ESTIMATION OF CROSSBREEDING PARAMETERS IN MIXED LINEAR-MODELS

Based on a complete 16 x 16 diallel in laboratory mice we investigated the influence of the transition from model I (no inclusion of covariances between relatives) to a mixed model (consideration of the complete relationship matrix) on the (co)variance matrix of the estimated crossbreeding parameters. The use of model I, ignoring the covariance between relatives, led to important differences in error variances of crossbreed parameters compared with the mixed model. To handle large sample problems and the relationship matrix we propose a two step procedure: Step 1: Estimation of effects of genotypic constructs and their error (co)variance matrix (using the package PEST). Step 2: Decomposition of effects of genotypic constructs into crossbreeding parameters. Furthermore, an algorithm is described to generate a part or the whole inverse of the mixed model equations in PEST.