Modelling phenotypic (co)variances of test day records in dairy ewe

Abstract

Two thousand six hundred and thirty test day (TD) records of dairy ewes on milk yield, fat and protein content were analysed by mixed linear models with different phenotypic (co)variance structures. In the first one (UN), (co)variances among TD records taken at different days in milk (DIM) intervals were not mathematically modelled. The second structure was the classical Compound Symmetry (CS) approach that assumes equal (co)variances among TD records within lactation. In the third one, the CS structure was combined with a first-order autoregressive process [AR(1)] in order to model both a time dependent and a time independent component of covariation between TDs. The different (co)variance models did not result in relevant variations in fixed effect estimates and in average lactation curves, whereas they gave contrasting results as far as individual TD covariations around the mean evolution pattern are concerned. The CS+AR(1) structure best fitted the data and pointed out the relevance of both a constant component of covariation within lactation and a lag dependent component, whose pattern exponentially decreases as the time interval between TD measures increases. Furthermore, interesting differences in the relative importance of these two components of covariation among the traits have been observed. Correlations among TD measures within lactation resulted more persistent for milk yield and protein content, whereas showed a lower magnitude and a more rapid decrease for fat content.