The main aim of this study was to analyze Norwegian somatic cell count (SCC) data using several statistical models, and to study the potential for use of SCC in selection for reduced incidence of mastitis. This thesis consists of four papers. The first three studies were based on data from the Norwegian Dairy Herd Recording System from the period 1978 to 1995, while the fourth paper is a theoretical study, using simulated data.
In the first study, SCC was analyzed as lactation mean of log-transformed test-day SCC (LSCS) for 1.3 million first-lactation cows. Estimated heritability for LSCS was 11%, and there was approximately no genetic change for sires born in the period from 1973 to 1990.
In the second study, efficiency of selection based on LSCS and on records of clinical mastitis was compared by using the correlation between predicted breeding values for the two traits, based on first-crop daughters and mastitis incidence in later daughters. Both traits had highly significant effect on mastitis incidence. However, clinical mastitis was 23-43% more efficient in selection than LSCS. Simultaneous use of both traits was slightly more efficient than using CM only. No non-linear genetic relationship between CM and LSCS was found.
In the third study, several random regression sire models for analyses of log transformed test-day SCC (SCS) were evaluated, using a subset of the total dataset. The preferred model had Legendre polynomials up to second order for sire effect and third order for cow-specific effect. Heritability of SCS (5-12%) was lowest in beginning and highest towards the end of lactation. Moderate to high genetic correlations were found between SCS in early and late lactation. The results may be used to predict random regression coefficients for each sire. How to use such coefficients in selection for reduced incidence of mastitis will be a topic of future studies.
Test-day SCC may be regarded as a mixture of observations from animals with unknown health status. A Bayesian two-component mixture model, implemented using Gibbs sampling, was therefore developed in the fourth study. Using mixture models, selection for reduced mastitis incidence may be based on the probability for mastitis given SCC, rather than selection for lowest possible SCC. Further development is necessary to implement mixture models in selection for reduced incidence of mastitis.
