Comparison of multivariate statistical methods for the
prediction of reproductive performance of Holstein Friesian cattle
B.L. HARRIS, J.E. PRYCE AND G. VERKERK
Livestock Improvement Corporation, Private Bag 3016, Hamilton,
New Zealand
E-Mail:
NZSAP 2003 Abstract No. 27
Proceedings of the New Zealand Society of Animal Production
63: 112-115
The objectives of this study were to predict success or failure
at mating (1/0) using information on 168 first lactation dairy cows from
the Dexcel Strain Trial. The farmlets were established by allocating
heifers from three genetic strains to eleven groups. The groups were
managed under a range of feeding systems, ranging from moderate to
generous allowances. Measurements were taken on milk, fat and protein
yields, liveweight and condition score and variables derived from these
used as predictors of success at mating in a series of models.
Reproductive measures used for analysis were cows in calf at 42, 49 and
56 days after the planned start of mating. Three statistical methods
were used to determine suitable predictive models of success at mating:
logistic regression, discriminant analysis and partition analysis.
Partition analysis was the best method of prediction. Kappa values,
measuring the degree of agreement between observations and predictions
ranged between 0.32 and 0.38 for discriminant analysis, 0.41 and 0.44
for logistic regression and 0.70 and 0.74 for partition analysis. The
partition analysis produced a set of rules for predicting likely success
at mating. Important rules were CS > 4.75 in month three of lactation
and calving before week four after the start of calving for the herd.
Keywords: NZSAPAB;
prediction models; fertility; dairy cattle; partition analysis
Last Updated 8/07/2004
