A link function approach to model heterogeneity of residual variances over time in lactation curve analyses.

Several studies with test-day models for the lactation curve show heterogeneity of residual variance over time. The most common approach is to divide the lactation length into subclasses, assuming homogeneity within these classes and heterogeneity between them. The main drawbacks of this approach are that it can lead to many parameters being estimated and that classes have to be arbitrarily defined, whereas the residual variance changes continuously over time. A methodology that overcomes these drawbacks is proposed here. A structural model on the residual variance is assumed in which the covariates are parametric functions of time. In this model, only a few parameters need to be estimated, and the residual variance is then a continuous function of time. The analysis of a sample data set illustrates this methodology.