Comparison of linear mixed models for genetic feather score analysis in laying hens kept in recurrent testing facilities.

Feather pecking, related feather damage, and mortality are crucial welfare and efficiency traits in laying hens. When individuals are kept in sib groups, genetic analysis of feather scores captures the performer and receiver components of feather damage due to pecking. Genetic parameters and breeding values estimated from such data with an ordinary linear mixed model refer to total breeding values. Thus, breeding against feather pecking-related mortality is possible by selecting breeding values for feather score when these are estimated from sib-groups. "Feather score" is part of the selection indexes of some breeding companies. However, there is no public information on the extensive evaluation and validation of the models used. Moreover, survival and feather score are genetically correlated, potentially biasing feather score-breeding value and genetic parameters estimation. This study compared and validated six models for genetic analysis of feather scores in the back and neck regions of laying hens at 45 and 70 weeks of age. We tested univariate models of feather score along with bivariate models for feather score and survival, and both sire and animal models, using individual or cage-level records. We compared the performance of the models based on the accuracy and dispersion of estimated breeding values. Additionally, genetic variances for all traits were estimated and compared. The individual-level univariate animal model showed the poorest performance for both accuracy and dispersion. Apart from the previous, no clear superiority regarding accuracy was observed between animal and sire models, nor between univariate and bivariate models. Breeding values estimated from cage-level observations tended to show less over-dispersion, and the estimates from the univariate cage-level sire model showed no significant over-dispersion for all traits. Since the univariate cage-level sire model was the simplest, as accurate as any of the other models, and showed no over-dispersion, it was considered the best model for feather score analysis with recurrent testing data.