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 covaria nces between relatives) to a mixed model (consideration of the complet e relationship matrix) on the (co)variance matrix of the estimated cro ssbreeding parameters. The use of model I, ignoring the covariance bet ween relatives, led to important differences in error variances of cro ssbreed parameters compared with the mixed model. To handle large samp le problems and the relationship matrix we propose a two step procedur e: Step 1: Estimation of effects of genotypic constructs and their err or (co)variance matrix (using the package PEST). Step 2: Decomposition of effects of genotypic constructs into crossbreeding parameters. Fur thermore, an algorithm is described to generate a part or the whole in verse of the mixed model equations in PEST.