Comparison on models for genetic evaluation of non-return rate and success in first insemination of the Danish Holstein cows

The aim of is study was to compare a linear Gaussian model with logit model and probit model for genetic evaluation of non-return rate within 56 d after first-insemination (NRR56) and success in first insemination (SFI). The whole dataset used in the analysis contained 471,742 records from the first lactation of the Danish Holstein cows, covering insemination year from 1995 to 2004. Model stability was evaluated by the correlation between sire EBV (estimated breeding values) from two sub-datasets. The predictive ability of models was assessed by two criteria: 1) the correlation between the EBV of proven bulls, obtained from the whole dataset and from a reduced dataset which only contains the first-crop daughters of sires; 2) χ2 statistic for the expected and observed frequency in a cross validation. Heritabilities estimated using linear, probit and logit models were 0.011, 0.014 and 0.036 for NRR56, and 0.017, 0.020 and 0.051 for SFI, respectively, when phenotypic variance excluded variance of herd–year interaction. The estimates were 0.011, 0.014 and 0.034 for NRR56, and 0.017, 0.019 and 0.048 for SFI, respectively, when phenotypic variance included variance of herd–year interaction. Model validation showed that there was no difference between probit model and logit model, but the two models were better than linear model in stability and predictive ability for genetic evaluation of NRR56 and SFI. However, based on the whole dataset, the correlations between EBV estimated using the three models were over 0.99, suggesting that if dataset is large enough, genetic evaluation of NRR56 and SFI using the three different models would not lead to serious re-rankings of candidates.