RESUMO
The aim of this study was to evaluate the effect of including calving age (CA) on genetic evaluation models for Holstein cattle. The evaluated models included the permanent environment, the sire-herd interactions, and the animals and residual effects as random. The fixed effects included the average production of milk, fat, and protein and the herd-year-season effect. The analyzed data included 603,521 records of milk production (in kg) corresponding to 438,098 animals from 527 herds. Additionally, there were 179,122 records of fat and protein components, corresponding to 148,930 animals from 137 herds. The records were classified by first lactation only (FL) and all available lactations (AL) for validation test (VT). The FL records corresponded to 275,487 milk production records with a mean of 10,874.1 ± 2773.9 kg at a mean CA of 25.6 ± 4.2 months. For FL, the milk components consisted of 78,955 records with a mean fat production of 392.86 ± 89.9 kg, a mean protein production of 362.8 ± 74.9 kg and a mean CA of 25.2 ± 4.1 months. For AL, the number of records was 603,521 for milk production with a mean of 10,802.8 ± 2905.9 kg and a mean CA of 35.6 ± 11.5 months. For the milk components, there were 179,122 records with a mean of 36.1 ± 9.5 months for CA and 388.3 ± 98.4 kg and 356.7 ± 82.6 kg for fat and protein, respectively. Three models were compared: the base model (M0) described above, and two alternative models that included CA in a linear and quadratic form (M1 and M2, respectively). Estimations of the variance components (VC) and breeding value (BV) were obtained using a repeatability animal model, with the same phenotypic and pedigree information used for all models. To select the best fit model for the data, a likelihood ratio test (LRtest) was used. A validation test (VT) was also applied to each model to evaluate the consistency of the genetic trends for females with information on AL and FL. The inclusion of CA in its linear form (M1) was the model that achieved the best results in the LRtest and an acceptable value for the VT. These results show that CA improves the model fit for BV prediction and reliability.