Screening for outliers in multiple trait genetic evaluation

Use of multivariate models in genetic evaluation requires a multivariate method for detecting erroneous outliers that cannot be detected using univariate methods. A simple rule for detecting outliers based on an approximated Mahanalobis distance was applied to Jersey data from the routine Nordic genetic evaluation in dairy cattle. Application of such is simple to implement and increased the accuracy of predicted breeding values for animals that has one or more records edited. Potential biases in evaluations for contemporary animals were also reduced. Optimum editing rules can be determined using the same data structures as used in the standard INTERBULL test for model verification.