Accurate estimation of genetic parameters and breeding values is a fundamental requirement for the successful implementation of animal breeding programs. For traits expressed as categorical or binary responses, the application of conventional linear models (LM) may result in biased estimates because these models assume a continuous and normally distributed phenotype. In contrast, threshold models (TM) are specifically designed to account for the discrete nature of categorical traits by assuming the existence of an underlying continuous liability. Therefore, evaluating the relative performance of these two statistical approaches is essential for improving the efficiency of genetic evaluation and selection strategies.
In the present study, two polygenic traits with heritabilities of 0.10 and 0.20 were simulated using QMSim software to compare the performance of linear and threshold models under selection. To generate a categorical phenotype, the top 20% of animals based on phenotypic values in each generation were assigned to category 2, whereas the remaining 80% were assigned to category 1. Selection and mating were performed over ten successive generations using both random selection and selection based on superior estimated breeding values. Breeding values were estimated using the Bayesian MCMCglmm package, incorporating both pedigree and phenotypic information. Model performance was evaluated by comparing heritability estimates, prediction accuracy, and genetic gain achieved under each analytical approach.
The estimated heritabilities obtained from the linear and threshold models were 0.0109 and 0.0109 for the simulated trait with true heritability of 0.10, whereas corresponding estimates for the trait with true heritability of 0.20 were 0.0152 and 0.0109, respectively. Although both models produced relatively low heritability estimates because of the categorical transformation of the simulated trait, the threshold model consistently yielded higher prediction accuracy than the linear model. Furthermore, cumulative genetic gain after ten generations was greater under the threshold model, showing approximately 21% and 11% higher genetic progress than the linear model for traits with heritabilities of 0.10 and 0.20, respectively.
These findings demonstrate that threshold models provide a more reliable framework for the genetic evaluation of categorical traits, particularly in breeding programs involving binary or ordinal phenotypes. The improved prediction accuracy and enhanced genetic gain obtained with the threshold model indicate its greater potential for accelerating genetic improvement and increasing the efficiency of selection in livestock populations.