.
PHILPOT Animal Improvement Programs
Laboratory Agricultural Research Service,
USDA Beltsville, MD 20705-2350,
USA wiggans_at_aipl.arsusda.gov ? 301-504-8334
2
Estimation of lactation yields

• Inclusion of abnormal yields may reduce accuracy
• Cow may be reported as sick on test day
• ICAR procedures exclude sick tests below a
  threshold

3
Calculation of lactation records

• Best prediction used since February 1999 for
  calvings during 1997 or later
• All sick tests have been excluded
• Many sick tests with milk yields gt45 kg
• Sick coding varies by herd and dairy records
  processing center
• No check for abnormally high test-day yield
  compared with cow's other test days

4
Objectives

• Check every test for abnormal yield (not just
  those coded as sick)
• Detect abnormally high or low test yields by
  comparison with yield from adjacent test days
• Define approximately 1 of each tail of the
  distribution as abnormal
• Adjust test-day yield to floor or ceiling

5
Methods

• Calculate predicted yield
• Compare test-day yield with thresholds
• Upper 150 of predicted yield
• Lower 60 of predicted yield
• If test-day yield detected to be abnormal,
  recalculate predicted yield based on yield from
  adjacent test days

6
Methods (continued)

• If test-day yield still detected to be abnormal,
  adjust to threshold
• Calculate lactation yields using best prediction
  with adjusted test-day yields

7
Predicted yield calculation

• For first test day
• predicted yield following yield slope
  (interval)

• For second test day and later
• predicted yield previous yield slope
  (interval)

8
Predicted yield calculation (continued)

• Slope is function of DIM, DIM2, test-day yield,
  and interaction
• Coefficients differ by
• Test day (first vs. second and later)
• Trait (milk, fat, protein)
• Parity (1, 2, 3)
• DIM group (5-49, 50) to account for peak yield
• Previously identified abnormal yields not used
• Sick tests with milk yields of lt4.5 kg ignored

9
Abnormal yield detected

• Predicted yield recalculated by interpolating
  between adjacent test days
• If test-day yield still detected as abnormal,
  predicted yield recalculated again using second
  test day after outlier
• Herd average used when only 2 test days available

10
Component percentage limits

• Detection of abnormal fat and protein yields
  based on weight (not percentage)
• Unlikely component percentages possible
• Component percentage limits
• Fat 2-7 and
• herd mean 1 to herd mean 2
• Protein 2-5 and
• herd mean 1 to herd mean 1.5

11
Outlier results

• Calvings during 1997 or later
• Among gt93 million test-day records, abnormal
  yields found
• Milk 1.8
• Fat 3.4
• Protein 1.9

12
Improved accuracy?

• Correlations of lactations with ?1 adjusted
  outlier with following lactation

• Adjustment of abnormal yields should increase
  similarity between consecutive lactations

13
August 2002 implementation

• Lactation records for calvings during 1997 or
  later recalculated
• Adjusted yields shown on industry web queries of
  test-day data
• Additional data may cause change in status of
  detected outlier and adjusted yield

14
Summary

• All test-day yields evaluated as potential
  outliers (not just sick tests)
• Improved accuracy in estimation of expected yield
• Floor or ceiling (instead of deletion) for
  abnormal yields to reflect actual yield better
• Improved correlation between consecutive
  lactations with outlier adjustment