Estimation of farm specific parameters in a longitudinal model for litter size with variance components and random dropout

Abstract

The estimation of farm specific parameters is often ignored in studies of decision support systems such as, e.g., sow replacement problems. This study concerns the specification and estimation of farm specific parameters in a model for litter size required by a replacement problem considering culling based on productive performance. This implies a joint distribution of the litter size at the parities under consideration. A non-linear parametric mean curve is used to model the expected litter size at different parities for sows within the same herd. The covariance matrix is modeled using variance components of random effects, serial correlation and measurement error. Data from a total of 43 herds are used to devise a likelihood based inference, using maximum likelihood to obtain initial estimates for maximum posterior estimation of farm specific parameters. The maximum posterior approach is required to obtain reasonable estimates in the cases where on-farm registrations are inadequate.