Comparison of linear and nonlinear models in estimation of variance components for reproductive traits in Markhoz...

The objective of this study was to estimate (co)variance components and genetic parameters for litter size at birth (LSB), litter size at weaning (LSW), and kid mortality from birth to weaning in Markhoz goats. Data were obtained from the Markhoz Goat Breeding Station in Sanandaj, Iran, and comprised 3439 records for LSB and LSW and 4087 records for kid mortality, collected over a 21-year period. Birth year and dam age had significant effects on LSB and LSW (P < 0.05), while birth year, dam age, birth type, and sex significantly affected kid mortality (P < 0.05). Genetic analyses for LSB and LSW were performed using linear and Poisson models, whereas linear and probit models were applied to mortality. Model performance was evaluated using predictive ability and goodness-of-fit statistics based on mean squared error of prediction and correlations between observed and fitted values. Heritability estimates ranged from 0.02 to 0.04 for LSB, 0.02 to 0.05 for LSW, and 0.16 to 0.35 for mortality. Estimated random effects, breeding values, and permanent environmental effects were highly correlated across models. For LSB, the linear model outperformed the Poisson model; for LSW, both models showed a similar fit, although the linear model had better predictive ability. For mortality, the linear model showed superior predictive performance, although the probit model uniquely identified maternal genetic and common litter effects. Overall, linear models are recommended for accurate animal ranking, whereas the probit model is preferred when the objective is to partition variance in binary traits into direct genetic, maternal, and litter components.