Estimation & Prediction

Modeling Asymmetrical Misclassification Errors for Genetic Evaluations of Binary Traits

---

Abstract

Threshold models for binary data have long been a cornerstone of quantitative genetics. These models posit an underlying continuous liability that, once exceeding a threshold, yields an observable binary outcome (e.g., diseased or healthy). Despite their power and widespread use, standard threshold formulations implicitly assume error-free phenotypic classification. In reality, binary phenotypes-particularly health and disease traits-are often affected by measurement error and misclassification. Animals may be incorrectly labeled as healthy when diseased (false negatives) or as diseased when healthy (false positives) due to imperfect diagnostic tests, inconsistent recording, or subjective evaluation. Ignoring such misclassification can attenuate genetic effects, inflate uncertainty, and bias estimates of heritability and breeding values. In this study, we develop a comprehensive Bayesian framework, implemented via Markov chain Monte Carlo, for the genetic evaluation of binary traits under asymmetric misclassification. The model is first formulated under a double-sampling design, which allows estimation of misclassification rates from a subset of individuals with verified phenotypes. It is then extended to the general setting in which misclassification probabilities are unknown and jointly estimated with all other model parameters. We evaluated the performance of this new model (model A) using a simulated binary disease trait with heritability 0.40, and compared it with models assuming symmetric misclassification (model S) and no misclassification (model N). The simulated data consisted of 643 cows connected through a true pedigree of 125 sires and 477 dams. Misclassification was introduced under both equal error rates (π₁|₀ = π₀|₁ = 0.15) and unequal error rates (π₁|₀ = 0.20; π₀|₁ = 0.10). Results showed that introducing error into the response variable reduced heritability estimates across all models. When asymmetric misclassification was present, model A outperformed the alternative models by producing more accurate estimates of disease incidence, more precise heritability estimates, and higher correlations between true and estimated breeding values. Under symmetric misclassification, models A and S performed similarly, and both outperformed the model that ignored misclassification. In summary, the proposed Bayesian framework provides a flexible and statistically rigorous solution for accounting for misclassification-especially asymmetric errors-in threshold models used for genetic evaluation.

Keywords: 2026

How to Cite:

Cole, J., Dà¼rr, J., Legarra, A., Parker Gaddis, K. & Wu, X., (2026) “Modeling Asymmetrical Misclassification Errors for Genetic Evaluations of Binary Traits”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2313316. doi: https://doi.org/10.31274/wcgalp.24389

Rights: 1