Estimation & Prediction

Fitting different nonlinear mixed models to the number of unlaid eggs

---

Abstract

Egg production traits in laying hens are typically recorded as discrete outcomes, which challenges the assumptions of Gaussian linear mixed models and motivates the use of generalized linear mixed models (GLMMs). The objectives of this study were: (1) to compare Poisson, binomial, negative binomial, and multiplicative binomial GLMMs in terms of model fit; and (2) to evaluate their impact on genetic parameter estimation, predictive ability, and selection decisions relative to the traditional Gaussian model.The analyzed trait was the egg yield gap (EYG), defined as the difference between the maximum possible production and the number of eggs laid within each 28‑day period. This variable shows a highly skewed distribution with a strong concentration at zero, reflecting the low frequency of non‑laying events. Data consisted of 20 consecutive periods (weeks 20-100) from 8,030 hens.The model included a fixed effect of laying period and two random effects: additive genetic and permanent environmental. In the alternative GLMMs, location parameters were sampled using the Metropolis-Hastings algorithm, whereas variance components were updated via Gibbs sampling; in the Gaussian model, all parameters were sampled with Gibbs sampling due to the availability of closed‑form full conditionals. Model fit was assessed using the Deviance Information Criterion (DIC) and the log‑conditional predictive ordinate (logCPO). Predictive ability was evaluated using the LR method, and agreement in selection decisions was quantified as the overlap in top‑ranked individuals.The negative binomial model provided the best fit, consistent with the marked overdispersion in the data. Heritability estimates on the observed scale were low but similar across models (0.046-0.089). Predictive ability was lowest for the Gaussian model (0.192), while the nonlinear GLMMs achieved higher and comparable accuracies (0.259-0.275). Concordance in selection decisions was higher among nonlinear GLMMs and notably lower when compared with the Gaussian model.Overall, the results support nonlinear GLMMs as a suitable framework for modeling count‑based egg production traits, improving both model fit and predictive performance relative to the conventional Gaussian approach.

Keywords: 2026

How to Cite:

Sánchez-Diaz, M., Varona, L., López-Carbonell, D., Cavero, D., Casto-Rebollo, C. & Ibañez-Escriche, N., (2026) “Fitting different nonlinear mixed models to the number of unlaid eggs”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2285597. doi: https://doi.org/10.31274/wcgalp.23746

Rights: 1